Description
Financial distress is a term in corporate finance used to indicate a condition when promises to creditors of a company are broken or honored with difficulty. If financial distress cannot be relieved, it can lead to bankruptcy. Financial distress is usually associated with some costs to the company; these are known as costs of financial distress.
ABSTRACT
Title of Document: Firm Decision Making under Financial Distress:
A Study of U.S. Air Fares and an Analysis of
Inventories in U.S. Manufacturing Industries
Christian Hofer, Ph.D. 2007
Directed By: Professor Martin E. Dresner, Ph.D.,
Robert H. Smith School of Business
Professor Robert J. Windle, Ph.D.,
Robert H. Smith School of Business
This dissertation investigates the effects of firm financial distress on two key firm
decision variables: sales prices and inventories. These analyses contribute to the
Structure-Conduct-Performance paradigm literature. Specifically, the feedback loop
between financial distress, a result of poor past performance, and two firm conduct
parameters, prices and inventories, is explored in great detail.
The first essay is motivated by the ambiguity of prior research on the relationship
between firm financial distress and prices. The extant economics, corporate finance and
strategic management literatures differentially approach this relationship, and empirical
research has found only limited, at times ambiguous support for any single theoretical
contention. These theoretical perspectives are reviewed and an attempt is made to
reconcile the apparent conflict by adopting a strategic contingency perspective that
identifies in which way and in what instances firm financial distress may impact prices.
The model is empirically tested using data from the U.S. airline industry. The results
indicate that firm financial distress and prices are generally negatively related. Moreover,
this effect is substantially stronger for firms operating under Chapter 11 protection than
for firms approaching bankruptcy. It is further shown that the magnitude of the effect of
financial distress on prices depends on firm factors such as operating costs, market
power, and firm size, as well as on competitive characteristics such as market
concentration and the financial condition of competitors.
The second essay analyzes the impact of firm distress on firm inventories and
investigates if this relationship is impacted by a firm’s power relative to its upstream and
downstream supply chain partners. Building on prior work in the economics field, this
research is not only based on microeconomics theory, but also draws on inventory theory
as well as on prior work on supply chain relationships. A comprehensive inventory
estimation model is specified, and novel measures of inventory determinants and power
are developed. The hypotheses are tested using panel data from the U.S. manufacturing
industry. It is shown that distressed firms hold less inventory and that a firm’s power
within the supply chain will determine to what extent inventory ownership is reduced
during times of financial distress. Implications for supplier selection and supply chain
cooperation are discussed.
In summary, this research significantly enhances researchers’ understanding of why,
how, and when firm financial distress affects prices and inventories.
Firm Decision Making Under Financial Distress: A Study of U.S. Air Fares and an
Analysis of Inventories in U.S. Manufacturing Industries
By
Christian Hofer
Dissertation submitted to the Faculty of the Graduate School of the
University of Maryland, College Park, in partial fulfillment
of the requirements for the degree of
Philosophical Doctor
2007
Advisory Committee:
Professor Martin E. Dresner, Co-Chair
Professor Robert J. Windle, Co-Chair
Professor Philip T. Evers
Professor Curtis M. Grimm
Professor Ali Haghani
© Copyright by
Christian Hofer
2007
ii
Dedication
To my parents,
in deep gratitude for all their love and support.
iii
Acknowledgements
The completion of this dissertation marks the endpoint of four tremendously exciting
years in the doctoral program at the University of Maryland’s Robert H. Smith School of
Business. Many faculty members, fellow doctoral students, and staff members have
contributed to making this time so enriching, enjoyable, and memorable.
I cannot possibly describe the many ways in which Adriana has enriched my life: She is a
best friend, an advisor, a critic, an inspiration, (…) and, thankfully, my wife.
I would like to express my deep gratitude to my advisors, Martin Dresner and Bob
Windle. Over the course of these four years, both Martin and Bob have read hundreds,
possibly thousands of pages of various drafts of this dissertation and other papers. We
have spent many dozens of hours in fruitful meetings and they have always given me
invaluable advice. I deeply appreciate all the time and effort they have invested in my
education and I am equally grateful for their genuine care. I hope that one day I will be
half as good an advisor as Martin and Bob have been for me.
I would also like to thank the entire logistics and supply chain faculty of the department
of Logistics, Business and Public Policy at the Robert H. Smith School of Business. I
particularly thank Phil Evers and Curt Grimm who have taught me much of the subject
matter that has inspired and shaped this dissertation. In addition, they have provided
valuable guidance and advice as members of my dissertation committee. I would also like
to thank Ali Haghani from the A. James Clark School of Engineering for kindly agreeing
to serve on my dissertation committee and generously sharing his thoughts and expertise.
My fellow doctoral students and friends have also contributed to making my time in the
doctoral program at Maryland a very special one. Toby Porterfield and Tashfeen Sohail,
in particular, have been wonderful colleagues during these years and I treasure their
friendship. I would also like to thank my friends Lorrie Westerlund and Joerg
Schnermann for all their care and support.
A final “thank you” goes to all the staff at the Smith School of Business, and to Mary
Slye, Dianne Fox, Anne Stevens, and Mary Crowe-Kokonis, in particular. Without them
this program would not run as smoothly as it does (and it would be much less fun, also!).
iv
Table of Contents
Dedication........................................................................................................................... ii
Acknowledgements............................................................................................................ iii
List of Tables .................................................................................................................... vii
List of Figures .................................................................................................................... ix
1. Introduction......................................................................................................... 1
2. The impact of firm financial distress on prices: A contingency approach ....... 12
2.1. Introduction....................................................................................................... 12
2.2. Theoretical background and hypothesis development...................................... 19
2.2.1. Financial distress as a driver of competitive pricing behavior ......................... 20
2.2.2. Conflicting theoretical arguments..................................................................... 26
2.2.3. The contingency approach................................................................................ 28
2.3. Data and methodology...................................................................................... 43
2.3.1. Data sample....................................................................................................... 44
2.3.2. Variables and measurement .............................................................................. 45
2.3.2.1. Dependent variable ........................................................................................... 46
2.3.2.2. Independent variables ....................................................................................... 47
2.3.2.3. Control variables............................................................................................... 51
2.3.3. Descriptive statistics ......................................................................................... 55
2.3.4. Empirical methodology..................................................................................... 60
2.4. Empirical results and discussion....................................................................... 65
2.4.1. First-stage regression ........................................................................................ 66
2.4.2. Second-stage regression.................................................................................... 68
2.4.3. Second-stage regression: Sensitivity analysis................................................... 77
2.5. Summary and discussion................................................................................... 82
3. The effect of firm financial distress on firm inventories: A supply chain
perspective ........................................................................................................ 87
3.1. Introduction....................................................................................................... 87
3.2. The financial distress-inventory relationship.................................................... 93
3.2.1. Economic theory............................................................................................... 93
v
3.2.2. Inventory theory................................................................................................ 99
3.2.3. The financial distress-inventory hypothesis.................................................... 107
3.3. The supply chain perspective.......................................................................... 108
3.3.1. Supply chain considerations in inventory decisions ....................................... 109
3.3.1.1. Inter-firm relationships: The role of power .................................................... 110
3.3.1.2. Supply chain power and inventory decisions.................................................. 112
3.3.1.3. The power-inventory hypotheses.................................................................... 115
3.3.2. Firm power as a moderator of the distress-inventory link .............................. 117
3.3.2.1. Prior research: Firm size as a moderator of the distress-inventory link ......... 118
3.3.2.2. The power moderator hypotheses ................................................................... 119
3.4. Data and methodology.................................................................................... 122
3.4.1. Sample selection ............................................................................................. 123
3.4.2. Model specification......................................................................................... 126
3.4.3. Variables and Measurement............................................................................ 130
3.4.3.1. Dependent variable ......................................................................................... 130
3.4.3.2. Independent variables ..................................................................................... 131
3.4.4. Data sources .................................................................................................... 141
3.4.5. Descriptive statistics ....................................................................................... 143
3.4.5.1. Descriptive statistics: Part I ............................................................................ 143
3.4.5.2. Descriptive statistics: Part II ........................................................................... 149
3.4.6. Methodology................................................................................................... 155
3.4.6.1. Overview of regression analyses .................................................................... 155
3.4.6.2. Empirical methodology: Part I........................................................................ 159
3.4.6.3. Empirical methodology: Part II ...................................................................... 161
3.5. Empirical results and discussion..................................................................... 162
3.5.1. Empirical results: Part I .................................................................................. 162
3.5.1.1. Regression results: Total inventory ................................................................ 163
3.5.1.2. Sensitivity analyses......................................................................................... 168
3.5.1.3. Regression results: Raw materials inventory.................................................. 174
3.5.1.4. Regression results: Finished goods inventory................................................. 176
3.5.2. Empirical results: Part II ................................................................................. 179
vi
3.5.2.1. Regression results: Total inventory ................................................................ 179
3.5.2.2. Regression results: Raw materials inventory.................................................. 184
3.5.2.3. Regression results: Finished goods inventory................................................. 187
3.6. Summary and discussion................................................................................. 188
4. Firm decision making under financial distress: Summary and outlook.......... 196
Appendix 1...................................................................................................................... 201
Appendix 2...................................................................................................................... 203
Appendix 3...................................................................................................................... 204
Appendix 4...................................................................................................................... 206
Appendix 5...................................................................................................................... 207
Appendix 6...................................................................................................................... 208
Bibliography ................................................................................................................... 209
vii
List of Tables
Table 1: Correlation matrix (n = 23,039) ........................................................................ 56
Table 2: Descriptive statistics for selected variables (n = 23,039).................................. 59
Table 3: First stage G2SLS regression estimates (n = 23,039)........................................ 68
Table 4: Second-stage G2SLS regression estimates......................................................... 71
Table 5: Second-stage G2SLS regression estimates using 1992, 1997, and 2002 data... 81
Table 6: Summary of results ............................................................................................. 83
Table 7: Sample composition (Part I)............................................................................. 145
Table 8: Pooled descriptive statistics (Part I) ................................................................ 146
Table 9: Descriptive statistics (Part I) – distressed vs. non-distressed firms................. 147
Table 10: Pairwise correlations (Part I) ........................................................................ 148
Table 11: Sample composition (Part II) ......................................................................... 151
Table 12: Pooled descriptive statistics (Part II)............................................................. 152
Table 13: Descriptive statistics (Part II) – distressed vs. non-distressed firms ............. 153
Table 14: Pairwise correlations (Part II)....................................................................... 154
Table 15: Overview of regression analyses .................................................................... 155
Table 16: Regression results: Total inventory (R1)........................................................ 165
Table 17: Split-sample regression results: Total inventory (R2, R3) ............................. 166
Table 18: Sensitivity analysis: Distressed vs. non-distressed firms ............................... 169
Table 19: Sensitivity analysis: Granularity of industry definitions................................ 171
Table 20: Sensitivity analysis: SalesSurprise vs. ForecastError.................................... 172
Table 21: Sensitivity analysis: Measurement of total inventories .................................. 173
Table 22: Regression results: Raw materials inventory (R4)......................................... 175
Table 23: Split-sample regression results: Raw materials inventory (R5, R6) .............. 176
Table 24: Regression results: Finished goods inventory (R7)........................................ 177
Table 25: Split-sample regression results: Finished goods inventory (R8, R9) ............. 178
Table 26: Regression results: Total inventory (R10)...................................................... 181
Table 27: Split-sample regression results: Total inventory (R11, R12) ......................... 183
Table 28: Regression results: Raw materials inventory (R13)....................................... 185
Table 29: Split-sample regression results: Raw materials inventory (R14, R15) .......... 186
viii
Table 30: Regression results: Finished goods inventory (R16)...................................... 187
Table 31: Split-sample regression results: Finished goods inventory (R17, R18) ......... 188
Table 32: Summary of results for entire data set............................................................ 190
Table 33: Summary of results for distressed firms ......................................................... 192
Table 34: Ranked residuals of regression of OpEx/ASM on avg. stage length (n=41).. 202
Table 35: OLS regression estimates (n = 23,039).......................................................... 203
Table 36: 2SLS regression estimates without fixed effects (n = 23,039)........................ 204
Table 37: 2SLS regression estimates with fixed effects (n = 23,039)............................. 206
Table 38: First-stage G2SLS regression estimates (n = 23,039) ................................... 207
Table 39: Mean comparisons between sampled firms and Compustat population ........ 208
ix
List of Figures
Figure 1: The structure-conduct-performance paradigm................................................... 3
Figure 2: The moderated distress-conduct feedback mechanism....................................... 5
Figure 3: The moderating effect of operating costs on the distress-price relationship.... 32
Figure 4: The moderating effect of firm size on the distress-price relationship .............. 35
Figure 5: The moderating effect of market shares on the distress-price relationship...... 38
Figure 6: The moderating effect of market concentration on the distress-price
relationship ............................................................................................................... 41
Figure 7: Research model................................................................................................. 42
Figure 8: Overview of Chapter 11 indicator variables .................................................... 50
Figure 9: Distribution of Distress scores prior to and during bankruptcy ...................... 58
Figure 10: Illustration of the r,Q policy ......................................................................... 103
Figure 11: The moderating effect of power on the distress-inventory relationship ....... 121
Figure 12: Research model............................................................................................. 122
Figure 13: Illustration of the construction of industrial supply chains.......................... 140
Figure 14: Alternative definitions of distressed and non-distressed firms..................... 168
1
“No matter what the state of the economy, no
company is immune from internal hard times—
stagnation or declining performance.” (Hofer 1980)
“Global competition, technological turbulence, high
costs of capital, and other nettlesome factors will
cause more and more businesses to face occasional
hard times.” (Hambrick and Schecter 1983)
1. Introduction
Firm financial distress is an omnipresent phenomenon in manufacturing and service
industries. While there is no unique definition of financial distress, distress firms are
generally loss-making and suffer from (severe) liquidity constraints. Based on these
criteria, Altman (2002, 1968) developed the Z score as a composite measure of a firm’s
financial condition. Altman suggests that firms with a Z score of less than 1.81 are
considered financially distressed and face a high risk of bankruptcy. Following this
definition, about one third of all U.S. manufacturing firms
1
and about half of all U.S.
airlines (The Economist 2005) were considered financially distressed in 2005. Most
recently, car manufacturers such as Ford and General Motors (McCracken 2006), and air
carriers like Northwest Airlines and Delta Airlines (Carey and Trottman 2005), to
mention but a few examples, have been experiencing financial difficulties. This
dissertation investigates the impact of financial distress on managerial decision variables
such as prices and inventories.
1
This estimate is based on the analysis of 2,323 manufacturing firms listed in the Compustat database.
Thirty-two percent of these firms had Z scores (Altman 1968) of less than 1.81.
2
Most research in the broad field of business management is concerned with
understanding how managerial decisions come about and how these decisions affect firm
and market performance. Many researchers therefore follow the tradition of the structure-
conduct-performance (SCP) paradigm which essentially posits that the structure of
markets impacts firms’ conduct which, in turn, is a key determinant of the performance of
firms and markets (Bain 1956, Mason 1949, 1939). The term “structure” thereby refers to
structural characteristics of markets that are indicators of the competitiveness of markets.
Commonly used measurement variables include industry concentration, the number of
firms in the market or barriers to entry and exit (Waldman and Jensen 2001). Firms
compete in the marketplace by means of actions that aim at maximizing firm
performance. These rivalrous activities are summarized by the term “conduct” which
may, for example, refer to pricing and product strategies (Waldman and Jensen 2001).
The aggregate performance of firms in a market can be measured in terms of allocative
efficiency or production efficiency, for example (Waldman and Jensen 2001). The
individual performance of firms, in turn, is typically evaluated based on financial (e.g.
profitability) or operating measures (e.g. productivity). Figure 1 graphically illustrates
the structure-conduct-performance paradigm.
3
Figure 1: The structure-conduct-performance paradigm
Firm-specific factors are added to the depiction of the structure-conduct-performance
paradigm in Figure 1 to indicate that not only (market) structural, but also other firm
characteristics (besides the firm’s financial condition) such as operating costs and firm
size, for example, may impact a firm’s conduct in the market (e.g. Spanos et al. 2004).
The structure-conduct-performance (SCP) paradigm, as presented by Waldman and
Jensen (2001), also recognizes that certain feedback loops may exist within the structure-
conduct-performance framework. An industry’s above-average performance, for
example, may attract new entrants, thus affecting the structure of markets. By the same
token, a firm’s past performance may impact future managerial decisions relating to, for
example, prices and sales quantities, thus linking the firm’s performance/distress to its
conduct. Also, a firm’s distress may ultimately impact other firm characteristics such as
the firm’s size and its cost structure. Figure 1 illustrates some of these feedback loops
Structure Performance
Other firm-specific
factors
Conduct Structure Performance
Other firm-specific
factors
Conduct
4
within the SCP paradigm
2
. While there are many such feedback mechanisms, one specific
link is of particular interest in this dissertation research: The effect of financial distress, a
direct result of poor past performance, on a firm’s conduct in terms of sales prices and
inventories.
Pricing and inventory decisions are important indicators of a firm’s competitive conduct
in the marketplace. Basic game-theoretic models suggest that firms compete on either
price (Bertrand competition) or quantities (Cournot competition) (Gibbons 1992). With
inventories being a function of sales quantities, both inventories and prices, thus, are
essential decision variables that reflect a firm’s competitive behavior. Consequently,
numerous researchers have investigated the competitive implications of firms’ pricing
(e.g. Busse 2002) and inventory (e.g. Cachon 2001, Mahajan and Ryzin 2001) decisions.
It is therefore deemed appropriate and relevant to investigate the effects of financial
distress on these two firm conduct parameters.
Clearly, a feedback mechanism between financial distress and conduct is intuitively
appealing: Managers of distressed firms must turn the situation around and ensure the
company’s future profitability. Given the widespread occurrence of financial distress,
researchers have been interested in understanding the effects of distress on firm conduct.
Specifically, researchers have examined the anatomy of corporate turnarounds: What do
financially troubled firms do to return to profitability?
2
Note that the changes in a firm’s conduct caused by a deretioration of the firm’s financial condition will
then impact the firm’s performance. The relationship between performance/distress and conduct (as well as
structure and firm-specific variables), thus, is iterative over time.
5
Hofer (1980) notes that price cutting is a popular measure implemented by distressed
firms. Arogyaswamy and Yasai-Ardekani (1995) and Sudarsanam and Lai (2001), in
turn, suggest that firms frequently reduce inventory levels as a part of their restructuring
efforts. While anecdotal evidence and conceptual work suggest that greater levels of
distress imply lower prices and inventories, ceteris paribus, empirical evidence in support
of this contention has been scant. In a similar vein, conceptual and empirical work has
arrived at the conclusion that there is no unique turnaround strategy and no single recipe
for turnaround success. Rather, different turnaround gestalts have emerged: Hofer (1980),
for example, distinguishes between revenue-generating, product-market refocusing, cost-
cutting, and asset reducing strategies. Both Hofer (1980) and Hambrick and Schecter
(1983) suggest that the choice of a turnaround strategy will be contingent on the gravity
of financial distress and other firm and market-related contingencies. This contention is
illustrated in Figure 2: The effect of financial distress (poor past performance) on
conduct is moderated by (market) structural characteristics and firm-specific factors.
Figure 2: The moderated distress-conduct feedback mechanism
Hofer (1980) and Hambrick and Schecter (1983), thus, contend that the effect of financial
Structure Conduct Performance
Other firm-specific
factors
Financial distress
Structure Conduct Performance
Other firm-specific
factors
Financial distress
6
distress on firm conduct is contingent on other factors. This contention is consistent with
the basic tenet of contingency theory. Contingency theory was originally motivated by
the observation that “[p]rominent theorists promote their ascribed frameworks as
conceptually valid and pragmatically applicable to all organizations in all situations”
(Luthans and Stewart 1977, p.182). This concept of universality, however, has been
questioned by researchers on the grounds of both theoretical and empirical
counterevidence. Instead, researchers have increasingly recognized the importance and
moderating role of situational characteristics in defining causal relationships (Hitt et al.
2004).
Proponents of the situational approach argue “that the most effective management
concept or technique depends on a set of circumstances at a particular point in time”
(Luthans and Stewart 1977, p.182) and that empirical research based on simple “linear
models [has generally] provided disappointing results” (Hitt et al. 2004, p.11).
Consequently, researchers have proposed a “general contingency theory of management”
(Luthans and Stewart 1977) which rests on the key premise that environmental, resource
and management variables intervene in cause-and-effect relationships in the context of
strategic management research.
There is, however, no defined set of contingency variables and no universal prescription
as to how, when and where contingencies ought to be considered (see e.g. Hofer 1975 for
a review of important control variables and contingency factors in the context of business
strategy research). Many researchers have therefore criticized contingency theory as an
7
“illusion” (Longenecker and Pringle 1978) and have attacked the theory’s vagueness and
“lack of clarity” (Schoonhoven 1981). The use of contingency frameworks has
nonetheless been popular in the strategic management literature (Hitt et al. 2004).
This dissertation follows this research tradition and defines context-specific contingency
variables that are expected to affect the relationship between firm financial distress and
prices and inventories, respectively. Specifically, it is suggested that structural and firm-
specific factors moderate the effects of firm distress on prices and inventories.
To date, the model shown in Figure 2 has not been subject to large-scale empirical
testing. While some researchers have investigated the effects of financial factors on firm
decision parameters such as prices (e.g. Borenstein and Rose 1995) and inventories (e.g.
Carpenter et al. 1994), the moderating effects of structural and firm-specific factors on
the distress-conduct relationship remain largely unexplored. The sole exception is the
work by Ferrier et al (2002): These authors investigate the effect of financial distress on
competitive aggressiveness as measured by the number and nature of firms’ competitive
actions. Ferrier et al (2002) thereby find evidence that the effect of distress on
competitive behavior is moderated by industry characteristics
3
and the educational and
functional heterogeneity of top management teams. This dissertation builds on the work
of Ferrier et al (2002) and extends it to the study of two particular firm conduct
parameters: sales prices and inventories.
3
Industry growth, industry concentration, and barriers to entry.
8
Clearly, gaining a better understanding of how financial distress impacts firms’ pricing
and inventory decision is a timely and relevant research endeavor. Prior research has
shown that linear models of the distress-price and distress-inventory relationships may be
overly simplistic and do not do justice to the complex nature of decision problems
relating to price and inventory management under financial distress (e.g. Singh 1986).
While most researchers contend that greater financial distress should result in lower
prices and lower inventory holdings, the empirical findings are largely inconsistent and
often times statistically insignificant. The basic premise of this research is that structural
and firm-specific characteristics moderate the distress-conduct relationship as shown in
Figure 2, thereby explaining why the distress-conduct effect may be substantial in some
instances and insignificant in other cases.
In summarizing, this dissertation investigates the following research questions:
? Does financial distress have an impact on prices and inventories, after controlling
for other relevant parameters?
? And how can these effects be characterized, i.e. what factors influence the
magnitude of the distress-price and distress-inventory relationships?
This dissertation addresses these questions and thereby makes a number of significant
contributions.
This is—to the best of the author’s knowledge—the first study to empirically investigate
the feedback loop between financial distress (poor past performance) and two key firm
conduct parameters: prices and inventories. Particular attention is paid to the moderating
9
effects of structural and firm factors on the distress-conduct relationship. This research,
thus, contributes to the SCP literature by analyzing the distress-conduct feedback loop
and empirically evaluating the effect of interactions between structural, firm, and
financial characteristics on firm conduct parameters.
This framework is empirically tested in two distinct contexts: Prices and inventories are
studied in the context of the U.S. airline industry and the U.S. manufacturing industry,
respectively. In both instances, context-specific contingency variables are proposed and
their moderating effects on the distress-price and distress-inventory relationships are
evaluated. This research draws on a broad array of theoretical arguments from the
strategic management, economics, and corporate finance literatures to identify these
contingency variables and to hypothesize about their impact on the distress-conduct
relationship. The validity of the theoretical arguments and models set forth in this
dissertation is underlined by solid estimation results. It is shown that financial distress is
an important explanatory variable that significantly impacts a firm’s sales prices and
inventories. This research thus also contributes to furthering empirical research on prices
and inventories.
In addition, answering these research questions also paves the way to exploring further
managerial implications of financial distress with respect to prices and inventories in
greater detail: When are pricing and inventory actions economically viable turnaround
strategies? And how will the distressed firm’s actions affect competition and inter-firm
cooperation?
10
This dissertation comprises four chapters. Following this introduction (Chapter 1),
Chapters 2 and 3 are devoted to the study of the effects of financial distress on prices and
inventories, respectively, while Chapter 4 provides a summary of the findings and
contributions of this dissertation research.
The impact of distress on prices is discussed in Chapter 2. Two specific research
questions are investigated in this essay: How does a firm’s financial distress impact its
pricing behavior? And what parameters moderate the effect of firm financial distress on
the firm’s prices? These questions arise upon reviewing a broad set of extant research
which is marked by ambiguous empirical findings. This conflict is addressed by
developing a contingency framework. It is suggested that firm factors such as operating
costs, firm size and market shares, as well as market characteristics such as market
concentration and competitors’ financial conditions determine to what extent financial
distress affects prices. A large-scale empirical analysis using panel data from the U.S.
airline industry is conducted. The results provide ample support for the proposed
contingency framework.
Chapter 3 focuses on the distress-inventory relationship. This essay is primarily
motivated by two observations: First, prior studies have approached the firm finance-
inventory link from an economics perspective only, thus ignoring the insights provided
by inventory theory. Second, most extant research has failed to put firm inventory
decisions into a supply chain context where inter-firm power balances may affect
11
inventory ownership in supply chains. Consequently, two research questions are
formulated: Does a firm’s financial situation have an impact on its inventories after
controlling for other relevant parameters prescribed by inventory theory and supply chain
research? And is the magnitude of the presumed effect of financial distress on inventories
impacted by power (im)balances in supply chain relationships? To investigate these
questions, a thorough review of related economics, inventory, and supply chain research
is provided and testable hypotheses are formulated. Based on this theoretical foundation,
an empirical estimation equation is specified. Data from U.S. manufacturing industries is
used to test the hypotheses. Specifically, it is shown that greater levels of firm financial
distress are associated with lower firm inventory levels, ceteris paribus. In addition, there
is some support for the hypothesis that greater levels of power over suppliers and buyers
not only reduce inventory ownership in general, but also increase the effect of financial
distress on inventories.
Chapter 4 presents a summary of this dissertation research and highlight its contributions.
In addition, a research agenda for further studies of the effects of firm financial distress is
outlined.
12
2. The impact of firm financial distress on prices: A contingency approach
Chapter 2 presents a theoretical and empirical analysis of the relationship between
financial distress and sales prices. This chapter is structured as follows: Section 2.1
provides a brief overview of prior research on the financial condition-prices link and
clearly states the research questions and contributions of this dissertation essay. In
Section 2.2, a comprehensive review of the literature and relevant theories is provided,
and hypotheses are derived. The research model is introduced in Section 2.3, the data and
variables are discussed, and econometric issues are addressed. In Section 2.4, the
regression results are presented. The article concludes with a summary of the study’s
findings and a discussion of their implications for managers and policy makers (Section
2.5). The study’s limitations are noted and directions for future research are provided as
well.
2.1. Introduction
The question of how a firm’s financial condition impacts the firm’s sales prices has been
investigated from multiple perspectives. Researchers from the economics, corporate
finance, and strategy fields have published a substantial amount of literature on this and
related issues (e.g. Borenstein and Rose 1995, Ferrier et al. 2002, Opler and Titman
1994). Yet, in summary, the findings have been largely inconclusive, not only across but
also within the respective research streams. Empirical research has found only limited, at
times ambiguous support for the contention that distressed firms’ sales prices tend to be
13
lower. This study draws on various theories from the economics, corporate finance and
strategic management fields to investigate this issue and attempts to reconcile the
apparent conflict by adopting a strategic contingency perspective that identifies in which
way and under what conditions firm financial distress may impact sales prices.
This research question is of particular interest given that firm financial distress is often
argued to lead to and result from price competition: Low market prices may drive firms
into bankruptcy, and the latter may, in turn, affect a firm’s competitive pricing behavior.
The so-called sick industry problem, thus, is intimately associated with the issues of
financial distress and price competition as repeatedly evidenced in the U.S. airline
industry. In recent years, many U.S. airlines have sought bankruptcy protection under
Chapter 11
4
, the ultimate manifestation of financial distress. Between 2001 and 2005
alone, seven of the top 20 U.S. carriers took advantage of the provisions of this code to
facilitate their restructuring processes
5
. An article in The Economist (2005) noted that “at
least half of America's airline industry has now been declared bankrupt” when Delta Air
Lines and Northwest Airlines declared bankruptcy in September 2005.
Airlines can achieve significant reductions in labor, leasing, and debt costs under Chapter
11 protection (McCafferty 1995), thus giving bankrupt firms a competitive advantage
over their non-bankrupt counterparts. Following Delta’s and Northwest’s bankruptcy
4
Title 11 of the U.S. code, commonly referred to as Chapter 11, is a form of interim bankruptcy and grants
the filing company protection from its creditors until a reorganization plan is developed and approved by
the creditor committees.
5
The top 20 U.S. commercial carriers were ranked based on 2001 passenger data (available from
www.transtats.bts.gov). The following carriers filed for Chapter 11 protection between 2001 and 2005:
TWA (2001), United (2002), US Airways (2002, 2004), Hawaiian (2003), ATA (2004), Delta (2005),
Northwest (2005).
14
filings, analysts therefore warned of potentially adverse consequences for other carriers
such as American Airlines and Continental Airlines (Trottman 2005). Consequently,
researchers (see e.g. Kennedy 2000, Rollman 2004) and managers of non-bankrupt firms
have repeatedly criticized the destructive implications of Chapter 11 protection. Gary
Kelly, then Chief Financial Officer with Southwest Airlines, for example, notes that “the
length of time an airline can go through bankruptcy protection and offer distressed prices
is very unsettling” (McCafferty 1995). Similarly, Robert Crandall, the former Chief
Executive Officer of American Airlines, argues that “Chapter 11 also undermines
responsible managements. In an intensely competitive industry providing a commodity
product, the ‘dumbest competitor’—unrestrained by fear of failure—sets the standard”
and hence calls for “bankruptcy laws designed to incentivize success and penalize
failure” (Crandall 2005). The criticism of Chapter 11 protection as unfair and destructive
is all but new: a 1989 article published in The Economist discusses the “uses and abuses”
of Chapter 11 and concludes that “what was designed as a shield has become a sword”
(Anonymous 1989).
Most of the previous statements make the explicit or implicit assumption that financially
distressed firms sell at lower prices than their healthier competitors. This contention,
however, has not found consistent theoretical and empirical support.
In the economics stream of research, Borenstein and Rose (1995) find that air fares
slightly decrease prior to bankruptcy filings, but do not further change in the time period
thereafter. Kennedy (2000) and Brander and Lewis (1986) assert that a firm’s financial
15
condition affects its market conduct, and Busse (2002) supports this contention,
indicating that financially distressed firms are more likely to start price wars than their
healthier competitors. The traditional economics literature, however, negates a
relationship between financial condition and firm output market behavior (e.g.
Modigliani and Miller 1958)
6
, and stresses the importance of demand fluctuations in
instigating price reductions.
From a corporate finance perspective, Baker (1973) argues that highly leveraged firms
are more risk-seeking than relatively profitable firms which take some of their “returns in
the form of reduced risk” (Hall and Weiss 1967, p.328). Along the same lines,
Maksimovic and Zechner (1991) suggest that financially distressed firms are more likely
to choose riskier (pricing) strategies. Opler and Titman (1994), in contrast, attribute the
lower performance of troubled firms to the (predatory) aggressiveness of competitors and
the costs of financial distress rather than to the firm’s own pricing behavior.
The strategy literature, finally, has focused the attention on the link between performance
distress and competitive behavior in general. Bowman (1982) contends that troubled
firms may be more risk-assertive (i.e. inclined to compete more aggressively) than
healthy firms, and Miller and Chen (1994) also relate past financial distress to
competitive aggressiveness. Ferrier et al (2002), however, find “that poor-performing
firms were less likely to exhibit aggressive competitive behavior” (p.311) when looking
at the direct relationship between performance distress and competitive aggressiveness. It
6
See also Brander and Lewis (1986) and Kennedy (2000).
16
is noteworthy that none of the studies in the strategy field have examined the link
between financial distress and prices in particular. Rather, competitive behavior has
typically been measured by counting and categorizing competitive actions and reactions
(Chen et al. 1992, Ferrier et al. 2002, Smith et al. 1991, Young et al. 1996).
These examples illustrate the inconclusiveness of prior research and suggest that the link
between financial distress and prices may be more complex (Singh 1986). The general
questions, thus, remain:
? How does a firm’s financial distress impact its pricing behavior?
? What parameters moderate the effect of firm financial distress on the firm’s prices?
As the research results referenced in the preceding paragraphs demonstrate, the answer to
these questions cannot be a straightforward one. There are multiple theoretical
perspectives and contingencies that may partly explain the variability of a troubled firm’s
pricing behavior. Focusing on competitive actions in general, Ferrier et al (2002) have
presented a first attempt to reconcile these conflicting views. They stress the importance
of context-specific contingencies such as industry growth and concentration, as well as
top management team heterogeneity in defining the relationship between performance
distress and competitive behavior. In fact, the strategy literature offers rich insights into
the contingencies that may moderate this relationship. This research builds on the work of
Ferrier et al (2002) in drawing on a broad theoretical basis and proposing a
comprehensive contingency framework that aims at characterizing the relationship
between financial distress and prices, and identifying factors that may affect the
17
magnitude of this relationship. In addition to developing and empirically testing this
contingency framework in the context of the U.S. airline industry, this research extends
the extant body of knowledge in three important respects:
First, price is used as a criterion variable. As mentioned earlier, none of the studies
published in strategic management journals examine the impact of financial distress on
prices. Yet, price is probably the single most important and relevant measure of
competitive behavior: From a consumer perspective, for example, prices are decisive in
determining consumer welfare ? the lower the prices, the greater the consumer surplus.
Consequently, prices are – under the assumption that the products and services offered by
firms are sufficiently homogenous – the primary driver of purchase decisions. From a
firm perspective, price is a key managerial decision variable affecting revenues and a
firm’s bottom line. Low prices may allow a firm to gain market share and obtain an
advantage over competitors, while a differentiation strategy may enable a firm to skim
the market and achieve higher prices (Porter 1980). Moreover, price is of interest from a
public policy point of view. Regulatory government bodies, such as the former Civil
Aeronautics Board (CAB) in the airline industry, and consumer interest groups survey
and screen markets for evidence of predatory pricing and intervene when free market
mechanisms of demand and supply fail to produce satisfactory market outcomes. Using
price as a dependent variable, rather than count and categorical variables such as number
and type of competitive actions, also allows for a more detailed evaluation of the
magnitude of a firm’s reaction to changes in its financial condition.
18
A second contribution lies in examining firm financial distress in more detail than has
been evident in most prior empirical work. While some studies focus on bankruptcy
filings (e.g. Borenstein and Rose 1995), others use measures such as Altman’s Z score
(Altman 1968) to evaluate a firm’s financial situation (e.g. Ferrier et al. 2002). There is,
however, substantial evidence that financial distress may differentially impact firm
behavior before, during, and after a Chapter 11 filing occurs (Borenstein and Rose 1995,
Busse 2002, Kennedy 2000). Therefore, both measures (a Z score-based distress measure
and bankruptcy dummy variables) are included in the empirical analyses to more
precisely sort out the effects of financial distress and bankruptcy per se. Furthermore, a
firm’s financial standing relative to its competitors in the market is considered. In fact,
financial distress in absolute terms may not necessarily imply any pricing actions if
competing firms find themselves in similar financial situations. More specifically, it is
expected that such pricing actions will be more pronounced when a distressed firm’s
financial situation is significantly different from that of its rivals.
Finally, this study is unique with respect to its empirical detail. A panel data set from the
U.S. airline industry is used to investigate the relationship between financial distress and
price. Unlike in many previous studies, the unit of observation in the analyses is a
specific route (i.e. “product”) market rather than a firm year or firm quarter (Busse 2002,
Chattopadhyay et al. 2001, Ferrier et al. 2002). This allows for a much more fine-grained
and statistically robust examination of the hypotheses.
This essay reports a comprehensive effort to understand if, when, and how firm financial
19
distress impacts prices. The empirical results suggest that financially distressed firms
offer lower prices than their healthier competitors, ceteris paribus. The magnitude of the
effect of firm financial distress on prices, however, is shown to decrease with unit
operating costs, increase with firm size, and decrease with firm market shares. The price
effects of financial distress are also stronger in more concentrated markets and when a
firm’s competitors are in significantly different financial situations. The insights provided
by this research will be useful to both firms and policy makers. Distressed firms and their
competitors gain a better understanding of how financial conditions typically impact
pricing decisions and customer demand. Managers of financially distressed firms may
benefit from this knowledge when developing turnaround strategies. Competing (healthy)
firms, on the other hand, can more accurately anticipate distressed firms’ pricing actions
and act accordingly. For policy makers, the findings of this study will help clarify if,
when, and to what extent financial distress and Chapter 11 protection impact sales prices
and the competitive behavior of firms. The findings presented here may help clarify if
current bankruptcy laws serve the purpose they were intended for, and contribute to
maintaining or improving the allocative efficiency of markets.
2.2. Theoretical background and hypothesis development
As briefly outlined above, there are competing perspectives on the relationship between
financial distress and prices. In this section, an overview of these theories from the
strategy, economics and corporate finance fields is provided and hypotheses are derived.
The research hypotheses are developed in two steps: In line with the Structure-Conduct-
20
Performance paradigm (see Figure 1), it is expected that there is a relationship between
financial distress and firm conduct in terms of prices. Several theories which further
support this contention are discussed in Section 2.2.1. Theories that may negate this
relationship are reviewed in Section 2.2.2. A contingency framework is proposed which
suggests that the relationship between firm financial distress and a firm’s pricing
behavior may be moderated by certain firm and structural characteristics (Section 2.2.3).
2.2.1. Financial distress as a driver of competitive pricing behavior
The strategy literature offers two theories, prospect theory
7
and organizational learning
theory that may support a negative relationship between financial distress and a firm’s
prices. Both theories are discussed in turn before empirical evidence and arguments from
standard microeconomic and corporate finance theory are set forth.
Prospect theory posits that decision makers are more risk seeking when facing situations
of likely loss while the inverse is true for decision makers operating in the domain of
profitability (Kahneman and Tversky 1979). Prospect theory can, thus, readily be applied
to evaluate the risk-taking behavior of financially troubled firms: Managers of low-
performing, troubled firms may be risk-assertive in their strategic choices in the
expectation of positive long-term returns to risk (in terms of increased market shares,
revenues, or profits, for example).
7
While prospect theory has its origins in the economics field, its concepts have been widely adopted by
strategic management researchers.
21
There is substantial support for the contention that troubled firms choose riskier strategies
in the strategic management literature (Bowman 1982, Moses 1992, Singh 1986,
Wiseman and Bromiley 1996). Chattopadhyay et al (2001) further investigate firms’
responses to threats such as declining organizational performance by considering
elements such as organizational characteristics and strategic type, and Wiseman and
Gomez-Mejia (1998) examine managerial risk taking across different governance modes.
While extending the basic framework of prospect theory, both papers still support the
hypothesized relationship between a firm’s level of distress and risk seeking behavior.
Authors have, thus, based their arguments on prospect theory when investigating the
relationship between organizational decline and risk taking behavior in general (Bowman
1982, Chattopadhyay et al. 2001, Shoham and Fiegenbaum 2002, Singh 1986, Wiseman
and Gomez-Mejia 1998), organizational adaptation (McKinley 1993) or innovation
(Mone et al. 1998).
With the connection between financial distress and risk taking behavior established, the
relationship between the latter and a firm’s pricing strategy can be characterized as
follows: As noted by Ferrier (2001), pricing actions represent a particular type of
competitive actions which have been associated with organizational risk taking (Ferrier et
al. 2002). Similarly, (Borenstein and Rose 1995) equate bankrupt firms’ “preference for
greater risk” (p.397) to competitive aggressiveness. Moses (1992) further notes that low
price strategies “sacrifice short-run profits in an attempt to establish a market and
generate profits over the long run” (p.40). He concludes that penetration strategies are
high risk strategies because the firm might incur further losses if costs fail to decrease
22
below price levels in the longer term. Pricing actions also entail the risk of imitation or
retaliation by competing firms. LeBlanc (1992), for example, suggests that low-cost
incumbents may choose to price aggressively in response to firms entering their (low-
price) markets. In more general terms, authors have investigated the dynamics of
competitive actions and responses and have found that a firm’s actions drive competitors’
responses (Chen et al. 1992), which in turn, determine the effectiveness and performance
effects of the focal firm’s actions (Chen 1996, Peteraf 1993, Smith et al. 1991). The risk
of choosing low price strategies in a homogenous competitive environment, thus, lies in
the possibility of unbalancing the competitive equilibrium (Xu and Tiong 2001) and the
potential loss resulting from aggressive competitive responses (Young et al. 1996). In
summary, prospect theory supports the argument that financial distress induces firms to
commit to a riskier, more aggressive pricing behavior, i.e. to lower prices.
Organizational learning theory also provides support for a positive relationship
between performance distress and strategic change or competitive aggressiveness (Ferrier
et al. 2002). Lant et al (1992) argue that previously unsuccessful firms undergo a learning
process which may lead to strategic reorientation, and Ferrier (2001) suggests that the
discrepancy between an organization’s goals and its actual performance provides
motivation for future actions and increases the likelihood of strategic change. To the
extent that pricing actions reflect changes in the underlying firm strategy, one may thus
argue that financially distressed firms are more likely to change their prices than are
healthy firms. Ferrier et al (2002), for example, note that “poor performance provides the
firm with strong incentives to aggressively search out new approaches to compete more
23
effectively in the marketplace” (p.304). It is thereby implicit that potential price changes
will typically involve lower prices (see also e.g. Ferrier 2001).
From a microeconomics and corporate finance perspective, Brander and Lewis
(Brander and Lewis 1986) argue that a firm’s “output market behavior will, in general, be
affected by [its] financial structure” (p.957, brackets added). Investigating the linkages
between financial and product markets, they demonstrate that highly leveraged firms will
likely compete more aggressively by increasing their output since riskier strategies with
(potentially) higher returns are more attractive to equity holders as a result of the limited
liability effect of equity financing, than are conservative strategies which primarily
appeal to debt holders. In a similar vein, Maksimovic and Zechner (1991) suggest that
highly leveraged firms choose technologies which are riskier in terms of their expected
cash flows. Hendel (1996) supports this assertion, arguing that “firms under financial
distress use aggressive pricing to generate cash” (p.309) and that prices are a function of
a firm’s liquidity.
A number of authors have empirically examined the relation between firm financial
condition and pricing behavior. Borenstein and Rose (1995) regress the change in prices
on a set of Chapter 11 indicator variables
8
and use a panel dataset from the U.S. airline
industry (1988-1992 data) to estimate their model. They find support for the theoretical
contentions summarized above, indicating that air fares drop by five to six percent in the
months preceding the carrier’s Chapter 11 filing. Kennedy (2000) demonstrates that a
8
As noted by the authors, the effects of many other variables typically included in price estimation
equations are assumed negligible and are excluded from the model specification.
24
distressed firm’s sales revenues and profits (and that of its rivals) decrease prior to
bankruptcy as a result of its altered product market conduct. He analyzes 51 bankruptcy
filings and uses Chapter 11 indicator variables and a small set of market and firm-specific
control variables to predict revenues and profit margins. Analyzing U.S. airline data from
the 1985 to 1992 period, Busse (2002) finds that highly leveraged firms are more likely to
start price wars. Busse also argues that “firms in poor financial condition discount future
revenues more heavily than do financially sound firms” (p.298), thus focusing on
boosting short term sales (by cutting prices, for example).
Taken together, there is theoretical and empirical support for the contention that
financially distressed firms choose riskier strategies and price more aggressively, i.e.
follow a low-price strategy in an effort to gain market shares and boost sales
9
. Hypothesis
1 is stated as follows:
Hypothesis 1: Financial distress negatively impacts prices.
It may also be argued that a firm’s prices will affect firm financial condition. If prices are
consistently below marginal costs, the firm’s financial situation will deteriorate. Prices
above marginal costs, in turn, will positively impact firm financial condition as long as
marginal costs are larger than average costs. The possibility of such reverse causality is
not further explored in this research. Firm financial condition is, of course, a firm-level
phenomenon while prices are market-specific. A firm’s financial distress may, as argued
9
See also Ferrier et al (2002) for a definition of competitive aggressiveness.
25
here, impact a firm’s pricing behavior in all markets, but the sales price in an individual
market will not necessarily affect the firm’s financial standing. In fact, the latter may only
be true if all prices are systematically lower (or higher) than marginal costs. This is,
however, a strong assumption which requires empirical and theoretical substantiation.
Such work is not within the scope of this analysis and is left for future research. This
research uses firm level financial distress to estimate multi-market firms’ market level
prices. It is therefore assumed that problems of endogeneity, caused by reverse causality,
do not arise.
Hypothesis 1 implies that financially distressed carriers may be expected to sell at lower
prices, all else equal. Prior research, however, suggests that the above hypothesized price
effect of firm financial distress may intensify as bankruptcy occurs. As Borenstein and
Rose (1995) and Kennedy (2000) have shown, firms try to prevent insolvency by
generating cash through aggressive competition prior to bankruptcy filings. Once these
firms operate under Chapter 11 protection, however, they benefit from lower operating
costs as debt payments are paused (Barla and Koo 1999, Rose-Green and Dawkins 2002)
to support the restructuring of the firm. This lower cost base may allow bankrupt firms to
charge even lower prices. Moreover, soft demand may force carriers to cut fares once
they operate under bankruptcy protection since the latter signals uncertainty to consumers
(Hofer et al. 2005). Barla and Koo (1999) further suggest that firms “under protection of
Chapter 11 are more likely to adopt short term profit maximization behaviors” which
equate to “prices that are well below long run marginal costs” (p.104) when demand is
low (see also Hofer et al. 2005).
26
In summary, there are three rationales which support the contention that the effect of
financial distress on prices should be stronger during bankruptcy than prior to the Chapter
11 filing (see also Hofer et al. 2005): First, when operating under bankrupt protection,
firms benefit from lower costs and may pass these savings on to consumers in the form of
lower prices. Second, bankrupt firms may experience lower demand due to the
uncertainty concerning the firms’ future operations. Third, bankrupt firms may focus on
short term profit maximization and thus offer lower prices, ceteris paribus. Consequently,
the following hypothesis is suggested:
Hypothesis 2: The negative impact of financial distress on prices is greater during
bankruptcy than prior to the Chapter 11 filing.
As indicated previously, a different set of theories suggest that a firm’s financial distress
may not significantly impact its prices. These perspectives are reviewed below, and
hypotheses that suggest that the relationship between financial distress and prices is
moderated by other factors are formulated.
2.2.2. Conflicting theoretical arguments
In this section, theoretical arguments and empirical results from the industrial
organization economics, game theory and finance literatures that do not provide support
for the financial distress-price relationship are reviewed.
27
The threat-rigidity model has emerged as a counterhypothesis to prospect theory. Staw
et al (1981) argue that individuals, groups, and organizations exhibit restrictive
information processing patterns, centralize control and conserve resources when faced
with threatening situations. These mechanisms result in increased rigidity which reduces
an organization’s ability to change and adapt to its environment (McKinley 1993). As
noted by McKinley (1993) and Mone et al (1998), there is broad empirical support for the
threat-rigidity model: Smart and Vertinsky (1984), for example, find that executives
consult fewer information sources during crises, and Chattopadhyay et al (2001) present
some evidence that organizations respond to control-reducing threats with low risk,
internally directed actions. From this perspective, firms in poor financial conditions may,
thus, be expected to not reduce prices in the short-term, but to behave passively and
conservatively (Ferrier et al. 2002).
Similar rigidity arguments can be found in the industrial organization economics, game
theoretic and finance literature. First, the kinked demand curve theory suggests that
firms in oligopolistic markets with few sellers and rather homogenous products face
highly inelastic demand for price decreases (Waldman and Jensen 2001). Put differently,
firms will refrain from price competition given that their rival firms may be expected to
match these moves, thus offsetting any profit gains (Scherer 1980). This argument is
further supported by game theory: Derfus et al (forthcoming) argue that pricing actions
are negative-sum actions since all competing firms will be worse off after implementing
successive price reductions. Consider, for example, a sequential game between
28
duopolists: Firm Two observes Firm One’s move and subsequently acts in response to
Firm One’s action. Firm One, in turn, observes Firm Two’s action and may choose to
react, etc. (Gibbons 1992). When such moves consist of price reductions, the price may
fall below average cost levels in the course of this competitive interaction of moves and
countermoves (see also Dasgupta and Titman 1998). These theories are, thus, in line with
the imitation/retaliation argument discussed earlier (Busse 2002, Chen 1996, Chen et al.
1992, Peteraf 1993, Smith et al. 1991).
In summary, the threat-rigidity model and arguments from the industrial organization,
game theoretic, and finance literatures suggest that financially distressed firms may
refrain from lowering prices as information processing and decision making processes are
altered in the face of threats or for fear of retaliation.
2.2.3. The contingency approach
This essay attempts to reconcile the apparent theoretical and empirical conflict that has
shaped previous research on the relationship between financial condition and prices. Each
of the groups of theoretical arguments – those supporting and those denying a negative
impact of financial distress on prices – may be valid under specific circumstances. As
will be discussed below, there are a number of contingencies that may impact the
relationship under investigation. Similar to Ferrier et al (2002), a contingency framework
which suggests moderating effects of organizational and market structural characteristics
is developed. This framework aims at defining in what instances the price effects of
29
financial distress are largest.
As shown in Figure 2, two groups of contingencies are hypothesized to impact the
relationship between financial distress and prices are presented: organizational
characteristics and market characteristics. Both groups of variables are discussed in turn,
and hypotheses are formulated.
Organizational characteristics
It is suggested that the relationship between firm financial distress and prices is
moderated by certain organizational characteristics. More specifically, a firm’s operating
costs, its size and market shares are hypothesized to influence the extent to which firm
financial distress impacts the firm’s pricing behavior. The importance of these factors has
been shown in prior research.
Prior research has suggested that a firm’s particular strategic type may impact its
behavior. Chattopadhyay et al (2001), for example, find that a firm’s propensity to
respond to threats with externally as opposed to internally oriented actions is impacted by
its strategic focus. They present empirical support for the contention that firms focusing
on product-market development (prospectors) are more likely to act externally (by
changing prices, for example) since the “effectiveness of a product-market development
strategy depends to a large extent on controlling or modifying the external environment”
(p.940/941). Firms focusing on domain defense (defenders), in turn, “are more likely to
act within themselves to become more efficient through standardizing organizational
30
processes” (p.941). Therefore, a differential impact of financial distress on a firm’s prices
by strategic type is expected, given the firms’ differential inclinations to act externally
versus internally in response to changes in financial situation.
Although there are multiple definitions and classifications of strategic types (see e.g.
Shoham and Fiegenbaum 2002), these can be simplified and synthesized as follows (see
also Chattopadhyay et al. 2001): Defenders are those firms that operate in a stable, well-
defined set of market segments, tend to act conservatively, and are characterized by
deadlocked organizational structures and operating routines. Prospectors, in turn, are
those firms that constantly seek opportunities to expand their business and whose most
distinctive features are their innovativeness and cost-leadership.
In the empirical practice, many operationalizations of strategic types have been
suggested, ranging from simple dichotomies (e.g. Peteraf 1993) to multidimensional
clusters (Smith et al. 1997). There is, however, substantial agreement in the literature that
a firm’s costs are an important differentiator with respect to its strategic type (see the
above definitions of prospectors and defenders). This is particularly true in the U.S.
airline industry: Both the academic and trade presses frequently refer to specific airlines
as either high-cost carriers or low-cost carriers. Peteraf (1993), for example distinguishes
between pre- and post-deregulation air carriers, the former being mostly high-cost firms
10
while the latter are virtually all low-cost airlines. A firm’s strategic type is therefore
identified by means of its operating costs. In fact, an airline’s relative cost
10
Southwest Airlines being a notable exception.
31
(dis)advantage may impact its choice of strategy. Assuming that lower operating costs
also imply higher profit margins, low-cost firms have some financial flexibility to allow
for price reductions and potentially ensuing price wars. Higher operating costs (and lower
profit margins), in turn, would imply that price cuts likely lead to increased operating
losses.
The crucial assumption for this reasoning to be valid is, of course, a negative correlation
between operating costs and profit margins. The empirical analyses will be conducted
using data from the U.S. airline industry
11
. Accordingly, financial data on U.S. airlines
were collected for a total of eight quarterly time periods (1992 and 2002) from the
Bureau of Transportation Statistics. An analysis of these data indicates that the
correlation coefficient between operating costs per available seat-mile and operating
profit per available seat-mile is equal to r = -0.1481 and is statistically significant at the
five percent level (p = 0.0485)
12
. This result provides some support for the contention that
firms with lower operating costs tend to achieve higher profits and may be able to operate
profitably even if prices are cut. Firms with higher costs and lower profit margins, in turn,
do not have this flexibility and may tend to refrain from lowering prices. The negative
effect of financial distress on prices may thus be expected to decrease with the magnitude
of the firm’s operating costs as depicted in Figure 3 below. The coefficient of the
associated interaction term is, thus, expected to be positive.
11
Further information about the data sources and the nature of the data set is provided in Chapter 2.3.
12
This correlation analysis is based on firm-level 228 observations.
32
Figure 3: The moderating effect of operating costs on the distress-price relationship
Accordingly, Hypothesis 3 is proposed as follows:
Hypothesis 3: The negative effect of financial distress on prices decreases with the
magnitude of the firm’s operating costs.
The effect of firm financial distress on prices may also be impacted by the firm’s size.
Commenting on the survivability of large firms, Tiras (2002) notes that creditors have
greater confidence in the turnaround performance of large distressed firms and thus grant
them more favorable loan conditions than to small firms. In a similar vein, Smith and
Graves (2005) suggest that “larger firms are likely to have a higher probability of
survival, as the potential losses to stakeholders are greater. Also, such firms are likely to
have a higher profile and therefore more likely to be kept alive” (p.306).
Financial
distress
Price ($)
a
v
e
r
a
g
e
lo
w
o
p
e
r
a
t
in
g
c
o
s
t
h
ig
h
o
p
e
ra
tin
g
c
o
s
t
positive
interaction
effect
Financial
distress
Price ($)
a
v
e
r
a
g
e
lo
w
o
p
e
r
a
t
in
g
c
o
s
t
h
ig
h
o
p
e
ra
tin
g
c
o
s
t
positive
interaction
effect
33
Looking at bankrupt firms, in particular, prior research suggests that larger bankrupt
firms have a bankruptcy cost advantage due to scale effects in reorganization costs
(Campbell 1996). The costs of bankruptcy consist of both direct and indirect costs, where
the former “include lawyers’ and accountants’ fees, other professional fees, and the value
of managerial time spent in administering the bankruptcy”, and the latter “include lost
sales, lost profits, and possibly the inability of the firm to obtain credit or to issue
securities” (Warner 1977, p.338)
13
. Numerous researchers have attempted to estimate
these costs. Their estimates vary significantly due to differences in cost definitions,
variable measurement, sample composition, and estimation methodology. The estimates
range from an average of 1.3% of the change in firm value during bankruptcy in the
railroad industry (Warner 1977) to 4% of the firm value in the retail business (Altman
1984), and up to 16.35% of the firm value for a cross-section of industries (Branch
2002)
14
.
Many researchers note, however, that there are significant scale economies in bankruptcy
costs: Warner (1977), for example, finds that bankruptcy costs are linearly decreasing
with firm size. His analyses indicate that bankruptcy costs may be as high as about nine
percent of the firm’s market value for firms with a market value of less than 30 million
dollars and as low as two percent for firms with a market value of around 120 million
dollars. Analyzing bankruptcies in the U.S. trucking industry, Guffey and Moore (1991)
also find a significant negative correlation between firm size (as measured by total asset
13
A more detailed discussion of the composition of bankruptcy costs can be found in Guffey and Moore
(1991) and Branch (2002).
14
See also Bradbury and Lloyd (1994) for a summary of prior research estimating bankruptcy costs.
34
values) and bankruptcy costs. Betker (1997), in turn, finds that the relationship between
firm size (total assets) and bankruptcy costs follows an inverted U shape: The direct
effect of assets on bankruptcy costs carries a positive coefficient while the coefficient of
the squared asset value carries a negative coefficient. The observation of the relationship
between firm size and bankruptcy costs has led researchers to conclude that such
bankruptcy costs may significantly impact smaller firms’ decisions while they may not
substantially impact large firms (Bradbury and Lloyd 1994). Consequently, it is expected
that larger bankrupt firms may be able to offer lower prices than smaller firms due to
their bankruptcy cost advantage.
Previous research has also found that larger firms tend to remain in bankruptcy for longer
periods of time and exhibit significantly higher survival rates than smaller firms (Queen
and Roll 1987, Rodgers 2000). The latter observation may be attributed to lower
bankruptcy costs (Campbell 1996), for example. These advantages in terms of credit
conditions, stakeholder confidence, and bankruptcy costs may allow larger distressed
firms to commit to riskier turnaround strategies that involve more aggressive pricing
behaviors. While detrimental in the short term, the latter may drive competitors out of the
market and result in greater long term returns. It is expected that the negative effect of
firm financial distress on prices will be stronger for larger firms. Consequently, the
interaction effect between financial distress and firm size is hypothesized to positively
affect prices, as noted in Hypothesis 4 and illustrated in Figure 4.
35
Hypothesis 4: The negative effect of financial distress on prices increases with firm
size.
Figure 4: The moderating effect of firm size on the distress-price relationship
The magnitude and direction of the effect of firm financial distress on prices may also
depend on the firm’s market share in the particular product (i.e. route) market. In the long
run, greater market shares may result in the achievement of lower marginal costs
through economies of density (Ferrier et al. 2002). Furthermore, high market shares may
be indicative of barriers to entry and mobility that isolate market-leading firms from
intense competition (Caves and Porter 1978, Caves and Ghemawat 1992). From this
perspective, high market shares may be considered a valuable firm resource that allows
for above-normal returns. Consequently, some researchers have argued that firms will
likely try to defend their market power. Busse (2002), for example, presents empirical
evidence that firms are more likely to enter price wars the greater their market shares, and
Financial
distress
Price ($)
a
v
e
r
a
g
e
la
r
g
e
r
f
ir
m
s
s
m
a
lle
r firm
s
negative
interaction
effect
Financial
distress
Price ($)
a
v
e
r
a
g
e
la
r
g
e
r
f
ir
m
s
s
m
a
lle
r firm
s
negative
interaction
effect
36
LeBlanc (1992) argues that firms strive to maintain monopoly profits by implementing
limit or predatory pricing.
These predictions may, however, not hold when explicitly considering distressed firms.
First, note that distressed firms typically focus on short term survival rather than on long
term strategic positioning. While the latter is the ultimate purpose of distressed firms’
turnaround efforts, generating sufficient cash flows is a mandatory obligation these firms
face in the immediate future. In this vein, bankrupt U.S. airlines frequently terminate
unfavorable aircraft leases and collective labor agreements right upon entry into Chapter
11 protection. If liquidity is the prime objective, however, price cuts in an effort to
maintain market shares may prove counter-productive for high market share firms: Any
price reductions will imply lower total revenues since the incremental increase in
customer demand likely will not outweigh the detrimental effect of lower sales prices.
Assuming (quasi-)fixed production costs in the short run, these revenue losses directly
affect the firm’s bottom line. Low-market share firms, in turn, may see a substantial
increase in customer demand when reducing prices. The prospect of increased volume
may, thus, offset the negative effect of lower sales prices. This implies that engaging in
price competition is more appealing to firms with smaller market shares: The potential
market shares to be gained are greater, and any pricing actions hurt the market leading
firm(s) significantly more than the smaller firm. This reasoning reflects the concepts of
Judo economics (Gelman and Salop 1983) and Judo strategy (Yoffie and Kwak 2002),
which essentially posit that a firm’s market shares (and/or size) may constitute a
competitive disadvantage when adequately leveraged against it by smaller firms (in terms
37
of market shares).
Standard microeconomic theory further suggests that firms with greater market shares
possess market power and can charge price premiums (see e.g. Borenstein 1989).
Extending this argument to the present research context, it is expected that distressed
firms with higher market shares have higher degrees of market power and will be
required to compete on prices to a lesser extent than firms with lower market shares and
little market power. Firms with higher market shares may be able to retain greater shares
of market demand due to customer retention instruments such as loyalty programs which
create higher switching costs for consumers. The latter may thus be reluctant to switch to
financially stronger competitors even though they may seem more reliable or offer lower
prices. From this perspective, demand inelasticity confers firms with greater market
shares greater degrees of market power. And such market power, in turn, enables even
distressed firms to maintain higher price levels, ceteris paribus.
In summary, these arguments thus suggest that higher market shares reduce the negative
effect of firm financial distress on prices (see Figure 5), and the associated interaction
effect is expected to be positive. Hypothesis 5 below formally states this contention:
Hypothesis 5: The negative effect of financial distress on prices decreases with the
firm’s market share.
38
Figure 5: The moderating effect of market shares on the distress-price relationship
A second set of contingencies, those relating to market characteristics, are discussed
below.
Market characteristics
Besides organizational characteristics, select market characteristics are hypothesized to
impact a distressed firm’s pricing strategy. Market concentration is one of the most
widely used measures of the competitiveness of markets in the extant literature. While
there are many alternative measures of market structure—the number of sellers in the
market and multi-market contact measures, for example, have been used to characterize
the structure of markets in prior research (e.g. Mazzeo 2002, Scott 1982)—the degree of
market concentration is likely highly correlated with these alternative measures and
Financial
distress
Price ($)
a
v
e
r
a
g
e
lo
w
m
a
r
k
e
t
s
h
a
r
e
h
ig
h
m
a
rk
e
t s
h
a
re
positive
interaction
effect
Financial
distress
Price ($)
a
v
e
r
a
g
e
lo
w
m
a
r
k
e
t
s
h
a
r
e
h
ig
h
m
a
rk
e
t s
h
a
re
positive
interaction
effect
39
appropriately captures the structural characteristics of a market. The second market-
specific factor included in this study is the financial condition of all the firms in a market.
This variable is included to evaluate how a firm’s financial condition differs from the
average distress level of the other firms in the market and how this relative difference
impacts the magnitude of a firm’s pricing actions.
First, market concentration will likely affect a firm’s pricing decision. More
specifically, the expectation of competitive responses and retaliatory moves in highly
concentrated markets impacts a firm’s valuation of the effects of any price changes. The
structure-conduct-performance paradigm posits that industry concentration reduces the
level of competition (Scherer 1980, Waldman and Jensen 2001). Young et al (1996) find
empirical support for this contention, noting that firms in concentrated markets or
industries carry out fewer competitive moves. The underlying assumption of this
reasoning is, however, that the competing firms are similar to one another and that their
products are largely homogeneous. Waldman and Jensen (2001) list a variety of factors
that violate this homogeneity assumption and may hinder effective collusion between
firms in concentrated markets. Cost differences between competing firms, for example,
may negatively affect the ease of collusion.
A deterioration in a firm’s financial position, and bankruptcy in particular, may bring
about such cost differences: firms operating under Chapter 11 protection, in particular,
may pause debt payments and shed financial obligations such as contributions to pension
plans, for example (Rose-Green and Dawkins 2002). This new cost structure may then
40
lead to the firm’s repositioning in the product market. Specifically, a change in a firm’s
operating costs changes the firm’s profit maximization problem, and consequently its
optimal price levels. The interaction of market concentration and financial distress may
therefore lead to a destabilization of collusive arrangements and increase pricing
competitiveness (Barla and Koo 1999). While market concentration is expected to be
positively related to prices, this research contends that this positive relationship will
diminish in magnitude in the light of an aggravation of a firm’s financial condition. Put
differently, the interaction of financial distress and market concentration is expected to
negatively affect prices, ceteris paribus (Hypothesis 6).
Hypothesis 6: The impact of financial distress on prices is greater, the higher the level
of market concentration.
Figure 6 illustrates the differential effect of financial distress on prices as a function of
the degree of market concentration.
41
Figure 6: The moderating effect of market concentration on
the distress-price relationship
A distressed firm’s pricing decisions will, in part, also depend on its competitors’
financial situations. If a firm’s rivals experience similar degrees of distress as the focal
firm does (and assuming that the firms’ products are undifferentiated), then these rivals
may be expected to exhibit comparable or symmetric pricing behaviors. A focal firm’s
price reductions would then be matched by the other firms, and no single firm could gain
a competitive advantage. In fact, game theory suggests that in a perfectly competitive
setting each firm will always have an incentive to slightly undercut its competitor’s
prices, thus eroding profit margins to zero (Gibbons 1992). Financially distressed firms
will, therefore, avoid competing on price when their competitors find themselves in
similar financial conditions. Conversely, Hypothesis 7 is stated as follows:
Financial
distress
Price ($)
a
v
e
r
a
g
e
lo
w
m
a
rk
e
t c
o
n
c
e
n
tra
tio
n
h
ig
h
m
a
r
k
e
t
c
o
n
c
e
n
t
r
a
t
io
n
negative
interaction
effect
42
Hypothesis 7: The greater a firm’s financial distress relative to its competitors, the
lower the firm’s sales prices.
In summarizing, a set of hypotheses on the link between firm financial distress and firm
prices has been formulated based on a variety of theoretical perspectives. Conflicting
viewpoints that may suggest the absence of any significant relationship respectively are
presented, and a contingency framework that more precisely defines for what type of
firms and under what circumstances changes in a firm’s financial situation may indeed
cause changes in the firm’s pricing behavior is proposed. The resulting model is shown in
Figure 7.
Figure 7: Research model
Financial
distress
Market
concentration
Relative fin.
distress
Control variables
Market share
Firm size
Operating costs
Price
M
a
r
k
e
t
(
s
t
r
u
c
t
u
r
a
l
)
c
h
a
r
a
c
t
e
r
i
s
t
i
c
s
F
i
r
m
c
h
a
r
a
c
t
e
r
i
s
t
i
c
s
Financial
distress
Market
concentration
Relative fin.
distress
Control variables
Market share
Firm size
Operating costs
Price
M
a
r
k
e
t
(
s
t
r
u
c
t
u
r
a
l
)
c
h
a
r
a
c
t
e
r
i
s
t
i
c
s
F
i
r
m
c
h
a
r
a
c
t
e
r
i
s
t
i
c
s
43
In the following section, information about the sample data that is used for the empirical
analyses is provided, and measurements of the variables in the research model as well as
methodological issues are discussed.
2.3. Data and methodology
The U.S. airline industry provides the setting for the empirical analyses. This selection is
particularly suitable for a number of reasons. First, the markets are clearly defined (Smith
et al. 1991), and all firms operating in these markets are dominant-business firms (Peteraf
1993), i.e. firm-specific data reflect the firms’ aviation activities and are not diluted by
non-aviation business activities. Second, the U.S. airline industry is highly competitive
and encompasses a large cross-section of routes that differ significantly with respect to
their market characteristics (Peteraf 1993, Smith et al. 1991). Third, the industry has
experienced periods of severe financial distress (Borenstein and Rose 1995), but is
sufficiently heterogeneous with respect to the airlines’ financial conditions. Finally, there
is a wealth of publicly available data on the U.S. airline industry due to the U.S.
Department of Transportation’s reporting requirements
15
.
15
Some sections of this chapter, particularly the sample data and variable descriptions, are similar or equal
to the corresponding sections of a related paper published by Hofer et al (2005).
44
2.3.1. Data sample
Data were collected on the top 1000 U.S. domestic origin and destination route markets
16
,
for all quarters in 1992 and 2002. These years were chosen because the airline industry
experienced serious distress in the early nineties (Barla and Koo 1999) and in the
aftermath of the 9/11 attacks. At the same time, limiting the analyses to two years only
allowed keeping the dataset at a manageable size. The sensitivity of the empirical results
with respect to the selection of these particular time periods is investigated by re-
estimating the regression models using an extended data set that also includes 1997 data.
These results will be discussed in Section 2.4.3.
Quarterly data are used to capture the short-term effects of financial distress and Chapter
11 filings on air fares. The top 1000 route markets cover a wide range of route
characteristics in terms of traffic volume, distance, and intensity of competition. The unit
of observation is a specific carrier’s fare on a particular route market in a given time
period.
The raw data were purchased from Database Products Inc. (DPI), a reseller of the
Department of Transportation’s DB 1A data which contain a 10% sample of all U.S.
domestic origin and destination tickets. DPI downloads the DB 1A data and screens them
for erroneous and redundant data entries. These entries and data points from non-revenue
16
Based on 2002 traffic figures, 48 contiguous states only.
45
transactions
17
are removed from the dataset. The data obtained from DPI thus are filtered
and quality-controlled and provide airline and route specific information on fares,
nonstop and itinerary miles, the number of passengers, and the number of coupons.
Additional air traffic and airline operating and financial data were gathered from the
DOT’s T-1
18
and Form 41
19
databases. Other data sources include the American
Transport Association (ATA; U.S. airline bankruptcy data), the Bureau of Labor
Statistics (BLS; income data and inflation indexes) and the Bureau of Economic Analysis
(BEA; population statistics).
Observations from carriers with less than five percent route market share were deleted
from the data set to keep the data set at a manageable size
20
. Furthermore, a total of 577
observations were excluded because of unidentified carriers
21
, or unavailable airport and
airline-specific data. A total of 23,039 observations were retained for the analyses. Each
observation indicates data for a specific carrier on a specific route market in a specific
time period.
2.3.2. Variables and measurement
This section provides detailed information on the variables used in this research model.
17
E.g. personnel travel and frequent flyer award travel.
18
Table T-1 provides summaries of T-100 data by carrier, aircraft type and service class and includes
information on available seat miles (ASM) and revenue passenger miles (RPM).
19
Form 41 (financial schedule) contains financial information on large U.S. certified air carriers including
data from balance sheets, income statements, and information on cash flows, and aircraft operating
expenses.
20
This is common practice: Borenstein and Rose (1995), for example, exclude all observations of carriers
with less than ten percent route market shares.
21
“XX – unduplicated commuters” and “UK – unknown carrier”.
46
The purpose of this research is to investigate the effect of financial distress on prices.
Consequently, ticket prices (fares) are used as the dependent variable. Among those
factors that may explain and predict variations in ticket prices, firm financial distress is of
particular interest here. Other independent variables include not only the aforementioned
moderating factors—operating costs, firm size, market shares, and market concentration
(see also Figure 7)—but also a set of airline-specific, route-specific, and airport-specific
characteristics that have been shown to impact air fares in prior research (Hofer et al.
2005). The dependent variable is discussed first, followed by the independent variables
of interest. In addition, information on the set of control variables included in the
empirical estimation model is provided. Descriptive statistics and a correlation table are
provided in Section 2.3.3.
2.3.2.1. Dependent variable
Previous studies published in the strategic management literature have measured the
impact of financial condition on firm behavior in terms of the number and type of
competitive actions, response speed and delay, for example (Chen et al. 1992, Ferrier
2001, Ferrier et al. 2002, Smith et al. 1997, Smith et al. 1991, Young et al. 1996). While
price data are commonly used as dependent variables in the economics literature, this is,
to the best of the author’s knowledge, the first study to investigate the impact of financial
distress on prices from a strategic management perspective. More specifically, Fare
kij
is
the average price carrier k charges on the route between airports i and j
22
. All fare values
22
Fare values are averages across all booking classes and do not include taxes and fees.
47
are one-way fares based on roundtrip purchases and are given in real 1992 U.S. dollars
23
.
2.3.2.2. Independent variables
The measurement of financial distress is of particular interest in the context of this study.
Previous studies of financial condition have generally relied on one of two measures.
Ferrier et al (2002) and Chakravarthy (1986), for example, relied on a composite measure
to evaluate a firm’s financial situation. Altman’s (1968) Z score is the most prominent
member of this group of measures and takes into account the firm’s past and present
profitability, its liquidity and its degree of activity. Other researchers have focused on
Chapter 11 filings
24
, the most visible and definite sign of financial distress, to investigate
the effects of firm financial distress (e.g. Borenstein and Rose 1995, Kennedy 2000).
While both measures have their merit, it is important to note that they capture different
aspects of financial distress. Z score-type measures are indicators of a firm’s financial
health (or distress), while Chapter 11 filings refer to a specific point in time at which the
firm is no longer able to meet its debt obligations. The model builds on both of these
indicators and includes four measures of financial distress to more precisely sort out its
effects on firm behavior in terms of pricing:
? Distress
k
is a measure of Airline k’s financial distress. The Distress variable is the
inversion of firms’ Z scores. More specifically, Z’’ scores (Altman 2002) are used, a
revised version of Altman’s original Z score formulation (1968) which is particularly
suitable for firms operating in service industries (such as the airline industry). The
23
All nominal values were converted to real 1992 dollars using the appropriate price indexes published by
the Bureau of Economic Analysis.
24
See Daily (1994) for a comprehensive explanation and discussion of the U.S. Code Chapter 11.
48
more recent Z’’ scores (Altman 2002) are also preferred over the original Z score
formulation (Altman 1968) since it has been shown that “the relation between
financial ratios and financial distress changes over time” (Grice and Ingram 2001)
such that more recent formulations are more reliable and effective in predicting a
firm’s financial distress. Based on discriminant analysis, Altman (2002) developed
the following model to estimate a firm’s financial fitness:
1 2 3 4
'' 6.56* 3.26* 6.72* 1.05* Z X X X X = + + + where X
1
= working capital / total
assets; X
2
= retained earnings / total assets; X
3
= Earnings Before Interests and Taxes
(EBIT) / total assets; X
4
= book value of equity / total liabilities. All airline financial
data needed to compute the Z’’ scores were obtained from the Department of
Transportation’s Form 41 data which are available online on a carrier-time period
basis. High Z’’ scores indicate financial health, while low and negative scores
indicate (serious) financial distress. Specifically, it has been suggested that scores of
2.60 or above indicate financial health, while scores of 1.10 or lower indicate severe
distress. To facilitate the interpretation of the estimation results, the Z scores are
inverted, i.e. ( ) 1 Distress ZScore = ? ? , such that higher (positive) Distress scores
indicate financial distress (see also Ferrier et al. 2002). This variable is used to test
Hypothesis 1. Moreover, the airlines’ Distress scores are used to test the moderating
effects from Hypothesis 3, Hypothesis 4, Hypothesis 5, and Hypothesis 6,
respectively.
? Chpt11Ops
k
is a binary (0/1) variable that identifies those carriers that operate under
Chapter 11 protection (“1”). It thus is an alternate, though rather coarse, measure of a
firm’s financial distress. All bankruptcy data were obtained from WebBRD, a
49
bankruptcy research database that is accessible online athttp://webbrd.com/. This
database is maintained by Professor Lynn M. LoPucki with the University of
California at Los Angeles. This database specifies the dates at which firms (airlines)
entered into and exited from Chapter 11 protection. These data were also double-
checked with the bankruptcy data which are available online at the Air Transport
Association’s website (http://www.airlines.org/econ/). No discrepancies were found.
? Pre4Chpt11
k
, identifies those carriers that will face bankruptcy within the following
four quarters. In the latter case, this binary variable takes on the value of “1”. This
variable is based on the same sources as the Chpt11Ops variable defined above. A
four-quarter period (prior to the Chapter 11 filing) is selected to best capture price
reactions to aggravating financial distress in the time period immediately preceding
bankruptcy.
? Post4Chpt11
k
is similar to the Pre4Chpt11
k
variable, but identifies those carriers that
filed for Chapter 11 within the past four quarters (“1”). This variable is based on the
same sources as the Chpt11Ops variable defined above. The inclusion of the
Pre4Chpt11
k
and Post4Chpt11
k
variables allows capturing the differential impact of
financial distress over time as stated in Hypothesis 2.
? The Chpt11
k
variable is an indicator variable which is equal to “1” if the focal carrier
filed for bankruptcy protection in the current quarter. The number of observations in
which this is the case is small such that this variable is only used for descriptive
purposes (see Figure 9) and is not included in the regression analysis. An overview of
the Chapter 11 dummy variables is provided in Figure 8 below. Note that the
Pre4Chpt11, Chpt11, and Post4Chpt11 variables are mutually exclusive.
50
Figure 8: Overview of Chapter 11 indicator variables
? DistressDiff
kij
is an indicator of an airline’s financial standing relative to its route
competitors. It is based on Altman’s Z’’ score and is computed for each carrier in
each route market for each time period. It is the difference between the focal carrier’s
Z’’ score and the route market share weighted average of its route competitors’ Z’’
scores:
( )
*
competitors competitors
focal
competitors
Distress route shares
DistressDiff Distress
route shares
= ?
?
?
. Higher scores,
thus, indicate that the focal carrier is financially better off relative to its route
competitors and vice versa. The DistressDiff
kij
variable is designed to test Hypothesis
7 which refers to an airline’s financial standing relative to its competitors. This
variable, thus, differs from the Distress and Chpt11 variables in that it indicates an
airline’s relative financial standing, i.e. the focal firm’s Distress relative to the market
share weighted average Distress of its competitors, rather than its absolute financial
distress. Positive DistressDiff values indicate that the airline is financially worse off
than its route competitors, while negative values indicate relative financial wellbeing.
Chapter 11
filing
Pre4Chpt11 Post4Chpt11
Chpt11
Chpt11Ops
time
(quarters)
Chapter 11
filing
Pre4Chpt11 Post4Chpt11
Chpt11
Chpt11Ops
time
(quarters)
51
Further, a set of moderating variables is included as suggested in Hypothesis 3 to
Hypothesis 6. More specifically, it is hypothesized that the impact of financial distress on
prices varies by strategic type/operating costs (Hypothesis 3), firm size (Hypothesis 4),
firm market shares (Hypothesis 5), and market concentration (Hypothesis 6). These
moderating variables are operationalized as follows:
? AirlineCost
k
is an indicator of an airline’s operating efficiency. It is defined by the
ratio of operating expenses to available seat miles (ASM).
? Size
k
indicates the firm’s size in terms of its total assets (measured in 000s of U.S. $).
? RouteShare
kij
measures an airline’s market share on a route market (based on its share
of route passengers).
? RouteHHI
ij
is a measure of route market concentration. It is based on the Herfindahl-
Hirschmann Index (HHI), the sum of the squared market shares of all firms
competing in the route market. The route HHI is computed on an airport-to-airport
basis rather than on a city-to-city basis. This allows capturing airport-specific effects.
These variables are interacted with the Distress variable to estimate their moderating
effects in the relationship between firm financial distress and prices.
2.3.2.3. Control variables
A set of firm and market specific control variables that have been shown to impact prices
in previous research (see e.g. Borenstein 1989) is included in the empirical model. The
firm-specific variables are the following:
? MaxAirportShare
kij
indicates an airline’s market share in the airport market i or j,
52
whichever is highest. The rationale for this approach is that a higher market share at
an airport conveys an airline some degree of market power in that airport market
which may be expected to impact fares in route markets involving that airport
(Borenstein 1989). Higher airport market shares likely imply higher fares.
? Circuity
kij
is another measure of the quality and convenience of carrier k’s service
between airports i and j. Circuity is the ratio of itinerary miles; i.e. the distance
actually flown, and nonstop miles between airport i and j. The higher this ratio; i.e.
the larger the detour, the lower the quality of the transportation service. At the same
time, however, higher circuities mean higher operating costs. The impact on fares is,
thus, undetermined.
? AirlinePass
kij
is the number of passengers carried by airline k between airports i and j
in a given time period. Higher numbers of passengers may be associated with
economies of density, and, thus, lower costs and lower fares. On the other hand, high
traffic volumes reflect high demand levels which may result in high prices.
? Loadfactor
k
is the average fill rate of a carrier k’s passenger aircraft during a given
time period. Note that this variable is not route specific since data were not available
on a route basis. Higher load factors may imply economies of density and utilization,
and fares may be expected to be lower for carriers with high load factors. At the same
time, high load factors may be associated with poorer service quality (e.g. possibly
lower frequency of service, less space for each traveler in a fully booked cabin, less
attentive/personalized cabin service) and lower fares.
The group of market specific control variables consists of the following variables:
53
? Distance
ij
is the distance (in miles) between airports i and j. In general, fares may be
expected to rise as distance increases.
? DistanceSquared
ij
is the square of the Distance
ij
variable. Its inclusion allows for a
nonlinear relationship between distance and fares.
? SlotRoute
ij
is a binary variable that indicates whether one or both airports i, j are slot-
controlled
25
. Such airports are typically highly congested and access is limited. Fares
are therefore expected to be higher on routes to or from these airports.
? MaxAirportHHI
ij
indicates the degree of concentration of an airport market. Rather
than including two values for both airports i and j, only the higher HHI value is
retained in this analysis. The rationale for this approach is that the more concentrated
airport is more likely to be the “bottleneck”, and fares on routes involving this airport
may be expected to be higher than fares on routes between “unconcentrated” airports.
? LCCCompForHCC
ij
is a binary variable. It takes on the value “1” when the carrier in
the observation is a high cost carrier and faces route competition by a low-cost
carrier. While some studies focused on Southwest Airlines only (e.g. Morrison,
2001), others employed a wider definition of low-cost carriers and defined all carriers
that started operations after deregulation as low-cost carriers (e.g. Dresner et al.
1996). In an effort to rigorously define LCCs in this research, financial data on all
airlines included in this analysis were collected. To account for the fact that operating
expenses per available seat mile (ASM) are likely higher for airlines operating
predominantly short haul flights, a carrier’s operating expenses per ASM were
regressed on its average stage length. The error terms, thus, reflect differences in
25
Presently, JFK, LGA, and DCA are the only slot-controlled airports in the U.S.; ORD was slot controlled
until June 2002
54
operating costs that cannot be attributed to differences in average stage length and are
indicators of an airline’s operating efficiency. A ranking of these error terms revealed
consistent patterns across all time periods considered in this research, and twelve
airlines were identified as low-cost carriers (see Appendix 1): Southwest Airlines,
Reno Air, Sun Country Airlines, Spirit Air Lines, JetBlue Airways, Western Pacific
Airlines, Airtran Airways, American Trans Air, Braniff Int'l Airlines, America West
Airlines, Frontier Airlines, Valujet Airlines.
? LCCCompForLCC
ij
is a binary variable which takes on the value “1” when the carrier
in the observation is a low-cost carrier and competes with another low-cost carrier in
the route market. LCCCompForHCC
ij
and LCCCompForLCC
ij
specify the presence
of low-cost carrier competition. These two variables are used to allow for differential
impacts in terms of pricing on LCCs and high cost carriers.
? AltRouteLCC1M
ij
is another dummy variable which indicates if there are one or more
adjacent route markets that are served by one or more low-cost carriers. The inclusion
of this variable builds on the work by Dresner, Lin and Windle (1996) and Morrison
(2001) who analyzed the impact of adjacent route competition on fares. Based on the
population statistics (i.e. PMSA, CMSA, MSA) published by the Bureau of Economic
Analysis (BEA), the following markets have been defined as metropolitan multi-
airport markets in this research: Boston (BOS, MHT, PVD), Chicago (ORD, MDW),
Cleveland (CLE, CAK), Dallas (DAL, DFW), Detroit (DTW, FNT), Houston (HOU,
IAH), Los Angeles (BUR, LAX, LGB, ONT, SNA), Miami (MIA, FLL), New York
(EWR, JFK, LGA, ISP, HPN), Norfolk (ORF, PHF), Philadelphia (PHL, ACY), San
Francisco (OAK, SFO, SJC), Tampa (TPA, PIE), Washington (BWI, DCA, IAD).
55
? Time variables are also included in the analysis to capture macroeconomic changes as
well as seasonal fluctuations (quarter dummies Quarter2, Quarter3, Quarter4) and
general trends over time (year dummy 2002).
? Population
ij
is used as a first-stage instrument and is the product of the metropolitan
area populations around airports i and j:
ij i j
Population Population Population = ? . A
first stage estimation of the AirlinePass variable is required to address the
endogeneity of the Fare and AirlinePass variables (see Section 2.3.4 for further detail
on this endogeneity issue). For econometric reasons, the first-stage estimation of the
endogenous variable (AirlinePass) requires the use of at least one instrumental
variable. Population is one of two instrumental variables used in this research.
? Income
ij
is also used as a first-stage instrument and is the population-weighted
average income in the metropolitan areas around airports i and j:
( ) ( )
( )
i i j j
ij
i j
Income Population Income Population
Income
Population Population
? + ?
=
+
.
2.3.3. Descriptive statistics
Correlations are presented in Table 1 below. Due to the large number of observations,
most correlations are statistically significant at the five percent level. Few correlations
coefficients, however, are larger than 0.50. The market share and market concentration
measures are highly correlated
26
, as are the measures of firm financial distress
27
.
26
Airport market shares (MaxAirportShare) and route market shares (RouteShare), for example, have a
correlation coefficient of 0.72.
27
The correlation coefficient for the Distress and DistressDiff variables is 0.79, for example.
56
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
1 Fare
2 Distance 0.52
3 SlotRoute 0.13 -0.04
4 RouteHHI -0.19 -0.59 0.01
5 MaxAirportHHI 0.15 -0.21 -0.10 0.50
6 RouteShare -0.12 -0.40 0.04 0.52 0.25
7 MaxAirportShare 0.05 -0.34 -0.01 0.44 0.46 0.72
8 Size 0.20 0.11 0.06 -0.06 0.01 -0.01 0.20
9 LCCCompForHCC -0.10 0.19 -0.12 -0.29 -0.19 -0.18 -0.14 0.28
10 LCCCompForLCC -0.27 -0.04 -0.10 -0.04 -0.11 -0.02 -0.06 -0.31 -0.20
11 AltRouteLCC1M -0.06 0.06 0.26 0.02 -0.03 0.04 -0.01 0.02 0.07 0.00
12 Circuity 0.00 0.03 -0.07 -0.06 -0.08 -0.37 -0.28 0.11 0.11 -0.07 -0.14
13 Distress 0.10 0.06 0.06 -0.05 0.01 -0.13 -0.16 -0.37 0.06 -0.13 -0.01 0.00
14 DistressDiff -0.03 0.00 0.00 0.00 0.00 -0.10 -0.13 -0.32 0.12 -0.06 0.00 0.00 0.79
15 Chpt11Ops 0.08 0.04 0.00 -0.04 -0.01 -0.09 -0.10 -0.23 -0.03 -0.03 -0.01 0.00 0.43 0.33
16 Pre4Chpt11 -0.06 0.03 0.02 -0.02 -0.02 -0.03 0.00 0.11 0.14 -0.06 0.07 -0.01 0.18 0.16 -0.08
17 Post4Chpt11 0.03 0.00 0.02 -0.01 0.00 -0.06 -0.08 -0.18 -0.04 0.00 -0.01 -0.01 0.23 0.18 0.55 -0.04
18 Loadfactor -0.17 0.12 0.03 -0.07 -0.07 -0.03 0.00 0.32 0.24 -0.05 0.10 0.05 -0.12 -0.02 -0.18 0.18 -0.14
19 AirlineCost 0.43 0.00 0.09 -0.03 0.06 -0.05 0.06 0.20 0.06 -0.37 -0.10 0.05 0.20 0.06 0.09 0.05 0.07 -0.31
20 AirlinePass -0.32 -0.41 0.12 0.35 0.08 0.71 0.56 -0.05 -0.10 0.09 0.18 -0.47 -0.10 -0.06 -0.10 0.02 -0.07 0.08 -0.17
(correlation coefficients in bold are significant at the 5% level)
Table 1: Correlation matrix (n = 23,039)
57
The change of firms’ Distress during bankruptcy is illustrated in Figure 9: For each
possible state with respect to bankruptcy, the airlines’ unweighted mean Distress scores
are graphed. The averages are computed for all carriers and across all time periods (eight
quarters in 1992 and 2002) included in the dataset. The sample size n indicates the
number of firm-quarter observations the respective statistics are based on. As can be seen
in Figure 9, non-bankrupt (i.e. comparatively healthy) carriers have a mean Distress
score of 0.8, with the minimum and maximum Distress scores being -3.0 and 18.6,
respectively. Financially sound airlines are thereby assumed to have negative Distress
scores while troubled (yet non-bankrupt) carriers have positive Distress scores. Mean
Distress scores for carriers approaching bankruptcy (Pre4Chpt11) are substantially
higher (12.2) and always positive, ranging from 0.4 to 69.5. A Welch-Aspen two-sample
t test for independent groups
28
is performed to evaluate whether these differences are
statistically significant. The test statistic is t = 2.55 with 15.1 degrees of freedom (df).
This result is statistically significant at the five percent level, indicating that firms
approaching bankruptcy have significantly higher distress scores. Carriers’ Distress
scores average 1.7 during the quarter in which the Chapter 11 filing occurs (Chpt11). It
should be noted, however, that this statistic is based on four carrier observations only
29
.
Bankrupt carriers’ Distress scores averaged 7.7 in the four quarters following the entry
into bankruptcy (Post4Chpt11). In this latter case, the Distress scores range from 0.4 to
26.8. Based on these descriptive statistics it is concluded that financially distressed
carriers will always have positive Distress scores. Financially sound airlines, in turn,
have negative Distress scores.
28
Given the difference in sample sizes, unequal variances are assumed.
29
TWA filed for Chapter 11 in the first quarter of 1992, Markair filed in the second quarter of 1992, US
Airways filed in the third quarter of 2002, and United Airlines filed in the fourth quarter of 2002.
58
Figure 9: Distribution of Distress scores prior to and during bankruptcy
Table 2 provides the mean and standard deviation, minimum and maximum values for
some selected variables included in the model. The mean Distress score of 0.59 indicates
that most carriers experienced some level of financial distress in 1992 and 2002
30
. While
there is no direct interpretation for this value, it implies that a significant proportion of
passengers used (severely) troubled carriers. This contention is further substantiated by
the mean of the Chpt11Ops variable (0.11) which suggests that eleven percent of all
passengers traveled with bankrupt airlines. It is further noted that approximately seven
percent of all passengers traveled with near-bankrupt carriers (Pre4Chpt11, 1571
observations), while approximately five percent of all passengers used airlines that filed
30
The mean of the Distress variable is significantly different from 0. A one-sample mean comparison test
yields a test statistic of t = 43.74 which is statistically significant at the one percent level.
18.6
12.2
1.7
7.7
-3.0
26.8
2.8
69.5
0.8 0.4 0.4
0.6
-10
0
10
20
30
40
50
60
70
80
No Chapter 11 Pre4Chpt11 Chpt11 Post4Chpt11
Distress
(n = 138) (n = 16) (n = 4) (n = 9)
Max.
Avg.
Min.
Max.
Avg.
Min.
Max.
Avg.
Min.
Max.
Avg.
Min.
59
for Chapter 11 protection within the past year (Post4Chpt11, 1196 observations).
Moreover, the data set contains 2,414 carrier-route market observations (out of a total of
23,039 observations) with DistressDiff values larger than 2.86 (one standard deviation
above the mean), indicating an airline’s severe financial distress relative to its route
competitors.
Variable Mean Std. dev. Min Max
Fare (1992 U.S. dollars) 114.26 60.61 25.17 1140.19
Distress 0.59 2.28 -3.02 69.53
DistressDiff 0.05 2.81 -31.97 68.59
Chpt11Ops 0.11 0.31 0 1
Pre4Chpt11 0.07 0.25 0 1
Post4Chpt11 0.05 0.21 0 1
AirlineCost ($) 0.08 0.02 0.04 0.24
RouteShare 55.47% 27.28% 5% 100%
RouteHHI 5298.78 2172.02 1259.66 10000
Size (1,000 $) 11,900,000 8,887,998 3,936 29,300,000
AirlinePass 1570.76 2408.01 1 29368
Distance (miles) 950.86 643.09 30 2717
SlotRoute 0.22 0.41 0 1
MaxAirportShare 46.09% 23.89% 0.015% 100%
MaxAirportHHI 3966.56 1885.11 1131.37 10000
Circuity 1.02 0.05 1 2.2
Loadfactor 67.81% 6.18% 35.1% 84.6%
Quarter1 0.23 0.42 0 1
Quarter2 0.26 0.44 0 1
Quarter3 0.27 0.44 0 1
Quarter4 0.25 0.43 0 1
1992 0.42 0.49 0 1
2002 0.58 0.49 0 1
Mean values weighted based on number of airline passengers, except for "AirlinePass"
F
i
n
a
n
c
i
a
l
d
i
s
t
r
e
s
s
M
o
d
e
r
a
t
o
r
s
C
o
n
t
r
o
l
v
a
r
i
a
b
l
e
s
(
s
e
l
e
c
t
e
d
)
Table 2: Descriptive statistics for selected variables (n = 23,039)
60
2.3.4. Empirical methodology
A log-linear price estimation equation forms the basis of the model used in this research.
More specifically, an airline’s fare on a route is modeled as a function of a set of route,
airport, and carrier specific variables, as well as a number of control variables. The
estimation of the model requires the implementation of a two-stage least squares
procedure since AirlinePass
kij
is an endogenous variable; i.e. the number of airline
passengers may impact airfares while at the same time the latter may have an effect on
the number of passengers. In a first stage regression, the number of airline passengers
(AirlinePass) is modeled as a function of all exogenous variables including two
instrumental variables, Income and Population. Fitted values for AirlinePass are then
used to estimate fares (Fare) in the second stage model. The basic estimating model is
defined as follows:
Equation 1: First-stage regression model
lnAirline Passengers = ?
0
+ ?
1
lnDistance + ?
2
(lnDistance)
2
+ ?
3
Slot Route
+ ?
4
lnRoute HHI + ?
5
lnMax Airport HHI + ?
6
Route Share + ?
7
Max Airport Share
+ ?
8
LCC Comp for HCC + ?
9
LCC Comp for LCC + ?
10
Alt Route LCC
+ ?
11
lnCircuity + ?
12
Distress + ?
13
Load Factor + ?
14
lnAirline Cost + ?
15
lnSize
+ ?
16
lnPopulation + ?
17
lnIncome + ?
18
2002 + ??
t
Quarter
t
61
Equation 2: Second-stage regression model
lnFare = ?
0
+ ?
1
lnAirline Passengers (fitted) + ?
2
lnDistance + ?
3
(lnDistance)
2
+ ?
4
Slot Route + ?
5
lnRoute HHI + ?
6
lnMax Airport HHI + ?
7
Route Share
+ ?
8
Max Airport Share + ?
9
LCC Comp for HCC + ?
10
LCC Comp for LCC
+ ?
11
Alt Route LCC + ?
12
lnCircuity + ?
13
Distress + ?
14
Load Factor
+ ?
15
lnAirline Cost + ?
16
lnSize + ?
17
2002 + ??
t
Quarter
t
The OLS assumptions of homoskedasticity and independence are frequently not met
when dealing with cross-sectional time series data (Greene 2003). Therefore, tests to
detect the potential problems of heteroskedasticity and autocorrelation of the error terms
are implemented.
First, the Breusch-Pagan/Cook-Weisberg Lagrange multiplier test (Breusch and Pagan
1979, Cook and Weisberg 1983) uses fitted values of the dependent variable (Fare) to
determine whether the residuals vary with the fitted values of the dependent variable; i.e.
violate the homoskedasticity assumption. This test is implemented after an OLS
regression similar to the second stage model described above (the sole difference being
that the actual values of Airline Passengers are used rather than fitted values; see
Appendix 2). The implementation of this test yields a test statistic of 614.33 which
follows a ?
2
distribution. The null hypothesis of constant variance is clearly rejected with
a significance level of less than one percent.
62
Second, the Wooldridge test for autocorrelation in panel data (Drukker 2003, Wooldridge
2002) suggests the presence of first-order auto-correlation with an F-statistic of
F = 829.37 which is statistically significant at the less than one percent level. Given the
presence of heteroskedasticity, autocorrelation and endogeneity (as discussed previously),
a generalized two-stage least squares (G2SLS) procedure is recommended (Greene
2003).
The generalized least squares procedure is typically implemented in one of two distinct
econometric specifications: fixed effects or random effects
31
. These two specifications
differentially address the heterogeneity of unobserved group and time specific effects,
which in the classical ordinary least squares approach, are subsumed in the error term.
In the fixed effects model, the constant term is adjusted for each group and each time
period such that the regression model becomes '
it i t kit
y x ? ? ? ? ? = + + + + . The first term
on the right-hand side of the equation is the constant term ( ) ? , and the second term
represents the sum of the products of the regressors ( ) x and their respective
coefficients ( ) ? . The third and fourth terms are the group ( )
i
? and time ( )
t
? fixed
effects which effectively adjust the constant term for group and time specific effects. The
last term in the model is the individual error term associated with the kth observation in
group i in time period t ( )
kit
? .
31
The reader is referred to any econometrics textbook for a detailed discussion of the econometric issues
revolving around generalized least squares models and the choice between fixed and random effects
specifications. The overview provided here is based on Greene (2003).
63
The random effects model proposes a different specification of the error term in the
econometric model. In this case, the unobserved individual heterogeneity is assumed
independent of the regressors ( ) x , and the group and time specific adjustments to the
constant term are assumed to be randomly distributed across cross-sectional units and
time. The benefit of the random effects procedure relative to the fixed effects procedure
lies in the preservation of a significant number of degrees of freedom since only two
random variables are needed (random group and time effects) rather than an exhaustive
set of group and time specific dummy variables. If, however, the group and time effects
are correlated with the regressors, the random effects procedure may produce inconsistent
estimates.
To decide whether it is appropriate to use the fixed effects or random effects procedure,
the Hausman specification test (Hausman 1978) is used to test for orthogonality between
the regressors and the random effects. If the null hypothesis of no correlation cannot be
rejected, the random effects model is both consistent and efficient and preferred over the
fixed effects model which, in this case, is inefficient. If, however, the null hypothesis is
rejected, only the fixed effects model is consistent and thus preferred over the random
effects model. The implementation of the Hausman specification test requires the
estimation of the model (Equation 1 and Equation 2) using the fixed effects and random
effects procedures, respectively. The test statistic W is based on the covariance matrix ?
of the difference vector of the respective coefficients [ ] b ? ? and is given by
[ ] [ ]
1
' W b b ? ? ?
?
= ? ? (Hausman 1978). The test produces a
2
? distributed statistic of
64
W = 995.43 which is significant at the less than one percent level. The null hypothesis of
no correlation is therefore clearly rejected, suggesting that the fixed effects model should
be selected.
As noted above, the fixed effects model has the disadvantage of consuming a large
number of degrees of freedom due to the inclusion of group and time specific dummy
variables in the regression analysis. Greene (2003) and Yaffee (2003) therefore suggest
carefully evaluating the benefits of the fixed effects G2SLS procedure relative to the
standard 2SLS procedure. The F test of joint significance of fixed effects (Greene 2003)
evaluates the contribution of the fixed group and time effects to the fit of the model. To
that end, two regression analyses must be performed: The baseline regression which does
not include any fixed effects, and the fixed effects regression. The improvement in the fit
of the model which is achieved by adding fixed effects is measured by the following F
statistic:
( ) ( )
( ) ( )
2 2
2
1
1
fixed effects no effects
fixed effects
no effects
R R n
F
R nT n k
? ?
=
? ? ?
(Greene 2003, Yaffee 2003), where n
is the number of groups, nT is the number of observations, and k is the number of
regressors.
In this study, the cross-section is defined by route-carrier combinations (a total of 4,508
groups), and there are eight distinct time periods (two years with four quarters each). This
implies that 4,514 dummy variables must be added to the baseline 2SLS regression
65
equation
32
. The statistical software package used for this research (Intercooled STATA
8.2) does not support such an operation due to the software’s insufficient matrix size. For
the purpose of this test, the number of dummy variables is therefore reduced by
estimating fixed carrier effects only as opposed to fixed route-carrier effects, thereby
reducing the number of cross-sectional indicators from 4,508 to 30. By constraining
seasonal (quarterly) effects to be constant over time (in 1992 and 2002), the number of
time indicators is reduced to four as specified in Equation 1 and Equation 2. This test is a
highly conservative approximation of the full fixed effects test with 4,514 fixed effects
and therefore provides a lower bound for the joint significance of the fixed effects
33
. The
baseline model (see Appendix 3) yields an R
2
of 0.376, while the reduced fixed effects
model (see Appendix 4) yields an R
2
of 0.512. The resulting F statistic is F = 194.22
which is significant at the less than one percent level. It is therefore concluded that the
fixed effects generalized two-stage least squares procedure is the most appropriate data
analysis technique.
2.4. Empirical results and discussion
The regression results are discussed in two stages: The first-stage regression, in which
AirlinePass is the dependent variable, is discussed, before the second-stage regression
results are presented. The second-stage regression uses Fare as the dependent variable
and tests the hypotheses set forth in this essay.
32
Note that this must be done manually to ensure consistency of the R
2
computation (the computation of
the R
2
statistic differs between the 2SLS and [fixed effect] G2SLS).
33
The breakdown of the 30 carrier indicator variables to 4,508 route-carrier indicator variables will
necessarily result in an increased R
2
statistic.
66
2.4.1. First-stage regression
In the first-stage regression, all independent variables (which are assumed exogenous)
and the Population and Income instruments are used to estimate exogenously determined
fitted values for AirlinePass. Table 3 presents the coefficient estimates of the first stage
regression as specified in Equation 1.
Most of the results displayed in Table 3 are in accordance with prior theoretical reasoning
and empirical research: The relationship between the number of passengers and route
distance is nonlinear as evidenced by the negative coefficient of the Distance variable
and the positive sign of the DistanceSquared coefficient. This suggests that the number of
passengers increases with the distance flown at a rate which increases in route length.
The positive coefficient of the SlotRoute dummy variable is indicative of congestion and
higher passenger volumes on slot-controlled routes. Moreover, greater firm market shares
at the route and airport market levels (RouteShare and MaxAirportShare) imply greater
numbers of passengers. Holding market shares constant, an increase in market
concentration (RouteHHI, MaxAirportHHI) then results in lower passenger numbers (see
also Ravenscraft 1983). Competition in adjacent route markets (AltRouteLCC1M) has a
slight negative effect on the number of passengers, while more circuitous routings
(Circuity) exhibit significantly decreased passenger numbers. Higher load factors
(LoadFactor) are, of course, associated with more passengers, and higher operating costs
(AirlineCost), presumably implying higher prices, negatively affect demand. The
67
coefficient of the Size variable is statistically insignificant, indicating that firm size per se
does not influence passenger demand. The positive coefficients of the quarter dummies
indicate seasonal effects (Quarter2-4), while the time trend variable (2002) carries a
negative, though statistically insignificant coefficient, thus hinting at the downturn in the
airline industry in 2002. The first instrumental variable, Population, carries a positive
coefficient indicating that passenger numbers increase as the potential market volume
increases. The coefficient of the Income variable is statistically insignificant which may
be attributed to its limited variability.
There are three variables with unexpected signs: First, the LCCCompForHCC and
LCCCompForLCC variables both have positive coefficients, indicating that the presence
of a low-cost competitor increases passenger demand for the focal carrier. This result is
most likely due to the focal airline lowering its prices as it faces aggressive competition.
These lower prices then translate into higher passenger demand. The Distress variable
carries a positive and statistically insignificant coefficient which suggests a firm’s
financial distress does not impact passenger demand. A potential explanation may be that
distressed carriers mitigate potentially negative demand effects by charging lower prices
or that passengers have few or no alternative carrier choices.
In summary, it is noted that the first-stage model is highly significant (F = 1,038.3,
significant at the less than one percent level), and that most independent variables are at
least marginally significant with most coefficients having the expected signs. Appendix 5
presents the first stage regression results for all five specifications of the model.
68
Dependent variable:
AirlinePass Coefficient P>|z|
Constant 444.458 0.000
Distance -136.296 0.000
DistanceSquared 10.181 0.000
SlotRoute 0.099 0.001
RouteHHI -0.412 0.000
MaxAirportHHI -0.367 0.000
RouteShare 0.027 0.000
MaxAirportShare 0.001 0.007
LCCCompForHCC 0.227 0.000
LCCCompForLCC 0.238 0.000
AltRouteLCC1M -0.022 0.061
Circuity -2.326 0.000
Distress 0.005 0.190
Loadfactor 0.016 0.000
AirlineCost -0.095 0.023
Size 0.028 0.150
Quarter 2 0.048 0.000
Quarter 3 0.074 0.000
Quarter 4 0.047 0.000
2002 -0.055 0.315
Population 0.579 0.000
Income -0.034 0.785
F 1038.3 0.000
R-squared (within) 0.541
Table 3: First stage G2SLS regression estimates (n = 23,039)
34
2.4.2. Second-stage regression
In this section, the results from five different second-stage regression analyses are
reported. The first and second second-stage analyses test Hypothesis 1 by including the
Distress and Chpt11Ops variables, respectively. The third regression tests the differential
34
The carrier fixed effects are omitted in this table.
69
effect of financial distress over time (Hypothesis 2) by estimating the model with the
Pre4Chpt11 and Post4Chpt11 variables. Hypothesis 1 and Hypothesis 2 are tested
separately to avoid confounding the results by including both the Distress and Chpt11
variables in a single regression (since all bankrupt airlines have high Distress scores).
The fourth second-stage regression model tests the moderating effects of firm costs, firm
market shares, and market concentration (Hypothesis 3 to Hypothesis 6). These
interactions are not included in model 1 to allow for a direct interpretation of the direct
effect of the Distress variable in model 1. As noted by Aiken and West (1991), when
interaction effects are present, a variable’s direct effect cannot be assessed by interpreting
the variable’s coefficient only, but it must be evaluated in conjunction with all its
interactions. The fifth and final model tests the importance of a firm’s relative financial
distress as discussed in Hypothesis 7. Similar to the argumentation above, the
DistressDiff variable is tested separately to avoid confounding the effects of absolute
(Distress) and relative (DistressDiff) financial distress. The second-stage regression
results are presented in Table 4.
Before focusing on the variables of interest in the respective regressions, it is noted that
all second-stage models are highly significant (Wald ?
2
? 23,900,000). Caution must be
used, however, when interpreting the R-squared statistics
35
. In the generalized least
squares (GLS) procedure, the total sums of squares are not broken down as in the
ordinary least squares procedure. The GLS R-squared, therefore, is not bounded between
zero and one and cannot be interpreted as the percentage of variability explained. In
35
The information on the use and meaning of R-squared statistics in GLS regressions was obtained from
the STATA manuals and the STATA website at www.stata.com.
70
addition, there are two sources of variation: within variation and between variation. When
fixed effects models (i.e. within estimators) are used, only the within R-squared should be
used
36
. The R squared for within variation indicates to what extent the model is able to
predict a new observation on one of the subjects already in the study. The R squared for
total variation indicates the quality of predictions relating to a new observation on a new
subject. While all R-squared (within, between, overall) statistics are reported, the reader’s
attention is directed toward the within R-squared measures which range between 0.741
and 0.783 as reported in Table 4.
36
This statistic is obtained by fitting a mean-deviated regression model where all the group effects are
assumed to be fixed. These group effects are subtracted out of the model and no attempt is made to quantify
their overall effect on the fit of the model.
71
Second-stage G2SLS regression Number of obs. 23039 Obs. per group: min. 1
(fixed effects) Number of groups 4508 avg. 5.1
max. 8
Dependent variable:
Fare Coefficient P>|z| Coefficient P>|z| Coefficient P>|z| Coefficient P>|z| Coefficient P>|z|
Constant -231.741 0.000 -237.998 0.000 -247.393 0.000 -245.434 0.000 -268.520 0.000
AirlinePass (fitted) -0.093 0.000 -0.076 0.006 -0.042 0.138 -0.141 0.000 -0.022 0.441
Distance 69.277 0.000 70.705 0.000 73.037 0.000 74.364 0.000 78.903 0.000
DistanceSquared -5.006 0.000 -5.108 0.000 -5.268 0.000 -5.420 0.000 -5.671 0.000
SlotRoute 0.090 0.000 0.096 0.000 0.083 0.000 0.083 0.000 0.088 0.000
RouteHHI -0.009 0.495 -0.003 0.795 0.008 0.553 -0.008 0.518 0.021 0.130
MaxAirportHHI 0.055 0.000 0.067 0.000 0.091 0.000 0.030 0.019 0.105 0.000
RouteShare 0.002 0.010 0.001 0.053 0.001 0.437 0.003 0.000 0.000 0.994
MaxAirportShare 0.002 0.000 0.002 0.000 0.002 0.000 0.002 0.000 0.002 0.000
LCCCompForHCC -0.110 0.000 -0.118 0.000 -0.130 0.000 -0.091 0.000 -0.132 0.000
LCCCompForLCC -0.024 0.034 -0.014 0.242 -0.001 0.914 -0.011 0.311 -0.005 0.684
AltRouteLCC1M -0.027 0.000 -0.026 0.000 -0.027 0.000 -0.021 0.000 -0.028 0.000
Circuity -0.443 0.000 -0.402 0.000 -0.320 0.000 -0.536 0.000 -0.269 0.001
Distress -0.036 0.000 0.496 0.000
Chpt11Ops -0.072 0.000
DistressDiff -0.009 0.000
Pre4Chpt11 -0.007 0.179
Post4Chpt11 -0.042 0.000
Loadfactor -0.015 0.000 -0.014 0.000 -0.014 0.000 -0.013 0.000 -0.015 0.000
AirlineCost 0.224 0.000 0.210 0.000 0.235 0.000 0.223 0.000 0.226 0.000
Size 0.055 0.000 0.119 0.000 0.152 0.000 -0.018 0.061 0.130 0.000
Quarter 2 -0.031 0.000 -0.040 0.000 -0.044 0.000 -0.034 0.000 -0.045 0.000
Quarter 3 -0.026 0.000 -0.034 0.000 -0.048 0.000 -0.028 0.000 -0.044 0.000
Quarter 4 -0.034 0.000 -0.046 0.000 -0.054 0.000 -0.027 0.000 -0.053 0.000
2002 -0.257 0.000 -0.306 0.000 -0.332 0.000 -0.208 0.000 -0.314 0.000
AirlineCost*Distress 0.062 0.000
Size*Distress -0.013 0.000
RouteShare*Distress 0.0002 0.000
RouteHHI*Distress -0.023 0.000
Wald ?
2
26,800,000 25,900,000 24,500,000 28,500,000 23,900,000
Prob > ?
2
0.000 0.000 0.000 0.000 0.000
R-squared: within 0.769 0.761 0.748 0.783 0.741
between 0.080 0.084 0.092 0.041 0.101
overall 0.087 0.091 0.097 0.049 0.104
5 2 4 1 3
Table 4: Second-stage G2SLS regression estimates
72
Turning to the control variables first, it is noted that most coefficient estimates are
consistent across all five second-stage models and are statistically significant at the less
than one percent level: Prices are shown to increase with Distance, but at a decreasing
rate, as evidenced by the negative coefficient of DistanceSquared. As expected, fares
tend to be higher in route markets involving one or two slot-controlled airports
(SlotRoute), and both airport market concentration (MaxAirportHHI) and airport market
shares (MaxAirportShare) are associated with higher fares, ceteris paribus. The presence
of low-cost carrier competition has a strong negative effect on a high cost carrier’s prices
(LCCCompForHCC), as does the presence of low-cost carriers in adjacent route markets
(AltRouteLCC1M). Prices for less convenient connecting traffic are shown to be lower
than for direct service (Circuity), and higher load factors (LoadFactor) – indicative of
economies of density – also tend to result in lower fares. An airline’s operating costs
(AirlineCost) and size (Size), finally, are both shown to positively impact air fares. The
time variables capture both seasonal price fluctuations (Quarter2-4) as well as a clearly
negative time trend (2002).
The following variables have either unexpected or statistically insignificant coefficients:
While the coefficient of the AirlinePass variable is negative as expected in all instances,
it is statistically insignificant in models 3 and 5. There is, nonetheless, at least some
evidence that higher passenger numbers – implying economies of density – result in
lower prices, all else equal. Note that the coefficients of the RouteShare variable, while
positive as expected, are also statistically insignificant in models 3 and 5. The two
variables (AirlinePass and RouteShare) are highly correlated as expected (? = 0.57, see
73
Table 1) with RouteShare being the ratio of AirlinePass and the total number of
passengers in the route market. It is, therefore, likely that multicollinearity between right-
hand side variables cause some degree of variance inflation. The RouteHHI variable
carries a statistically insignificant coefficient in all model specifications, suggesting that
route market concentration does not have a direct effect on prices. Also, the presence of
LCC competitors does not appear to impact other low-cost carriers’ prices as indicated by
the insignificant coefficient estimates of the LCCCompForLCC variable in models 2-5.
Only in the baseline model (1) can the expected negative effect be observed.
The attention is now directed to the variables of interest that test the hypotheses set forth
in this paper.
The negative and significant coefficient of the Distress variable in the first second-stage
regression (? = -0.036, p = 0.000) provides clear support for the contention that greater
levels of financial distress result in lower prices, ceteris paribus (Hypothesis 1). This
result thus confirms the basic finding in the extant literature that financially distressed
firms behave more aggressively in the output market. More specifically, this result
suggests that the reduction of a firm’s Distress score by one unit leads to a price
reduction of 3.6 percent, all else held constant. The second regression presents an
alternative test of Hypothesis 1 using the ChptOps variable. The latter carries a
statistically significant coefficient of -0.072 (p = 0.000) which implies that, on average,
airlines operating under Chapter 11 protection charge about seven percent less than their
non-bankrupt competitors, ceteris paribus. This finding is consistent with the result of the
74
Distress variable and clearly in support of Hypothesis 1.
The third regression presented in Table 4 tests the differential impact of financial distress
prior to and after Chapter 11 filings. The coefficient of the Pre4Chpt11, while negative,
is statistically insignificant (? = -0.007, p = 0.170) which suggests that there are no
significant price changes as an airline approaches bankruptcy. The Post4Chpt11 variable,
however, carries a negative and statistically significant coefficient (? = -0.042,
p = 0.000). This indicates that airlines tend to lower prices upon declaring bankruptcy and
that the effect of firm financial distress on prices is substantially larger (-4.2%) once the
airline operates under bankruptcy protection. This finding supports the contention that
passengers may be reluctant to choose bankrupt carriers given the uncertainty about its
reliability and future operations. This may entice such firms to cut prices in an effort to
stimulate or maintain passenger demand. Moreover, bankrupt carriers may simply pass
some of the cost savings that result from operating under bankruptcy protection
37
on to
consumers. Hypothesis 2 is thus supported.
The fourth column in Table 4 presents a test of the hypothesized interaction effects
(Hypothesis 3 to Hypothesis 6). Hypothesis 3 argues that a firm’s operating costs
positively moderate the relationship between financial distress and prices, meaning that
the effect of firm financial distress on prices will be of lesser magnitude for high-cost
firms than for lower-cost firms (see Figure 3). The rationale for this contention is that
low-cost firms likely have higher profit margins and can more easily (and profitably)
37
Due to paused leasing and debt payments, for example.
75
afford price cuts than high-cost firms. The strategic management literature further argues
that operating costs are a good proxy for a firm’s strategic type: Low-cost firms are often
referred to as prospectors, and high-cost firms have been termed defenders. Prior
research has shown that prospectors tend to act more aggressively (in terms of prices, for
example), while defenders tend to behave more conservatively and focus on internally-
oriented rather than market-oriented actions which involve price and product changes.
The coefficient of the interaction term AirlineCost*Distress is positive and statistically
significant at the less than one percent level (? = 0.062, p = 0.000). As discussed above,
the effect of financial distress on prices is generally negative, implying that distressed
firms sell at lower prices, all else equal. The interaction with operating costs
(AirlineCost) then adds a positive term to the distressed firm’s price, where the value of
this addition increases in the firm’s operating costs. The analyses, thus, present some
evidence for the contention that distressed firms will tend to refrain from competing on
price when their operating costs are higher, as suggested in Hypothesis 3.
It has been suggested in Hypothesis 4 that firm size will increase a distressed firm’s
tendency to compete on price (see Figure 4). More specifically, it has been argued that
larger firms benefit from greater reputation, creditor trust and resource availability which
increase their survivability. Consequently, it is expected that larger distressed firms
leverage their size advantage and do not avoid price competition to the extent smaller,
more fragile airlines do: Larger firms can afford the detrimental short-term effects of
price cuts and may pursue such aggressive pricing strategies in an effort to eliminate
smaller competitors and thus enhance their long-term profitability prospects. The
76
interaction term between Distress and Size is negative and statistically significant (? = -
0.013, p = 0.000). This result thus implies that larger distressed firms will price more
aggressively than smaller distressed firms, all else equal. Consequently, Hypothesis 4 is
supported.
In Hypothesis 5, it was argued that the impact of firm financial distress on prices is
moderated by firm market shares (see Figure 5). Firm with higher market shares may
have higher degrees of market power and therefore experience less pressure to lower
prices in the light of financial distress. In addition, for firms with high market shares, the
potential benefits of cutting prices are limited since the expected gains in terms of market
volume may not offset the losses due to lower sales prices. The coefficient of the
interaction term of the RouteShare and Distress variables is positive and significant
(? = 0.0002, p = 0.000). Higher route market shares, thus, reduce a distressed firm’s
pricing aggressiveness as stated in Hypothesis 5.
As to the moderating effect of (route) market concentration, it was hypothesized that the
interaction of financial distress and route market concentration will positively impact
prices (Hypothesis 6), ceteris paribus (see Figure 6). While high market concentration
per se may facilitate collusive price fixing among firms, deteriorations in a firm’s
financial condition and ensuing changes in that firm’s cost structure may lead to the
breakdown of collusive arrangements with competitors and greater degrees of price
competition. The interaction term of RouteHHI and Distress has a negative and
statistically significant coefficient (? = -0.023, p = 0.000). As stated in Hypothesis 6, this
77
implies that greater levels of distress and market concentration increase a heavily
troubled firm’s tendency to compete aggressively and sell at lower prices (after
controlling for the moderating effects of route market shares).
Hypothesis 7 suggests that the difference between a focal firm’s Distress score and that of
its (route market) competitors affects the focal firm’s prices. This hypothesis is motivated
by the fact that firms that are in similar financial conditions may be expected to behave
symmetrically. In this case, no single firm would benefit from price reductions and
reinforced price competition. It is, therefore, expected that a focal firms pricing actions
will be more pronounced the greater the focal firm’s financial distress relative to its
competitors. To test Hypothesis 7 the coefficient of the DistressDiff variable from the
fourth second-stage regression can be interpreted straightforwardly. The negative and
significant coefficient (? = -0.009, p = 0.000) indicates that a firm’s financial distress
relative to its competitors negatively impacts the focal firm’s prices as stated in
Hypothesis 7
38
.
2.4.3. Second-stage regression: Sensitivity analysis
The results discussed above are based on the analysis of 1992 and 2002 data. These time
periods were chosen since the airline industry experienced substantial financial distress
during those years. To investigate the sensitivity of the results with respect to the
38
Recall that positive DistressDiff values indicate relative financial distress, while negative values indicate
relative financial wellbeing.
78
selection of the time period studied, the regression models are re-estimated using data
from 1992, 1997, and 2002. The addition of 1997 data brings the total number of
observations to 34,097. 1997 data were selected since this year is in the middle of the
1992-2002 time period. Also, the airline industry as a whole performed relatively well
during that year. It is therefore expected that the findings with respect to the effect of
financial distress on prices will be weaker when 1997 data are included in the analyses.
Nonetheless, the empirical results should be consistent with the contentions set forth in
Hypothesis 1 to Hypothesis 7.
Table 5 presents the second-stage regression results which are based on the analysis of
the extended data set (including 1997 data). It is noted that the fit of the regression
models is generally inferior compared to the results presented in Table 4 which were
based on 1992 and 2002 data only. Specifically, the R-squared within statistics shown in
Table 5 suggest that the models explain only about fifty to sixty percent of the variability
as compared to the seventy to eighty percent variability explained for the 1992 and 2002
data (see Table 4). While most variables have statistically significant coefficients with the
expected signs, the Distance and DistanceSquared variables have insignificant coefficient
estimates in all models.
The hypothesis testing results can be summarized as follows:
? Hypothesis 1: The negative coefficient of the Distress variable (? = -0.020,
p = 0.000) in the first regression, provides support for the contention that greater
levels of financial distress result in lower prices. This contention is further
79
corroborated by the negative and significant coefficient of the Chpt11Ops variable
in the second regression (? = -0.047, p = 0.000).
? Hypothesis 2: The differential effect of financial distress over time (prior to versus
during bankruptcy) is empirically examined in the third regression where the
Pre4Chpt11 and Post4Chpt11 variables are included in the model. While the
Pre4Chpt11 variable carries a statistically significant negative coefficient, the
coefficient of the Post4Chpt11 variable is statistically insignificant. This suggests
that, on average, carriers approaching bankruptcy tend to cut prices, while carriers
operating under bankruptcy protection do not cut prices. This finding is contrary
to Hypothesis 2 and inconsistent with the results shown in Table 4. It is noted that
virtually no airline bankruptcies were observed in 1997. As a result, it is not
surprising that adding 1997 data to the regression analysis weakens the robustness
of the regression results with respect to the effect of bankruptcy on prices.
? Hypothesis 3: Hypothesis 3 suggests that a firm’s operating costs positively
moderate the relationship between financial distress and prices. The positive and
significant coefficient of the AirlineCost*Distress interaction effect (? = 0.015,
p = 0.000) confirms this expectation. This finding is consistent with the results
shown in Table 4.
? Hypothesis 4: The distress-price effect was hypothesized to be stronger for larger
firms than for smaller firms. In line with the regression results reported earlier,
this hypothesis is supported even when 1997 data are included: The Size*Distress
interaction effect carries a negative and significant coefficient (? = -0.009,
p = 0.000).
80
? Hypothesis 5: The results shown in Table 5 suggest that the distress-price effect
does not change in magnitude as a firm’s route market share increases. The
RouteShare*Distress interaction effect does not yield a statistically significant
coefficient (? = 0.0000, p = 0.495), whereas this interaction effect was positive
and significant in the analysis of 1992 and 2002 data (see Table 4). Again, the
lack of a significant finding may potentially be attributed to the fact that the
addition of 1997 data tends to dilute statistical effects of financial distress since
the airline industry experienced little distress in that year.
? Hypothesis 6: The RouteHHI*Distress interaction carries the expected negative
coefficient (? = -0.008, p = 0.001), suggesting that the distress-price effect is
greater in more concentrated markets than in less concentrated markets. This
finding is consistent with Hypothesis 6 and the previously reported results (see
Table 4).
?
? Hypothesis 7: A firm’s financial distress relative to its competitors in the route
market is also shown to significantly impact prices (? = -0.003, p = 0.000).
Hypothesis 7 is, thus, supported.
In summary, five out of seven hypotheses are supported when 1997 data are included in
the analyses. The lower model fit statistics and smaller coefficient values, however,
confirm the contention that adding 1997 data—a period of relative financial health in the
airline industry—tends to dilute the results. Nonetheless, the hypothesis testing results are
shown to be largely robust.
81
Second-stage G2SLS regression Number of obs. 34097 Obs. per group: min. 1
(fixed effects) Number of groups 4798 avg. 7.1
max. 12
Dependent variable:
Fare Coefficient P>|z| Coefficient P>|z| Coefficient P>|z| Coefficient P>|z| Coefficient P>|z|
Constant -30.762 0.477 -16.919 0.707 -27.484 0.531 -40.643 0.364 -29.798 0.489
AirlinePass (fitted) -0.505 0.000 -0.544 0.000 -0.515 0.000 -0.550 0.000 -0.493 0.000
Distance 13.024 0.318 8.645 0.525 11.527 0.384 16.686 0.217 12.118 0.351
DistanceSquared -1.025 0.295 -0.690 0.498 -0.895 0.368 -1.315 0.195 -0.936 0.338
SlotRoute 0.188 0.000 0.194 0.000 0.179 0.000 0.183 0.000 0.188 0.000
RouteHHI -0.161 0.000 -0.177 0.000 -0.168 0.000 -0.172 0.000 -0.157 0.000
MaxAirportHHI -0.079 0.000 -0.087 0.000 -0.073 0.000 -0.105 0.000 -0.067 0.000
RouteShare 0.013 0.000 0.014 0.000 0.013 0.000 0.014 0.000 0.013 0.000
MaxAirportShare 0.003 0.000 0.003 0.000 0.003 0.000 0.003 0.000 0.003 0.000
LCCCompForHCC -0.054 0.000 -0.047 0.000 -0.055 0.000 -0.036 0.000 -0.060 0.000
LCCCompForLCC 0.073 0.000 0.087 0.000 0.090 0.000 0.078 0.000 0.084 0.000
AltRouteLCC1M -0.043 0.000 -0.043 0.000 -0.044 0.000 -0.039 0.000 -0.045 0.000
Circuity -1.266 0.000 -1.359 0.000 -1.285 0.000 -1.373 0.000 -1.231 0.000
Distress -0.020 0.000 0.194 0.000
Chpt11Ops -0.047 0.000
DistressDiff -0.003 0.000
Pre4Chpt11 -0.041 0.000
Post4Chpt11 -0.005 0.222
Loadfactor -0.011 0.000 -0.009 0.000 -0.009 0.000 -0.011 0.000 -0.010 0.000
AirlineCost 0.062 0.000 0.048 0.000 0.056 0.000 0.055 0.000 0.052 0.000
Size 0.049 0.000 0.086 0.000 0.098 0.000 -0.010 0.225 0.094 0.000
Quarter 2 0.005 0.126 0.001 0.796 -0.002 0.582 0.006 0.047 -0.002 0.483
Quarter 3 0.027 0.000 0.024 0.000 0.017 0.000 0.029 0.000 0.020 0.000
Quarter 4 -0.018 0.000 -0.018 0.000 -0.025 0.000 -0.010 0.001 -0.023 0.000
2002 -0.157 0.000 -0.183 0.000 -0.188 0.000 -0.115 0.000 -0.192 0.000
AirlineCost*Distress 0.015 0.000
Size*Distress -0.009 0.000
RouteShare*Distress 0.0000 0.495
RouteHHI*Distress -0.008 0.001
Wald ?
2
25,300,000 23,300,000 24,600,000 23,600,000 25,500,000
Prob > ?
2
0.000 0.000 0.000 0.000 0.000
R-squared: within 0.565 0.528 0.552 0.533 0.569
between 0.005 0.001 0.007 0.033 0.010
overall 0.001 0.006 0.015 0.019 0.019
5 2 4 1 3
Table 5: Second-stage G2SLS regression estimates using 1992, 1997, and 2002 data
82
2.5. Summary and discussion
The study’s results are summarized in Table 6 below. Ample support for the theoretical
arguments set forth in this paper is found. The implications of these findings are
discussed in this section, and some limitations and directions for future research are
noted.
The primary objective of this research is to reconcile the extant theoretical conflict
revolving around the impact of firm financial distress on prices. Based on a review of
varied theoretical perspectives and numerous empirical studies, it is suggested that
financial distress is negatively related to prices. It is noted, however, that this may not be
true in all cases. More specifically, it is hypothesized that operating costs, firm size and
market shares, as well as market concentration and a firm’s financial standing relative to
its competitors may impact the magnitude of a troubled firm’s pricing actions. A strategic
contingency framework which incorporates these moderating effects is developed and
tested using a comprehensive panel dataset from the U.S. airline industry.
The empirical results provide clear statistical support for all hypotheses: Firm financial
distress negatively impacts prices, and it is shown that these price effects are greatest for
carriers that operate under bankruptcy protection. The empirical results further suggest
that this is particularly true for firms with lower operating costs and smaller market
shares, and for firms operating in highly concentrated markets. The difference between a
83
focal firm’s financial distress and that of its competitors is also shown to impact the
magnitude of airlines’ pricing actions. All hypothesized direct and moderating effects are
thus supported.
Hypothesized effect
on prices
H
y
p
o
t
h
e
s
i
s
Testing variable Direct
Interaction
w/ Distress
variable Finding
Empirical
support for
hypothesis?
1 Distress – – Yes
2
(Post4Chpt11 –
Pre4Chpt11)
< 0
< 0 Yes
3 AirlineCost + + Yes
4 Size – – Yes
5 RouteShare + + Yes
6 RouteHHI
– – Yes
7 DistressDiff – – Yes
Table 6: Summary of results
This study’s results suggest that passengers traveling on distressed or bankrupt carriers
pay nearly four percent less than other passengers, all else equal. This is, of course, a
desirable outcome from a consumer perspective as a distressed firm’s lower prices
implies increases in consumer welfare. Since bankruptcy only sometimes results in a
firm’s liquidation, there is no indication for longer term negative effects of Chapter 11
protection on consumer welfare through, for example, reduced competition or reduced
service levels.
84
For managers and policy makers, however, this finding may be troubling. Financial
distress appears to negatively affect a firm’s revenue streams by virtue of lower prices
and, in some instances, lower demand
39
and may ultimately reduce the firm’s profitability
(see also Kennedy 2000). Taken together, these findings raise questions about the
rationality of a distressed firm’s pricing behavior and the adequacy of Chapter 11
protection. The results suggest that financial distress is both at the beginning and at the
end of a vicious circle of literally destructive price competition (see also Moulton and
Thomas 1993 for a discussion of the success rates of reorganizations under bankruptcy).
Managers and policy makers try to avoid organizational failure by offering lower prices
and supporting reorganization efforts respectively, but the very opposite effect may be
observed in at least some instances: Financial distress, and Chapter 11 protection in
particular, lead to an increase in the competitive pressures, thus increasing the firm’s
distress and spreading it beyond the firm’s boundaries. While it is not an objective of this
research to make any managerial or public policy prescriptions, the findings presented in
this study may be useful in gaining a greater understanding of the effects of financial
distress on prices by considering the moderating effects of firm and market
characteristics.
The key message of this study is clear: Microeconomic and corporate finance theory
alone cannot fully explain the relationship between a firm’s capital structure and its
output market behavior. The diversity of firms and circumstantial characteristics have to
be considered when investigating the effect of financial distress on prices. Strategic
39
See Table 38: the negative coefficient of the Post4Chpt11 variable implies that bankrupt firms face lower
demand, all else equal.
85
management research offers an array of theoretical approaches to further explore this
issue, and a contingency framework appears to be an appropriate means to do so. In that
vein, the hypotheses reflect and the results present evidence for elements of prospect
theory, organizational learning theory, and strategic groups research, for example. The
author is unaware of any other research that has examined the research question at hand
from a strategic management perspective. By combining multiple theoretical perspectives
and incorporating them in a single, comprehensive contingency framework, the
understanding of the link between firm financial distress and prices is advanced.
Data from the U.S. airline industry are used for the empirical analyses. While this
selection has many desirable qualities in terms of the detail and availability of data, one
must consider the possibility that these findings may not be generalizable to other
industries. The U.S. airline business is particularly competitive and, to some extent, still
marked by the era of regulation
40
. The exploration of the effects of financial distress on
prices in a cross-section of industries is left for future research. Moreover, the DOT
airline data do not contain any information about booking and service classes. As noted
by Lee and Luengo-Prado (2005), the failure to recognize these distinctive attributes of
the tickets purchased is a potentially critical shortfall of any empirical analysis of air
fares.
Research of the impact of financial distress faces a general dilemma: While financial
distress is a firm-level phenomenon, prices are clearly market-specific. In this research,
40
Regulation by the Civil Aeronautics Board ended in 1978, but has shaped the industry in many ways.
Although formally deregulated, regulatory controls (e.g. slot controls, antitrust rulings) continue to impact
the industry.
86
the impact of firm-level financial distress on individual product market prices is
investigated. This approach presents some challenges in that it is more difficult to isolate
statistical effects, and it may be desirable to investigate this research question in the
context of single-market firms. The latter are, however, hard to find nowadays.
On a final note, it should be stated that this study’s results may also depend upon the
measurement of financial distress. This study employed distress measures based on Z
scores and Chapter 11 dummy variables given that they have been widely applied in the
extant literature. The finance literature offers numerous variations of these measures as
well as entirely different ones (see e.g. Gritta 2004 for a comprehensive review of some
of these measures). Future research may explore the sensitivity of the results with respect
the measurement of financial distress.
This research contributes to the literature on the link between firm financial distress and
output market behavior. It is shown that this issue is far from being fully understood and
that strategic management theory offers avenues for further exploration of the impact of
financial distress on prices. This study has made a first step in this direction by estimating
the moderating effect of a number of strategic contingencies on this relationship.
87
3. The effect of firm financial distress on firm inventories: A supply chain
perspective
In this chapter, the effects of financial distress on inventory holdings are discussed and
tested empirically. The structure of this chapter is similar to that of Chapter 2 of this
dissertation: The subject matter of this essay is introduced in Section 3.1. The latter
includes a statement of the research questions and contributions of this research. Section
3.2 presents a review of theories and prior research on the relationship between firm
financial condition and inventories, and a baseline hypothesis is formulated. A supply
chain perspective is discussed in Section 3.3. It is hypothesized that firm power not only
directly impacts firm inventories, but also moderates the effect of financial distress on
inventories. Details about the data sample and the empirical methodology are provided in
Section 3.4. The empirical results are presented and discussed in Section 3.5, and the
study’s findings are summarized in Section 3.6. Managerial implications are discussed,
and suggestions for future research are provided, while the study’s limitations are noted.
3.1. Introduction
Financial considerations play an important role in inventory decision making. The survey
results presented by Osteryoung et al (1986), for example, indicate that 73.5% of all
respondents consider the firm’s cash position, and 57.3% factor in anticipated changes in
interest rates when making inventory decisions. It is intuitively appealing to assume that
firms under financial distress will shed inventories to generate liquidity. For American
88
car manufacturer Chrysler Corp., for example, reducing inventories was a major
component of its turnaround efforts (Stundza and Milligan 2001). Case Corporation, a
U.S. manufacturer of construction and agricultural equipment, also drastically cut
inventories when it restructured its business in the early 1990s (Buxbaum 1995).
While anecdotal evidence suggests that declining firm financial condition implies lower
inventory levels, prior empirical research on this relationship has produced ambiguous
results (see e.g. Corbett et al. 1999, Guariglia 1999). Upon closer examination of
previously published work, which relies exclusively on finance and economic theory, it
becomes clear that the link between firm finances and inventories is not yet well
understood, both in theoretical and empirical terms. Insights from inventory theory and
supply chain research will be useful to better understand this relationship and to improve
upon the specification of empirical estimation models.
A supply chain perspective on the link between financial distress and inventories is of
particular interest in this research. More specifically, this research is concerned with the
effect of a (distressed) firm’s power relative to buyers and suppliers on the firm’s
inventory decisions. In other words, can a distressed firm with greater levels of power
push greater amounts of inventory onto suppliers and buyers? If so, firms may want to
pay more attention to the financial condition of potential supply chain partners and be
aware of the potentially adverse impact of distressed firms’ inventory decisions. The
interplay between financial distress, supply chain power, and inventories remains
unexplored. The following paragraphs summarize the state of knowledge in this area and
89
outline the agenda of this research.
A sizeable literature in the economics and finance fields deals with the effects of financial
parameters on inventories. Some researchers have taken a rather macroeconomic
approach, analyzing the impact of monetary policy on aggregate inventory levels across
industries (see e.g. Corbett et al. 1999). Another set of research papers has investigated
the relationship between firm financial parameters such as bank lending rates or cash
flows and firm inventories (see e.g. Carpenter et al. 1998, Gertler and Gilchrist 1994).
While approaching the phenomenon from different theoretical and methodological
angles, many researchers contend that unfavorable financial conditions are associated
with lower inventory levels across an economy and within firms. The empirical findings,
however, provide only partial support for the researchers’ contentions. Corbett et al
(1999), for example, find that interest rates are a significant predictor of inventory levels
in certain industries only. Similarly, the results presented by Gertler and Gilchrist (1994)
suggest that the coverage ratio, i.e. the ratio of a firm’s cash flow and short term interest
expenses, explains the inventory behavior of small firms but not that of large firms. A
study by Guariglia (1999), finally, establishes a significant relationship between firm
finances and firm inventories during recessionary periods only.
The inconsistency of prior findings may, in part, be explained by differences in variable
measurement, the composition of data samples, estimation techniques, and perhaps most
importantly, variations in model specification. As the relationship between firm financial
factors and firm inventories appears to be more complex than previously assumed,
90
important explanatory variables may have been omitted in past research. In fact, most
published articles in this area rely exclusively on corporate finance and economic theory.
As pointed out by Roumiantsev and Netessine (2007), authors thereby ignore the insights
provided by inventory theory. Classical inventory models suggest that firm inventories
are a function of factors such as average demand, average lead times, holding costs,
demand and lead time variability, for example. Out of these factors, only demand has
been incorporated in the models of the articles referenced in the previous paragraph.
Potential specification problems encountered in prior research may therefore be alleviated
by drawing on inventory theory to a greater extent than has been done before.
Another shortcoming of the extant literature relating firm financial factors to inventories
may be the myopic treatment of inventories as firm decision parameters and the neglect
of the supply chain context in which most firms operate. While a firm’s managers
ultimately decide on the amount of inventory they order and sell, firms typically operate
within the confines of the terms and conditions negotiated with buyers and suppliers.
Some firms, for example, commit to specified service levels and must hold more
inventory to meet these performance targets. In other instances, buyers and sellers closely
cooperate by implementing Vendor-Managed Inventory (VMI) programs, for example.
Under this regime, a firm physically holds inventory that is managed (and possibly
owned) by a supplier until items are used in production or sold. Regardless of the
inventory policy in place, a firm’s bargaining power relative to its buyers and suppliers
will significantly impact the extent to which the firm exerts control over its inventories
(Wallin et al. 2006). Supply chain considerations, thus, may have a substantial impact on
91
a firm’s inventory holdings and on the degree to which a firm’s financial distress affects
inventories. This research adds to prior work in the firm finance-inventory area by
drawing on the supply chain power literature and incorporating associated measures in
the empirical analysis.
In summarizing, this essay theoretically and empirically revisits the link between firm
financial distress and firm inventories. An objective of this research thereby is to gain a
refined understanding of why a firm’s financial situation may have an impact on
inventories. This relationship is also tested empirically. Particular attention is paid to the
specification of the regression equation using not only microeconomic theory, but also
inventory theory, and insights from supply chain research. Also investigated is how the
nature of supply chain relationships, i.e. inter-firm power (im)balances, impact the extent
to which firms can reduce inventory holdings when experiencing financial distress. The
following research questions, thus, emerge:
1. Does a firm’s financial situation have an impact on its inventories after controlling for
other relevant parameters prescribed by inventory theory and supply chain research?
2. How does a firm’s (supply chain) power impact its inventory holdings?
3. Is the magnitude of the presumed effect of financial distress on inventories impacted
by power (im)balances in supply chain relationships?
The contributions of this research are manifold. First, it is shown that firm inventories
should respond to changes in firm financial condition. Prior research has not provided
such theoretical rationales. This is, to the best of the author’s knowledge, the first attempt
92
to investigate the research question at hand from both a microeconomic and an inventory
theory perspective, thus providing a broader, more complete theoretical basis for the
empirical analyses. Second, the impact of supply chain relationship variables on
inventory management is investigated. To date, few researchers have empirically
analyzed how the nature of buyer-supplier power balances impact firm inventory levels
41
.
In this essay, the role of firm power and concentration in both the upstream (supplier) and
downstream (buyer) markets in explaining focal firm inventories is examined. Moreover,
the analysis of the relationship between firm financial distress and inventories is extended
beyond the boundaries of the firm and is approached from a supply chain perspective,
thus more appropriately capturing the external influences on firms’ (inventory) decisions
(Cox et al. 2003, Dobson 2005). More specifically, it is argued that a firm’s buying and
selling power moderates the distress-inventory relationship. This contingency framework
may help reconcile prior findings by defining when and under what conditions the effect
of firm distress on inventories is greatest. This research thus adds to both the inventory
and supply chain literatures by analyzing the relationships between firm financial
distress, supply chain power, and firm inventories.
Besides its academic theoretical appeal, this research also has potentially important
managerial implications for supplier selection. If, for example, distressed firms are shown
to use their power to push inventory ownership to buyers or suppliers, firms may want to
carefully evaluate a potential partner firm’s financial condition and determine how the
41
Amihud and Mendelson (1989) study how firm market power affects firm inventory. They do not,
however, consider a firm’s power over suppliers or market concentration measures. Blazenko and
Vandezande (2003), in turn, investigate the relationship between market concentration and inventories only
and also ignore characteristics of the upstream supplier market.
93
partner’s distress might affect inventory ownership in the supply chain. In addition,
Halley and Nollet (2002) note that supplier selection and supplier development become
increasingly strategic, long-term firm decisions. An investigation of the role and impact
of financial considerations on such decisions, therefore, seems timely and managerially
relevant.
3.2. The financial distress-inventory relationship
In this section, the theoretical bases for a link between firm financial distress and
inventories are reviewed. Most prior research relied on economic theory when
investigating this relationship. This literature and the underlying theoretical rationales are
reviewed below. The second subsection discusses the firm finance-inventory link from an
inventory theory perspective. It is also suggested that inventory theory offers various
determinants of firm inventories that have not been included in prior economics research.
This section concludes with the formulation of a baseline hypothesis.
3.2.1. Economic theory
Within the economics stream of research, three articles, all first published in 1994, merit
particular attention. Gertler and Gilchrist (1994) are among the first to explore the
relationship between monetary policy (interest rates) and firm inventory levels. Kashyap
et al (1994) present a very similar study but use firm liquidity rather than security-market
interest rates as a measure of financial condition. Both papers support the lending view
94
which suggests that a firm’s dependence on external finance drives the strength of the
relationship between firm financial condition and inventories. Carpenter et al (1994),
finally, focus uniquely on the availability of internal finance as a determinant of
inventory (dis)investments and disregard macroeconomic factors such as security-market
interest rates. All three papers are discussed in more detail below.
Gertler and Gilchrist (1994) investigate the relationship between monetary policy and
firm behavior with respect to sales and inventories. The authors present two theoretical
rationales which suggest that tight monetary policy (i.e. an increase in interest rates)
negatively affects firm output and inventories. First, it is noted that rising interest rates
weaken firms’ balance sheet positions by reducing cash flows (net of interest) and
lowering the value of collateral assets. Consequently, borrowers reduce their spending
which implies output and inventory contractions. Second, monetary policy regulates the
pool of funds that is available to bank-dependent borrowers. The effect of monetary
policy on firm behavior is argued to be particularly strong for firms with limited access to
public capital markets. Both rationales, thus, suggest that monetary policy may affect
firm sales and inventories, and that firm financial factors, the access to capital markets in
particular, influence this relationship.
Gertler and Gilchrist (1994) use firm size to approximate a firm’s access to capital
markets and use industry-level time series data disaggregated by firm size classes to
estimate the effects of monetary policy on firm behavior. Descriptive analyses and the
estimation of structural inventory equations with the firm’s coverage ratio (cash flow
95
over total interest payments) as the key independent variable of interest indicate that
small firms’ sales and inventories decline more significantly during and after periods of
tight monetary policy. This result is shown to be significant and quantitatively
meaningful for small firms but not for large firms which supports the contention that tight
monetary policy particularly affects small firms with limited access to public capital
markets.
The work of Kashyap et al (1994) is closely related to that of Gertler and Gilchrist (1994)
and is motivated by the observation that there has been little empirical support for a
relationship between real interest rates and inventory investment. Yet, the observations
that inventory movements explain a substantial portion of the variability in aggregate
output, and that economic downturns typically follow periods of tight credit strongly
suggest such a relationship.
Kashyap et al (1994) attribute the lack of empirical support to measurement
imperfections. More specifically, the authors suggest that measures such as security-
market interest rates do not fully capture firm financial conditions or the cost of external
finance (e.g. bank loans). The latter, however, is argued to have a greater impact on
inventories than security-market interest rates. The authors’ key hypothesis thus states
that firms that depend on external finance should see their inventories fall more sharply
than firms with higher levels of internal funds and better access to public debt markets.
This contention is frequently referred to as the “lending view” in extant research.
96
Kashyap et al (1994) seek to empirically validate their hypothesis by regressing the
change in inventories on a set of firm-level determinants which include most notably the
inventory/sales ratio, the change in sales over the current and preceding years, and a
measure of liquidity (cash and marketable securities over total assets). A series of
different regression analyses using time series data indicate that firm liquidity is
consistently positively and significantly related to changes in inventory. This is, however,
only true for the 1974-75 and 1981-82 time periods when there were substantial liquidity
constraints. Data from 1985-86 are used as a control sample, and for this time period the
coefficient of the “liquidity” variable is statistically insignificant. In summarizing, the
authors thus conclude that financial factors influence inventory movements during tight
money (recessionary) episodes but not otherwise.
Most prior research relating inventory investments to financial parameters focuses on the
effects of monetary policy (e.g. Gertler and Gilchrist 1994) or financial factors such as
commercial paper spread and the mix of bank loans and commercial paper on firm
inventories (e.g. Kashyap et al. 1993). Carpenter, Fazzari and Petersen (1994) build on
this stream of research and add to it on two accounts: First, they focus on the flow of
internal finance as opposed to on monetary policy effects (see Gertler and Gilchrist 1994)
and external (bank) finance (see Kashyap et al. 1994). Moreover, Carpenter et al (1994)
test the importance of financing constraints using high-frequency (i.e. quarterly) panel
data and are thus able to observe short term changes in inventory investment levels. The
perspective of financing constraints, as adopted by Carpenter et al (1994), builds on the
notion that external finance (e.g. loans, bonds, commercial paper) is substantially more
97
expensive than internal finance (e.g. earnings and depreciation flows). The latter, thus, is
the preferred means of financing (inventory) investments.
Internal finance, however, is extremely volatile over the business cycle as it is
immediately affected by a slow-down in sales revenues given fixed or quasi-fixed
production costs in the short-run. As a consequence, comparatively liquid assets with
relatively low adjustment costs, such as inventories, are likely to absorb most of the
internal finance fluctuations of financially constrained firms. Carpenter et al (1994) argue
that this is particularly true for small firms whose access to external finance alternatives
such as corporate bonds and commercial paper is impeded by the lack of publicly
available information and the ensuing information asymmetry, adverse selection and
moral hazard problems. Small firms, the authors suggest, are thus forced to rely on
expensive bank loans as a last recourse to compensate for fluctuations in internal finance.
The effect of internal finance constraints on inventory (dis)investment is therefore
expected to be even greater for small firms than for large firms. The authors further
suggest that the magnitude of this effect depends on the optimality of inventory levels at
the beginning of the period. This contention builds on the idea that the marginal cost of
liquidating inventory stocks increases as current inventory levels (negatively) deviate
from optimal inventory levels.
Carpenter et al (1994) use Compustat data from the U.S. manufacturing industry (1981-
1992) to perform a series of regression analyses. The results generally indicate that the
level of cash flows is positively related to inventory investment, or put differently,
98
internal finance flows account for a significant portion of the variability in inventory
investment. While this is found to be true for both small and large firms, the authors note
that the effect tends to be greater in magnitude for small firms. The general result, thus, is
in line with the authors’ theoretical expectations. It is further noted, that the movements
of cash flows are highly procyclical, which, combined with the identified cash flow-
inventory link, provides a rationale for the high volatility of inventory investment over
the business cycle.
More recent empirical research also finds partial support for the contention that (firm)
financial factors impact firm inventory. Corbett et al (1999), for example, present a study
of UK and Japanese industries. They find that interest rates are significant predictors of
inventory investments in the paper, chemicals, and non-electric machinery industries
(UK), as well as in the Japanese chemicals, steel and iron, and metal manufacturing
industries. A study by Guariglia (1999) of UK manufacturing firms further explores the
effect of financial factors – Guariglia uses the coverage ratio as a measure of a firm’s
financial condition – on inventories. Her findings indicate a significant positive
relationship between coverage ratios and inventory levels during recessions and periods
of tight monetary policy.
In summary, it is noted that researchers in the economics field expect that less favorable
financial conditions will result in lower inventory levels. The significance levels of
empirical findings, however, vary greatly from study to study, depending on the
measures, data sets, and time periods used. It is also noted that researchers have used a
99
broad range of financial variables (interest rates, coverage ratios, and cash levels, for
example). No prior research has attempted to more comprehensively measure the
multifaceted firm financial distress construct and relate the latter to firm inventories. This
essay fills this gap. In addition, it will be argued in this research that previously
unobserved factors may also impact firm inventories and moderate the magnitude of the
financial distress-inventory relationship.
3.2.2. Inventory theory
Firms hold inventory for at least two reasons. First, delivery and production cycles are
typically not perfectly aligned. Natural stocks of raw materials as well as intermediate
and finished products therefore occur at various points throughout the production and
distribution process. These inventories are typically referred to as cycle stocks. Second,
inventories buffer against uncertainty. Specifically, unexpectedly high demand or longer
than usual lead times may lead to costly disruptions in manufacturing and delivery. Safety
stocks are a means of mitigating this risk by holding extra inventory that will be used
only if the need arises.
Determining the magnitude of cycle and safety stocks is a crucial task in inventory
management. While holding inventory is costly due to warehousing and opportunity
costs, not holding inventory may result in substantial stockout costs. The latter can take
the form of backorder (e.g. expediting) or lost sales costs, for example. Inventory theory
has been concerned with developing optimal, i.e. cost-minimizing or profit-maximizing
100
inventory policies. Multiple models have been proposed for different settings and
assumptions (see e.g. Tersine 1994). It is not the focus of this research to provide a
comprehensive review of these models. Rather, two questions are asked. First, which
determinants of inventories are proposed by inventory theory? And second, how may
firm financial distress be related to inventories from an inventory theory perspective? To
address these questions, two widely used and commonly known inventory models, the
r,Q model and the s,S model are briefly reviewed below. Particular attention is paid to the
r,Q model, and most of the subsequent discussion refers to this inventory policy. The
general results relating to the determinants of inventory levels and the relationship
between financial distress and inventory levels do, however, hold for most other
inventory models as well.
The r,Q inventory model is an extension of the well-known and widely used economic
order quantity (EOQ) model which, in its most basic form, balances ordering and
inventory holding costs
42
. First developed by Harris (1913), the EOQ and its variants
have been prominently featured in inventory management research and practice for over
ninety years (see Erlenkotter 1990 for a review of the early history of the EOQ model).
This model’s appeal lies in its relative simplicity and ease of use, as well as in its
robustness (Alstrom 2001). Reuter (1978) surveyed a total of 228 firms in five states in
the U.S. and finds that 75.4% of all respondents use the EOQ on a continuing basis with
an additional 9.6% indicating occasional use of the EOQ. In a study conducted by
McLaughlin et al (1994), 28% out of 236 survey respondents reported using the EOQ. In
42
See any textbook on inventory management for a detailed discussion of the economic order quantity
model and its variants (e.g. Tersine 1994)
101
a more recent survey, Rabinovich and Evers (2002) find that the EOQ is deemed
important in managerial practice and is commonly used by logistics managers to
determine optimal order quantities
43
. Zinn and Charnes (2005) note that quick response
(QR) inventory policies have become increasingly popular in modern inventory
management and therefore analyze the relative merits of the EOQ and QR methods,
respectively. Based on a series of numerical analyses, Zinn and Charnes (2005) conclude
that the EOQ continues to be the preferred inventory policy when order costs are
relatively high
44
. Numerous researchers have conducted sensitivity analyses and have
found that moderate deviations from the EOQ’s assumptions do not have a substantial
impact on order quantities and associated total inventory costs (e.g. Sun and Queyranne
2002). Its popularity, simplicity, and robustness make the EOQ a good starting point for
developing an inventory theory perspective on the financial distress-inventory
relationship.
The classical EOQ is based on the following assumptions: the demand rate is constant,
continuous and known, and lead times are zero. Replenishments are received
instantaneously and all at once, and the cost of placing an order as well as unit holding
costs are constant. The classical EOQ model considers only a single product and assumes
that there are no interactions with other inventory items. Moreover, it is assumed that the
firm has sufficient capital and capacity to purchase the economic order quantity. In the
r,Q model, the rather unrealistic assumptions of constant demand and zero lead times are
43
On a five point scale (1 = unimportant, 5 = very important) 256 survey respondents ascribe an average
weight of 3.27 to the EOQ, and 19.61% report the use of the EOQ for determining finished goods orders.
44
See Zinn and Charnes (2005) for a summary of their study’s results. Table 6 (p.139) identifies the
conditions under which QR and EOQ policies are preferred, respectively.
102
relaxed. With stochastic demand, nonzero but constant lead times, and per-unit backorder
costs, the total inventory cost equation is defined by
( )
2
I
S Q
TC A B E M r H r M
Q
(
= + ? > + + ? (
¸ ¸ (
¸ ¸
, where
S is the expected sales volume
45
over the planning horizon, A is the order cost
46
, H is the unit holding cost, and B is the
unit backorder cost. M is lead time demand (i.e. the sales volume during lead time) and r
is the reorder point which, along with order quantity Q, is the decision variable of
interest. ( ) E M r > is the expected stockout quantity, while
( )
r M ? represents the
average size of the safety stock. Taking the derivative of
I
TC with respect to Q and
setting the expression equal to zero yields the cost minimizing order quantity:
( )
*
2S A B E M r
Q
H
+ ? > (
¸ ¸
= . Similarly, the optimal reorder point
*
r is obtained by
setting the derivative of
I
TC with respect to r to zero. This yields the cost-minimizing
stockout probability ( )
*
HQ
P M r
BS
> =
47
. Under the assumption of normally distributed
demand, the latter value converts to standard normal deviate k , and the reorder point is
defined as
*
LTS S
r M k SL k L ? ? = + ? = + ? where
LTS
? is the standard deviation of lead
time sales, and
S
? is the standard deviation of sales
48
.
The control parameters of the r,Q inventory policy are thus defined by the expected sales
45
The inventory literature commonly uses the term Demand instead of sales volume. It is noted however,
that inventory decisions are made a priori based on forecasts.
46
In a manufacturing context, these order costs may also be thought of as production setup costs.
47
Since Q is a function of r and vice versa, the optimal solutions for these parameters are found by
iteration.
48
Lead times are assumed constant. See Tersine (1994) for more detail.
103
volume, sales variability, lead times, ordering costs, holding costs and backorder costs.
Inventory theory suggests that these parameters appropriately predict a firm’s inventory
decisions and thereby firm inventory levels. Figure 10 provides a graphic illustration of
the r,Q policy.
Figure 10: Illustration of the r,Q policy
The s,S policy is similar to the r,Q inventory policy but may differ from the latter in that
the order quantity is variable when inventories are reviewed periodically only, or when
demand is lumpy, i.e. does not follow a pure Poisson process with unit demand. Each
time the inventory level drops below a threshold level (or reorder point) s, on order of
size (S-s) is placed. For details on the mathematical derivation of the inventory control
parameters (s,S) the interested reader is referred to Denardo (2003), for example. In this
context, it shall suffice to note that the s,S policy is defined by the magnitude of demand,
the item’s unit costs, per-unit holding and backorder costs, as well as ordering costs.
Randomness of demand and lead times can also be incorporated in s,S type inventory
Q*
time
r*
Safety
stock
M
average inventory level
Inventory
Q*
time
r*
Safety
stock
M
average inventory level
Inventory
104
models. The determinants of an s,S inventory control policy, thus, are essentially the
same as the determinants of the r,Q policy.
In summary, inventory theory suggests that firm inventories should be a function of
average demand, demand variability, lead times, ordering costs, holding costs and
backorder costs (or lost sales costs), regardless of the specific inventory control policy in
place.
Next, the potential effects of firm financial distress on inventories are discussed from an
inventory theory perspective. Financial costs are most directly reflected in a firm’s
holding costs. Holding costs include a financial cost component representing the capital
cost of inventories (Followill et al. 1990). While holding costs also comprise a noncapital
carrying charge
49
, Timme (2003) notes that the financial component of holding costs
usually exceeds noncapital carrying charges. In accordance with this contention, the
survey results presented by Fraser and Gaither (1984) suggest that 68% of all firms
approximate inventory carrying costs with borrowing costs. The latter are a function of a
firm’s financial condition (see e.g. Buzacott and Zhang 2004, Wiersema 2005).
Specifically, a deterioration of a firm’s financial condition implies higher borrowing costs
and thereby higher inventory holding costs. Returning to the inventory control parameters
of the r,Q policy, it is evident that higher holding costs entail lower inventory levels, all
49
Noncapital carrying costs comprise the costs of warehousing, obsolescence, pilferage, damage, and
insurance, as well as taxes and administrative charges (see Timme, 2003).
105
else equal
50
. The optimal order quantity
( )
*
2S A B E M r
Q
H
| |
+ ? > (
¸ ¸
|
=
|
\ ¹
is decreasing in
holding costs
*
i.e. 0
Q
H
| | ?
<
|
?
\ ¹
, as is the reorder point: The optimal stockout probability
increases in H ( )
*
HQ
P M r
BS
| |
> =
|
\ ¹
51
. This translates to a lower safety factor k and
consequently to a lower reorder point r
( )
*
LTS S
r M k SL k L ? ? = + ? = + ? . A negative
relationship between firm financial distress and optimal firm inventories can therefore
straightforwardly be established from an inventory theory perspective.
While the author is unaware of any empirical inventory research relating firm financial
condition to inventory levels, there is some analytical research on the relationship
between various financial factors in a broader sense and inventory decisions. For
completeness, a few examples of such research are discussed below.
One literature stream, for example, investigates the effects of trade credits, permissible
payment delays granted by suppliers, on economic order quantities. Haley and Higgins
(1973) analyze the interdependence of inventory decisions and credit terms, and
determine jointly optimal order quantities and payment schedules. Most subsequent
research assumed credit terms as exogenously given (i.e. unilaterally defined by the
supplier) and focused on the effects of trade credits on order quantities. Chapman et al
50
In addition, higher holding costs may imply lower firm output choices and hence lower demand.
51
Note that the optimal stockout probability is also a function of Q, which, in turn, decreases in H .
*
P
therefore increases in H and decreases in H , Overall, the optimal stockout probability increases in H .
106
(1984), for example, conduct an average cost analysis and conclude that trade credit
periods, while significantly impacting total costs, do not affect optimal order quantities.
Chand and Ward (1987), on the contrary, find that order quantities increase as payment
delay times increase. Rachamadugu (1989) reconciles these contradictory findings and
ascribes them to differences in the assumptions and setup of the respective models. In
summary, Rachamadugu’s (1989) analyses corroborate Chand and Ward’s (1987)
intuitively appealing findings, as do the results of a more recent study conducted by
Chang and Teng (2004). For a review of some earlier works on inventory models with
consideration of permissible payment delays the interested reader is referred to Kim and
Chung (1990).
Another stream of research is concerned with the impact of budget constraints of
inventory decisions. Financially distressed firms are likely to operate under budgetary
constraints. Rosenblatt (1981) formulates a constrained inventory optimization problem
with limited budget availability. He uses the Lagrangian procedure to demonstrate the
intuitive result that the optimal order quantity will be restricted to the maximum
affordable level when the budget constraint is tight. A multi-item newsvendor problem
with a budget constraint is analyzed by Moon and Silver (2000). The authors’ attention
focuses on rules for optimally allocating scarce resources to different products. In the
context of this research, however, it is sufficient to note that a restriction on total
expenditure is shown to lead to lower than optimal order quantities and increased overall
costs (Moon and Silver 2000). Abdel-Malek and Montanari (2005) extend Moon and
Silver’s (2000) work by conducting an analysis of the multi-product newsvendor problem
107
with two (generic) constraints. Rustenburg et al (2000) also present a study similar to that
of Moon and Silver (2000) in the context of spare parts logistics, where resupply
decisions for multiple items must be made under limited budgets. One of the basic
finding’s of Rustenburg et al (2000) is that budget constraints result in lower part
availability levels.
Empirical inventory research is challenging from a data collection standpoint and
therefore rather scarce (examples include Ballou 1981, Roumiantsev and Netessine
2007). Inventory theory is, however, indispensable when empirically explaining
inventory levels and analyzing the relationship between firm financial distress and
inventories. This research builds on prior work in the economics area by drawing on
inventory theory to explain this relationship and by incorporating a set of previously
ignored inventory variables in the regression model.
3.2.3. The financial distress-inventory hypothesis
The theoretical link between firm financial distress and inventories has been discussed
from both an economics perspective and an inventory theory perspective in the previous
subsections. Clearly, both theories suggest that greater levels of financial distress (i.e. less
favorable financial conditions) result in lower inventory levels, all else equal. Most prior
research argues that budgetary constraints and increased borrowing costs lead distressed
firms to hold less inventory. Prior research has found some support for this hypothesized
relationship (Carpenter et al. 1998, Carpenter et al. 1994, Gertler and Gilchrist 1994,
108
Kashyap et al. 1994). As Roumiantsev and Netessine (2007) point out, however, this
body of work “might contain biases because many important micro-economic data points
that affect inventories have been left out, including lead times, demand uncertainty,
inventory holding costs, etc.” (p.6). This study reexamines the financial distress-
inventory link while controlling for these inventory determinants.
Besides the previously discussed rationale that financial distress results in budgetary
constraints and increased borrowing costs, it may be argued that managers of distressed
firms have an incentive to liquidate assets (Hofer 1980) such as inventories in an effort to
increase liquidity and improve key firm performance measures such as the Return on
Assets (RoA). In summary, there appears to be clear theoretical and at least some
empirical support for Hypothesis 8:
Hypothesis 8: Holding demand constant, greater levels of financial distress result in
lower inventories.
So far, the focus has been on firm level determinants of inventories. In the next section,
this focus is expanded to include a firm’s supply chain partners. Specifically, the effect of
power on inventories and the financial distress-inventory link is discussed.
3.3. The supply chain perspective
In this section, the relationship between distress and inventories is analyzed from a
109
supply chain perspective. Specifically, the role of power in inter-firm relationships and
firm (inventory) decision making is reviewed in Section 3.3.1. In line with prior research
in the industrial organization economics field, it is suggested that a firm’s power position
impacts its inventory decisions. The second subsection analyzes the moderating role of
inter-firm power in the financial distress-inventory relationship. It is hypothesized that
power determines to what extent financial distress affects firm inventories. The resulting
contingency framework is subsequently tested using U.S. industry data.
3.3.1. Supply chain considerations in inventory decisions
Many parameters influence managerial decision making. While firm-level variables such
as holding and purchasing costs, for example, naturally have a strong impact on
managerial decisions relating to sales prices and inventories, market factors and inter-
firm relationship variables cannot be ignored.
First, competitors’ actions clearly impact a firm’s choices. Researchers from both the
economics and strategy fields have contended that managers must anticipate competitive
reactions and evaluate their implications when deciding on sales prices (see e.g. Chen et
al. 1992, Gibbons 1992). By the same token, firms also compete on inventories. Cachon
(2001), for example, analyzes competitive inventory policies and, for a given set of
assumptions, defines a competitive Nash equilibrium in inventories (see also e.g.
Mahajan and Ryzin 2001). As a consequence, a firm’s inventory decisions are a direct
function of competitors’ inventory choices.
110
Second, firms are typically a part of supply chains that extend across many companies
from raw material suppliers to the end customer. As firm decisions impact the
functioning of the entire supply chain, supply chain firms are necessarily interdependent
(Cox et al. 2001). This interdependence is of particular interest in this research. When the
Case Corporation reduced its inventories as a part of its restructuring efforts, for example,
suppliers had to bear the burden, but were willing to do so to improve customer service
levels (Buxbaum 1995). Chrysler’s aggressive cost-cutting measures implemented in
2000 and 2001, in turn, were considered “acts of war” (p.32) by some suppliers who
agreed to cooperate only because they had little choice (Stundza and Milligan 2001).
These examples illustrate how firms’ (inventory) decision making can be constrained by
cooperative arrangements and coercive pressure exerted by buying and supplying firms.
The extent to which firms are willing or forced to yield to these constraints is a function
of the inter-firm power balance. It is argued in this research that power not only impacts a
firm’s inventory decision but also moderates the link between financial distress and
inventories. Following a brief review of the role of power in inter-firm relationships and
some related literature, the corresponding hypotheses are derived below.
3.3.1.1. Inter-firm relationships: The role of power
There exists a sizeable literature base on the nature, drivers, and consequences of power
in inter-firm relationships. Gaski (1984) provides a review of the early work in this field
111
and, in summarizing, defines power as the “ability to evoke change in another’s
behavior” (p.10). Emerson (1962) relates power to dependence and suggests that Firm
A’s power over Firm B equals Firm B’s dependence on Firm A. The sources of (firm)
power in (inter-firm) relationships were first analyzed by French and Raven (1959).
According to French and Raven (1959) these bases of power include:
? Reward power: A can motivate B by granting rewards;
? Coercive power: A can effectively punish B;
? Legitimate power: A has a legitimate right to prescribe B’s behavior;
? Referent power: A serves as a model to B;
? Expert power: A’s expertise conveys A the power to influence B.
The term power often carries a negative connotation (Hingley 2005). French and Raven’s
power bases, however, suggest that power may be used both collaboratively and
coercively. Along the same lines, Frazier and Antia (1995) suggest distinguishing
between the possession and the application of inter-firm power. Frazier and Antia (1995)
argue that the channel context and the specific inter-firm power constellation drive the
communication style between firms. The latter can be either threatening (as seen above in
the case of Chrysler) or collaborative (as evidenced in the previously mentioned example
of Case Corp.). Firms with some degree of power can, thus, exert either coercive control
or collaborative control to affect other firms’ forced or voluntary behavioral change,
respectively (Frazier and Antia 1995, Hingley 2005).
Cox et al (2003) note that power is an element of every buyer-supplier relationship. The
112
authors suggest that each relationship can be characterized by one of four power
structures: buyer dominance, supplier dominance, buyer-supplier interdependence, and
buyer-supplier independence. Cox (2001) further notes that firms strive to be in a
dominant position over buyers and suppliers so as to extract the maximum amount of
value generated in the supply chain. Cox et al (2001) have coined the term “value
appropriation” to describe this mechanism which can take the form of cost squeezing on
the supply side or high-margin pricing on the sales side, for example. As the previously
cited examples of Chrysler and Case Corp. (Buxbaum 1995, Stundza and Milligan 2001)
have illustrated, firms may also use their dominant power position to shift inventory
ownership to suppliers or buyers. In this vein, Wallin et al (2006) contend that “if a firm
within a specific buyer-supplier relationship were to hold bargaining power, this would
greatly enhance its ability to dictate to and make certain demands of a specific supplier”
(p.59) with respect to the inventory management approach used in the supply chain (see
also Dobson 2005). This research empirically tests the contention that a firm’s power
relative to its suppliers and buyers will impact firm inventory levels, ceteris paribus.
Prior work in this area is reviewed in the following subsection.
3.3.1.2. Supply chain power and inventory decisions
Few researchers have investigated the effect of power on inventories. Blazenko and
Vandezande (2003) ascribe the lack of power-inventory research to the fact that “the
academic literature on inventory focuses on production and procurement as the principal
determinants of […] inventory […] management” (p.256) while “the principal focus of
113
the study of inventory in the economics literature is on the macroeconomic role of
inventory as a stabilizing or destabilizing factor for production in business cycles”
(p.256). Much of the research relating dyadic power, i.e. a firm’s power vis-à-vis another
firm, to inventories remains descriptive in nature and is mostly based on case studies (see
e.g. Dobson 2005). Within the supply chain management literature, articles on power and
its implications are, for the most part, purely conceptual. The author is aware of only two
papers that empirically investigate the effects of power on inventories. Both papers are
housed within the industrial organization economics literature and are discussed in turn.
Amihud and Mendelson (1989) suggest that “a firm with market power will use inventory
as a wedge between the quantity available for sale and the quantity shipped to market”
(pp.269-270). According to Amihud and Mendelson (1989), firms build up inventories
when supply exceeds demand in an effort to maintain higher prices and keep production
at constant levels. With demand greater than supply, in turn, firms deplete inventories to
maximize revenues. A firm’s motivation to use inventories to smooth price fluctuations
thereby increases with the firm’s market power as “greater market power implies a
stronger effect of the firm’s sales quantity on price” (p.270). Amihud and Mendelson
(1989) test the market power-inventory relationship using Compustat data from the U.S.
manufacturing industry. The results suggest that firm market power, measured by either
the Lerner index ([price – marginal cost]/price) or the firm’s market share, positively
affects firm inventories after controlling for firm sales, sales trends, sales variability, and
average industry inventory levels. The authors therefore conclude that “market power has
a sizeable effect on inventory, which has been overlooked so far” (p.275).
114
Blazenko and Vandezande (2003) build their article on the contention that stockout costs
should be represented in inventory estimation models. The authors suggest that greater
levels of competition erode profit margins and thereby reduce the amount of profits
foregone in case of a stockout, while, at the same time, more competition also increases
stockout costs due to the greater availability of alternative sources of supply. According
to Blazenko and Vandezande (2003) the effect of market concentration (an indicator of
the level of power firms possess in a given market) on inventories is ambiguous and
depends on whether the effects of lower foregone profit or increased lost sales costs
prevail. The authors empirically investigate the effect of industry concentration
(measured by the two-firm concentration ratio) on finished goods inventory levels (at the
industry level). The control variables included in the model are, most notably, industry
gross-margins and a set of industry indicator variables. Data from the U.S. manufacturing
industry are used for the empirical analyses. The results suggest that higher industry
concentration levels result in lower inventory levels, ceteris paribus. Blazenko and
Vandezande (2003) conclude that “a less competitive product market reduces the adverse
consequences of stock outs and firms respond by reducing inventories” (p.263).
In summary, Amihud and Mendelson (1989) suggest that greater levels of market power
imply higher inventory levels, while Blazenko and Vandezande (2003) find that
inventories are lower in more concentrated markets (implying more powerful firms). This
conflict may, in part, be explained by different levels of analysis (firm vs. industry), and
differences in measurement. In addition, it is noted that both models fail to include
115
variables prescribed by inventory theory, such as lead times and the cost of holding
inventory, for example. The results may, therefore, be biased (Roumiantsev and
Netessine 2007). Also, neither article attempts to relate focal firm or industry power to
the power levels of buyers and suppliers. Yet, power is dyadic in nature (Cox et al. 2001,
Emerson 1962, Frazier and Antia 1995, Gaski 1984), and a complete evaluation of power
must consider a firm’s power relative to another firm or industry. Prior research has
focused uniquely on downstream power vis-à-vis buyers but has ignored the upstream
supply side. Power, however, is “Janus-faced”, i.e. double-sided (Cox 2001), as firms are
engaged in power relationships with both their buyers and their suppliers (as well as with
their competitors). This research addresses this shortcoming in terms of measurement of
power and proposes a comprehensive set of power measures (see Chapter 3.4) capturing
not only focal firm power, but also power levels in the buying and supplying industries.
3.3.1.3. The power-inventory hypotheses
As outlined previously, prior research on the role of power in supply chain relationships
has suggested that greater levels of power allow firms to obtain more favorable terms and
conditions in negotiations with their buyers and suppliers (Blazenko and Vandezande
2003, Cox 2001, Cox et al. 2001, Wallin et al. 2006). More powerful firms may thus be
able to push the burden of inventory ownership onto buyers and suppliers to a greater
extent than less powerful firms. Hypothesis 9 is therefore proposed as a baseline power-
inventory hypothesis:
116
Hypothesis 9: Greater firm power results in lower inventory levels.
Hypothesis 9 can be refined by distinguishing between firm power relative to suppliers
and buyers, respectively. Accordingly, Hypothesis 10 and Hypothesis 11 are introduced
below.
Wallin et al (2006), for example, argue that a firm with bargaining power may impose
item availability targets on suppliers, thus forcing suppliers to hold larger inventories to
meet these targets while reducing the need to hold inventory at the buying firm (see also
Cox et al. 2001). In addition, a powerful firm may be able to demand inventory
consignments from its suppliers, thus providing the buying firm with improved item
availability without incurring the cost of inventory ownership (Wallin et al. 2006).
Hypothesis 10 therefore suggests that greater power over suppliers implies lower
inventory levels, all else equal.
Hypothesis 10: Greater firm power relative to suppliers results in lower inventory
levels.
Hypothesis 11 mirrors the reasoning underlying Hypothesis 10 and projects it to the
downstream relationship between a firm and its buyers. Accordingly, greater levels of
power over buying firms are expected to be associated with lower inventory levels, all
else equal. While the work of Blazenko and Vandezande (2003) presents some evidence
in support of this contention, the results of the study published by Amihud and
117
Mendelson (1989) appear to contradict this expectation. The latter researchers implicitly
assumed that firms can use inventories to mitigate price fluctuations only if they directly
own these inventories. Powerful firms in supply chains, however, may be able to dictate
the release and buildup of inventories even when these inventories are not under direct
ownership and control. From a supply chain perspective, Hypothesis 11, therefore, does
not necessarily disagree with the arguments and results presented by Amihud and
Mendelson (1989).
Hypothesis 11: Greater firm power relative to buyers results in lower inventory levels.
Besides the direct effect of power on firm inventories, it is also contended that firm
power impacts the extent to which firms can reduce inventories when experiencing
financial distress. These moderating hypotheses are developed below.
3.3.2. Firm power as a moderator of the distress-inventory link
The inconsistency of the results presented by prior research on the link between financial
variables and inventories may be an indication that there are factors that affect the
magnitude and significance of this relationship. Prior research has suggested that firm
size may be such a moderator. This rationale is briefly reviewed below. This essay, in
turn, focuses on the moderating role of firm power. The related reasoning is discussed in
Section 3.3.2.2 and the corresponding hypotheses are formulated.
118
3.3.2.1. Prior research: Firm size as a moderator of the distress-inventory link
As discussed in Section 3.2.1, most studies on the link between financial factors and
inventories have suggested that the magnitude of this relationship may differ by firm size.
Gertler and Gilchrist (1994) and Carpenter et al (1998, 1994), for example, perform
separate regression analyses by firm size classes (small vs. large). These authors find
empirical support for their contention that smaller financially constrained firms
experience stronger inventory contractions due to their limited access to capital markets
and, thus, means of financing inventory investments. Kashyap et al (1994) find that the
effect of firm liquidity on inventories is stronger for firms without bond ratings than for
firms with bond ratings. Since unrated firms typically are smaller firms, their results also
suggest that the effect of financial constraints on inventories differs by firm size.
Firm size may be a proxy for a firm’s power, with larger firms being more powerful than
smaller firms, all else equal. Following this reasoning, the negative effect of firm distress
on inventories (as hypothesized in Hypothesis 8) may be expected to decrease with the
firm’s power. It is noted, however, that the arguments set forth in prior research focus
uniquely on the operating implications of financial constraints, suggesting that firms with
limited resources must reduce inventory investments, particularly when external funds
can be procured at high costs only. This research, in turn, suggests that financially
distressed firms want to reduce inventories and will do so to the largest extent possible. In
other words, reducing inventories is considered desirable as long as potentially negative
119
consequences of inventory cutbacks, such as stockouts and decreases in customer service
levels, can be mitigated by increased buyer and supplier efforts (e.g. in terms of increased
inventory holdings, shorter lead times, etc.). This contention is discussed in more detail in
the following subsection.
3.3.2.2. The power moderator hypotheses
While firms consistently strive to increase efficiency and profitability, these efforts are
reinforced during corporate turnarounds (e.g. Hofer 1980). Tom Sidlik, then Executive
Vice President with Chrysler, for example, indicated that “we’ve accelerated our ongoing
cost-reduction programs so that we can take 15% costs out of the system by the end of
2002.” (Stundza and Milligan 2001, p. 30). Sidlik continued to note that “in the current
business situation, we are counting on our supplier partners to stand with our company
[…] in these difficult times” (p.31). The importance of concessions and support offered
by suppliers during corporate turnarounds is further illustrated by Arogyaswamy and
Yasai-Ardekani (1995) who argue that cutting inventory can only be a successful
turnaround strategy if potentially resulting delivery delays can be mitigated through
suppliers’ or buyers’ increased efforts, for example. Finkin (1985) also notes that during
company turnarounds “[t]erms and conditions of sale are worth fighting over” (p.17) and
that a supplier’s agreement to shorter lead times may help reduce inventory levels.
Clearly, a firm’s bargaining power vis-à-vis its suppliers and buyers will determine to
what extent such concessions will be made. In a similar vein, Hambrick and Schecter
(1983) note that a firm’s power might affect its choice of turnaround strategy as “strong
120
channels of distribution […] could allow [the distressed firm] to solve [the] problems at
less human and organizational cost” (p.234), with loyal and obedient distributors carrying
larger shares of the burden.
These arguments and examples lend support for the contention that distressed firms may
be able to reduce inventories to a greater extent when they have higher degrees of power
relative to their suppliers and buyers. Hypothesis 12 is formulated accordingly:
Hypothesis 12: The effect of firm financial distress on inventories increases with the
firm’s power.
Hypothesis 12 can be specified for a firm’s power relative to buyers and suppliers,
respectively:
Hypothesis 13: The effect of firm financial distress on inventories increases with the
firm’s power relative to suppliers.
Hypothesis 14: The effect of firm financial distress on inventories increases with the
firm’s power relative to buyers.
The moderating effect of firm power on the distress-inventory relationship is graphically
illustrated in Figure 11. On average, a negative relationship between the magnitude of
121
firm financial distress and inventories is expected (Hypothesis 8). This relationship,
however, is hypothesized to be stronger the greater the firm’s power (Hypothesis 12-
Hypothesis 14).
Figure 11: The moderating effect of power on the distress-inventory relationship
An overview of the resulting model is given in Figure 12. In summarizing, a set of
hypotheses on the link between firm financial distress and inventories has been
formulated based on a variety of theoretical perspectives. Particular attention is given to
the role of power as a determinant of firm inventories and as a moderator of the distress-
inventory relationship.
Financial
distress
Inventory
a
v
e
r
a
g
e
m
o
r
e
p
o
w
e
r
f
u
l
f
ir
m
s
le
s
s
p
o
w
e
rfu
l firm
s
negative
interaction
effect
Financial
distress
Inventory
a
v
e
r
a
g
e
m
o
r
e
p
o
w
e
r
f
u
l
f
ir
m
s
le
s
s
p
o
w
e
rfu
l firm
s
negative
interaction
effect
122
Figure 12: Research model
3.4. Data and methodology
The hypotheses set forth in the previous sections are tested using data sets comprising
information on a cross-section of U.S. industries. Details on the data samples,
specification of the model, variable measurement, and data sources are provided in the
following subsections.
Financial Distress Inventory
Power
• Total inventory
• Raw materials
• Finished goods
• Firm power
• Firm power relative to suppliers
• Firm power relative to buyers
Control
variables
Financial Distress Inventory
Power
• Total inventory
• Raw materials
• Finished goods
• Firm power
• Firm power relative to suppliers
• Firm power relative to buyers
Control
variables
123
3.4.1. Sample selection
The empirical tests are conducted using data from U.S. manufacturing firms. Two data
sets from 1997 and the time period from 1998 to 2004, respectively, are used for the
analyses. This subsection provides information on the sample selection criteria.
Industries
Most empirical inventory research has focused on manufacturing industries for the
obvious reason that manufacturing firms are likely to hold substantial inventories
(Carpenter et al. 1998, Carpenter et al. 1994, Corbett et al. 1999, Gertler and Gilchrist
1994, Guariglia and Schiantarelli 1998, Kashyap et al. 1994, Roumiantsev and Netessine
2007). Manufacturing industries are defined by the North American Industry
Classification System (NAICS). The NAICS system has replaced the U.S. Standard
Industrial Classification (SIC) system and all U.S. government agencies commonly report
industry statistics by NAICS codes. NAICS codes have between two and six digits and
are structured hierarchically. The first two digits of a NAICS code designate the
“economic sector” and the third digit identifies the “subsector”. The fourth, fifth, and
sixth digits designate the “industry group”, “NAICS industry”, and “national industry”,
respectively. Manufacturing industries are part of the economic sectors 31-33. All firms
in these sectors for which complete data are available are included in the empirical
analyses.
This study approaches the analysis of inventories from a supply chain perspective and
124
investigates, among other things, the role of a firm’s power on firm inventories and on
the relationship between financial distress and inventories (see Figure 12). Given the
dyadic nature of power, the data set also comprises information on the wholesale and
retail trade industries
52
as these industries likely are manufacturing firms’ principal
suppliers and buyers (besides buyers and suppliers within the manufacturing industries).
Other industries that may potentially buy from or sell to manufacturing industries are not
included in the analysis for the two following reasons:
? Service industries: Service industries
53
are of limited interest in the context of
inventory studies and are not considered in this research (see also Roumiantsev
and Netessine 2007).
? Insufficient data availability: The remaining economic sectors are Agriculture,
Forestry, Fishing and Hunting (11), Mining (21), Utilities (22), Construction (23).
While these industries may potentially be involved in the exchange of goods (i.e.
inventories) with manufacturing firms, these industries must be excluded from the
data analysis due to insufficient data availability at the industry level.
Specifically, industry sales data and industry concentration ratios are available at
highly aggregated levels only and, thus, are not usable in the empirical analyses.
52
NAICS codes 42, 44, and 45.
53
Service industries are found in the following economic sectors (two-digit NAICS codes are given in
parentheses): Transportation and Warehousing (48,49), Information (51), Finance and Insurance (52), Real
Estate, Rental and Leasing (53), Professional, Scientific, and Technical Services (54), Management of
Companies and Enterprises (55), Administrative and Support and Waste Management and Remediation
Services (56), Educational Services (61), Health Care and Social Assistance (62), Arts, Entertainment, and
Recreation (71), Accommodation and Food Services (72), Other Services (except Public Administration)
(81), Public Administration (92).
125
Time periods
This research uses data from 1997 to 2004. This time period is selected for two reasons.
First, consistent data at the industry level are available for this time period only. More
recent data (after 2004) were not available at the time of writing, and older data (prior to
1997) were aggregated differently as the industry classification system was revised in
1997 with the move from the Standard Industry Classification (SIC) system to the
NAICS system. Second, the selection of a relatively recent time period is adequate given
that inventory dynamics may have been significantly different in earlier time periods
prior to the widespread adoption of information systems and Just-In-Time practices, for
example (Roumiantsev and Netessine 2007).
In the U.S., an Economic Census is conducted every five years (years ending with “2”
and “7”). During the time period considered here (1997-2004), Economic Census data
were thus collected in 1997 and 2002. While the 2002 Economic Census data were not
available at the time of writing, data from the 1997 Economic Census could be obtained
from the website of the Bureau of Economic Analysis (BEA). The Economic Census data
provide more detailed industry information, for example on industry sales and
concentration ratios, than the data that are collected by the BEA during years in which no
Economic Census is conducted. Therefore, the empirical analyses are performed using
two different datasets: First, a panel data set is constructed. This panel data set contains
information on a cross-section of U.S. manufacturing industries for the time period from
1998 to 2004. Second, a cross-sectional data set using data from 1997 only is constructed.
126
Using two distinct data sets for the empirical analyses has several advantages. The time
series data set, henceforth denoted “data set I”, is relatively large with multiple
observations per firm. Larger sample sizes generally facilitate the empirical analyses and
typically result in more robust coefficient estimates. The 1997 data set, in turn, provides
more fine-grained industry level data. This data set, henceforth denoted “data set II”, thus
is particularly useful when attempting to evaluate the relative power balances between
industries. In addition, the robustness and validity of the model are underlined if both
data sets produce consistent coefficient estimates.
Frequency
As noted by Carpenter et al (1998, 1994), high-frequency quarterly data may be desirable
for the analysis of firm inventories and financial factors due to their dynamic and volatile
nature. Many firms, however, report only selected parameters on a quarterly basis. Raw
materials and finished goods inventory data, for example, are often available on an
annual basis only. In line with prior research and due to greater data availability, annual
data are used in this study (Guariglia 1999).
3.4.2. Model specification
The purpose of this section is to derive an empirical inventory estimation model which is
grounded in inventory theory and supply chain management research. This research
thereby enhances prior economics research which generally modeled inventories as a
function of (lagged) sales, financial indicators, and a small set of control variables only.
127
According to inventory theory, firm inventory decisions should be a function of order
quantities (Q) and safety stock (SS) (see also Figure 10). The specific magnitude of end-
of-period inventories will then also be a function of sales realization ( )
t
S . In addition, it
is argued in this essay that a firm’s distress and power will affect firm inventories.
Dummy variables to account for inventory accounting differences (LIFO, AvgCost
54
) are
included as well (Carpenter et al. 1994, Gertler and Gilchrist 1994, Kashyap et al. 1994,
Roumiantsev and Netessine 2007). This yields the following inventory model:
(1) ( ) , , , , , ,
t
Inv f Q SS S Distress Power LIFO AvgCost = .
As seen in Chapter 3.2.2, the order quantity Q is a function of expected sales (
t S ), order
costs (A), backorder costs (B) and holding costs (H):
(2)
( )
, , , t Q f S A B H = .
Similarly, safety stocks are shown to be a function of lead times (L), sales (
t S ), sales
variability (
S
? ), and the safety factor k which, in turn, is a function of the optimal
stockout probability ( )
*
t
HQ
P M r
BS
| |
> =
|
\ ¹
(see Chapter 3.2.2). Safety stocks can, thus, be
represented as follows:
(3)
( )
, , , , t
S
SS f L H B S ? = .
The author is unaware of prior empirical inventory research that measured order and
backorder costs. For lack of suitable proxy measures, it is common practice to exclude
order and backorder costs from empirical inventory analyses (see e.g. Lieberman et al.
54
Further detail is provided in Section 3.4.3.2.
128
1999, Roumiantsev and Netessine 2007). While an attempt is made to approximate
order/setup costs (see section 3.4.3.2), backorder costs are not further considered in this
research. Holding costs are a function of the cost of the item that is purchased or
produced and the holding cost rate. Unit cost measures are not readily available due to a
lack of output indicators. Total costs of goods sold, however, are highly correlated with
sales, and are therefore not included in the regression model. A proxy for a firm’s capital
carrying charge will be used to approximate holding costs. Substituting Equations (2),
and (3) in Equation (1) and dropping the above mentioned variables, Equation (1) can be
rewritten as follows:
(4)
( )
, , , , , , , , , t
t S
Inv f S S A H L Distress Power LIFO AvgCost ? = .
Expected sales (
t S ) and realized sales (
t
S ) are naturally highly correlated with
S
t
t t
S S ? = + , where
S
t
? is the forecast error. To avoid excessive multicollinearity
55
,
Equation (4) is therefore restated as follows:
(5)
( )
, , , , , , , , ,
S S
Inv f S A H L Distress Power LIFO AvgCost ? ? = .
The resulting basic empirical estimation equation is defined in Equation (6) below:
(6) Inventory
itf
= ?
0
+ ?
1
SalesForecast
itf
+ ?
2
ForecastError
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
LeadTime
itf
+ ?
7
Distress
itf
+ ?
8
Power
itf
+ ?
9
LIFO
itf
+ ?
10
AvgCost
itf
+ ?
11
Distress
itf
* Power
itf
+ ?
itf
The subscripts i, t, and f designate the industry, time period, and firm respectively.
55
Not only are actual and expected sales highly correlated, but actual sales and the standard deviation of
sales are correlated as well.
129
Inventory
itf
, for example, indicates firm f’s inventory level in time period t, where firm f
operates in industry i. The interaction term (Distress
itf
* Power
itf
) is included to test
Hypothesis 1 Hypothesis 12, Hypothesis 13, and Hypothesis 14.
This inventory model specification differs from prior specifications on multiple accounts:
First, the model presented here controls for important predictors of firm inventories as
prescribed by inventory theory. The author is aware of only one study that controlled for
sales variability and lead times (Roumiantsev and Netessine 2007). The latter study,
however, does not investigate the effect of financial distress on inventories, nor does it
consider the role of firm power in inventory management. Second, measures of a firm’s
buying power and selling power are included. While few prior studies analyzed the
impact of market power on inventories (Amihud and Mendelson 1989, Blazenko and
Vandezande 2003), this is the first study to differentiate between power over buyers and
power over suppliers. In addition, a more comprehensive measure of financial distress is
proposed. Prior research relied on one-dimensional measures such as market interest rates
or firm cash flows to estimate holding costs or a firm’s financial situation. Market interest
rates, however, are poor approximations of holding costs (or firm financial condition, for
that matter), as such measures do not account for the heterogeneity of firms’ borrowing
rates. Measures such as cash flows, in turn, may not comprehensively evaluate firm
financial condition. Consequently, this is—to the best of the author’s knowledge—the
first study to investigate the effects of firm financial distress on inventories from an
inventory theory and supply chain management perspective.
130
3.4.3. Variables and Measurement
Variable measurement has been a major challenge in inventory research and probably is
the most important reason for the scarcity of empirical inventory studies. The measures
used in this research are in part based on the work of Roumiantsev and Netessine (2007)
and Carpenter et al (1994). The dependent and independent variables are discussed in
turn.
3.4.3.1. Dependent variable
The dependent variable in this study is firm-level inventory, i.e. firm f’s inventory in
period t. Specifically, three distinct firm inventory measures are used: total inventory,
raw materials inventory, and finished goods inventory
56
. Several researchers have
previously used these inventory variables in empirical analyses (e.g. Blazenko and
Vandezande 2003, Guariglia 1999). In line with prior research, absolute inventory values
are used (see e.g. Roumiantsev and Netessine 2007). Total inventories of firm f (which is
affiliated with industry i) in time period t are denoted
itf
TotalInv and are measured in
U.S. dollars as reported on the balance sheet (see e.g. Amihud and Mendelson 1989)
57
.
Analogously, raw materials inventories and finished goods inventories are denoted
itf
RawMatInv and
itf
FinGoodsInv , respectively.
56
Work-in-process inventories are reported by few firms only and are therefore not analyzed separately.
57
Roumiantsev and Netessine (2007) note that it is generally not necessary to adjust dollar values in time
series data since inflation has been at very low levels in the United States over the past decade.
131
All inventory data are obtained from Standard & Poor’s Compustat database. Only, firms
with at least $5,000 worth of (total) inventory are included in the dataset to ascertain that
only inventory-carrying firms are analyzed. The regression analyses are performed with
all three inventory measures.
3.4.3.2. Independent variables
The set of independent variables is discussed next. Variables suggested by inventory
theory are discussed first, followed by a review of the measures of firm power. Unless
otherwise stated, all data are obtained from Standard & Poor’s Compustat database.
? SalesForecast
itf
Inventory ordering decisions are made based on expected demand. For each firm and
time period, annual sales are forecast as follows:
( )
1
1 2 t
t
S S g
?
= ? + , where the
average growth rate over the past two years
( )
g is defined as
( ) ( )
2 3 3 1 2 2
2
t t t t t t
S S S S S S
g
? ? ? ? ? ?
? + ?
= . When only incomplete prior sales data are
available or prior sales were impacted by merger and acquisition (M&A) activity, the
average growth rate equals the growth rate for the years for which data are available
and no M&A activity was observed.
? SalesSurprise
itf
While firms make inventory decisions based on expected demand, the magnitude of
inventories at the end of the year is impacted by actual demand. If actual demand
exceeds expected demand, year-end inventory levels should be lower. Conversely,
132
lower than expected sales should result in larger year-end inventories. The
SalesSurprise variable provides some control for the difference between expected
demand (SalesForecast) and realized demand. Following the procedure suggested by
Roumiantsev and Netessine (2007), a binary variable is created. Specifically, this
variable equals “1” if actual demand is greater than expected demand and is
“0”otherwise.
? SalesVariability
itf
SalesVariability is measured as the coefficient of variation of sales and is a proxy for
demand variability. The more variable firm demand, the more inventory a firm will
hold, ceteris paribus. While sales may not be equal to actual demand in case of
stockouts, demand variability is approximated with sales variability. The coefficient
of variation of sales is computed as the ratio of the standard deviation of sales over
the past three periods and the mean of sales over the past three periods:
( )
( )
1, 2, 3
1, 2, 3
Sales i t t t f
itf
Sales i t t t f
CVS
?
µ
? ? ?
? ? ?
= . The SalesVariability variable thus is a standardized
measure of the variability of sales.
? SetupCost
itf
Information on firms’ average cost of setting up production or placing orders is not
readily available. The magnitude of setup costs may, however, be reflected in the
magnitude of firms’ order backlogs. Clearly, there are many reasons why firms
backlog orders: high demand, long lead times, or manufacturing problems are just a
few potential causes of order backlogs. On average, however, larger backlogs may
simply reflect higher order setup costs: Firms may prefer to accumulate orders before
starting production if the cost of setting up production is high. Since the absolute
133
value of backlogged orders is likely highly correlated with sales, the standardized
value of order backlogs is used as a proxy for production setup costs:
itf
itf
itf
OrderBacklog
SetupCost
Sales
= .
? HoldingCost
itf
Inventory holding costs consist of warehousing/handling costs and capital carrying
costs (Timme 2003). While the former component cannot be estimated based on
available accounting information, the latter can be approximated as follows: The
capital cost of holding inventory is a function of the capital interest rate which
represents either the opportunity cost of internally financed (inventory) investments
or the borrowing cost of externally financed (inventory) investments. The firm-
specific interest rate is approximated by dividing the firm’s interest expenses by total
debts:
itf
itf
itf
InterestExpenses
HoldingCost
TotalDebt
= .
? LeadTime
itf
The measure of lead times follows the novel procedure suggested by Roumiantsev
and Netessine (2007). Roumiantsev and Netessine propose the following measure:
365
itf
itf
itf
AP
LeadTime
COGS
?
= where
itf
AP stands for Accounts Payable and
itf
COGS
stands for the Cost of Goods Sold. While not a measure of physical lead times, this
proxy captures the quarterly cash conversion cycle which may, to some extent reflect
physical shipment times. Roumiantsev and Netessine (2007) further justify the use of
this measure as follows:
“[A]ccounts payable are credited, then [the] product is shipped and is typically debited,
134
then it is received and [payment is made]. Hence, financial transactions are correlated with
times of shipment and delivery of inputs and therefore are correlated with the lag a
company has to respond to changing market environment by adjusting inventories.“ (p.13).
Roumiantsev and Netessine (2007) empirically verify “that the lead time proxy is not
dominated by standard payment terms (e.g. 30 or 60 days)” (p.14), and that lead times
are not merely a function of firm power (as measured by the firm’s market share)
58
. In
the data sets used here, the correlation coefficients between LeadTime and the firm
power measures are small and negative ( 0.06
LeadTime MarketShare
r
?
= ? and
0.08
LeadTime IndSalesNet
r
?
= ? ; see Table 10), suggesting that more powerful firms tend to
have shorter permissible payment delays. It may also be argued that longer payment
lead times result from a distressed firm’s inability to pay suppliers (implying inflated
accounts payable). The correlation coefficients between LeadTime and Distress are
positive and statistically significant in both data sets, although limited in magnitude
( 0.19 r ? ; see Table 10). This suggests that the lead time proxy used here may indeed
be a function of, amongst other factors, financial distress. Given the lack of suitable
alternative lead time proxies and the limited size of the distress-lead time correlation,
the LeadTime proxy is included in the subsequent analyses. The lead time proxy
yields the expected positive sign in Roumiantsev and Netessine’s (2007) analyses of
firm inventories.
? Distress
itf
Distress is a measure of firm f’s financial distress. The Distress variable is the
negative value of a firm’s Z score, a measure which was first developed by Altman
58
Firms with greater levels of power may be able to squeeze their suppliers and obtain longer permissible
payment delays, thus increasing accounts payable.
135
(1968). Based on discriminant analysis, Altman (1968) developed the following
model to estimate a firm’s financial fitness:
1 2 3 4 5
0.012* 0.014* 0.033* 0.006* 0.999* Z X X X X X = + + + + ,
where X
1
= working capital / total assets, X
2
= retained earnings / total assets, X
3
=
Earnings Before Interests and Taxes (EBIT) / total assets, X
4
= market value of equity
/ total liabilities, and X
5
= sales / total assets. The information needed to compute the
Z scores is included in the firms’ balance sheets and profit and loss statements. These
data, as well as stock market data are obtained from Standard & Poor’s Compustat
database. High Z scores indicate financial health, while low and negative scores
indicate (serious) financial distress. Specifically, a score of 2.67 or above indicates
financial health, and a score of 1.81 or below suggests (severe) financial distress
(Altman 2002). The Z scores are then rescaled to indicate the level of financial
distress, i.e. ( ) 1
itf
Distress ZScore = ? ? , such that higher (positive) Distress scores
indicate greater financial distress (see also Ferrier et al. 2002). The Distress variable
is included to test the effect of firm financial distress on inventories as hypothesized
in Hypothesis 8.
? DistressDummy
itf
While Distress is a continuous variable, DistressDummy is a binary variable which
indicates whether a carrier is considered financially distressed. Firms are categorized
as distressed and non-distressed based on the above-mentioned cutoff levels
suggested by Altman (1968). Specifically, firms with Z scores of less than 1.81 (i.e.
Distress scores of greater than -1.81) are considered financially distressed
(DistressDummy equals “1”). The sensitivity of the results with respect to the
136
definition of this cutoff value is investigated in Section 3.5. The DistressDummy
variable is used to investigate if distressed firms, on average, hold less inventory
(Hypothesis 8), and to split the data samples into distressed and non-distressed firms
(see Chapter 3.4.6 for further detail).
This study adds to prior research by investigating the effects of firm buying and selling
power on inventories and on the distress-inventory relationship. In the past, researchers
have used relatively simple measures of firm power and have ignored the inherently
dyadic nature of power (Cox 2001, Cox et al. 2001). Amihud and Mendelson (1989), for
example approximate firm power with either the firm’s market share or the firm’s gross
profit margin. Blazenko and Vandezande (2003), in turn, use a market concentration
measure to approximate the average level of power firms possess in a particular industry.
These measures may not fully capture inter-firm power balances. Unlike prior research,
this study uses a set of firm power measures which proxy not only the focal firm’s power,
but also the power levels in the supplying and buying industries. Since a focal firm’s
specific supply chain transaction partners (i.e. buyers and suppliers) cannot be identified
using accounting data, buyer and supplier industry characteristics are used as proxies of
buyer and supplier power. The measures of focal firm, supplier industry and buyer
industry power are presented below.
? IndustrySalesNet
itf
Many prior studies have used market shares
FirmSales
MarketShare
IndustrySales
| |
=
|
\ ¹
to
approximate a firm’s power (e.g. Amihud and Mendelson 1989). The regression
model established in Section 3.4.2, however, already contains the SalesForecast
137
variable and thus controls for the magnitude of firm sales. Including market shares in
the regression model would thus entail two potential problems: First, multicollinearity
problems may arise given the high correlation between sales forecasts and market
shares, thus resulting in inefficient estimates. Second, and perhaps more importantly,
the market share variable might then pick up a size effect (larger firms hold more
inventory) rather than the firm power effect it is intended to measure. This issue is
addressed by transforming the market share variable. Since firm sales are already
controlled for by means of the SalesForecast variable, the size of the firm’s
competitors may indicate the level of power a firm exerts in a market. Specifically,
the IndustrySalesNet variable indicates the sales volume (measured in U.S. $) of all
the other firms in the market (excluding the focal firm). To simplify the interpretation
of the coefficient estimates, the industry sales volume (net of firm sales) is inverted so
as to represent a proxy measure of a firm’s power:
( )
1
itf
it itf
IndustrySalesNet
IndustrySales FirmSales
=
?
. This variable thus indicates the
effect of an increase (decrease) in the sales volume of a firm’s competitors on the
firm’s inventory holdings after controlling for firm sales (~ SalesForecast).
Specifically, a positive coefficient estimate of the IndustrySalesNet variable would
suggest the following: The smaller the firm’s competitors, i.e. the more powerful the
focal firm, the more inventory the (focal) firm will hold. Conversely, a negative
coefficient would confirm the expectation expressed in Hypothesis 9: More powerful
firms, on average, hold less inventory.
For the empirical analyses, industries are defined at the six-digit NAICS level. While
some researchers have computed market shares at the four-digit NAICS level
138
(Amihud and Mendelson 1989), it is believed that the more fine-grained six-digit
NAICS industry data are better suited for the purpose of the analyses. The sensitivity
of the empirical results with respect to the granularity of industry definitions will be
investigated in Section 3.5. Industry sales data are obtained from the website of the
Bureau of Economic Analysis (BEA). These data include the values of exports by
U.S. firms, but do not include imports from foreign firms.
? IndCR4
it
Researchers in the industrial organization economics area have suggested that the
level of market concentration is an indicator of the competitiveness of markets (e.g.
Ravenscraft 1983). Specifically, firms in more concentrated markets are believed to
be more powerful since there are fewer competitors and collusion between firms is
easier to achieve (Waldman and Jensen 2001). While Blazenko and Vandezande
(2003) use two-firm concentration ratios (i.e. the sum of sales of the two largest firms
divided by total sales in the industry) as a measure of market concentration, the
Bureau of Economic Analysis (BEA) publishes 4, 8, 20, and 50 firm industry
concentration ratios. Given its widespread use in the extant literature (see e.g. Pryor
2001, Ravenscraft 1983), the four-firm concentration ratio (CR4) is used here.
Holding all else constant, greater values of the four-firm concentration ratio imply
greater firm power. IndCR4 thus is one of the measures of firm power used to test
Hypothesis 9 to Hypothesis 14 in the analysis of the second data set (part II) in which
generic industrial supply chains are constructed.
As with IndustrySales, industry concentration is measured at the six-digit NAICS
level. Industry concentration data are provided by the Bureau of Economic Analyses
139
and are available in Economic Census years only. The IndCR4 measure is therefore
included in the second data set (part II) only.
? SupplyCR4
(i-1)t
Analogous to the IndCR4 measure, SupplyCR4 is an indicator of the weighted
average concentration levels of those industries that sell to a focal industry. The
Input-Output Tables published by the Bureau of Economic Analysis, illustrate the
flow of goods (and services) between industries. Specifically, the I-O Tables not only
identify those industries that sell goods to another industry, but also indicate the value
of the respective transactions. As a result, the relative importance of supplying
industries to a focal industry can be evaluated and the weighted average four-firm
concentration ratio of the supplying industries can be computed (SupplyCR4) as
illustrated in Figure 13 below. As discussed in Section 3.4.1, this study focuses on
inventory-carrying industries. The average supplying industry concentration
measures, therefore, are based on the four-firm concentration ratios of manufacturing,
wholesale and retail trade industries only. Other (e.g. service) industries are not
considered in the computation of the SupplyCR4 measures. Moreover, only domestic
suppliers are considered when computing the supplying industry concentration ratio;
imports from foreign suppliers are not included.
Holding all else constant, an increase in the SupplyCR4 measure, suggests a relative
decrease in the focal industry’s (and focal firm’s) power. This variable is thus used to
test Hypothesis 10 and Hypothesis 13.
? BuyCR4
(i+1)t
Symmetrical to the SupplyCR4 measure, the BuyCR4 measure is the weighted average
140
of the four-firm concentration ratios of a focal industry’s buying industries (see
Figure 13). Holding all else constant, an increase in the BuyCR4 measure, suggests a
relative decrease in the focal industry’s (and focal firm’s) power. This variable is thus
used to test Hypothesis 11 and Hypothesis 14.
Figure 13: Illustration of the construction of industrial supply chains
There are three widely used methods of inventory accounting: The First In, First Out
(FIFO) method values inventories assuming that items are sold out of inventory in the
same order they were inventoried. Hence, the cost of the most recently added items
determines the value of end-of-period inventories. The Last In, First Out (LIFO) method
values inventories assuming that the most recently inventoried items are sold first.
Consequently, at the end of the accounting period, the oldest items are left over in
inventory. The Average Cost method values inventories at the weighted average cost of
40%
25%
20%
15%
Industry B
Industry C
Industry D
Industry E
40%
25%
20%
15%
Industry B
Industry C
Industry D
Industry E
40%
25%
20%
15%
Industry B
Industry C
Industry D
Industry E
4Supply CR
Industry F
35%
20%
15%
15%
15%
Industry G
Industry H
Industry K
Industry L
4Buy CR
Industry F
35%
20%
15%
15%
15%
Industry G
Industry H
Industry K
Industry L
4Buy CR
Industry A
Industry A
Focal Firm
Industry A
Industry A
Focal Firm
141
all units available for sale during the accounting period. As prices typically change over
time, each inventory accounting method results in different inventory valuations at the
end of the accounting period. Specifically, with generally increasing prices, the use of the
LIFO method will understate the true value of inventories, whereas the FIFO method
more appropriately reflects the value of ending inventories. The average cost procedure
results in inventory values that lie between LIFO and FIFO. To account for these
differences, it is common practice to include an indicator variable which identifies those
firms that use one of the “extreme” accounting methods. Following the example of prior
research (e.g. Blazenko and Vandezande 2003), two binary variables are included in the
model to account for differences in inventory accounting methods:
? LIFO
itf
This indicator variable equals “1” if the firm uses LIFO as the primary inventory
accounting method and equals “0” otherwise (see also Roumiantsev and Netessine
2007).
? AvgCost
itf
This indicator variable equals “1” if the firm uses the average cost method as the
primary inventory accounting method and equals “0” otherwise. (see also
Roumiantsev and Netessine 2007)
3.4.4. Data sources
All firm-level data are obtained from the Compustat database which is maintained by
Standard & Poor’s. This database includes accounting information on publicly traded
142
firms. While the focus on public companies excludes smaller, not publicly traded firms
from the analyses, this selection also ensures that all reported operating and financial data
conform to Generally Accepted Accounting Principles (GAAP) (Roumiantsev and
Netessine 2007). The Compustat database contains firm specific accounting data which
are commonly found in balance sheets and profit and loss statements, including all the
information that is required to construct the firm-specific variables.
Industry level data, most notably industry sales, are obtained from the Bureau of
Economic Analysis. Specifically, the BEA provides annual estimates of total industry
shipments (in U.S. dollars) for U.S. manufacturing industries. At the time of writing, data
were available for the time period from 1998 to 2004
59
. As noted above, the availability
of these data thereby defines the timeframe studied in the first part of the data analysis.
Industry level data for the year 1997 were obtained from the U.S. Census Bureau. This
agency’s website provides access to detailed industry statistics collected through the
Economic Census survey (1997). Total industry sales and industry concentration ratios
were collected from the Economic Census website.
The information relating focal firms to buying and supplying industries is found in the
Input-Output Tables, which are also published by the Bureau of Economic Analysis. The
data in these tables summarize the trade flows between industries. Specifically, the “Use”
tables indicate from which industries firms in a particular industry purchased goods and
59
These data are found in the file “GDPbyInd_SHIP_NAICS.xls” available on the BEA website.
143
services and indicate the respective dollar volumes such that the relative importance of
supplying industries can be evaluated. Conversely, the I-O Tables also identify the
industries that purchase from firms in a focal industry, and the relative importance of
buying industries in terms of the shares of total focal industry sales can be inferred.
The major limitation of using I-O Tables is that these tables focus on U.S. domestic firms
only and disregard foreign buyers and suppliers. As a consequence, the industry power
proxies used here (concentration ratios) may overstate the true power levels in these
industries, particularly when foreign firms hold significant market shares in these
industries. By the same token, industry sales data do not include imports from foreign
firms. This may result in incorrect estimates of the true size of industries and of firms’
market shares. It is believed, however, that the I-O Tables provide at least reasonable
estimates of industry characteristics for the purpose of inter-industry comparisons.
3.4.5. Descriptive statistics
This section provides descriptions of the data samples used in this research. Both data
sets (Part I and Part II) are discussed in turn.
3.4.5.1. Descriptive statistics: Part I
As discussed in Section 3.4.1, data from U.S. manufacturing industries (NAICS 3xxxxx)
for the time period from 1998 to 2004 are used for the empirical analyses. All
144
manufacturing firms for which information on all relevant variables were available for
any or all years in the 1998-2004 time period were included in the data set. The firm
observations in this data set represent about 8.5% of total sales and 9.9% of total
inventory holdings by all publicly traded U.S. manufacturing firms that are included in
the Compustat database.
A two-sample Hotelling T-squared test is implemented to evaluate to what extent the
firms included in this data set differ from those firms for which data are available in
Compustat but which are not included in the analyses due to missing data on one or more
variables. Specifically, the Hotelling test compares these two groups on the following
variables: Total inventories, sales, cost of goods sold, total assets and total debt. The test
yields a test statistic of 2.89. This statistic follows an F distribution and, thus, is
statistically significant at the five percent level. This result suggests that the data sample
used for the empirical analysis differs significantly from the population of firms included
in the Compustat database. Upon closer examination of the data, it becomes apparent
that, on average, the sample firms tend to be smaller (in terms of inventories, sales, costs,
assets, and debt) than those firms that are not included in the data sample (see Appendix
6). The results of the analyses presented here may, therefore, not be generalizable to
firms of all size classes.
The composition of the final data set is shown in Table 7. About forty percent of all firm
observations are in the computer and electronics industry. The second largest industry in
this data set is the machinery industry with 852 observations or about sixteen percent of
145
all observations. While the remainder of the data set comprises firms from a broad array
of manufacturing industries, it cannot be ascertained that the empirical results of this
study will be generalizable to all manufacturing industries, given the dominance of the
computer and electronics, and machinery industries
60
.
NAICS Industry N %
334 Computer and electronics 1983 37.9%
333 Machinery 852 16.3%
336 Transportation equipment 341 6.5%
339 Miscellaneous 275 5.3%
335 Electrical equipment 273 5.2%
332 Fabricated metal 252 4.8%
325 Chemical 203 3.9%
315 Apparel 187 3.6%
331 Primary metal 159 3.0%
316 Leather 137 2.6%
326 Plastics and rubber 136 2.6%
337 Furniture 111 2.1%
313 Textile mills 78 1.5%
327 Nonmetallic mineral 69 1.3%
323 Printing 57 1.1%
321 Wood 44 0.8%
322 Paper 36 0.7%
311 Food 26 0.5%
314 Textile products 11 0.2%
312 Beverage and tobacco 6 0.1%
Total 5236 100%
Table 7: Sample composition (Part I)
Table 8 presents the descriptive statistics of this sample. It is noted that raw materials and
finished goods data were not available for all firms. Hence, the sample size is smaller for
these particular variables. There is substantial variability in all variables. In some
instances, however, the standard deviations are larger than the means suggesting
skewness in the data. Consequently, all inventory variables, as well as SalesForecasts
60
In future research, within-industry analyses could be performed.
146
and DaysPayable (LeadTime) are log-transformed prior to the empirical estimation. It is
also noted that, on average, sales forecasts closely approximate actual sales and that
about 27 percent of all observations are for financially distressed firms.
Variable Mean Std. dev. Min Max N
Inventory Total (million $) 103.9 483.36 0.005 12,207 5236
Inventory RawMat (million $) 24.17 78.74 0.002 1,802 4505
Inventory FinGood (milion $) 43.98 249.23 0.001 7,319 4307
Sales (million $) 835.6 5,164.7 0.05 155,974 5236
Sales Forecast (million $) 847.2 5,149.6 0.01 158,827 5236
SalesSurprise 0.49 0.5 0 1 5236
Coeff. of Variation of Sales 0.20 0.19 0.001 1.73 5236
OrderBacklog/Sales 0.32 1.43 0 90.34 5236
Interest Rate 0.18 0.28 0 1 5236
Days Payable 49.69 50.57 1.32 1,248 5236
Distress -3.79 13.90 -333.2 220.9 5236
Distress Dummy 0.27 0.45 0 1 5236
Market Share (6 dig. NAICS) 0.07 0.18 0.000001 1 5236
LIFO 0.14 0.35 0 1 5236
AvgCost 0.09 0.28 0 1 5236
Table 8: Pooled descriptive statistics (Part I)
Table 9 presents the descriptive statistics for distressed and non-distressed firms
separately. The most striking differences are found in raw materials inventories and days
payable outstanding. Specifically, distressed firms appear to hold less raw materials
inventory and have larger accounts payable than non-distressed firms. The latter
observation can likely be attributed to distressed firms’ lower ability to pay. The former
observation, however, is interesting and lends some support for the contention that
147
distressed firms try to reduce inventories
61
. This is most easily done with raw materials
inventories since extant stock can be reduced by consuming materials while not placing
any new raw materials orders.
Variable Mean Std. dev. Min Max N Mean Std. dev. Min Max N
Inventory Total (million $) 104.8 388.85 0.011 8,349 3804 101.5 672.94 0.005 12,207 1432
Inventory RawMat (million $) 27.32 85.02 0.005 1,802 3278 15.76 57.99 0.002 993 1227
Inventory FinGood (milion $) 42.44 142.07 0.001 2,209 3180 48.35 424.88 0.001 7,319 1127
Sales (million $) 829.4 3,340.5 0.05 58,198 3804 851.8 8,241.6 0.05 155,974 1432
Sales Forecast (million $) 836.8 3,380.2 0.09 68,849 3804 875.0 8,163.7 0.01 158,827 1432
SalesSurprise 0.51 0.5 0 1 3804 0.42 0.5 0 1 1432
Coeff. of Variation of Sales 0.18 0.16 0.001 1.73 3804 0.25 0.23 0.002 1.73 1432
OrderBacklog/Sales 0.30 0.75 0 34.0 3804 0.40 2.46 0 90.34 1432
Interest Rate 0.18 0.30 0 1 3804 0.15 0.22 0 1 1432
Days Payable 41.74 27.02 1.78 630 3804 70.82 82.46 1.32 1,248 1432
Distress -6.57 11.82 -333.2 -1.8 3804 3.60 16.14 -1.8 220.9 1432
Market Share (6 dig. NAICS) 0.08 0.19 0.000001 1 3804 0.04 0.12 0.000002 1 1432
LIFO 0.15 0.36 0 1 3804 0.11 0.31 0 1 1432
AvgCost 0.08 0.28 0 1 3804 0.10 0.30 0 1 1432
Non-distressed firms Distressed firms
Table 9: Descriptive statistics (Part I) – distressed vs. non-distressed firms
A two-sample Hotelling T-squared test is performed to assess whether distressed firms
are statistically significantly different from non-distressed firms based on the variables
listed in Table 9. The test yields a test statistic of F = 58.23 which is statistically
significant at the one percent level, indicating that distressed firms, on average, differ
from non-distressed firms.
61
An alternative explanation may be that suppliers are reluctant to sell to distressed firms, especially when
the latter purchase on credit.
148
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1 Inventory Total (million $)
2 Inventory RawMat (million $) 0.89
3 Inventory FinGood (million $) 0.91 0.74
4 Sales (million $) 0.94 0.85 0.87
5 Sales Forecast (million $) 0.92 0.83 0.85 0.98
6 Sales Surprise 0.06 0.05 0.04 0.08 -0.06
7 Coeff. of Variation of Sales -0.25 -0.19 -0.24 -0.25 -0.23 -0.06
8 OrderBacklog/Sales -0.03 -0.05 -0.05 -0.06 -0.05 -0.01 0.06
9 Interest Rate -0.11 -0.12 -0.11 -0.10 -0.11 0.00 0.05 0.03
10 Days Payable -0.14 -0.09 -0.10 -0.17 -0.17 0.03 0.21 0.01 -0.02
11 Distress -0.14 -0.10 -0.07 -0.15 -0.14 -0.05 0.04 -0.01 -0.02 0.19
12 Distress Dummy -0.26 -0.23 -0.21 -0.28 -0.26 -0.08 0.17 0.03 -0.05 0.28 0.33
13 Market Share (6 dig. NAICS) 0.53 0.45 0.51 0.54 0.53 0.05 -0.12 0.00 -0.04 -0.06 -0.02 -0.11
14 Net Industry Sales (inverted) 0.22 0.20 0.22 0.22 0.21 0.01 -0.10 0.00 -0.03 -0.08 0.03 -0.07 0.75
15 LIFO 0.29 0.24 0.29 0.29 0.29 0.00 -0.18 -0.03 -0.04 -0.15 0.01 -0.06 0.15 0.09
16 AvgCost 0.01 -0.03 0.00 0.01 0.00 0.00 0.00 0.01 0.02 0.02 -0.01 0.02 0.01 -0.03 -0.12
(Values in bold are significant at the 5% level)
Table 10: Pairwise correlations (Part I)
149
Pairwise correlations are displayed in Table 10. Several observations are worth noting:
? As expected, all size variables (inventories, sales, forecasts) are highly and positively
correlated. Market shares are also highly correlated with sales and forecasts (as
discussed previously in Section 3.4.3.2), with correlation coefficients of up to 0.54.
While NetIndustrySales (inverted) are also significantly and positively correlated with
the size variables, the correlation coefficients are much smaller in magnitude (about
0.20 to 0.22).
? The correlation coefficient of 0.98 between actual and forecasted sales is an
indication of the good quality of the sales forecasts.
? In line with the hypotheses presented here, the Distress variable is negatively
correlated with the inventory variables. There are, however, no excessive correlations
between the distress measures and other independent variables.
3.4.5.2. Descriptive statistics: Part II
The second data set (Part II) differs from the first data set (Part I) in that it comprises
observations from a cross-section of U.S. manufacturing firms for the year 1997 only.
The firm observations in this data set represent about 10.7% of total sales and 12.4% of
total inventory holdings by all publicly traded U.S. manufacturing firms that are included
in the Compustat database.
A two-sample Hotelling T-squared test is implemented to investigate potential
150
differences between those firms that are included in the data sample and those firms that
are not included in the empirical analyses due to missing data. The Hotelling test
compares these two groups on the following variables: Total inventories, sales, cost of
goods sold, total assets and total debt. The implementation of this test yields a test
statistic of 4.15 which is statistically significant at the one percent level. On average, the
sampled firms tend to be smaller (in terms of inventories, sales, costs, assets, and debt)
than those firms that are not included in the data sample (see Appendix 6). It is therefore
noted that the results of the analyses presented here may not be generalizable to firms of
all size classes.
The composition of the second data set is very similar to that of the first data set: 446 out
of 755 observations are from firms in the computer and electronics, and machinery
industries (see Table 11). The remainder of the data sample comprises observations of
firms from broad variety of U.S. manufacturing industries.
151
NAICS Industry N %
334 Computer and electronics 291 38.5%
333 Machinery 155 20.5%
335 Electrical equipment 48 6.4%
339 Miscellaneous 47 6.2%
336 Transportation equipment 46 6.1%
332 Fabricated metal 34 4.5%
331 Primary metal 28 3.7%
337 Furniture 21 2.8%
313 Textile mills 16 2.1%
327 Nonmetallic mineral 14 1.9%
325 Chemical 14 1.9%
321 Wood 11 1.5%
323 Printing 8 1.1%
322 Paper 6 0.8%
311 Food 5 0.7%
316 Leather 3 0.4%
326 Plastics and rubber 3 0.4%
315 Apparel 3 0.4%
314 Textile products 2 0.3%
Total 755 100%
Table 11: Sample composition (Part II)
Table 12 presents the descriptive statistics of this sample. The conclusions that can be
drawn upon observing these statistics are consistent with what was noted about the first
data set. There is substantial variability in all variables. The inventory variables,
SalesForecasts and DaysPayable (LeadTime), however, have particularly large standard
deviations relative to the means and are log-transformed. It is further noted that the sales
forecasts are, on average, substantially larger than actual sales. This result is driven by a
relatively small set of observations for which the particular forecasting technique
employed
62
here resulted in substantial overpredictions. The log-transformation of the
SalesForecast variable deemphasizes the impact these outliers have on the regression
62
Forecasts were calculated based on prior year sales which were progressed using the average sales
growth rate over the previous three years (see Section 3.4.3.2 for more detail).
152
estimates, such that the inferior quality of the sales forecasts is not a great concern
63
.
Compared to the first data set, this data sample also contains three new variables:
IndCR4, SupplyCR4, BuyCR4. The data in Table 12 indicate that, on average, the four
largest firms in the focal, supplying and buying industries control between 29 and 38
percent of the market.
Variable Mean Std. dev. Min Max N
Inventory Total (million $) 110.9 623.42 0.036 12,102 755
Inventory RawMat (million $) 20.92 51.38 0 728 678
Inventory FinGood (milion $) 39.02 296.35 0 7,347 656
Sales (million $) 835.2 6,161.3 0.44 154,329 755
Sales Forecast (million $) 1412.4 17,899.8 0.02 465,806 753
SalesSurprise 0.51 0.5 0 1 755
Coeff. of Variation of Sales 0.23 0.22 0.005 1.72 755
OrderBacklog/Sales 0.38 1.42 0 36.99 755
Interest Rate 0.16 0.58 0 11.33 755
Days Payable 38.26 41.79 2.67 736 755
Distress -4.80 9.62 -114.4 50.5 755
Distress Dummy 0.18 0.39 0 1 755
Market Share (6 dig. NAICS) 0.05 0.13 0.00002 1 755
IndCR4 37.74 16.37 4.6 94.5 755
SupplyCR4 28.96 6.71 14.8 83.2 755
BuyCR4 37.38 15.08 6.8 86.9 755
LIFO 0.16 0.36 0 1 755
AvgCost 0.08 0.27 0 1 755
Table 12: Pooled descriptive statistics (Part II)
63
The correlation coefficient between (logged) Sales and (logged) SalesForecasts is r = 0.97 (see Table
36).
153
Table 13 presents the split-sample comparison between distressed and non-distressed
firms. In this sample, distressed firms appear to be larger than non-distressed firms and
therefore tend to hold more inventory. At the same time, distressed firms, on average,
have smaller market shares than non-distressed firms. This may be an indication that the
distressed firms tend to be concentrated in some (larger) industry sectors.
The result of a two-sample Hotelling T-squared test suggests that, overall, distressed
firms are statistically significantly different from non-distressed firms. The test statistic is
F = 8.0767 which is statistically significant at the one percent level.
Variable Mean Std. dev. Min Max N Mean Std. dev. Min Max N
Inventory Total (million $) 101.0 442.36 0.044 8,967 617 155.0 1121.07 0.036 12,102 138
Inventory RawMat (million $) 22.26 50.21 0 728 561 14.49 56.45 0 509 117
Inventory FinGood (milion $) 31.41 84.39 0 1,078 544 76.01 694.06 0 7,347 112
Sales (million $) 712.7 2,790.0 1.52 45,800 617 1382.9 13,174.0 0.44 154,329 138
Sales Forecast (million $) 1425.7 18,863.6 0.11 465,806 617 1351.9 12,692.2 0.02 147,672 136
SalesSurprise 0.53 0.5 0 1 617 0.42 0.5 0 1 138
Coeff. of Variation of Sales 0.22 0.21 0.005 1.46 617 0.26 0.27 0.01 1.72 138
OrderBacklog/Sales 0.33 0.48 0 5.86 617 0.59 3.17 0 36.99 138
Interest Rate 0.15 0.44 0 6.66 617 0.22 0.98 0.01 11.33 138
Days Payable 34.36 36.43 2.67 736 617 55.70 57.18 5.78 438 138
Distress -6.38 9.41 -114.4 -1.8 617 2.27 6.98 -1.8 50.5 138
Market Share (6 dig. NAICS) 0.06 0.14 0.00002 1 617 0.03 0.09 0.00003 1 138
IndCR4 38.03 16.20 4.6 94.5 617 36.45 17.09 4.6 88.3 138
SupplyCR4 29.10 6.81 16.1 83.2 617 28.34 6.25 14.8 54.3 138
BuyCR4 37.78 15.15 6.8 86.9 617 35.60 14.68 6.8 83.2 138
LIFO 0.16 0.37 0 1 617 0.12 0.33 0 1 138
AvgCost 0.07 0.26 0 1 617 0.11 0.31 0 1 138
Distressed firms Non-distressed firms
Table 13: Descriptive statistics (Part II) – distressed vs. non-distressed firms
Pairwise correlations are presented in Table 14. Again, all size variables are highly
correlated, but no excessive correlations between independent variables are found. Given
the relatively small sample size, few correlation coefficients are statistically significant.
154
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1 Inventory Total (million $)
2 Inventory RawMat (million $) 0.92
3 Inventory FinGood (million $) 0.87 0.76
4 Sales (million $) 0.94 0.88 0.84
5 Sales Forecast (million $) 0.91 0.86 0.81 0.97
6 Sales Surprise -0.12 -0.04 -0.10 -0.13 -0.09
7 Coeff. of Variation of Sales 0.05 0.03 0.05 0.05 -0.09 -0.09
8 OrderBacklog/Sales -0.07 -0.10 -0.08 -0.07 -0.08 0.01 -0.01
9 Interest Rate -0.08 -0.07 -0.06 -0.10 -0.11 0.01 0.06 0.00
10 Days Payable -0.14 -0.11 -0.12 -0.22 -0.21 0.27 -0.03 0.02 0.04
11 Distress -0.03 -0.06 0.00 -0.05 -0.05 -0.15 0.02 0.02 0.01 0.16
12 Distress Dummy -0.31 -0.32 -0.25 -0.33 -0.31 0.07 -0.08 0.07 0.05 0.24 0.35
13 Market Share (6 dig. NAICS) 0.55 0.48 0.52 0.58 0.55 -0.08 0.09 0.00 -0.05 -0.09 0.03 -0.09
14 Net Industry Sales (inverted) 0.11 0.08 0.15 0.11 0.08 -0.04 0.05 -0.04 -0.06 -0.11 0.04 -0.02 0.58
15 IndCR4 0.11 0.04 0.02 0.11 0.11 -0.02 0.05 0.07 0.04 0.07 -0.06 -0.04 0.13 -0.19
16 SupplyCR4 0.08 0.07 -0.01 0.07 0.07 -0.01 0.02 0.09 0.02 0.03 0.00 -0.04 -0.01 -0.31 0.51
17 BuyCR4 0.15 0.12 0.02 0.12 0.12 0.00 0.03 0.09 0.02 0.04 -0.03 -0.06 0.05 -0.24 0.55 0.41
18 LIFO 0.29 0.25 0.31 0.30 0.29 -0.19 0.06 -0.05 -0.06 -0.20 0.05 -0.04 0.16 0.09 -0.11 -0.04 -0.14
19 Avg. Cost 0.01 -0.01 0.05 0.00 -0.01 -0.01 -0.05 0.01 0.06 0.04 -0.05 0.05 0.02 -0.01 0.04 0.00 0.05 -0.13
(Values in bold are significant at the 5% level)
Table 14: Pairwise correlations (Part II)
155
3.4.6. Methodology
The empirical methodology is discussed in this section. A series of regression analyses
are performed to test the hypotheses developed in this essay. An overview of these
regressions is presented in the following subsection. Both the panel data set (Part I) and
the cross-sectional data set (Part II) present particular econometric challenges that have to
be considered when choosing an empirical estimation procedure. The methodologies for
the analyses of both data sets are discussed in turn in Subsections 3.4.6.2 and 3.4.6.3.
3.4.6.1. Overview of regression analyses
The hypotheses set forth in this essay are tested by means of a series of regression
analyses. Table 15 provides an overview of the regressions that are performed.
Data Part I Data Part II
Dependent
variable
Baseline Split-sample Distressed firms
with interaction
effect
Baseline Split-sample Distressed firms
with interaction
effect
Total
inventory
R1 R2 R3 R10 R11 R12
Raw
materials
inventory
R4 R5 R6 R13 R14 R15
Finished
goods
inventory
R7 R8 R9 R16 R17 R18
Table 15: Overview of regression analyses
156
As described previously, two separate data sets are used for the empirical analyses. Nine
regressions are performed to analyze each data set (R1-R9 and R10-R18). For the
analysis of each data set, three lines of regressions are required to estimate the model for
three different dependent variables: Total inventory, raw materials inventories, and
finished goods inventories. For each dependent variable, a baseline regression using the
full data set is implemented first. In a second step, the data set is split into distressed and
non-distressed firms, and the regression is implemented for both subsamples separately.
The Distress*Power interaction effect is included in the third regression model which is
implemented using the subsample of distressed firms only and designed to test
Hypothesis 12 to Hypothesis 14.
The regression models are further discussed in the following paragraphs. The models
below show TotalInventory as the dependent variable. The same models are also
analyzed with raw materials inventories and finished goods inventories as the dependent
variables.
The first regression (R1) estimates the baseline model shown below. This regression is
performed using the entire data sample (part I). The measures of financial distress
(DistressDummy) and of firm power (IndSalesNet) are of particular interest. It is expected
that, on average, distressed firms hold less inventory than non-distressed firms.
157
(R1) lnTotalInventory
itf
= ?
0
+ ?
1
lnSalesForecast
itf
+ ?
2
SalesSurprise
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
lnLeadTime
itf
+ ?
7
DistressDummy
itf
+ ?
8
lnIndSalesNet
itf
+ ?
9
LIFO
itf
+ ?
10
AvgCost + ?
itf
The second regression (R2) is nearly identical to R1. This regression however, is
performed for distressed and non-distressed firms separately, using the DistressDummy
variable to split the sample into these groups. The continuous Distress variable then
replaces the DistressDummy variable in the model. It is expected that greater levels of
financial distress result in lower inventory levels for distressed firms
64
.
(R2) lnTotalInventory
itf
= ?
0
+ ?
1
lnSalesForecast
itf
+ ?
2
SalesSurprise
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
lnLeadTime
itf
+ ?
7
Distress
itf
+ ?
8
lnIndSalesNet
itf
+ ?
9
LIFO
itf
+ ?
10
AvgCost + ?
itf
The third regression (R3) is similar to R2 for distressed firms, the only difference being
that the Distress*IndSalesNet interaction term is included to test the contention that the
(negative) effect of financial distress on inventories increases with the firm’s power.
(R3) lnTotalInventory
itf
= ?
0
+ ?
1
lnSalesForecast
itf
+ ?
2
SalesSurprise
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
lnLeadTime
itf
+ ?
7
Distress
itf
+ ?
8
lnIndSalesNet
itf
+ ?
9
LIFO
itf
+ ?
10
AvgCost
+ ?
11
Distress
itf
* lnIndSalesNet
itf
+ ?
itf
64
This study focuses on the analysis of financially distressed firms’ inventories. The effect of financial
health on inventories is not investigated here.
158
The regression models used to analyze the second data sample (part II) are similar to
those described above.
Regression 10 (R10) estimates the baseline model which includes the focal industry’s
four-firm concentration ratio, as well as the weighted average concentration ratios of the
supplying and buying industry in addition to the variables included in R1. The new
variables are added to approximate firms’ supply chain power. R10 is performed using
the entire data sample (part II).
(R10) lnTotalInventory
itf
= ?
0
+ ?
1
lnSalesForecast
itf
+ ?
2
SalesSurprise
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
lnLeadTime
itf
+ ?
7
DistressDummy
itf
+ ?
8
lnIndSalesNet
itf
+ ?
9
IndCR4
itf
+ ?
10
SupplyCR4
itf
+ ?
11
BuyCR4
itf
+ ?
12
LIFO
itf
+ ?
13
AvgCost + ?
itf
Regression 11 (R11) is performed for distressed and non-distressed firms separately,
similar to R2. The continuous Distress variable then replaces the DistressDummy variable
in model R10.
(R11) lnTotalInventory
itf
= ?
0
+ ?
1
lnSalesForecast
itf
+ ?
2
SalesSurprise
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
lnLeadTime
itf
+ ?
7
Distress
itf
+ ?
8
lnIndSalesNet
itf
+ ?
9
IndCR4
itf
+ ?
10
SupplyCR4
itf
+ ?
11
BuyCR4
itf
+ ?
12
LIFO
itf
+ ?
13
AvgCost + ?
itf
159
Building on R11, regression 12 (R12) adds the interaction terms between the Distress and
IndSalesNet, IndCR4, SupplyCR4, and BuyCR4 variables, respectively.
(R12) lnTotalInventory
itf
= ?
0
+ ?
1
lnSalesForecast
itf
+ ?
2
SalesSurprise
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
lnLeadTime
itf
+ ?
7
Distress
itf
+ ?
8
lnIndSalesNet
itf
+ ?
9
IndCR4
itf
+ ?
10
SupplyCR4
itf
+ ?
11
BuyCR4
itf
+ ?
12
Distress
itf
* lnIndSalesNet
itf
+ ?
13
Distress
itf
* IndCR4
itf
+ ?
14
Distress
itf
* SupplyCR4
itf
+ ?
15
Distress
itf
* BuyCR4
itf
+ ?
16
LIFO
itf
+ ?
17
AvgCost + ?
itf
3.4.6.2. Empirical methodology: Part I
As discussed in Chapter 2, the OLS assumptions of homoskedasticity and independence
are frequently not met when dealing with cross-sectional time series data (Greene 2003).
Tests for heteroskedasticity and autocorrelation of the error terms are implemented prior
to selecting the appropriate empirical estimation procedure.
The Breusch-Pagan/Cook-Weisberg Lagrange multiplier test (Breusch and Pagan 1979,
Cook and Weisberg 1983) evaluates the correlation between the residuals of an OLS
regression and the dependent variable (e.g. TotalInventory). If no such correlation is
found, the homoskedasticity assumption is valid and OLS regressions can be assumed to
provide efficient and unbiased estimates. The test is implemented after estimating
160
regression R1 (see Table 15) using the OLS procedure. The resulting test statistic is
844.18 which follows a ?
2
distribution. This result is statistically significant at the less
than one percent level and suggests that the magnitude of the residuals varies with the
levels of the dependent variable (heteroskedasticity).
The Wooldridge test for autocorrelation in panel data (Drukker 2003, Wooldridge 2002)
is implemented to determine if the residuals are serially correlated over time. This test is
particularly suitable for panel data sets since it evaluates serial correlations within panels
only. The test statistic is F = 144.745 and is significant at the one percent level. This
suggests the presence of first-order autocorrelation.
A generalized least squares procedure (GLS) is recommended for the analysis of panel
data with heteroskedastic and serially correlated error terms (Greene 2003). As noted in
Section 2.3.4, the GLS procedure can be implemented with the unobserved cross-
sectional and time effects modeled as either random or fixed effects. The appropriate
procedure (fixed effects or random effects) is determined by implementing the Hausman
specification test (Hausman 1978). This test analyzes whether the error terms are
independent of the independent variables. If that is the case, the random effects procedure
is preferred, while the fixed effects procedure should be selected otherwise. The test
produces a ?
2
distributed statistic of W = 845.51 which is significant at the less than one
percent level. The null hypothesis of no correlation is therefore clearly rejected,
suggesting that the fixed effects model should be selected.
161
The STATA software package is used for the empirical analyses. This software lets users
specify the way in which the first-order autocorrelation of the error terms should be
modeled. The default method is to compute the autocorrelation based on the Durbin-
Watson statistic. This method is applied here, although the results are found to be largely
insensitive to the way in which autocorrelation is computed.
3.4.6.3. Empirical methodology: Part II
The second data set used for the empirical analyses contains firm-level observations from
the year 1997. Given that there is only one observation per firm (rather than a time series
of firm-level observations), serial correlation of the error terms is not a concern with this
data set. Heteroskedasticity may, however, be observed. Therefore, the Breusch-
Pagan/Cook-Weisberg Lagrange multiplier test (Breusch and Pagan 1979, Cook and
Weisberg 1983) is implemented after an OLS regression (R10, see Table 15). The test
statistic is 23.81 with a ?
2
distribution. This result is statistically significant at the one
percent level. This indicates that the magnitude of the residuals varies with the levels of
the dependent variable (heteroskedasticity). This constitutes a violation of the OLS
assumption of homoskedasticity.
Robust estimation techniques provide a mechanism to control the heteroskedasticity of
errors. While the coefficient estimates themselves remain unchanged relative to the
standard OLS estimation procedure, the values of standard errors are adjusted for
correlations across observations. The robust regression procedure in STATA uses Huber-
162
White sandwich estimators to compute robust standard errors (White 1980). The
empirical results are presented and discussed in the following section.
3.5. Empirical results and discussion
The empirical analyses are performed for both data sets (Part I and Part II) separately.
The regression results are discussed in Subsections 3.5.1 and 3.5.2, respectively, and the
empirical support for the hypotheses set forth in this paper is evaluated.
3.5.1. Empirical results: Part I
In this section, the results of the analyses of data set Part I are discussed in four
subsections:
? First, the regression results for TotalInventory as the dependent variable are
presented. Specifically, the baseline regression (R1, see Table 15) and the split-
sample regression (R2) results are reported, and the interaction between distress
and power is evaluated for distressed firms (R3).
? Second, the sensitivity of the regression results (R1) with respect to the definition
of the DistressDummy variable and the granularity of industry definitions (6-digit
NAICS versus 4-digit NAICS) is also evaluated.
? Third, the regression results for RawMatInventory (raw materials inventory) as
the dependent variable are discussed (R4-R6, see Table 15).
? Fourth, the regression results for FinGoodInventory (finished goods inventory) as
163
the dependent variable are discussed (R7-R9, see Table 15).
3.5.1.1. Regression results: Total inventory
As discussed previously, the baseline regression is specified as follows:
(R1) lnTotalInventory
itf
= ?
0
+ ?
1
lnSalesForecast
itf
+ ?
2
SalesSurprise
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
lnLeadTime
itf
+ ?
7
DistressDummy
itf
+ ?
8
lnIndSalesNet
itf
+ ?
9
LIFO
itf
+ ?
10
AvgCost + ?
itf
This model is tested using the panel data set described in Section 3.4.5.1 and the
autoregressive linear regression estimation procedure outlined in Section 3.4.6.2. The
empirical estimation results are presented in Table 16.
The model’s F statistic (F = 135.8) is statistically significant at the one percent level, and
the R-squared within statistic is 0.33 indicating that the model explains about one third of
the variability in the dependent variable. The coefficient estimates are discussed below:
? Forecast: Higher expected demand should result in larger order quantities and
larger average cycle stocks. This expectation is confirmed by the positive and
significant coefficient ( ) 0.361 ? = . Specifically, this result suggests that a one
percent increase in expected demand should result in an increase in total inventory
holdings by 0.361%. It is noted that this result is consistent with Ballou’s (1981)
contention that inventories should increase as the square-root of demand.
? SalesSurprise: Greater than expected demand should result in lower end-of-period
inventory holdings. This variable’s coefficient ( ) 0.202 ? = , however, is positive
164
and significant. One explanation may be that firms build up inventory once it
becomes apparent that demand may exceed expectations.
? Coefficient of Variation of Sales: Greater demand variability should result in
larger safety stocks and, thus, greater inventory levels. The coefficient estimate
( ) 0.027 ? = ? , however, is statistically insignificant.
? OrderBacklog/Sales: The standardized value of order backlogs is used as a proxy
for production setup costs. The higher this cost, the higher production quantities
and average cycle stocks should be. The coefficient estimate ( ) 0.007 ? = is
positive as expected although only marginally significant.
? InterestRate: The InterestRate measure is used as a proxy for inventory carrying
costs. Higher carrying costs should equal lower inventory levels. The coefficient
estimate ( ) 0.038 ? = ? has the expected sign but is statistically insignificant.
? DaysPayable: Days payable outstanding is used as a proxy for lead times. The
longer the lead times, the more inventory firms should hold. This expectation is
confirmed by the positive and significant coefficient estimate ( ) 0.120 ? = .
? DistressDummy: As discussed previously, the DistressDummy variable identifies
those firms that have high Distress scores and thus find themselves in situations
of financial distress. The key contention of this research is that distressed firms
will hold less inventory, all else equal (Hypothesis 8). The coefficient estimate is
negative and statistically significant ( ) 0.065 ? = ? . This result suggests that, on
average, distressed firms hold 6.5 percent less inventory than financially healthier
firms. This finding provides support for Hypothesis 8.
165
? IndSalesNet: This variable measures a firm’s power in a market relative to its
competitors. As stated in Hypothesis 9, more powerful firms are expected to hold
less inventory, ceteris paribus. The coefficient estimate ( ) 0.105 ? = ? provides
strong support for this hypothesis.
? LIFO and AvgCost: The coefficient estimates of both the LIFO and AvgCost
variables are not statistically significantly different from 0. The results therefore
suggest that in this particular data sample, differences in inventory accounting
methods did not significantly affect inventory valuations.
Total Inv Coef. P>t
Constant -0.317 0.000
Forecast 0.361 0.000
SalesSurprise 0.202 0.000
Coeff. of Variation -0.027 0.621
OrderBacklog/Sales 0.007 0.095
InterestRate -0.038 0.188
DaysPayable 0.120 0.000
DistressDummy -0.065 0.003
IndSalesNet -0.105 0.000
LIFO -0.005 0.928
AvgCost 0.016 0.803
Number of obs 3,862
F(10,2758) 135.8
Prob > F 0.000
R-sq. within 0.330
R-sq. between 0.806
R-sq. overall 0.781
Table 16: Regression results: Total inventory (R1)
The regression model discussed above is also implemented for both distressed and non-
distressed firms separately (R2, see Table 15), using the DistressDummy variable to split
the sample into these groups. In addition, the moderating effect of power (IndSalesNet)
166
on the Distress-Inventory relationship is evaluated by estimating the corresponding
interaction effect for distressed firms (R3). Table 17 presents these regression results.
Total Inv Coef. P>t Coef. P>t Coef. P>t
Constant -0.121 0.014 -0.689 0.000 -0.691 0.000
Forecast 0.432 0.000 0.191 0.000 0.187 0.000
SalesSurprise 0.197 0.000 0.164 0.000 0.162 0.000
Coeff. of Variation 0.044 0.509 -0.025 0.850 -0.022 0.868
OrderBacklog/Sales 0.077 0.058 0.005 0.387 0.005 0.383
InterestRate -0.026 0.369 -0.269 0.004 -0.268 0.005
DaysPayable 0.201 0.000 0.014 0.706 0.013 0.724
Distress -0.001 0.344 -0.011 0.000 -0.015 0.272
IndSalesNet -0.032 0.009 -0.197 0.000 -0.199 0.000
LIFO 0.002 0.970 0.010 0.954 0.010 0.955
AvgCost 0.076 0.235 0.073 0.688 0.076 0.678
Distress*IndSalesNet 0.000 0.733
Number of obs 2,701 857 857
F 157.0 19.4 17.6
Prob > F 0.000 0.000 0.000
R-sq. within 0.458 0.289 0.289
R-sq. between 0.863 0.627 0.616
R-sq. overall 0.842 0.604 0.593
Distressed firms Non-distressed firms
w/o interaction with interaction
Table 17: Split-sample regression results: Total inventory (R2, R3)
The leftmost column of Table 17 shows the regression results for non-distressed, i.e.
healthy firms. It is noted that the coefficient estimates are generally consistent with the
results for the entire data sample (Table 16). The key difference is that the model shown
in Table 17 contains the (continuous) Distress variable. It is interesting to note that the
level of Distress (or financial health in this case
65
) does not appear to impact non-
distressed firms’ inventory holdings.
65
Note that non-distressed firms will have low or negative Distress scores, indicating financial health.
167
The right part of Table 17 shows the regression results for distressed firms. Given the
smaller number of observations (n = 857), the model fit is lower than for non-distressed
firms (F = 19.4, 17.6; R-squared = 0.289). The coefficient estimates are, however,
generally consistent with those for the entire sample (Table 16) and those for non-
distressed firms (Table 17, left column). A few results of the analysis of distressed firms
merit further discussion:
? The coefficients of the OrderBacklog/Sales and DaysPayable variables are
statistically insignificant for distressed firms.
? The InterestRate variable, on the other hand, has a statistically significant
negative coefficient ( ) 0.269 ? = ? .
? The Distress variable carries a negative and statistically significant coefficient
( ) 0.011 ? = ? in the model without the interaction effect. This suggests that, for
distressed firms, greater levels of distress result in even lower inventory levels.
This finding provides further support for Hypothesis 8
66
.
? The interaction effect between Distress and IndSalesNet is added to the model in
the rightmost column of Table 17. The corresponding coefficient estimate is close
to zero and does not add any explanatory power to the model. Hypothesis 12 is,
thus, not supported.
In summary, the regression model is of at least reasonable quality and most coefficient
estimates have the expected signs. In particular, the results indicate that distressed firms
66
A squared term of the Distress variable was also tested to investigate if the relationship between financial
distress and inventories is non-linear. While the results are not reported here, it is noted that the squared
Distress variable carries a negative and significant coefficient, suggesting that the magnitude of the effect
of distress on prices increases with the severity of financial distress.
168
hold less inventory than financially healthy firms (Hypothesis 8), and that greater levels
of distress equate to lower inventory levels. The contention that power moderates the
distress-inventory relationship (Hypothesis 12) is not supported.
3.5.1.2. Sensitivity analyses
The sensitivity of the regression results with respect to the definition of the
DistressDummy, IndSalesNet and SalesSurprise variables is assessed in this section. In
addition, it is investigated if the results hold if average inventories rather than end-of-year
inventories are used as dependent variables
The DistressDummy variable indicates whether a firm has a Distress score of greater than
-1.81. This cutoff level, initially proposed by Altman (1968), is, of course, somewhat
arbitrary. The effect of alternative cutoff definitions on the estimation results is
investigated by comparing the regression results for three distinct cutoff levels.
Figure 14: Alternative definitions of distressed and non-distressed firms
0 -1.81 -2.80
Financial
distress
Financial
health
(Altman 1968) (median)
Distress
Dummy
DistressMed
Dummy
DistressNegZ
Dummy
0 -1.81 -2.80
Financial
distress
Financial
health
(Altman 1968) (median)
Distress
Dummy
DistressMed
Dummy
DistressNegZ
Dummy
169
Figure 14 illustrates the cutoff levels that are proposed here. The standard cutoff level of
-1.81 is shown in the middle of the graph. As seen in Table 12, this results in 27 percent
of the firms in the full data sample being classified as financially distressed. An equal
split into relatively distressed and relatively healthy firm is obtained by using the median
Distress value (-2.80) as a cutoff level. The resulting indicator variable is named
DistressMedDummy. Conversely, a stricter definition of financial distress is obtained by
moving the cutoff level to the right. The DistressNegZDummy variable identifies all
severely distressed firms with negative Z scores (i.e. positive Distress scores). With this
cutoff level, about twelve percent of all firms are considered distressed. Table 18
juxtaposes the regression results for all three cutoff definitions.
TotalInventory Coef. P>t Coef. P>t Coef. P>t
Constant -0.317 0.000 -0.312 0.000 -0.314 0.000
Forecast 0.361 0.000 0.360 0.000 0.361 0.000
SalesSurprise 0.202 0.000 0.200 0.000 0.203 0.000
Coeff. of Variation -0.027 0.621 -0.024 0.665 -0.028 0.617
OrderBacklog/Sales 0.007 0.095 0.005 0.203 0.007 0.094
InterestRate -0.038 0.188 -0.035 0.223 -0.036 0.207
DaysPayable 0.120 0.000 0.121 0.000 0.119 0.000
DistressDummy -0.065 0.003 -0.150 0.000 -0.033 0.113
IndSalesNet -0.105 0.000 -0.105 0.000 -0.106 0.000
LIFO -0.005 0.928 -0.005 0.928 -0.007 0.903
AvgCost 0.016 0.803 0.020 0.760 0.013 0.842
Number of obs 3,862 3,862 3,862
F 135.8 139.0 134.5
Prob > F 0.000 0.000 0.000
R-sq. within 0.330 0.335 0.328
R-sq. between 0.806 0.809 0.804
R-sq. overall 0.781 0.784 0.778
DistressMedDummy DistressDummy DistressNegZDummy
Table 18: Sensitivity analysis: Distressed vs. non-distressed firms
The coefficient estimates are robust across all three cases and there are only minimal
170
variations in the overall fit of the model. Specifically, the only notable result is that the
DistressMedDummy variable does not yield a statistically significant coefficient. This is
not surprising considering that the DistressMedDummy variable is based on a very broad
definition of financial distress. The alternative variables (DistressDummy and
DistressNegZDummy) both carry negative and statistically significant coefficients. It is
therefore concluded that the results are largely insensitive to variations in the definition
of financial distress. The DistressDummy variable (as defined initially) is retained for the
remainder of the analyses.
A second sensitivity analysis is performed to evaluate how the regression results change
as the definition of industries changes. Specifically, industries may be defined at different
levels of granularity. While a narrow definition based on six-digit NAICS codes is used
as a default, the model is re-estimated using the broader four-digit NAICS definition.
These definitions affect the magnitude and variability of the IndSalesNet variable.
Table 19 presents the regression results for both the six and four-digit NAICS industry
definitions. While the broader four-digit NAICS definition provides slightly better results
in terms of model fit than the six-digit NAICS definition, the results are robust and
consistent across both regressions. The six-digit NAICS industry definition is retained to
ensure consistency with the granularity of industry definitions in the second part of the
data analysis
67
.
67
The analysis of the second data set (Part II) introduces the concept of supply chain power by adding
weighted average industry concentration levels in the buying and supplying industries. To obtain sufficient
variability in these variables, industries must be defined at the full six-digit NAICS level.
171
TotalInventory Coef. P>t Coef. P>t
Constant -0.317 0.000 -0.986 0.000
Forecast 0.361 0.000 0.272 0.000
SalesSurprise 0.202 0.000 0.159 0.000
Coeff. of Variation -0.027 0.621 -0.076 0.161
OrderBacklog/Sales 0.007 0.095 0.006 0.128
InterestRate -0.038 0.188 -0.043 0.126
DaysPayable 0.120 0.000 0.077 0.000
DistressDummy -0.065 0.003 -0.072 0.001
IndSalesNet -0.105 0.000 -0.204 0.000
LIFO -0.005 0.928 -0.015 0.779
AvgCost 0.016 0.803 -0.003 0.965
Number of obs. 3,862 3,862
F 135.8 151.4
Prob > F 0.000 0.000
R-sq. within 0.330 0.354
R-sq. between 0.806 0.743
R-sq. overall 0.781 0.721
6-digit NAICS 4-digit NAICS
Table 19: Sensitivity analysis: Granularity of industry definitions
Following the example of Roumiantsev and Netessine (2007), the SalesSurprise variable
is included in the regression to capture the effect of unexpectedly large sales on
inventories. Specifically, this indicator variable takes on the value “1” if actual sales
exceed forecasted sales. Alternatively, the actual value of the difference between actual
sales and forecasted sales, i.e. the forecast error (ForecastError), can be included in the
regression model as a more finegrained measure of the magnitude of the deviation of
actual sales from forecasted sales. The regression results using the SalesSurprise and
ForecastError variables, respectively, are compared in Table 20. The model with the
ForecastError variable is of poorer quality than the model with the SalesSurprise
variable. The signs of the coefficient estimates, however, are consistent across both
models. The SalesSurprise variable is retained for all subsequent analyses.
172
TotalInventory Coef. P>t Coef. P>t
Constant -0.317 0.000 -0.082 0.134
Forecast 0.361 0.000 0.234 0.000
SalesSurprise / Error 0.202 0.000 0.000 0.000
Coeff. of Variation -0.027 0.621 0.001 0.985
OrderBacklog/Sales 0.007 0.095 0.006 0.197
InterestRate -0.038 0.188 -0.039 0.186
DaysPayable 0.120 0.000 0.132 0.000
DistressDummy -0.065 0.003 -0.104 0.000
IndSalesNet6D -0.105 0.000 -0.154 0.000
LIFO -0.005 0.928 0.015 0.798
AvgCost 0.016 0.803 0.053 0.427
Number of obs 3,862 3,862
F 135.8 97.3
Prob > F 0.000 0.000
R-sq. within 0.330 0.261
R-sq. between 0.806 0.561
R-sq. overall 0.781 0.536
SalesSurprise ForecastError
Table 20: Sensitivity analysis: SalesSurprise vs. ForecastError
As noted in Section 3.4.3.1, end-of-year inventories as reported in firms’ balance sheets
are the dependent variables used in this research. It may be argued that end-of-year
inventory values are biased estimates of true average inventory levels as firms may
reduce inventory levels toward the end of the year in order to improve key financial and
operating performance indicators. This concern is addressed as follows: For each firm in
the dataset, an average annual inventory value is approximated by averaging the firm’s
inventory levels at the end of the first, second, third, and fourth quarters. As shown in
Table 8, the mean total inventory (end-of-year inventory values) in the panel data set
(Part I) is $103.9 million. The mean average total inventory, in contrast, is $107.78
million. A paired two-sample t test indicates that end-of-year total inventories and
173
average total inventories are statistically significantly different (t = 4.86, p = 0.000). This
result, thus, is consistent with the above mentioned contention that end-of-year inventory
values may be biased proxies for average inventories. A regression analysis with average
total inventories as the dependent variable is performed and compared to the results with
end-of-year total inventories as the dependent variable. This comparison in shown in
Table 21 below.
Coef. P>t Coef. P>t
Constant -0.317 0.000 -0.279 0.000
Forecast 0.361 0.000 0.376 0.000
SalesSurprise 0.202 0.000 0.174 0.000
Coeff. of Variation -0.027 0.621 0.007 0.903
OrderBacklog/Sales 0.007 0.095 0.006 0.159
InterestRate -0.038 0.188 -0.029 0.299
DaysPayable 0.120 0.000 0.060 0.000
DistressDummy -0.065 0.003 -0.033 0.135
IndSalesNet6D -0.105 0.000 -0.119 0.000
LIFO -0.005 0.928 0.022 0.681
AvgCost 0.016 0.803 0.021 0.740
Number of obs 3,862 3,862
F 135.8 135.0
Prob > F 0.000 0.000
R-sq. within 0.330 0.329
R-sq. between 0.806 0.799
R-sq. overall 0.781 0.774
Total Inventory Average Inventory
Table 21: Sensitivity analysis: Measurement of total inventories
The coefficient estimates are generally consistent across both the TotalInventory and
AverageInventory regressions. The significance levels, however, are weaker when
average inventories are used as the dependent variable. It is noted that there is no
indication suggesting that the use of end-of-year inventories results in biased estimation
results. As has been done in prior research (e.g. Carpenter et al 1994, Roumiantsev and
Netessine 2007), end-of-year inventories are, therefore, retained for the empirical
174
analyses.
The regression results for raw materials inventories are discussed next.
3.5.1.3. Regression results: Raw materials inventory
The full data sample, comprising both distressed and non-distressed firms is used to
estimate the regression model using raw materials inventory as the dependent variable
(R4, see Table 15). The results are reported in Table 22.
The estimation results are consistent with the previously presented results for total
inventories. The sales variability and production setup cost proxies (Coeff. of Variation
and OrderBacklog/Sales, respectively), as well as the LIFO and AvgCost control
variables are the only variables that have statistically insignificant coefficient estimates.
All other variables carry significant coefficients with the expected signs. The
DistressDummy variable is of particular interest. The negative coefficient ( ) 0.096 ? = ?
indicates that distressed firms tend to hold less inventory than their healthier counterparts
(Hypothesis 8).
175
Raw Mat Inv Coef. P>t
Constant -0.724 0.000
Forecast 0.330 0.000
SalesSurprise 0.164 0.000
Coeff. of Variation 0.013 0.858
OrderBacklog/Sales -0.001 0.778
InterestRate -0.136 0.000
DaysPayable 0.073 0.001
DistressDummy -0.096 0.002
IndSalesNet -0.063 0.001
LIFO -0.004 0.951
AvgCost -0.090 0.331
Number of obs 3,288
F(10,2758) 49.1
Prob > F 0.000
R-sq. within 0.174
R-sq. between 0.709
R-sq. overall 0.675
Table 22: Regression results: Raw materials inventory (R4)
The split-sample regression results for non-distressed and distressed firms are shown in
Table 23. Focusing on the results for non-distressed firms in the leftmost column first, it
is interesting to note that greater levels of financial health (Distress) do not impact
inventory holdings ( ) 0.000 ? = . In addition, the insignificant coefficient of the
IndSalesNet variable indicates that financially sound firms do not leverage their power to
push inventory ownership up or down the supply chain. The results for distressed firms
(without interaction effect), in contrast, suggest that greater levels of financial distress
(Distress) and greater levels of power (IndSalesNet) result in lower raw materials
inventory holdings. These findings provide support for Hypothesis 8 and Hypothesis 9,
respectively. Moreover, the significant and negative coefficient of the interaction effect
between the Distress and IndSalesNet variables ( ) 0.004 ? = ? suggests that the
magnitude of the effect of distress on raw materials inventories increases with the firm’s
176
power. This result is consistent with Hypothesis 12.
Raw Mat Inv Coef. P>t Coef. P>t Coef. P>t
Constant -0.195 0.004 -0.902 0.000 -0.920 0.000
Forecast 0.393 0.000 0.160 0.006 0.139 0.019
SalesSurprise 0.180 0.000 0.082 0.098 0.071 0.154
Coeff. of Variation 0.126 0.176 -0.029 0.872 -0.012 0.948
OrderBacklog/Sales 0.088 0.189 0.001 0.904 0.001 0.866
InterestRate -0.131 0.001 -0.394 0.001 -0.396 0.001
DaysPayable 0.073 0.006 0.094 0.057 0.091 0.063
Distress 0.000 0.947 -0.013 0.000 -0.054 0.026
IndSalesNet 0.015 0.458 -0.096 0.073 -0.108 0.044
LIFO -0.012 0.867 -0.044 0.837 -0.046 0.831
AvgCost -0.018 0.855 -0.047 0.824 -0.027 0.899
Distress*IndSalesNet -0.004 0.088
Number of obs 2,304 731 731
F 41.1 7.6 7.2
Prob > F 0.000 0.000 0.000
R-sq. within 0.207 0.160 0.166
R-sq. between 0.687 0.591 0.517
R-sq. overall 0.658 0.552 0.481
Non-distressed firms Distressed firms
w/o interaction with interaction
Table 23: Split-sample regression results: Raw materials inventory (R5, R6)
The regression results for finished goods inventories are discussed in the following
subsection.
3.5.1.4. Regression results: Finished goods inventory
In this section, the regression model is estimated using finished goods inventories as the
dependent variable. The results for the full data set, consisting of both distressed and non-
distressed firms, are provided in Table 24.
177
Fin Good Inv Coef. P>t
Constant -0.677 0.000
Forecast 0.357 0.000
SalesSurprise 0.164 0.000
Coeff. of Variation -0.271 0.008
OrderBacklog/Sales 0.004 0.561
InterestRate -0.076 0.151
DaysPayable 0.041 0.172
DistressDummy -0.035 0.402
IndSalesNet -0.055 0.025
LIFO -0.056 0.562
AvgCost 0.285 0.025
Number of obs 3,130
F(10,2758) 29.0
Prob > F 0.000
R-sq. within 0.117
R-sq. between 0.749
R-sq. overall 0.711
Table 24: Regression results: Finished goods inventory (R7)
It is noted that the overall quality of the model is markedly lower for finished goods
inventories, than for total and raw materials inventories. The F statistic is 29.0 and the R-
squared within is only 0.117. Many independent variables, including the DistressDummy
variable, carry statistically insignificant coefficient estimates. While financially distressed
firms, on average, appear to hold less total and raw materials inventory this is not found
to be true for finished goods inventories. Hypothesis 8, thus, is not supported for finished
goods inventories.
The regression results for non-distressed and distressed firms (without and with
interaction effect) are shown in Table 25. Surprisingly, the Distress variable carries a
positive and marginally significant coefficient in the regression analysis of non-distressed
firms (leftmost column). For distressed firms, however, the Distress variable carries the
178
expected negative sign ( ) 0.024 ? = ? . The IndSalesNet variable also has a negative and
significant coefficient ( ) 0.200 ? = ? . The Distress*IndSalesNet interaction effect,
however, is not statistically significant (rightmost column).
FinGoodInv Coef. P>t Coef. P>t Coef. P>t
Constant -0.402 0.001 -1.889 0.000 -1.833 0.000
Forecast 0.472 0.000 0.173 0.062 0.186 0.045
SalesSurprise 0.184 0.000 0.160 0.036 0.167 0.029
Coeff. of Variation -0.216 0.088 -0.409 0.127 -0.443 0.099
OrderBacklog/Sales 0.276 0.009 0.001 0.935 0.001 0.953
InterestRate -0.074 0.183 -0.243 0.217 -0.266 0.178
DaysPayable 0.029 0.450 0.077 0.325 0.077 0.324
Distress 0.003 0.076 -0.024 0.001 0.055 0.340
IndSalesNet 0.022 0.376 -0.200 0.011 -0.192 0.016
LIFO -0.076 0.460 -0.148 0.621 -0.146 0.627
AvgCost 0.361 0.009 0.147 0.652 0.145 0.658
Distress*IndSalesNet 0.009 0.17
Number of obs 2,228 661 661
F 31.8 4.9 4.7
Prob > F 0.000 0.000 0.000
R-sq. within 0.174 0.123 0.128
R-sq. between 0.730 0.434 0.440
R-sq. overall 0.701 0.410 0.419
Non-distressed firms Distressed firms
w/o interaction with interaction
Table 25: Split-sample regression results: Finished goods inventory (R8, R9)
The analysis of finished goods inventories, thus, provides only limited support for the
hypotheses set forth in this paper. It generally seems as though the hypothesized
relationships between financial distress, power and inventories are strongest for total and
raw materials inventories. A summary of the empirical results is presented in Section 3.6.
179
3.5.2. Empirical results: Part II
The second data set (Part II) used for the empirical analyses is described in Section
3.4.5.2. This data set comprises observations from 1997 only, but provides more detailed
industry level statistics. Specifically, focal industry, buying industry, and supplying
industry four-firm concentration ratios are added to the model.
The analysis of the second data set (Part II) is also conducted in three parts: Total
inventories are analyzed, followed by raw materials, and finished goods inventories,
respectively. For each of these dependent variables, three regression analyses are
performed. First, the model is estimated using the full data set (comprising both
financially healthy and financially distressed firms). Then, the model is estimated for
healthy and distressed firms, separately. In a third step, the interaction effects between the
Distress and Power variables are added to the model which is then estimated using the
subsample of financially distressed firms only. The reader is referred to Table 15 for an
overview of the regression analyses.
3.5.2.1. Regression results: Total inventory
The basic regression model used to analyze total inventories is shown below (R10, see
Table 15). This model includes the focal industry’s four-firm concentration ratio, as well
as the weighted average concentration ratios of the supplying and buying industries in
addition to the variables included in R1 (see Table 15). The new variables are added to
180
approximate a firm’s supply chain power.
(R10) lnTotalInventory
itf
= ?
0
+ ?
1
lnSalesForecast
itf
+ ?
2
SalesSurprise
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
lnLeadTime
itf
+ ?
7
DistressDummy
itf
+ ?
8
lnIndSalesNet
itf
+ ?
9
IndCR4
itf
+ ?
10
SupplyCR4
itf
+ ?
11
BuyCR4
itf
+ ?
12
LIFO
itf
+ ?
13
AvgCost + ?
itf
This model is estimated using an OLS regression procedure with robust standard errors.
Specifically, the Huber-White sandwich estimator of standard errors is used to provide
some control for heteroskedasticity.
The regression results for TotalInventory are shown in Table 26. The model explains
about 86 percent of the variability in total inventories, and the model’s F statistic is
215.33 which is statistically significant at the less than one percent level.
While some variables have statistically insignificant or unexpected coefficients
(SalesSurprise, Coefficient of Variation, OrderBacklog/Sales, InteresRate), many
variables have significant coefficients with the expected signs. Specifically, total
inventories are shown to increase with forecasted sales (? = 0.868) and days payable
outstanding (the lead time proxy, ? = 0.247). The coefficient of the DistressDummy
variable is negative (? = -0.182), thus supporting Hypothesis 8 which states that
financially distressed firms should hold less inventory than their healthier counterparts.
181
Total Inv Coef. P>t
Constant -1.982 0.000
Forecast 0.868 0.000
SalesSurprise 0.451 0.000
Coeff. of Variation -0.335 0.168
OrderBacklog/Sales 0.006 0.729
InterestRate 0.036 0.230
DaysPayable 0.247 0.000
DistressDummy -0.182 0.023
IndSalesNet 0.052 0.005
IndCR4 -0.003 0.091
SupplyCR4 0.004 0.349
BuyCR4 0.007 0.002
LIFO 0.185 0.013
AvgCost 0.204 0.107
N 753
F( 13, 739) 215.33
Prob > F 0.000
R-squared 0.863
Table 26: Regression results: Total inventory (R10)
Some of the coefficient estimates of the power variables also have the expected signs:
Greater levels of (focal) industry concentration, suggesting greater firm power, are shown
to be associated with lower inventory levels (? = -0.003, Hypothesis 9). In addition, the
buying industry power (BuyCR4) has a positive and statistically significant coefficient (?
= 0.007). An increase in the buying industry’s concentration level (holding focal industry
concentration levels constant), thus, implies that firms will hold more inventory. This
finding supports Hypothesis 11. The SupplyCR4 variable, however, does not have a
statistically significant coefficient, and the coefficient of the IndSalesNet variable, while
significant, does not have the expected sign.
In summary, the analysis of the second data set (Part II) is generally consistent with the
results from the analysis of the first data set (Part I, see Table 16).
182
Table 27 shows the split-sample regressions (non-distressed and distressed firms) and the
interaction effects between Distress and the Power variables are included in the
regression analysis of distressed firms in the rightmost column.
Despite the relatively small sample sizes all models explain 85 percent of the variability
in the dependent variable. Many of the coefficient estimates, however, are statistically
insignificant which may be a function of the small number of observations, especially in
the case of distressed firms (n = 136).
Focusing on the results for non-distressed firms first (Table 27, leftmost column), it is
noted that the statistically significant coefficient estimates are consistent with the results
shown in Table 26. The only unexpected result is the positive and significant coefficient
of the Distress variable (? = 0.006). The other coefficients, including those of the four-
firm concentration ratio variables are statistically insignificant.
183
Total Inv Coef. P>t Coef. P>t Coef. P>t
Constant -1.909 0.000 -2.780 0.001 -2.715 0.002
Forecast 0.857 0.000 0.829 0.000 0.831 0.000
SalesSurprise 0.384 0.000 0.594 0.000 0.562 0.000
Coeff. of Variation -0.306 0.308 -0.207 0.659 -0.247 0.613
OrderBacklog/Sales 0.165 0.014 0.365 0.002 -0.016 0.129
InterestRate 0.016 0.733 -0.014 0.151 0.071 0.381
DaysPayable 0.263 0.000 0.069 0.368 0.356 0.003
Distress 0.006 0.026 -0.036 0.015 -0.038 0.799
IndSalesNet 0.054 0.008 0.591 0.039 0.008 0.917
IndCR4 -0.001 0.654 0.143 0.710 -0.010 0.114
SupplyCR4 0.006 0.152 0.010 0.897 -0.009 0.587
BuyCR4 0.002 0.283 -0.012 0.042 0.019 0.007
LIFO 0.136 0.042 -0.006 0.685 0.569 0.050
AvgCost 0.182 0.060 0.020 0.004 0.116 0.771
Distress * IndShipValueNet -0.014 0.273
Distress * IndCR4 -0.002 0.066
Distress * SupplyCR4 -0.004 0.450
Distress * BuyCR4 0.001 0.464
Number of obs 617 136 136
F 248.02 105.47 90.11
Prob > F 0.000 0.000 0.000
R-squared 0.859 0.851 0.857
Non-distressed firms Distressed firms
w/o interaction with interaction
Table 27: Split-sample regression results: Total inventory (R11, R12)
The results for distressed firms (second column in Table 27) closely resemble the results
for non-distressed firms. The significance are, however, generally lower due to the small
sample size (n = 136). The Distress variable carries a negative and significant coefficient
(? = -0.036) which is in line with Hypothesis 8. This hypothesis suggests that distressed
firms hold less inventory than healthier firms.
The only power measure with a statistically significant coefficient is BuyCR4. The
coefficient is negative (? = -0.012) which suggests that greater buying industry power
results in lower focal firm inventory holding. This finding is surprising and inconsistent
with Hypothesis 11 and the results for the entire data sample (see Table 26).
184
The only distress-power interaction effect that is significant is that of the IndCR4 variable
(? = -0.002, see rightmost column in Table 27). This result indicates that the negative
effect of (focal) industry concentration—a proxy for a firm’s power—on the firm’s
inventory holdings is greater the more distressed the firm is. This finding lends some
support to Hypothesis 12. There are, however no statistically significant interaction
effects between Distress and the IndSalesNet, BuyCR4, and SupplyCR4 variables.
3.5.2.2. Regression results: Raw materials inventory
The key results for the analysis of raw materials inventories are similar to those for total
inventories. Table 28 presents the regression results for the entire sample of non-
distressed and distressed firms (n = 676).
Again, distressed firms are shown to hold less raw materials inventory than healthy firms
(Distress, ? = -0.206), thus confirming Hypothesis 8. It is also interesting to note that all
power variables have statistically significant coefficients:
? IndSalesNet: Greater levels of firm power, as approximated by the IndSalesNet
variable, are shown to be associated with greater inventory holdings (? = 0.094).
The same unexpected result was found in the analysis of total inventories.
? IndCR4: The negative coefficient of the IndCR4 variable (? = -0.007), in turn, is
consistent Hypothesis 9 and suggests that greater power, as approximated by focal
industry concentration levels, should result in lower inventory holdings.
185
? BuyCR4 and SupplyCR4: Both the BuyCR4 and SupplyCR4 variables have
positive and significant coefficients (? = 0.009 for BuyCR4 and ? = 0.013 for
SupplyCR4). These results support Hypothesis 11 and Hypothesis 10, suggesting
that, while holding focal industry power levels constant, greater buying and
supplying industry power levels result in larger inventory holdings.
Raw Mat Inv Coef. P>t
Constant -2.515 0.000
Forecast 0.786 0.000
SalesSurprise 0.363 0.000
Coeff. of Variation -0.106 0.648
OrderBacklog/Sales -0.027 0.014
InterestRate 0.075 0.000
DaysPayable 0.263 0.000
DistressDummy -0.206 0.034
IndSalesNet 0.094 0.000
IndCR4 -0.007 0.008
SupplyCR4 0.013 0.005
BuyCR4 0.009 0.001
LIFO 0.081 0.384
AvgCost 0.122 0.374
N 676
F( 13, 662) 141.38
Prob > F 0.000
R-squared 0.769
Table 28: Regression results: Raw materials inventory (R13)
The results for non-distressed firms only (see Table 29) are largely consistent with the
results for the entire data set. The analysis of distressed firms (second and third columns
in Table 29), in turn, yields only few statistically significant coefficient estimates.
Specifically, the coefficient of the Distress variable (? = -0.034) is not statistically
significant in these regressions, and the only power variables with significant coefficients
are IndCR4 (? = - 0.017) and BuyCR4 (? = 0.015). The Distress*IndCR4 interaction
186
effect also carries a negative and statistically significant coefficient estimate (? = -0.003).
This finding suggests that financial distress only affects firms’ raw materials inventories
when these firms operate in highly concentrated industries, i.e. when firms possess some
degree of market power. This finding thus is consistent with Hypothesis 12. There is,
however, no support for the contention that the distress-inventory effect increases with
the level of buying industry or supplying industry concentration (Hypothesis 13 and
Hypothesis 14, respectively).
Raw Mat Inv Coef. P>t Coef. P>t Coef. P>t
Constant -1.952 0.000 -6.013 0.000 -5.839 0.000
Forecast 0.759 0.000 0.844 0.000 0.807 0.000
SalesSurprise 0.309 0.000 0.662 0.000 0.595 0.001
Coeff. of Variation 0.056 0.845 -0.260 0.551 -0.260 0.552
OrderBacklog/Sales -0.033 0.805 -0.027 0.001 -0.033 0.000
InterestRate 0.093 0.002 0.052 0.377 0.037 0.558
DaysPayable 0.213 0.006 0.642 0.000 0.597 0.000
Distress 0.003 0.436 -0.034 0.140 -0.109 0.557
IndSalesNet 0.123 0.000 -0.066 0.481 -0.094 0.313
IndCR4 -0.005 0.114 -0.017 0.002 -0.011 0.082
SupplyCR4 0.014 0.006 0.025 0.140 0.010 0.621
BuyCR4 0.007 0.017 0.015 0.028 0.019 0.011
LIFO 0.073 0.445 0.277 0.397 0.310 0.347
AvgCost 0.108 0.425 0.203 0.613 0.160 0.701
Distress * IndShipValueNet -0.006 0.746
Distress * IndCR4 -0.003 0.002
Distress * SupplyCR4 0.004 0.576
Distress * BuyCR4 0.000 0.922
Number of obs 561 115 115
F 87.19 136.42 107.48
Prob > F 0.000 0.000 0.000
R-squared 0.731 0.825 0.842
Non-distressed firms Distressed firms
w/o interaction with interaction
Table 29: Split-sample regression results: Raw materials inventory (R14, R15)
187
3.5.2.3. Regression results: Finished goods inventory
The results for finished goods inventories are shown in Table 30 and in Table 31.
Unfortunately, the estimation results are of generally poorer quality than the results for
total and raw materials inventories. None of the hypotheses set forth in this essay are
empirically supported for finished goods inventories and few variables carry statistically
significant coefficients. The lack of significant findings may be attributable to, among
other factors, the particularly small sample sizes.
Fin Good Inv Coef. P>t
Constant -2.885 0.000
Forecast 0.882 0.000
SalesSurprise 0.464 0.000
Coeff. of Variation -0.504 0.129
OrderBacklog/Sales 0.005 0.926
InterestRate 0.127 0.191
DaysPayable 0.302 0.001
DistressDummy 0.090 0.500
IndSalesNet 0.065 0.044
IndCR4 -0.005 0.209
SupplyCR4 -0.004 0.562
BuyCR4 -0.004 0.352
LIFO 0.465 0.001
AvgCost 0.440 0.052
N 654
F( 13, 640) 81.28
Prob > F 0.000
R-squared 0.701
Table 30: Regression results: Finished goods inventory (R16)
188
Fin Good Inv Coef. P>t Coef. P>t Coef. P>t
Constant -3.231 0.000 -1.578 0.246 -1.393 0.360
Forecast 0.885 0.000 0.847 0.000 0.817 0.000
SalesSurprise 0.445 0.000 0.588 0.028 0.556 0.045
Coeff. of Variation -0.772 0.061 0.270 0.700 0.308 0.656
OrderBacklog/Sales -0.645 0.000 0.067 0.000 0.059 0.000
InterestRate 0.088 0.338 -0.099 0.874 -0.139 0.795
DaysPayable 0.388 0.000 0.120 0.547 0.101 0.628
Distress 0.003 0.419 0.015 0.664 0.135 0.630
IndSalesNet 0.057 0.090 0.098 0.491 0.093 0.551
IndCR4 -0.001 0.723 -0.012 0.197 -0.007 0.514
SupplyCR4 -0.001 0.868 -0.012 0.646 -0.025 0.378
BuyCR4 -0.002 0.586 -0.001 0.896 0.002 0.841
LIFO 0.412 0.003 0.631 0.240 0.595 0.266
AvgCost 0.519 0.013 0.415 0.464 0.430 0.468
Distress * IndShipValueNet -0.006 0.841
Distress * IndCR4 -0.001 0.508
Distress * SupplyCR4 -0.001 0.891
Distress * BuyCR4 -0.003 0.282
Number of obs 544 110 110
F 107.14 15.12 14.7
Prob > F 0.000 0.000 0.000
R-squared 0.717 0.672 0.686
Non-distressed firms Distressed firms
w/o interaction with interaction
Table 31: Split-sample regression results: Finished goods inventory (R17, R18)
3.6. Summary and discussion
This paper develops a comprehensive theoretical perspective of the firm distress-
inventory relationship, drawing on theories and prior research from the economics,
inventory theory, and supply chain management fields. Previously, researchers generally
ignored the role of firm financial distress when investigating inventories. This study
contends that financial distress plays a significant role in inventory management and that
a firm’s power relative to its buyers and suppliers will impact the magnitude of this
distress-inventory effect. Specifically, the hypotheses set forth in this essay contend that
189
? financially distressed firms hold less inventory than healthier firms
(Hypothesis 8),
? more powerful firms hold less inventory than less powerful firms (Hypothesis 9),
? greater power relative to suppliers results in lower inventory holdings
(Hypothesis 10),
? greater power relative to buyers results in lower inventory holdings
(Hypothesis 11),
? the effect of financial distress on inventories increases with the firm’s power
(Hypothesis 12),
? greater power relative to suppliers increases the magnitude of the distress-
inventory effect (Hypothesis 13),
? greater power relative to buyers increases the magnitude of the distress-inventory
effect (Hypothesis 14).
The results of the empirical analyses of total, raw materials and finished goods
inventories are summarized in Table 32 and Table 33. Table 32 shows the hypothesis
testing results obtained from the analyses of both distressed and non-distressed firms.
Table 33, in turn, focuses on the results for distressed firms only.
190
Data Part I
(1998-2004)
Data Part II
(1997)
H
y
p
o
t
h
e
s
i
s
Testing variable(s)
E
x
p
e
c
t
a
t
i
o
n
T R F T R F
8 DistressDummy – – – 0 – – 0
9 IndSalesNet – – – – + + +
9 IndCR4 – – – 0
10 SupplyCR4 + 0 + 0
11 BuyCR4 +
+ + 0
T = Total inventory, R = Raw materials inventory, F = Finished goods inventory
0 = statistically insignificant result
Table 32: Summary of results for entire data set
Focusing on the results for the entire data sets (both Part I and Part II, see Table 32) first,
it is evident that there is strong support for the contention that distressed firms, on
average, hold less inventory than financially healthy firms (Hypothesis 8). This finding,
however, is not confirmed for finished goods inventories. This is not surprising since raw
materials inventories can easily be reduced by consuming extant stock without reordering
further supplies.
The hypothesis that greater levels of power should be associated with lower inventory
levels (Hypothesis 9) also finds some empirical support. As shown in Table 32, the
IndSalesNet variable carries the expected negative coefficients in the analysis of the panel
data set (Part I), thus indicating that inventories decrease as firm power (as measured by
the IndSalesNet variable) increases. The analysis of 1997 data (Part II), in turn,
consistently yields (unexpected) positive coefficients for the IndSalesNet variable. At the
191
same time, however, the industry concentration variable (IndCR4), carries negative
coefficients (for total and raw materials inventories). While this latter finding is
consistent with Hypothesis 9, the positive coefficients of the IndSalesNet variable are not
consistent with Hypothesis 9. The multicollinearity between the power variables
(IndSalesNet, IndCR4, SupplyCR4, BuyCR4) may partly explain this contradictory result.
Hypothesis 10 and Hypothesis 11 suggest that greater levels of power over suppliers and
buyers, respectively, should result in lower inventory holdings. Table 32 shows positive
coefficients for SupplyCR4 (raw materials inventory only) and BuyCR4 (total and raw
materials inventory). These results imply that greater supplying and buying industry
concentration levels—i.e. lower focal firm power when focal industry concentration
levels are held constant—equate to greater firm inventory holdings. This is, to the best of
the author’s knowledge, the first study to present empirical evidence for the contention
that inter-firm power balances in the supply chain affect the location and ownership of
inventories in supply chains. The results also indicate that power levels affect raw
materials inventories to a much greater extent than finished goods inventories which do
not appear to be impacted by supply chain power.
The results of the analysis of distressed firms only are summarized in Table 33. The
negative coefficients of the Distress variable further support Hypothesis 8. This result
suggests that the magnitude of financial distress impacts the magnitude of the distressed
firm’s inventory reductions.
192
Data Part I
(1998-2004)
Data Part II
(1997)
H
y
p
o
t
h
e
s
i
s
Testing variable(s)
E
x
p
e
c
t
a
t
i
o
n
T R F T R F
8 Distress – – – – – 0 0
9 IndSalesNet – – – – + 0 0
9 IndCR4 – 0 – 0
10 SupplyCR4 + 0 0 0
11 BuyCR4 +
– + 0
12 Distress*IndSalesNet – 0 – 0 0 0 0
12 Distress*IndCR4 – – – 0
13 Distress*SupplyCR4 + 0 0 0
14 Distress*BuyCR4 +
0 0 0
T = Total inventory, R = Raw materials inventory, F = Finished goods inventory
0 = statistically insignificant result
Table 33: Summary of results for distressed firms
The results for the power-inventory hypothesis (Hypothesis 9) are mixed. In the analysis
of the panel data set (Part I), the IndSalesNet variable carries negative coefficients as
expected, suggesting that more powerful distressed firms tend to hold less inventory. As
seen in Table 32, however, the analysis of the second data set yields unexpected (or
insignificant) coefficient estimates for the IndSalesNet variable. The four-firm
concentration ratio (IndCR4), an alternative proxy for power, is shown to significantly
impact distressed firms’ inventory holdings only in the case of raw materials inventories.
Specifically, distressed firms in more concentrated industries are found to hold less raw
materials inventory, all else equal.
193
The supplying and buying industry concentration variables (SupplyCR4 and BuyCR4)
mostly carry insignificant or unexpected coefficients. Only the BuyCR4 variable has a
positive and significant coefficient in the raw materials inventory regression. This result
suggests that distressed firms facing more powerful buyers may be forced to hold greater
raw materials inventories and provides some support for Hypothesis 11.
There is, finally, only scant evidence that distressed firms reduce inventories to a greater
extent when they are more powerful. Only in three instances did the interaction effects
between Distress and the power variables have the expected negative coefficient
estimates. In the first part of the data analysis (Part I), the effect of financial distress on
raw materials inventories is shown to increase with the firm’s power (IndSalesNet). The
same result is obtained in the second part of the data analysis (Part II) when power is
approximated with the industry concentration ratio (IndCR4). These findings provide
some evidence in support of Hypothesis 12. There is, however, no support for the
contention that the distress-inventory effect depends on the levels of supplying and
buying industry power (Hypothesis 13 and Hypothesis 14).
In summary, many of the hypotheses set forth in this study are empirically supported. It is
shown that a firm’s financial condition significantly impacts a firm’s inventory decisions.
Moreover, it is shown that power balances in supply chains may impact the distribution
of inventory ownership in supply chains. At the same time, the data provide only limited
evidence for the contention that power moderates the distress-inventory relationship.
194
This research contributes to the extant literature on multiple accounts: Different
theoretical perspectives are synthesized to investigate the financial distress-inventory
relationship. Specifically, insights from inventory theory and supply chain management
research are used to improve upon the specification of empirical estimation models
presented in prior economics research. Novel proxies for variables such as order and
holding costs are proposed to overcome measurement problems that have previously
hindered empirical inventory research.
The analyses presented in this essay not only refine the extant knowledge of the distress-
inventory relationship but also provide new insights on inventory management issues in a
supply chain context. Specifically, this is, to the best of the author’s knowledge, the first
study to empirically explore the role of inter-firm power in inventory management. In
addition, the moderating role of power in the financial distress-inventory relationship is
investigated.
Understanding the effect of a firm’s financial condition on its inventory decisions may
also have important managerial implications in terms of supplier selection, for example.
Managers should be aware of how a supply chain partner’s distress and power may affect
inventory ownership in the supply chain. While this research does not evaluate how
financial distress and power imbalances in supply chains affect overall supply chain
performance, it is conceivable that the shifting of inventory ownership from the
distressed firm to suppliers and buyers may reduce a supply chain’s effectiveness in
terms of, for example, service levels and responsiveness. The investigation of these
195
questions is left for future research.
This research investigates if and how financial distress affects firm inventories and it is
shown that distressed firms tend to reduce inventory holdings. Future research may also
investigate if and when cutting inventories is a viable turnaround strategy.
As noted previously, this study adds to the small, emerging body of empirical inventory
research. While efforts have been made to overcome the difficulties of data collection and
variable measurement that are associated with doing research in this field, the work
presented here must be considered exploratory. Secondary accounting data from public
firms only were used for the empirical analyses. The generalizability of the results to the
entire population of manufacturing firms, both public and private, can not be ascertained.
In addition, buyer and supplier power levels could only be approximated using rather
crude measures such as buying industry and supplying industry concentration ratios. The
computation of these ratios relies on the Input-Output Tables and industry concentration
data published by the Bureau of Economic Analysis. As noted previously, the omission
of international firms in the construction of these data may lead to a misrepresentation of
inter-industry power constellations. Moreover, these industry power levels are only rough
approximations of firm power levels. Future research may use qualitative methods and
different data collection techniques, such as dyadic surveys, for example, to further
investigate how power affects supply chain inventories and how it moderates the distress-
inventory relationship.
196
4. Firm decision making under financial distress: Summary and outlook
The structure-conduct-performance (SCP) paradigm is a theoretical framework that is
widely used in the industrial organization and strategic management literatures. The basic
tenet of the SCP paradigm is that the structure of markets influences firms’ conduct, and
the latter then is a determinant of firm and market performance. In addition, it is also
recognized that feedback mechanisms may exist within this framework (Waldman and
Jensen 2001, see also Figure 1). The performance observed in a market, for example, may
attract new entrants, thus changing the market structure. Similarly, firms may change
their conduct in the light of poor past performance. This dissertation is concerned with
this particular feedback mechanism: How does a firm’s financial distress affect its
conduct in terms of sales prices and inventories. While the potential existence of such
relationships has been recognized previously, the author is unaware of any study that has
systematically investigated the nature of these causal links. This dissertation addresses
this gap in the literature by investigating the following two research questions:
? Does financial distress have an impact on prices and inventories, after controlling
for other relevant parameters?
? If so, how can these effects be characterized, i.e. what factors influence the
magnitude of the distress-price and distress-inventory relationships?
These questions are investigated through analyses of prices in the U.S. airline industry
and inventories in U.S. manufacturing industries. Upon reviewing the literature, two sets
197
of hypotheses relating financial distress to prices and inventories, respectively, are
formulated. These hypotheses reconcile the conflicts revolving around prior
conceptualizations of the distress-price, and distress-inventory links. More precisely, it is
suggested that firm-specific and structural contingencies moderate these relationships. As
a consequence, it is implied that financial distress may have a strong influence on prices
and inventories in some instances, but not in others.
Large-scale empirical analyses are conducted to test the hypotheses set forth in this
research. Data from the U.S. airline industry are used to investigate how financial distress
affects prices. The results present substantial evidence in support of the hypotheses.
Financial distress is found to be negatively related to air fares, with the magnitude of this
relationship depending on the distressed firm’s operating costs, market shares, and size.
In addition, the degree of market concentration and the competitors’ financial situations
are shown to impact the distress-price relationship.
As to the effect of financial distress on inventories, data from the U.S. manufacturing
industry are used for the empirical tests. It is shown that greater degrees of financial
distress will result in lower inventory levels, ceteris paribus. In some instances, this
effect is found to be stronger the greater the distressed firm’s power over its buyers and
suppliers.
Both the price and inventory studies thus suggest the following:
? Firm financial distress is an important determinant of a firm’s actions.
198
? The nature of the distress-conduct relationship is further impacted by market
structural and firm characteristics. Specifically such factors impact the occurrence
and magnitude of the effect of distress on firm conduct parameters.
This research thus helps refine and enhance researchers’ understanding of the relationship
between structure, conduct and performance. While only one particular feedback
mechanism within the structure-conduct-performance paradigm—the effect of distress on
firm conduct—is investigated here, it is expected that there exist further, previously
unexplored links between structural, conduct, and performance parameters. The analyses
of such relationships are suggested for future research.
Managers may benefit from this work through an enhanced understanding of how firms’
financial conditions may impact (competing) firms’ behavior. Competitors of distressed
firms, for example, may be able to better anticipate distressed firms’ competitive moves,
and as a consequence, may be in a better position to implement preemptive measures or
respond to distressed firms’ actions. Moreover, the findings of this work may be of
interest to cooperation partners of distressed firms. Specifically, managers may want to
understand how a distressed firm’s actions may ultimately impact the cooperating firm, in
terms of service levels, costs, or required inventory holdings, for example. While the
findings presented here do not provide direct evidence for the implications of a firm’s
distress on its cooperation partners, there are some indications that a distressed firm’s
supply chain partners will be affected by the changes in the distressed firm’s conduct. It
is suggested that future research further explore these issues.
199
This dissertation research, thus, enhances researchers’ and managers’ understanding of
how firm financial distress affects prices and inventories. Following these descriptive
causal analyses, a normative approach to the research question at hand is suggested for
future research. Specifically, the following questions could be addressed:
? Is the cutting of prices or the reduction of inventories a viable turnaround strategy,
i.e. do distressed firms that lower prices or reduce inventories exhibit greater
turnaround performance?
? In what specific instances is price or inventory cutting advisable? Are there
certain organizational or situational characteristics that influence the extent to
which lower prices or inventories result in distressed firms’ performance
improvements?
? How does distress affect firm and supply chain operating performance? Are there
any effects in customer service levels or purchasing lead times, for example?
These and more questions may be of great interest to both the academic and practitioner
communities. While these issues are not within the scope of this dissertation, the work
presented here provides a solid basis for further investigations of the managerial
implications and consequences of firm financial distress.
This dissertation empirically investigates the relationship between financial distress and
select firm conduct parameters using secondary data from the U.S. airline and
manufacturing industries. The use of secondary data is desirable in that advanced
statistical methods can be utilized to analyze large data sets and obtain robust and
200
generalizable estimation results. At the same time, however, the use of secondary data
often requires researchers to approximate variables for which no direct measures are
available or to omit explanatory factors from empirical models altogether should data not
be available. In this dissertation, variables such as order costs and sourcing lead times, for
example, could not be measured directly but could only be approximated. In addition,
some data sources present inherent deficiencies and limitations. The data from the Input-
Output tables, which are used to construct industrial supply chains for the analyses of the
distress-inventory relationship, for example, do not include information on trade flows
involving foreign buyers and suppliers. While a research design based on the analysis of
secondary data is deemed suitable for an initial study of the distress-conduct link, future
research could employ qualitative methods such as case studies, for example, to gain an
in-depth understanding of the managerial decision processes that are triggered by the
deterioration of a firm’s financial condition. At the same time, insights could be gained
into when and why specific turnaround strategies and actions result in performance
improvements.
201
Appendix 1
Table 34 below presents the residuals of the regression of airlines’ operating expenses per
available seat-mile (ASM) on average stage length. This regression was performed to
evaluate U.S. carriers’ operating costs after controlling for differences in the airlines’
average stage length. Negative residuals indicate relative cost advantages, while positive
residuals suggest relative cost disadvantages. Based on the results displayed in Table 34,
twelve carriers were identified as low-cost carriers (LCC). While the cut-off between
low-cost and high-cost carriers (HCC) is arbitrary, it is noted that there is a sizeable
difference in the magnitude of the residuals between the highest-cost LCC (Valujet
Airlines: -0.142), and the lowest-cost HCC (Carnival Airlines: -0.105). In the empirical
analyses, the top twelve airlines in Table 34 are, therefore, identified as LCCs.
202
Carrier code Carrier name Ranked residuals LCC indicator
WN Southwest Airlines, Co. -0.371 1
QQ Reno Air, Inc. -0.303 1
SY Sun Country Airlines -0.268 1
NK Spirit Air Lines -0.237 1
B6 Jetblue Airways -0.231 1
W7 Western Pacific Airlines -0.223 1
FL Airtran Airways Corporation -0.216 1
TZ American Trans Air, Inc. -0.214 1
BE Braniff Int'l Airlines, Inc -0.172 1
HP America West Airlines, Inc. -0.157 1
F9 Frontier Airlines, Inc. -0.142 1
J7 Valujet Airlines, Inc. -0.142 1
KW Carnival Air Lines, Inc. -0.105 0
AS Alaska Airlines, Inc. -0.099 0
NJ Vanguard Airlines, Inc. -0.081 0
KP Kiwi International -0.073 0
N7 National Airlines -0.051 0
TW Trans World Airlines, Inc. -0.039 0
XJ Mesaba Airlines -0.036 0
NW Northwest Airlines, Inc. -0.022 0
BF Markair, Inc. -0.010 0
WV Air South, Inc. -0.002 0
HQ Business Express -0.001 0
DL Delta Air Lines, Inc. 0.012 0
CO Continental Air Lines, Inc. 0.018 0
EV Atlantic Southeast Airlines 0.031 0
JI Midway Airlines, Inc. 0.086 0
YV Mesa Airlines, Inc. 0.090 0
ZN Key Airlines, Inc. 0.090 0
OE Westair Airlines 0.093 0
AQ Aloha Airlines, Inc. 0.112 0
RU Continental Express Airline 0.125 0
FF Tower Air, Inc. 0.132 0
AA American Airlines, Inc. 0.140 0
ZW Air Wisconsin Airlines Corp 0.166 0
UA United Air Lines, Inc. 0.170 0
US US Airways, Inc. 0.171 0
YX Midwest Express Airlines 0.179 0
HA Hawaiian Airlines, Inc. 0.195 0
TB USAir Shuttle 0.263 0
PA Pan American World Airways 1.243 0
Table 34: Ranked residuals of regression of OpEx/ASM on avg. stage length (n=41)
203
Appendix 2
Table 35 presents the OLS regression estimates of the empirical model presented in
Chapter 2 of this dissertation. These basic regression results are used solely to investigate
the presence of heteroskedasticity. This is done by means of the Breusch-Pagan/Cook-
Weisberg Lagrange multiplier test (Breusch and Pagan 1979, Cook and Weisberg 1983).
The test result suggests the presence of heteroskedasticity and motivates the choice of a
generalized least squares procedure for the empirical analyses.
Source SS df MS Number of obs 23039
Model 3522.49 20 176.12 F( 21, 23017) 2754.5
Residual 1471.77 23018 0.06 Prob > F 0.000
Total 4994.26 23038 0.22 R-squared 0.705
Adj R-squared 0.705
Root MSE 0.253
Fare Coefficient Std. error P>|t|
Constant 1.473 0.171 0.000
AirlinePass -0.048 0.002 0.000
Distance 0.713 0.048 0.000
DistanceSquared -0.023 0.004 0.000
SlotRoute 0.134 0.005 0.000
RouteHHI -0.005 0.006 0.416
MaxAirportHHI 0.107 0.005 0.000
RouteShare 0.001 0.000 0.000
MaxAirportShare 0.004 0.000 0.000
LCCCompForHCC -0.148 0.004 0.000
LCCCompForLCC -0.103 0.008 0.000
AltRouteLCC1M -0.016 0.004 0.000
Circuity -0.162 0.024 0.000
Distress 0.005 0.001 0.000
Loadfactor -0.004 0.000 0.000
AirlineCost 0.462 0.013 0.000
Size 0.034 0.002 0.000
Quarter 2 -0.086 0.005 0.000
Quarter 3 -0.120 0.006 0.000
Quarter 4 -0.060 0.005 0.000
2002 -0.260 0.006 0.000
Table 35: OLS regression estimates (n = 23,039)
204
Appendix 3
The regression results shown in Table 36 and Table 37 are used to evaluate the benefit of
adding fixed effects to the regression model. This benefit is measured by means of an F
statistic as proposed by (Greene 2003). The test returns a statistically significant F value,
suggesting that a fixed effects model should be used for the empirical analyses.
Source SS df MS Number of obs 23039
Model 1878.66 16 117.42 F( 17, 23021) 1582.87
Residual 3115.60 23022 0.14 Prob > F 0.000
Total 4994.26 23038 0.22 R-squared 0.3762
Adj R-squared 0.3757
Root MSE 0.3679
Fare Coefficient Std. error P>|t|
AirlinePass (fitted) 0.272 0.012 0.000
Distance -0.033 0.074 0.654
DistanceSquared 0.043 0.006 0.000
SlotRoute 0.051 0.008 0.000
RouteHHI 0.031 0.008 0.000
MaxAirportHHI 0.254 0.008 0.000
RouteShare -0.006 0.000 0.000
MaxAirportShare 0.000 0.000 0.103
LCCCompForHCC -0.223 0.006 0.000
LCCCompForLCC -0.158 0.011 0.000
AltRouteLCC1M -0.136 0.007 0.000
Circuity 1.136 0.057 0.000
Distress -0.001 0.001 0.277
Loadfactor -0.022 0.001 0.000
AirlineCost 0.960 0.016 0.000
Size 0.030 0.003 0.000
Constant 2.536 0.242 0.000
Table 36: 2SLS regression estimates without fixed effects (n = 23,039)
205
The Tables in Appendix 4 and Appendix 5 provide further details of the empirical
estimation results.
Table 37 (Appendix 4) presents the second-stage estimation results of the regression of air
fares on the set of independent variables as specified in Sections 2.3.2.2 and 2.3.2.3. In
addition, air carrier fixed effects are included in this regression to evaluate the contribution
of these fixed firm effects to the explanatory power of the model. This analysis is needed to
determine the appropriate econometric estimation technique as discussed in Section 2.3.4.
Table 38 (Appendix 5) presents the first-stage estimation results for all five empirical
models. This table, thus, is an extension of the baseline first-stage estimation results shown
in Table 3. It is noted that the estimation results are generally consistent across all five
models such that the discussion of the baseline first-stage results (see Section 2.4.1) also
apply to the results shown in Table 38.
206
Appendix 4
Source SS df MS Number of obs 23039
Model 2558.23 50 51.16 F( 51, 22987) 715.34
Residual 2436.04 22988 0.11 Prob > F 0.000
Total 4994.26 23038 0.22 R-squared 0.5122
Adj R-squared 0.5112
Root MSE 0.3255
Fare Coefficient Std. error P>|t|
AirlinePass (fitted) 0.247 0.011 0.000
Distance -0.033 0.066 0.618
DistanceSquared 0.040 0.005 0.000
SlotRoute 0.034 0.007 0.000
RouteHHI 0.069 0.008 0.000
MaxAirportHHI 0.192 0.007 0.000
RouteShare -0.006 0.000 0.000
MaxAirportShare 0.000 0.000 0.353
LCCCompForHCC -0.194 0.006 0.000
LCCCompForLCC -0.106 0.010 0.000
AltRouteLCC1M -0.118 0.006 0.000
Circuity 0.960 0.050 0.000
Distress -0.040 0.003 0.000
Loadfactor -0.022 0.001 0.000
AirlineCost 0.300 0.041 0.000
Size 0.090 0.017 0.000
Constant 0.527 0.376 0.161
Quarter2 -0.045 0.007 0.000
Quarter3 -0.048 0.009 0.000
Quarter4 -0.048 0.006 0.000
2002 -0.301 0.020 0.000
aq 0.097 0.143 0.499
as 0.002 0.050 0.961
b6 0.217 0.074 0.003
be 0.414 0.131 0.002
co 0.243 0.024 0.000
dl 0.060 0.010 0.000
f9 0.007 0.084 0.929
ff 0.190 0.182 0.295
fl 0.048 0.075 0.525
hp 0.172 0.050 0.001
hq 0.006 0.136 0.965
ji 0.794 0.121 0.000
kp -0.074 0.206 0.721
kw 0.656 0.165 0.000
n7 0.629 0.104 0.000
nj 1.345 0.116 0.000
nk 0.098 0.094 0.295
nw 0.133 0.015 0.000
oe 0.287 0.133 0.030
qq 0.287 0.146 0.048
sy -0.012 0.152 0.935
tb -0.479 0.140 0.001
tw 0.300 0.032 0.000
tz 0.162 0.068 0.017
ua 0.184 0.009 0.000
us 0.074 0.019 0.000
wn -0.393 0.037 0.000
yx 0.400 0.091 0.000
zn 3.164 0.374 0.000
zw 0.191 0.163 0.242
Table 37: 2SLS regression estimates with fixed effects (n = 23,039)
207
Appendix 5
First-stage G2SLS regression Number of obs. 23039 Obs. per group: min. 1
(fixed effects) Number of groups 4508 avg. 5.1
max. 8
Dependent variable:
AirlinePass Coefficient P>|t| Coefficient P>|t| Coefficient P>|t| Coefficient P>|t| Coefficient P>|t|
Constant 444.458 0.000 440.911 0.000 446.816 0.000 454.492 0.000 477.332 0.000
Distance -136.296 0.000 -134.880 0.000 -136.560 0.000 -139.647 0.000 -145.605 0.000
DistanceSquared 10.181 0.000 10.067 0.000 10.186 0.000 10.444 0.000 10.821 0.000
SlotRoute 0.099 0.001 0.101 0.001 0.103 0.000 0.104 0.000 0.085 0.004
RouteHHI -0.412 0.000 -0.411 0.000 -0.411 0.000 -0.420 0.000 -0.421 0.000
MaxAirportHHI -0.367 0.000 -0.373 0.000 -0.375 0.000 -0.366 0.000 -0.375 0.000
RouteShare 0.027 0.000 0.027 0.000 0.027 0.000 0.027 0.000 0.027 0.000
MaxAirportShare 0.001 0.007 0.001 0.005 0.001 0.010 0.001 0.004 0.001 0.013
LCCCompForHCC 0.227 0.000 0.229 0.000 0.229 0.000 0.223 0.000 0.221 0.000
LCCCompForLCC 0.238 0.000 0.226 0.000 0.224 0.000 0.236 0.000 0.249 0.000
AltRouteLCC1M -0.022 0.061 -0.021 0.068 -0.021 0.065 -0.022 0.055 -0.019 0.094
Circuity -2.326 0.000 -2.327 0.000 -2.326 0.000 -2.344 0.000 -2.335 0.000
Distress 0.005 0.190 -0.138 0.038
Chpt11Ops -0.015 0.141
DistressDiff 0.018 0.000
Pre4Chpt11 0.018 0.107
Post4Chpt11 -0.046 0.001
Loadfactor 0.016 0.000 0.015 0.000 0.015 0.000 0.015 0.000 0.017 0.000
AirlineCost -0.095 0.023 -0.097 0.021 -0.089 0.034 -0.083 0.053 -0.091 0.029
Size 0.028 0.150 0.008 0.666 0.015 0.370 0.042 0.059 0.065 0.000
Quarter 2 0.048 0.000 0.051 0.000 0.053 0.000 0.051 0.000 0.044 0.000
Quarter 3 0.074 0.000 0.079 0.000 0.079 0.000 0.076 0.000 0.062 0.000
Quarter 4 0.047 0.000 0.051 0.000 0.053 0.000 0.049 0.000 0.044 0.000
2002 -0.055 0.315 -0.036 0.509 -0.038 0.485 -0.057 0.306 -0.108 0.046
Population 0.579 0.000 0.567 0.000 0.568 0.000 0.588 0.000 0.577 0.000
Income -0.034 0.785 -0.035 0.780 -0.048 0.702 -0.033 0.791 -0.014 0.912
AirlineCost*Distress -0.025 0.020
Size*Distress -0.002 0.458
RouteShare*Distress 0.000 0.185
RouteHHI*Distress 0.013 0.033
F 1038.3 1038.3 992.3 873.7 1051.3
Prob > F 0.000 0.000 0.000 0.000 0.000
R-squared: within 0.541 0.541 0.541 0.541 0.544
between 0.006 0.007 0.008 0.005 0.009
overall 0.006 0.007 0.007 0.004 0.010
1 3 4 5 2
Table 38: First-stage G2SLS regression estimates (n = 23,039)
208
Appendix 6
Table 39 presents the means of select variables for those firms that are included in the
statistical analyses and those firms for which data are available in the Compustat database
but which have been deleted from the data sample due to missing data on one or more
variables. As discussed in Sections 3.4.5.1 and 3.4.5.2, the sampled firms, on average, tend
to be smaller than those firms firms that are not included in the data sample. A Hotelling T-
squared test confirms that the two groups are statistically significantly different in both
time periods studied (1997 and 1998-2004).
Variable Sample mean Compustat Sample mean Compustat
Total Inventory (million $) 110.9 190.5 103.9 186.2
Sales (million $) 835.2 1698.7 835.6 1788.2
COGS (million $) 626.0 1162.5 537.4 1217.0
Total Debt (million $) 148.0 488.4 158.1 574.9
Total Assets (million $) 760.1 1792.9 733.8 2023.6
1997 1998-2004
Table 39: Mean comparisons between sampled firms and Compustat population
209
Bibliography
Abdel-Malek, Layek L, and Roberto Montanari. "On the Multi-Product Newsboy Problem
with Two Constraints." Computers & Operations Research 32, no. 8 (2005): 2095.
Aiken, Leona S, and Stephen G West. Multiple Regression: Testing and Interpreting
Interactions. Newbury Park, CA: Sage Publications, 1991.
Alstrom, Poul. "Numerical Computation of Inventory Policies, Based on the Eoq/Ox Value
for Order-Point Systems." International Journal of Production Economics 71, no.
1-3 (2001): 235.
Altman, Edward. Bankruptcy, Credit Risk, and High Yield Junk Bonds. Malden, MA:
Blackwell Publishers, 2002.
Altman, Edward I. "Financial Ratios, Discriminant Analysis and the Prediction of
Corporate Bankruptcy." The Journal of Finance 23, no. 4 (1968): 589-609.
———. "A Further Empirical Investigation of the Bankruptcy Cost Question." The Journal
of Finance 39, no. 4 (1984): 1067.
Amihud, Yakov, and Haim Mendelson. "Inventory Behaviour and Market Power: An
Empirical Investigation." International Journal of Industrial Organization 7, no. 2
(1989): 269.
Anonymous. "American Bankruptcy: The Uses and Abuses of Chapter 11." The Economist
310, no. 7594 (1989): 72.
———. "America's Airlines, Flying on Empty." Economist.com / Global Agenda (2005):
1.
Arogyaswamy, Kamala, Vincent L III Barker, and Masoud Yasai-Ardekani. "Firm
Turnarounds: An Integrative Two-Stage Model." The Journal of Management
Studies 32, no. 4 (1995): 493.
Bain, Joe S. Barriers to New Competition. Cambridge, MA: Harvard University Press,
1956.
Baker, Samuel A. "Risk, Leverage and Profitability - an Industry Analysis." The Review of
Economics and Statistics 55, no. 4 (1973): 503.
Ballou, Ronald H. "Estimating and Auditing Aggregate Inventory Levels at Multiple
Stocking Points." Journal of Operations Management 1, no. 3 (1981): 143.
210
Barla, Philippe, and Bonchun Koo. "Bankruptcy Protection and Pricing Strategies in the Us
Airline Industry." Transportation Research. Part E, Logistics & Transportation
Review 35 (1999): 101.
Betker, Brian L. "The Administrative Costs of Debt Restructurings: Some Recent
Evidence." Financial Management 26, no. 4 (1997): 56.
Blazenko, George W, and Kirk Vandezande. "Corporate Holding of Finished Goods
Inventories." Journal of Economics and Business 55, no. 3 (2003): 255.
Borenstein, Severin. "Hubs and High Fares: Dominance and Market Power in the U.S."
The Rand Journal of Economics 20, no. 3 (1989): 344.
Borenstein, Severin, and Nancy L Rose. "Bankruptcy and Pricing Behavior in U.S. Airline
Markets." The American Economic Review 85, no. 2 (1995): 397.
Bowman, Edward H. "Risk Seeking by Troubled Firms." Sloan Management Review (pre-
1986) 23, no. 4 (1982): 33.
Bradbury, Michael E, and Suzanne Lloyd. "An Estimate of the Direct Costs of Bankruptcy
in New Zealand." Asia Pacific Journal of Management 11, no. 1 (1994): 103.
Branch, Ben. "The Costs of Bankruptcy: A Review." International Review of Financial
Analysis 11, no. 1 (2002): 39.
Brander, James A., and Tracy R. Lewis. "Oligopoly and Financial Structure: The Limited
Liability Effect." The American Economic Review 76, no. 5 (1986): 956.
Breusch, T S, and A R Pagan. "A Simple Test for Heteroskedasticity and Random
Coefficient Variation." Econometrica 47, no. 5 (1979): 1287.
Busse, Maghan. "Firm Financial Condition and Airline Price Wars." The Rand Journal of
Economics 33, no. 2 (2002): 298.
Buxbaum, Peter A. "A Case of Corporate Turnaround." Distribution 94, no. 1 (1995): 30.
Buzacott, John A, and Rachel Q Zhang. "Inventory Management with Asset-Based
Financing." Management Science 50, no. 9 (2004): 1274.
Cachon, Gerard P. "Stock Wars: Inventory Competition in a Two-Echelon Supply Chain
with Multiple Retailers." Operations Research 49, no. 5 (2001): 658.
Campbell, Steven V. "Predicting Bankruptcy Reorganization for Closely Held Firms."
Accounting Horizons 10, no. 3 (1996): 12.
Carey, Susan, and Melanie Trottman. "Airlines, Battered by Losses, Look Set for Rebound;
Falling Costs, Fewer Seats May Be Ticket to Recovery for Some Major Carriers."
Wall Street Journal, Nov 26 2005, A.1.
211
Carpenter, Robert E, Steven M Fazzari, and Bruce C Petersen. "Financing Constraints and
Inventory Investment: A Comparative Study with High-Frequency Panel Data." The
Review of Economics and Statistics 80, no. 4 (1998): 513.
Carpenter, Robert E, Steven M Fazzari, Bruce C Petersen, Benjamin M Friedman, and Anil
K Kashyap. "Inventory Investment, Internal-Finance Fluctuations, and the Business
Cycle; Comments and Discussion." Brookings Papers on Economic Activity, no. 2
(1994): 75.
Caves, R. E., and M. E. Porter. "Market Structure, Oligopoly, and Stability of Market
Shares." The Journal of Industrial Economics 26, no. 4 (1978): 289.
Caves, Richard E., and Pankaj Ghemawat. "Identifying Mobility Barriers." Strategic
Management Journal 13, no. 1 (1992): 1.
Chakravarthy, Balaji S. "Measuring Strategic Performance." Strategic Management
Journal 7, no. 5 (1986): 437.
Chand, Suresh, and James Ward. "A Note on "Economic Order Quantity" under Conditions
of Permissible Delay in Payments"." Journal of the Operational Research Society
38, no. 1 (1987): 83-84.
Chang, Chun-Tao, and Jinn-Tsair Teng. "Retailer's Optimal Ordering Policy under Supplier
Credits." Mathematical Methods of Operations Research 60, no. 3 (2004): 471.
Chapman, C. B., S. C. Ward, Dale F. Cooper, and M. J. Page. "Credit Policy and Inventory
Control." The Journal of the Operational Research Society 35, no. 12 (1984): 1055.
Chattopadhyay, Prithviraj, William H Glick, and George P Huber. "Organizational Actions
in Response to Threats and Opportunities." Academy of Management Journal 44,
no. 5 (2001): 937.
Chen, Ming-Jer. "Competitor Analysis and Interfirm Rivalry: Toward a Theoretical
Integration." Academy of Management. The Academy of Management Review 21,
no. 1 (1996): 100.
Chen, Ming-Jer, Ken G. Smith, and Curtis M. Grimm. "Action Characteristics as Predictors
of Competitive Responses." Management Science 38, no. 3 (1992): 439.
Cook, R. Dennis, and Sanford Weisberg. "Diagnostics for Heteroskedasticity in
Regression." Biometrika 70, no. 1 (1983): 1-10.
Corbett, Jenny, Donald Hay, and Helen Louri. "A Financial Portfolio Approach to
Inventory Behavior: Japan and the Uk." International Journal of Production
Economics 59, no. 1-3 (1999): 43.
212
Cox, Andrew. "Understanding Buyer and Supplier Power: A Framework for Procurement
and Supply Competence." Journal of Supply Chain Management 37, no. 2 (2001):
8.
Cox, Andrew, Chris Lonsdale, Glyn Watson, and Hong Qiao. "Supplier Relationship
Management: A Framework for Understanding Managerial Capacity and
Constraints." European Business Journal 15, no. 3 (2003): 135.
Cox, Andrew, Joe Sanderson, and Glyn Watson. "Supply Chains and Power Regimes:
Toward an Analytic Framework for Managing Extended Networks of Buyer and
Supplier Relationships." Journal of Supply Chain Management 37, no. 2 (2001): 28.
Crandall, Robert. "Coffee, Tea ... Or Bankruptcy?" Wall Street Journal, Sep 16 2005, A.14.
Dasgupta, Sudipto, and Sheridan Titman. "Pricing Strategy and Financial Policy." The
Review of Financial Studies 11, no. 4 (1998): 705.
Denardo, Eric V. Dynamic Programming: Models and Applications. first ed. Mineola, New
York: Dover Publications Inc., 2003.
Derfus, Pamela J., Curtis M. Grimm, Ken G. Smith, and Patrick G. Maggitti. "The Red
Queen Effect: Competitive Action and Performance." Academy of Management
Journal (forthcoming).
Dobson, Paul W. "Exploiting Buyer Power: Lessons from the British Grocery Trade."
Antitrust Law Journal 72, no. 2 (2005): 529.
Dresner, Martin, Jiun-Sheng Chris Lin, and Robert Windle. "The Impact of Low-Cost
Carriers on Airport and Route Competition." Journal of Transport Economics and
Policy 30, no. 3 (1996): 309.
Drukker, D.M. "Testing for Serial Correlation in Linear Panel-Data Models." Stata Journal
3 (2003): 168-77.
Emerson, Richard M. "Power-Dependence Relations." American Sociological Review 27,
no. 1 (1962): 31-41.
Erlenkotter, Donald. "Ford Whitman Harris and the Economic Order Quantity Model."
Operations Research 38, no. 6 (1990): 937.
Ferrier, Walter J. "Navigating the Competitive Landscape: The Drivers and Consequences
of Competitive Aggressiveness." Academy of Management Journal 44, no. 4
(2001): 858.
Ferrier, Walter J, Cormac Mac Fhionnlaoich, Ken G Smith, and Curtis M Grimm. "The
Impact of Performance Distress on Aggressive Competitive Behavior: A
Reconciliation of Conflicting Views." Managerial and Decision Economics 23, no.
4,5 (2002): 301.
213
Finkin, Eugene F. "Company Turnaround." The Journal of Business Strategy 5, no. 4
(1985): 14.
Followill, Richard A., Michael Schellenger, and Patrick H. Marchand. "Economic Order
Quantities, Volume Discounts, and Wealth Maximization." The Financial Review
25, no. 1 (1990): 143.
Fraser, Donald R., and Norman Gaither. "Estimating Financing Costs in Managing
Inventory Positions at Large Corporations: Practice and Implications." Managerial
Finance 10, no. 2 (1984): 21.
Frazier, Gray L, and Kersi D Antia. "Exchange Relationships and Interfirm Power in
Channels of Distribution." Academy of Marketing Science. Journal 23, no. 4
(1995): 321.
French, John R. P., and Bertram Raven. The Bases of Social Power. Edited by Dorwin
Cartwright. Ann Arbor: University of Michigan Press, 1959.
Gaski, John F. "The Theory of Power and Conflict in Channels of Distribution." Journal of
Marketing (pre-1986) 48, no. 3 (1984): 9.
Gelman, Judith R., and Steven C. Salop. "Judo Economics: Capacity Limitation and
Coupon Competition." The Rand Journal of Economics 14, no. 2 (1983): 315.
Gertler, Mark, and Simon Gilchrist. "Monetary Policy, Business Cycles, and the Behavior
of Small Manufacturing Firms." The Quarterly Journal of Economics 109, no. 2
(1994): 309.
Gibbons, Robert. Game Theory for Applied Economists. First ed. Princeton, NJ: Princeton
University Press, 1992.
Greene, William H. Econometric Analysis. Fifth ed. Delhi/India: Pearson Education Inc.,
2003.
Grice, John Stephen, and Robert W. Ingram. "Tests of the Generalizability of Altman's
Bankruptcy Prediction Model." Journal of Business Research 54 (2001): 53-61.
Gritta, Richard D. "Air Carrier Financial Condition: A Review of Models for Measuring
the Financial Fitness of the U.S. Airline Industry." Proceedings of the 2004
Conference of the Air Transport Research Society (2004).
Guariglia, Alessandra. "The Effects of Financial Constraints on Inventory Investment:
Evidence from a Panel of Uk Firms." Economica 66, no. 261 (1999): 43.
Guariglia, Alessandra, and Fabio Schiantarelli. "Production Smoothing, Firms'
Heterogeneity, and Financial Constraints: Evidence from a Panel of Uk Firms."
Oxford Economic Papers 50, no. 1 (1998): 63.
214
Guffey, Daryl M., and William T. Moore. "Direct Bankruptcy Costs: Evidence from the
Trucking Industry." The Financial Review 26, no. 2 (1991): 223.
Haley, Charles W, and Robert C Higgins. "Inventory Policy and Trade Credit Financing."
Management Science (pre-1986) 20, no. 4 (1973): 464.
Hall, M., and L. Weiss. "Firm Size and Profitability." Review of Economics and Statistics
49 (1967): 319-31.
Halley, Alain, and Jean Nollet. "The Supply Chain: The Weak Link for Some Preferred
Suppliers?" Journal of Supply Chain Management 38, no. 3 (2002): 39.
Hambrick, Donald C., and Steven M. Schecter. "Turnaround Strategies for Mature
Industrial-Product Business Units." Academy of Management Journal 26, no. 2
(1983): 231.
Harris, F.W. "How Many Parts to Make at Once." Factory, The Magazine of Management
10 (1913): 135-36, 52.
Hausman, J A. "Specification Tests in Econometrics." Econometrica (pre-1986) 46, no. 6
(1978): 1251.
Hendel, Igal. "Competition under Financial Distress." The Journal of Industrial Economics
44, no. 3 (1996): 309.
Hingley, M K. "Power to All Our Friends? Living with Imbalance in Supplier-Retailer
Relationships." Industrial Marketing Management 34, no. 8 (2005): 848.
Hingley, Martin K. "Power Imbalanced Relationships: Cases from Uk Fresh Food Supply."
International Journal of Retail & Distribution Management 33, no. 8/9 (2005): 551.
Hitt, Michael A., Brian K. Boyd, and Dan Li. "The State of Strategic Management
Research and Vision of the Future." In Research Methodology in Strategy and
Management, 1, edited by David J. Ketchen and Donald D. Bergh, 1-32: JAI Press,
2004.
Hofer, Charles W. "Toward a Contingency Theory of Business Strategy." Academy of
Management Journal (pre-1986) 18, no. 4 (1975): 784.
Hofer, Charles W. "Turnaround Strategies." The Journal of Business Strategy 1, no. 1
(1980): 19.
Hofer, Christian, Martin Dresner, and Robert Windle. "Financial Distress and Us Airline
Fares." Journal of Transport Economics and Policy 39, no. 3 (2005): 323.
Kahneman, Daniel, and Amos Tversky. "Prospect Theory: An Analysis of Decision under
Risk." Econometrica (pre-1986) 47, no. 2 (1979): 263.
215
Kashyap, Anil K, Owen A Lamont, and Jeremy C Stein. "Credit Conditions and the
Cyclical Behavior of Inventories." The Quarterly Journal of Economics 109, no. 3
(1994): 565.
Kashyap, Anil K, Jeremy C Stein, and David W Wilcox. "Monetary Policy and Credit
Conditions: Evidence from the Composition of External Finance." The American
Economic Review 83, no. 1 (1993): 78.
Kennedy, Robert E. "The Effect of Bankruptcy Filings on Rivals' Operating Performance:
Evidence from 51 Large Bankruptcies." International Journal of the Economics of
Business 7, no. 1 (2000): 5.
Kim, Yong H., and Kee H. Chung. "An Integrated Evaluation of Investment in Inventory
and Credit: A Cash Flow Approach." Journal of Business Finance & Accounting
17, no. 3 (1990): 381.
Lant, Theresa K., Frances J. Milliken, and Bipin Batra. "The Role of Managerial Learning
and Interpretation in Strategic Persistence and Reorientation: An Empirical
Exploration." Strategic Management Journal 13, no. 8 (1992): 585.
LeBlanc, Greg. "Signalling Strength: Limit Pricing and Predatory Pricing." The Rand
Journal of Economics 23, no. 4 (1992): 493.
Lee, Darin, and María José Luengo-Prado. "The Impact of Passenger Mix on Reported
"Hub Premiums" in the U.S. Airline Industry." Southern Economic Journal 72, no.
2 (2005): 372.
Lieberman, Marvin B, Susan Helper, and Lieven Demeester. "The Empirical Determinants
of Inventory Levels in High-Volume Manufacturing." Production and Operations
Management 8, no. 1 (1999): 44.
Longenecker, Justin G, and Charles D Pringle. "The Illusion of Contingency Theory as a
General Theory." Academy of Management. The Academy of Management Review
(pre-1986) 3, no. 000003 (1978): 679.
Luthans, Fred, and Todd I Stewart. "A General Contingency Theory of Management."
Academy of Management. The Academy of Management Review 2, no. 2 (1977):
181.
Mahajan, Siddharth, and Garrett van Ryzin. "Inventory Competition under Dynamic
Consumer Choice." Operations Research 49, no. 5 (2001): 646.
Maksimovic, Vojislav, and Josef Zechner. "Debt, Agency Costs, and Industry
Equilibrium." The Journal of Finance 46, no. 5 (1991): 1619.
Mason, Edward S. "The Current State of the Monopoly Problem in the United States."
Harvard Law Review 62, no. June 1949 (1949): 1265-85.
216
———. "Price and Production Policies of Large-Scale Enterprise." American Economic
Review 29, no. March 1939 (1939): 61-74.
Mazzeo, M.J. "Product Choice and Oligopoly Market Structure." The Rand Journal of
Economics 33, no. 2 (2002): 221-42.
McCafferty, Joseph. "Is Bankruptcy an Unfair Advantage?" CFO 11, no. 6 (1995): 28.
McCracken, Jeffrey. "Auto Maker Recovery Bid Intensifies; Ford to Accelerate, Deepen Its
Restructuring Efforts; Gm Cites Rosier Fiscal View." Wall Street Journal, Aug 10
2006, A.2.
McKinley, William. "Organizational Decline and Adaptation: Theoretical Controversies."
Organization Science 4, no. 1 (1993): 1-9.
McLaughlin, Curtis P, Gyula Vastag, and D Clay Whybark. "Statistical Inventory Control
in Theory and Practice." International Journal of Production Economics 35, no. 1-3
(1994): 161.
Miller, Danny, and Ming-Jer Chen. "Sources and Consequences of Competitive Inertia: A
Study of the U.S. Airline Industry." Administrative Science Quarterly 39, no. 1
(1994): 1.
Modigliani, Franco, and Merton H. Miller. "The Cost of Capital, Corporation Finance and
the Theory of Investment." The American Economic Review 48, no. 3 (1958): 261-
97.
Mone, Mark A, William McKinley, and Vincent L Barker III. "Organizational Decline and
Innovation: A Contingency Framework." Academy of Management. The Academy
of Management Review 23, no. 1 (1998): 115.
Moon, I, and E A Silver. "The Multi-Item Newsvendor Problem with a Budget Constraint
and Fixed Ordering Costs." The Journal of the Operational Research Society 51,
no. 5 (2000): 602.
Morrison, Steven A. "Actual, Adjacent, and Potential Competition." Journal of Transport
Economics and Policy 35, no. 2 (2001): 239.
Moses, O Douglas. "Organizational Slack and Risk-Taking Behaviour: Tests of Product
Pricing Strategy." Journal of Organizational Change Management 5, no. 3 (1992):
38.
Moulton, Wilbur N, and Howard Thomas. "Bankruptcy as a Deliberate Strategy:
Theoretical Considerations and Empirical Evidence." Strategic Management
Journal 14, no. 2 (1993): 125.
Opler, Tim C, and Sheridan Titman. "Financial Distress and Corporate Performance." The
Journal of Finance 49, no. 3 (1994): 1015.
217
Osteryoung, Jerome S., Eldon Nosari, Daniel E. McCarty, and Walter J. Reinhart. "Use of
the EOQ Model for Inventory Analysis." Production and Inventory Management
27, no. 3 (1986): 39.
Peteraf, Margaret A. "Intra-Industry Structure and the Response toward Rivals."
Managerial and Decision Economics (1986-1998) 14, no. 6 (1993): 519.
Porter, Michael. Competitive Strategy. New York, NY: Free Press, 1980.
Pryor, Frederic L. "New Trends in U.S. Industrial Concentration." Review of Industrial
Organization 18, no. 3 (2001): 301.
Queen, Maggie, and Richard Roll. "Firm Mortality: Using Market Indicators to Predict
Survival." Financial Analysts Journal 43, no. 3 (1987): 9.
Rabinovich, Elliot, and Philip T Evers. "Enterprise-Wide Adoption Patterns of Inventory
Management Practices and Information Systems." Transportation Research. Part E,
Logistics & Transportation Review 38E, no. 6 (2002): 389.
Rachamadugu, Ram. "Effect of Delayed Payments (Trade Credit) on Order Quantities."
Journal of the Operational Research Society 40, no. 9 (1989): 805-13.
Ravenscraft, David J. "Structure-Profit Relationships at the Line of Business and Industry
Level." The Review of Economics and Statistics 65, no. 1 (1983): 22.
Reuter, Vincent G. "Are We Using the Inventory Management Tools?" Industrial
Management 20, no. 3 (1978): 26.
Rodgers, Kimberly Jo Guthrie. "Chapter 11 Reorganization Policy and the Economic
Viability of Filing Firms." 3017860, Purdue University, 2000.
Rollman, Daniel P. "Flying Low: Chapter 11's Contribution to the Self-Destructive Nature
of the Airline Industry Economics." Emory Bankruptcy Developments Journal 21,
no. 1 (2004): 381.
Rose-Green, Ena, and Mark Dawkins. "Strategic Bankruptcies and Price Reactions to
Bankruptcy Filings." Journal of Business Finance & Accounting 29, no. 9/10
(2002): 1319.
Rosenblatt, Meir J. "Multi-Item Inventory System with Budgetary Constraint: A
Comparison between the Lagrangian and the Fixed Cycle Approach." International
Journal of Production Research 19, no. 4 (1981): 331-39.
Roumiantsev, Serguei, and Serguei Netessine. "What Can Be Learned from Classical
Inventory Models: A Cross-Industry Empirical Investigation." Manufacturing &
Service Operations Management forthcoming (2007).
218
Rustenburg, W D, G J van Houtum, and W H M Zijm. "Spare Parts Management for
Technical Systems: Resupply of Spare Parts under Limited Budgets." IIE
Transactions 32, no. 10 (2000): 1013.
Scherer, F M. Industrial Market Structure and Economic Performance. Second ed.
Chicago: Rand McNally College Publishing Company, 1980.
Schoonhoven, Claudia Bird. "Problems with Contingency Theory: Testing Assumptions
Hidden within the Language of Contingency ''Theory''." Administrative Science
Quarterly 26, no. 3 (1981): 349.
Scott, John T. "Multimarket Contact and Economic Performance." Review of Economics
and Statistics 64 (1982): 368-75.
Shoham, Aviv, and Avi Fiegenbaum. "Competitive Determinants of Organizational Risk-
Taking Attitude: The Role of Strategic Reference Points." Management Decision
40, no. 1/2 (2002): 127.
Singh, Jitendra V. "Performance, Slack, and Risk Taking in Organizational Decision
Making." Academy of Management Journal 29, no. 3 (1986): 562.
Smart, Carolyne, and Ilan Vertinsky. "Strategy and the Environment: A Study of Corporate
Responses to Crises." Strategic Management Journal 5, no. 3 (1984): 199.
Smith, Ken G, Curtis M Grimm, Stefan Wally, and Greg Young. "Strategic Groups and
Rivalrous Firm Behavior: Towards a Reconciliation." Strategic Management
Journal 18, no. 2 (1997): 149.
Smith, Ken G., Curtis M. Grimm, Martin J. Gannon, and Ming-Jer Chen. "Organizational
Information Processing, Competitive Responses, and Performance in the U.S.
Domestic Airline Industry." Academy of Management Journal 34, no. 1 (1991): 60.
Smith, Malcolm, and Christopher Graves. "Corporate Turnaround and Financial Distress."
Managerial Auditing Journal 20, no. 3 (2005): 304.
Spanos, Y., G. Zaralis, and S.Lioukas. "Strategy and Industry Effects on Profitability:
Evidence from Greece." Strategic Management Journal 25, no. 2 (2004): 139-66.
Staw, Barry M., Lance E. Sandelands, and Jane E. Dutton. "Threat-Rigidity Effects in
Organizational Behavior: A Multilevel Analysis." Administrative Science Quarterly
26, no. 4 (1981): 501.
Stundza, Tom, and Brian Milligan. "How Chrysler Will Cut Costs." Purchasing 130, no. 3
(2001): 30.
Sudarsanam, Sudi, and Jim Lai. "Corporate Financial Distress and Turnaround Strategies:
An Empirical Analysis." British Journal of Management 12, no. 3 (2001): 183.
219
Sun, Daning, and Maurice Queyranne. "Production and Inventory Model Using Net Present
Value." Operations Research 50, no. 3 (2002): 528.
Tersine, Richard J. Principles of Inventory and Materials Management. Fourth ed. Upper
Saddle River, NJ: Prentice-Hall, 1994.
Timme, Stephen G. "The Real Cost of Holding Inventory." Supply Chain Management
Review 7, no. 4 (2003): 30.
Tiras, Samuel L. "Which Firms Are Likely to Survive Chapter 11?" Commercial Lending
Review 17, no. 6 (2002): 18.
Trottman, Melanie. "Airline Bankruptcy Filings Put Continental, American at Risk." Wall
Street Journal, Sep 22 2005, B.4.
Waldman, Don E, and Elizabeth J Jensen. Industrial Organization: Theory and Practice.
Second ed: Addison Wesley Longman, 2001.
Wallin, Cynthia, M. Johnny Rungtusanatham, and Elliot Rabinovich. "What Is the "Right"
Inventory Management Approach for a Purchased Item?" International Journal of
Operations & Production Management 26, no. 1 (2006): 50.
Warner, Jerold B. "Bankruptcy Costs - Some Evidence." The Journal of Finance 32, no. 2
(1977): 337.
White, Halbert. "A Heteroskedasticity-Consistent Covariance Matrix Estimator and a
Direct Test for Heteroskedasticity." Econometrica 48, no. 4 (1980): 817-38.
Wiersema, William H. "Financing the Unfinanceable." Electrical Apparatus 58, no. 5
(2005): 42.
Wiseman, Robert M, and Philip Bromiley. "Toward a Model of Risk in Declining
Organizations: An Empirical Examination of Risk, Performance and Decline."
Organization Science 7, no. 5 (1996): 524.
Wiseman, Robert M, and Luis R Gomez-Mejia. "A Behavioral Agency Model of
Managerial Risk Taking." Academy of Management. The Academy of Management
Review 23, no. 1 (1998): 133.
Wooldridge, J M. Econometric Analysis of Cross Section and Panel Data. Cambridge,
MA: MIT Press, 2002.
Xu, Tianji, and Robert L K Tiong. "Risk Assessment on Contractors' Pricing Strategies."
Construction Management and Economics 19, no. 1 (2001): 77.
Yaffee, Robert. "A Primer for Panel Data Analysis." New York University, NY:http://www.nyu.edu/its/pubs/connect/fall03/yaffee_primer.htm, 2003.
220
Yoffie, David B, and Mary Kwak. "Judo Strategy: 10 Techniques for Beating a Stronger
Opponent." Business Strategy Review 13, no. 1 (2002): 20.
Young, Greg, Ken G Smith, and Curtis M Grimm. ""Austrian" and Industrial Organization
Perspectives on Firm-Level Competitive Activity and Performance." Organization
Science 7, no. 3 (1996): 243.
Zinn, Walter, and John M Charnes. "A Comparison of the Economic Order Quantity and
Quick Response Inventory Replenishment Methods." Journal of Business Logistics
26, no. 2 (2005): 119.
doc_294730646.pdf
Financial distress is a term in corporate finance used to indicate a condition when promises to creditors of a company are broken or honored with difficulty. If financial distress cannot be relieved, it can lead to bankruptcy. Financial distress is usually associated with some costs to the company; these are known as costs of financial distress.
ABSTRACT
Title of Document: Firm Decision Making under Financial Distress:
A Study of U.S. Air Fares and an Analysis of
Inventories in U.S. Manufacturing Industries
Christian Hofer, Ph.D. 2007
Directed By: Professor Martin E. Dresner, Ph.D.,
Robert H. Smith School of Business
Professor Robert J. Windle, Ph.D.,
Robert H. Smith School of Business
This dissertation investigates the effects of firm financial distress on two key firm
decision variables: sales prices and inventories. These analyses contribute to the
Structure-Conduct-Performance paradigm literature. Specifically, the feedback loop
between financial distress, a result of poor past performance, and two firm conduct
parameters, prices and inventories, is explored in great detail.
The first essay is motivated by the ambiguity of prior research on the relationship
between firm financial distress and prices. The extant economics, corporate finance and
strategic management literatures differentially approach this relationship, and empirical
research has found only limited, at times ambiguous support for any single theoretical
contention. These theoretical perspectives are reviewed and an attempt is made to
reconcile the apparent conflict by adopting a strategic contingency perspective that
identifies in which way and in what instances firm financial distress may impact prices.
The model is empirically tested using data from the U.S. airline industry. The results
indicate that firm financial distress and prices are generally negatively related. Moreover,
this effect is substantially stronger for firms operating under Chapter 11 protection than
for firms approaching bankruptcy. It is further shown that the magnitude of the effect of
financial distress on prices depends on firm factors such as operating costs, market
power, and firm size, as well as on competitive characteristics such as market
concentration and the financial condition of competitors.
The second essay analyzes the impact of firm distress on firm inventories and
investigates if this relationship is impacted by a firm’s power relative to its upstream and
downstream supply chain partners. Building on prior work in the economics field, this
research is not only based on microeconomics theory, but also draws on inventory theory
as well as on prior work on supply chain relationships. A comprehensive inventory
estimation model is specified, and novel measures of inventory determinants and power
are developed. The hypotheses are tested using panel data from the U.S. manufacturing
industry. It is shown that distressed firms hold less inventory and that a firm’s power
within the supply chain will determine to what extent inventory ownership is reduced
during times of financial distress. Implications for supplier selection and supply chain
cooperation are discussed.
In summary, this research significantly enhances researchers’ understanding of why,
how, and when firm financial distress affects prices and inventories.
Firm Decision Making Under Financial Distress: A Study of U.S. Air Fares and an
Analysis of Inventories in U.S. Manufacturing Industries
By
Christian Hofer
Dissertation submitted to the Faculty of the Graduate School of the
University of Maryland, College Park, in partial fulfillment
of the requirements for the degree of
Philosophical Doctor
2007
Advisory Committee:
Professor Martin E. Dresner, Co-Chair
Professor Robert J. Windle, Co-Chair
Professor Philip T. Evers
Professor Curtis M. Grimm
Professor Ali Haghani
© Copyright by
Christian Hofer
2007
ii
Dedication
To my parents,
in deep gratitude for all their love and support.
iii
Acknowledgements
The completion of this dissertation marks the endpoint of four tremendously exciting
years in the doctoral program at the University of Maryland’s Robert H. Smith School of
Business. Many faculty members, fellow doctoral students, and staff members have
contributed to making this time so enriching, enjoyable, and memorable.
I cannot possibly describe the many ways in which Adriana has enriched my life: She is a
best friend, an advisor, a critic, an inspiration, (…) and, thankfully, my wife.
I would like to express my deep gratitude to my advisors, Martin Dresner and Bob
Windle. Over the course of these four years, both Martin and Bob have read hundreds,
possibly thousands of pages of various drafts of this dissertation and other papers. We
have spent many dozens of hours in fruitful meetings and they have always given me
invaluable advice. I deeply appreciate all the time and effort they have invested in my
education and I am equally grateful for their genuine care. I hope that one day I will be
half as good an advisor as Martin and Bob have been for me.
I would also like to thank the entire logistics and supply chain faculty of the department
of Logistics, Business and Public Policy at the Robert H. Smith School of Business. I
particularly thank Phil Evers and Curt Grimm who have taught me much of the subject
matter that has inspired and shaped this dissertation. In addition, they have provided
valuable guidance and advice as members of my dissertation committee. I would also like
to thank Ali Haghani from the A. James Clark School of Engineering for kindly agreeing
to serve on my dissertation committee and generously sharing his thoughts and expertise.
My fellow doctoral students and friends have also contributed to making my time in the
doctoral program at Maryland a very special one. Toby Porterfield and Tashfeen Sohail,
in particular, have been wonderful colleagues during these years and I treasure their
friendship. I would also like to thank my friends Lorrie Westerlund and Joerg
Schnermann for all their care and support.
A final “thank you” goes to all the staff at the Smith School of Business, and to Mary
Slye, Dianne Fox, Anne Stevens, and Mary Crowe-Kokonis, in particular. Without them
this program would not run as smoothly as it does (and it would be much less fun, also!).
iv
Table of Contents
Dedication........................................................................................................................... ii
Acknowledgements............................................................................................................ iii
List of Tables .................................................................................................................... vii
List of Figures .................................................................................................................... ix
1. Introduction......................................................................................................... 1
2. The impact of firm financial distress on prices: A contingency approach ....... 12
2.1. Introduction....................................................................................................... 12
2.2. Theoretical background and hypothesis development...................................... 19
2.2.1. Financial distress as a driver of competitive pricing behavior ......................... 20
2.2.2. Conflicting theoretical arguments..................................................................... 26
2.2.3. The contingency approach................................................................................ 28
2.3. Data and methodology...................................................................................... 43
2.3.1. Data sample....................................................................................................... 44
2.3.2. Variables and measurement .............................................................................. 45
2.3.2.1. Dependent variable ........................................................................................... 46
2.3.2.2. Independent variables ....................................................................................... 47
2.3.2.3. Control variables............................................................................................... 51
2.3.3. Descriptive statistics ......................................................................................... 55
2.3.4. Empirical methodology..................................................................................... 60
2.4. Empirical results and discussion....................................................................... 65
2.4.1. First-stage regression ........................................................................................ 66
2.4.2. Second-stage regression.................................................................................... 68
2.4.3. Second-stage regression: Sensitivity analysis................................................... 77
2.5. Summary and discussion................................................................................... 82
3. The effect of firm financial distress on firm inventories: A supply chain
perspective ........................................................................................................ 87
3.1. Introduction....................................................................................................... 87
3.2. The financial distress-inventory relationship.................................................... 93
3.2.1. Economic theory............................................................................................... 93
v
3.2.2. Inventory theory................................................................................................ 99
3.2.3. The financial distress-inventory hypothesis.................................................... 107
3.3. The supply chain perspective.......................................................................... 108
3.3.1. Supply chain considerations in inventory decisions ....................................... 109
3.3.1.1. Inter-firm relationships: The role of power .................................................... 110
3.3.1.2. Supply chain power and inventory decisions.................................................. 112
3.3.1.3. The power-inventory hypotheses.................................................................... 115
3.3.2. Firm power as a moderator of the distress-inventory link .............................. 117
3.3.2.1. Prior research: Firm size as a moderator of the distress-inventory link ......... 118
3.3.2.2. The power moderator hypotheses ................................................................... 119
3.4. Data and methodology.................................................................................... 122
3.4.1. Sample selection ............................................................................................. 123
3.4.2. Model specification......................................................................................... 126
3.4.3. Variables and Measurement............................................................................ 130
3.4.3.1. Dependent variable ......................................................................................... 130
3.4.3.2. Independent variables ..................................................................................... 131
3.4.4. Data sources .................................................................................................... 141
3.4.5. Descriptive statistics ....................................................................................... 143
3.4.5.1. Descriptive statistics: Part I ............................................................................ 143
3.4.5.2. Descriptive statistics: Part II ........................................................................... 149
3.4.6. Methodology................................................................................................... 155
3.4.6.1. Overview of regression analyses .................................................................... 155
3.4.6.2. Empirical methodology: Part I........................................................................ 159
3.4.6.3. Empirical methodology: Part II ...................................................................... 161
3.5. Empirical results and discussion..................................................................... 162
3.5.1. Empirical results: Part I .................................................................................. 162
3.5.1.1. Regression results: Total inventory ................................................................ 163
3.5.1.2. Sensitivity analyses......................................................................................... 168
3.5.1.3. Regression results: Raw materials inventory.................................................. 174
3.5.1.4. Regression results: Finished goods inventory................................................. 176
3.5.2. Empirical results: Part II ................................................................................. 179
vi
3.5.2.1. Regression results: Total inventory ................................................................ 179
3.5.2.2. Regression results: Raw materials inventory.................................................. 184
3.5.2.3. Regression results: Finished goods inventory................................................. 187
3.6. Summary and discussion................................................................................. 188
4. Firm decision making under financial distress: Summary and outlook.......... 196
Appendix 1...................................................................................................................... 201
Appendix 2...................................................................................................................... 203
Appendix 3...................................................................................................................... 204
Appendix 4...................................................................................................................... 206
Appendix 5...................................................................................................................... 207
Appendix 6...................................................................................................................... 208
Bibliography ................................................................................................................... 209
vii
List of Tables
Table 1: Correlation matrix (n = 23,039) ........................................................................ 56
Table 2: Descriptive statistics for selected variables (n = 23,039).................................. 59
Table 3: First stage G2SLS regression estimates (n = 23,039)........................................ 68
Table 4: Second-stage G2SLS regression estimates......................................................... 71
Table 5: Second-stage G2SLS regression estimates using 1992, 1997, and 2002 data... 81
Table 6: Summary of results ............................................................................................. 83
Table 7: Sample composition (Part I)............................................................................. 145
Table 8: Pooled descriptive statistics (Part I) ................................................................ 146
Table 9: Descriptive statistics (Part I) – distressed vs. non-distressed firms................. 147
Table 10: Pairwise correlations (Part I) ........................................................................ 148
Table 11: Sample composition (Part II) ......................................................................... 151
Table 12: Pooled descriptive statistics (Part II)............................................................. 152
Table 13: Descriptive statistics (Part II) – distressed vs. non-distressed firms ............. 153
Table 14: Pairwise correlations (Part II)....................................................................... 154
Table 15: Overview of regression analyses .................................................................... 155
Table 16: Regression results: Total inventory (R1)........................................................ 165
Table 17: Split-sample regression results: Total inventory (R2, R3) ............................. 166
Table 18: Sensitivity analysis: Distressed vs. non-distressed firms ............................... 169
Table 19: Sensitivity analysis: Granularity of industry definitions................................ 171
Table 20: Sensitivity analysis: SalesSurprise vs. ForecastError.................................... 172
Table 21: Sensitivity analysis: Measurement of total inventories .................................. 173
Table 22: Regression results: Raw materials inventory (R4)......................................... 175
Table 23: Split-sample regression results: Raw materials inventory (R5, R6) .............. 176
Table 24: Regression results: Finished goods inventory (R7)........................................ 177
Table 25: Split-sample regression results: Finished goods inventory (R8, R9) ............. 178
Table 26: Regression results: Total inventory (R10)...................................................... 181
Table 27: Split-sample regression results: Total inventory (R11, R12) ......................... 183
Table 28: Regression results: Raw materials inventory (R13)....................................... 185
Table 29: Split-sample regression results: Raw materials inventory (R14, R15) .......... 186
viii
Table 30: Regression results: Finished goods inventory (R16)...................................... 187
Table 31: Split-sample regression results: Finished goods inventory (R17, R18) ......... 188
Table 32: Summary of results for entire data set............................................................ 190
Table 33: Summary of results for distressed firms ......................................................... 192
Table 34: Ranked residuals of regression of OpEx/ASM on avg. stage length (n=41).. 202
Table 35: OLS regression estimates (n = 23,039).......................................................... 203
Table 36: 2SLS regression estimates without fixed effects (n = 23,039)........................ 204
Table 37: 2SLS regression estimates with fixed effects (n = 23,039)............................. 206
Table 38: First-stage G2SLS regression estimates (n = 23,039) ................................... 207
Table 39: Mean comparisons between sampled firms and Compustat population ........ 208
ix
List of Figures
Figure 1: The structure-conduct-performance paradigm................................................... 3
Figure 2: The moderated distress-conduct feedback mechanism....................................... 5
Figure 3: The moderating effect of operating costs on the distress-price relationship.... 32
Figure 4: The moderating effect of firm size on the distress-price relationship .............. 35
Figure 5: The moderating effect of market shares on the distress-price relationship...... 38
Figure 6: The moderating effect of market concentration on the distress-price
relationship ............................................................................................................... 41
Figure 7: Research model................................................................................................. 42
Figure 8: Overview of Chapter 11 indicator variables .................................................... 50
Figure 9: Distribution of Distress scores prior to and during bankruptcy ...................... 58
Figure 10: Illustration of the r,Q policy ......................................................................... 103
Figure 11: The moderating effect of power on the distress-inventory relationship ....... 121
Figure 12: Research model............................................................................................. 122
Figure 13: Illustration of the construction of industrial supply chains.......................... 140
Figure 14: Alternative definitions of distressed and non-distressed firms..................... 168
1
“No matter what the state of the economy, no
company is immune from internal hard times—
stagnation or declining performance.” (Hofer 1980)
“Global competition, technological turbulence, high
costs of capital, and other nettlesome factors will
cause more and more businesses to face occasional
hard times.” (Hambrick and Schecter 1983)
1. Introduction
Firm financial distress is an omnipresent phenomenon in manufacturing and service
industries. While there is no unique definition of financial distress, distress firms are
generally loss-making and suffer from (severe) liquidity constraints. Based on these
criteria, Altman (2002, 1968) developed the Z score as a composite measure of a firm’s
financial condition. Altman suggests that firms with a Z score of less than 1.81 are
considered financially distressed and face a high risk of bankruptcy. Following this
definition, about one third of all U.S. manufacturing firms
1
and about half of all U.S.
airlines (The Economist 2005) were considered financially distressed in 2005. Most
recently, car manufacturers such as Ford and General Motors (McCracken 2006), and air
carriers like Northwest Airlines and Delta Airlines (Carey and Trottman 2005), to
mention but a few examples, have been experiencing financial difficulties. This
dissertation investigates the impact of financial distress on managerial decision variables
such as prices and inventories.
1
This estimate is based on the analysis of 2,323 manufacturing firms listed in the Compustat database.
Thirty-two percent of these firms had Z scores (Altman 1968) of less than 1.81.
2
Most research in the broad field of business management is concerned with
understanding how managerial decisions come about and how these decisions affect firm
and market performance. Many researchers therefore follow the tradition of the structure-
conduct-performance (SCP) paradigm which essentially posits that the structure of
markets impacts firms’ conduct which, in turn, is a key determinant of the performance of
firms and markets (Bain 1956, Mason 1949, 1939). The term “structure” thereby refers to
structural characteristics of markets that are indicators of the competitiveness of markets.
Commonly used measurement variables include industry concentration, the number of
firms in the market or barriers to entry and exit (Waldman and Jensen 2001). Firms
compete in the marketplace by means of actions that aim at maximizing firm
performance. These rivalrous activities are summarized by the term “conduct” which
may, for example, refer to pricing and product strategies (Waldman and Jensen 2001).
The aggregate performance of firms in a market can be measured in terms of allocative
efficiency or production efficiency, for example (Waldman and Jensen 2001). The
individual performance of firms, in turn, is typically evaluated based on financial (e.g.
profitability) or operating measures (e.g. productivity). Figure 1 graphically illustrates
the structure-conduct-performance paradigm.
3
Figure 1: The structure-conduct-performance paradigm
Firm-specific factors are added to the depiction of the structure-conduct-performance
paradigm in Figure 1 to indicate that not only (market) structural, but also other firm
characteristics (besides the firm’s financial condition) such as operating costs and firm
size, for example, may impact a firm’s conduct in the market (e.g. Spanos et al. 2004).
The structure-conduct-performance (SCP) paradigm, as presented by Waldman and
Jensen (2001), also recognizes that certain feedback loops may exist within the structure-
conduct-performance framework. An industry’s above-average performance, for
example, may attract new entrants, thus affecting the structure of markets. By the same
token, a firm’s past performance may impact future managerial decisions relating to, for
example, prices and sales quantities, thus linking the firm’s performance/distress to its
conduct. Also, a firm’s distress may ultimately impact other firm characteristics such as
the firm’s size and its cost structure. Figure 1 illustrates some of these feedback loops
Structure Performance
Other firm-specific
factors
Conduct Structure Performance
Other firm-specific
factors
Conduct
4
within the SCP paradigm
2
. While there are many such feedback mechanisms, one specific
link is of particular interest in this dissertation research: The effect of financial distress, a
direct result of poor past performance, on a firm’s conduct in terms of sales prices and
inventories.
Pricing and inventory decisions are important indicators of a firm’s competitive conduct
in the marketplace. Basic game-theoretic models suggest that firms compete on either
price (Bertrand competition) or quantities (Cournot competition) (Gibbons 1992). With
inventories being a function of sales quantities, both inventories and prices, thus, are
essential decision variables that reflect a firm’s competitive behavior. Consequently,
numerous researchers have investigated the competitive implications of firms’ pricing
(e.g. Busse 2002) and inventory (e.g. Cachon 2001, Mahajan and Ryzin 2001) decisions.
It is therefore deemed appropriate and relevant to investigate the effects of financial
distress on these two firm conduct parameters.
Clearly, a feedback mechanism between financial distress and conduct is intuitively
appealing: Managers of distressed firms must turn the situation around and ensure the
company’s future profitability. Given the widespread occurrence of financial distress,
researchers have been interested in understanding the effects of distress on firm conduct.
Specifically, researchers have examined the anatomy of corporate turnarounds: What do
financially troubled firms do to return to profitability?
2
Note that the changes in a firm’s conduct caused by a deretioration of the firm’s financial condition will
then impact the firm’s performance. The relationship between performance/distress and conduct (as well as
structure and firm-specific variables), thus, is iterative over time.
5
Hofer (1980) notes that price cutting is a popular measure implemented by distressed
firms. Arogyaswamy and Yasai-Ardekani (1995) and Sudarsanam and Lai (2001), in
turn, suggest that firms frequently reduce inventory levels as a part of their restructuring
efforts. While anecdotal evidence and conceptual work suggest that greater levels of
distress imply lower prices and inventories, ceteris paribus, empirical evidence in support
of this contention has been scant. In a similar vein, conceptual and empirical work has
arrived at the conclusion that there is no unique turnaround strategy and no single recipe
for turnaround success. Rather, different turnaround gestalts have emerged: Hofer (1980),
for example, distinguishes between revenue-generating, product-market refocusing, cost-
cutting, and asset reducing strategies. Both Hofer (1980) and Hambrick and Schecter
(1983) suggest that the choice of a turnaround strategy will be contingent on the gravity
of financial distress and other firm and market-related contingencies. This contention is
illustrated in Figure 2: The effect of financial distress (poor past performance) on
conduct is moderated by (market) structural characteristics and firm-specific factors.
Figure 2: The moderated distress-conduct feedback mechanism
Hofer (1980) and Hambrick and Schecter (1983), thus, contend that the effect of financial
Structure Conduct Performance
Other firm-specific
factors
Financial distress
Structure Conduct Performance
Other firm-specific
factors
Financial distress
6
distress on firm conduct is contingent on other factors. This contention is consistent with
the basic tenet of contingency theory. Contingency theory was originally motivated by
the observation that “[p]rominent theorists promote their ascribed frameworks as
conceptually valid and pragmatically applicable to all organizations in all situations”
(Luthans and Stewart 1977, p.182). This concept of universality, however, has been
questioned by researchers on the grounds of both theoretical and empirical
counterevidence. Instead, researchers have increasingly recognized the importance and
moderating role of situational characteristics in defining causal relationships (Hitt et al.
2004).
Proponents of the situational approach argue “that the most effective management
concept or technique depends on a set of circumstances at a particular point in time”
(Luthans and Stewart 1977, p.182) and that empirical research based on simple “linear
models [has generally] provided disappointing results” (Hitt et al. 2004, p.11).
Consequently, researchers have proposed a “general contingency theory of management”
(Luthans and Stewart 1977) which rests on the key premise that environmental, resource
and management variables intervene in cause-and-effect relationships in the context of
strategic management research.
There is, however, no defined set of contingency variables and no universal prescription
as to how, when and where contingencies ought to be considered (see e.g. Hofer 1975 for
a review of important control variables and contingency factors in the context of business
strategy research). Many researchers have therefore criticized contingency theory as an
7
“illusion” (Longenecker and Pringle 1978) and have attacked the theory’s vagueness and
“lack of clarity” (Schoonhoven 1981). The use of contingency frameworks has
nonetheless been popular in the strategic management literature (Hitt et al. 2004).
This dissertation follows this research tradition and defines context-specific contingency
variables that are expected to affect the relationship between firm financial distress and
prices and inventories, respectively. Specifically, it is suggested that structural and firm-
specific factors moderate the effects of firm distress on prices and inventories.
To date, the model shown in Figure 2 has not been subject to large-scale empirical
testing. While some researchers have investigated the effects of financial factors on firm
decision parameters such as prices (e.g. Borenstein and Rose 1995) and inventories (e.g.
Carpenter et al. 1994), the moderating effects of structural and firm-specific factors on
the distress-conduct relationship remain largely unexplored. The sole exception is the
work by Ferrier et al (2002): These authors investigate the effect of financial distress on
competitive aggressiveness as measured by the number and nature of firms’ competitive
actions. Ferrier et al (2002) thereby find evidence that the effect of distress on
competitive behavior is moderated by industry characteristics
3
and the educational and
functional heterogeneity of top management teams. This dissertation builds on the work
of Ferrier et al (2002) and extends it to the study of two particular firm conduct
parameters: sales prices and inventories.
3
Industry growth, industry concentration, and barriers to entry.
8
Clearly, gaining a better understanding of how financial distress impacts firms’ pricing
and inventory decision is a timely and relevant research endeavor. Prior research has
shown that linear models of the distress-price and distress-inventory relationships may be
overly simplistic and do not do justice to the complex nature of decision problems
relating to price and inventory management under financial distress (e.g. Singh 1986).
While most researchers contend that greater financial distress should result in lower
prices and lower inventory holdings, the empirical findings are largely inconsistent and
often times statistically insignificant. The basic premise of this research is that structural
and firm-specific characteristics moderate the distress-conduct relationship as shown in
Figure 2, thereby explaining why the distress-conduct effect may be substantial in some
instances and insignificant in other cases.
In summarizing, this dissertation investigates the following research questions:
? Does financial distress have an impact on prices and inventories, after controlling
for other relevant parameters?
? And how can these effects be characterized, i.e. what factors influence the
magnitude of the distress-price and distress-inventory relationships?
This dissertation addresses these questions and thereby makes a number of significant
contributions.
This is—to the best of the author’s knowledge—the first study to empirically investigate
the feedback loop between financial distress (poor past performance) and two key firm
conduct parameters: prices and inventories. Particular attention is paid to the moderating
9
effects of structural and firm factors on the distress-conduct relationship. This research,
thus, contributes to the SCP literature by analyzing the distress-conduct feedback loop
and empirically evaluating the effect of interactions between structural, firm, and
financial characteristics on firm conduct parameters.
This framework is empirically tested in two distinct contexts: Prices and inventories are
studied in the context of the U.S. airline industry and the U.S. manufacturing industry,
respectively. In both instances, context-specific contingency variables are proposed and
their moderating effects on the distress-price and distress-inventory relationships are
evaluated. This research draws on a broad array of theoretical arguments from the
strategic management, economics, and corporate finance literatures to identify these
contingency variables and to hypothesize about their impact on the distress-conduct
relationship. The validity of the theoretical arguments and models set forth in this
dissertation is underlined by solid estimation results. It is shown that financial distress is
an important explanatory variable that significantly impacts a firm’s sales prices and
inventories. This research thus also contributes to furthering empirical research on prices
and inventories.
In addition, answering these research questions also paves the way to exploring further
managerial implications of financial distress with respect to prices and inventories in
greater detail: When are pricing and inventory actions economically viable turnaround
strategies? And how will the distressed firm’s actions affect competition and inter-firm
cooperation?
10
This dissertation comprises four chapters. Following this introduction (Chapter 1),
Chapters 2 and 3 are devoted to the study of the effects of financial distress on prices and
inventories, respectively, while Chapter 4 provides a summary of the findings and
contributions of this dissertation research.
The impact of distress on prices is discussed in Chapter 2. Two specific research
questions are investigated in this essay: How does a firm’s financial distress impact its
pricing behavior? And what parameters moderate the effect of firm financial distress on
the firm’s prices? These questions arise upon reviewing a broad set of extant research
which is marked by ambiguous empirical findings. This conflict is addressed by
developing a contingency framework. It is suggested that firm factors such as operating
costs, firm size and market shares, as well as market characteristics such as market
concentration and competitors’ financial conditions determine to what extent financial
distress affects prices. A large-scale empirical analysis using panel data from the U.S.
airline industry is conducted. The results provide ample support for the proposed
contingency framework.
Chapter 3 focuses on the distress-inventory relationship. This essay is primarily
motivated by two observations: First, prior studies have approached the firm finance-
inventory link from an economics perspective only, thus ignoring the insights provided
by inventory theory. Second, most extant research has failed to put firm inventory
decisions into a supply chain context where inter-firm power balances may affect
11
inventory ownership in supply chains. Consequently, two research questions are
formulated: Does a firm’s financial situation have an impact on its inventories after
controlling for other relevant parameters prescribed by inventory theory and supply chain
research? And is the magnitude of the presumed effect of financial distress on inventories
impacted by power (im)balances in supply chain relationships? To investigate these
questions, a thorough review of related economics, inventory, and supply chain research
is provided and testable hypotheses are formulated. Based on this theoretical foundation,
an empirical estimation equation is specified. Data from U.S. manufacturing industries is
used to test the hypotheses. Specifically, it is shown that greater levels of firm financial
distress are associated with lower firm inventory levels, ceteris paribus. In addition, there
is some support for the hypothesis that greater levels of power over suppliers and buyers
not only reduce inventory ownership in general, but also increase the effect of financial
distress on inventories.
Chapter 4 presents a summary of this dissertation research and highlight its contributions.
In addition, a research agenda for further studies of the effects of firm financial distress is
outlined.
12
2. The impact of firm financial distress on prices: A contingency approach
Chapter 2 presents a theoretical and empirical analysis of the relationship between
financial distress and sales prices. This chapter is structured as follows: Section 2.1
provides a brief overview of prior research on the financial condition-prices link and
clearly states the research questions and contributions of this dissertation essay. In
Section 2.2, a comprehensive review of the literature and relevant theories is provided,
and hypotheses are derived. The research model is introduced in Section 2.3, the data and
variables are discussed, and econometric issues are addressed. In Section 2.4, the
regression results are presented. The article concludes with a summary of the study’s
findings and a discussion of their implications for managers and policy makers (Section
2.5). The study’s limitations are noted and directions for future research are provided as
well.
2.1. Introduction
The question of how a firm’s financial condition impacts the firm’s sales prices has been
investigated from multiple perspectives. Researchers from the economics, corporate
finance, and strategy fields have published a substantial amount of literature on this and
related issues (e.g. Borenstein and Rose 1995, Ferrier et al. 2002, Opler and Titman
1994). Yet, in summary, the findings have been largely inconclusive, not only across but
also within the respective research streams. Empirical research has found only limited, at
times ambiguous support for the contention that distressed firms’ sales prices tend to be
13
lower. This study draws on various theories from the economics, corporate finance and
strategic management fields to investigate this issue and attempts to reconcile the
apparent conflict by adopting a strategic contingency perspective that identifies in which
way and under what conditions firm financial distress may impact sales prices.
This research question is of particular interest given that firm financial distress is often
argued to lead to and result from price competition: Low market prices may drive firms
into bankruptcy, and the latter may, in turn, affect a firm’s competitive pricing behavior.
The so-called sick industry problem, thus, is intimately associated with the issues of
financial distress and price competition as repeatedly evidenced in the U.S. airline
industry. In recent years, many U.S. airlines have sought bankruptcy protection under
Chapter 11
4
, the ultimate manifestation of financial distress. Between 2001 and 2005
alone, seven of the top 20 U.S. carriers took advantage of the provisions of this code to
facilitate their restructuring processes
5
. An article in The Economist (2005) noted that “at
least half of America's airline industry has now been declared bankrupt” when Delta Air
Lines and Northwest Airlines declared bankruptcy in September 2005.
Airlines can achieve significant reductions in labor, leasing, and debt costs under Chapter
11 protection (McCafferty 1995), thus giving bankrupt firms a competitive advantage
over their non-bankrupt counterparts. Following Delta’s and Northwest’s bankruptcy
4
Title 11 of the U.S. code, commonly referred to as Chapter 11, is a form of interim bankruptcy and grants
the filing company protection from its creditors until a reorganization plan is developed and approved by
the creditor committees.
5
The top 20 U.S. commercial carriers were ranked based on 2001 passenger data (available from
www.transtats.bts.gov). The following carriers filed for Chapter 11 protection between 2001 and 2005:
TWA (2001), United (2002), US Airways (2002, 2004), Hawaiian (2003), ATA (2004), Delta (2005),
Northwest (2005).
14
filings, analysts therefore warned of potentially adverse consequences for other carriers
such as American Airlines and Continental Airlines (Trottman 2005). Consequently,
researchers (see e.g. Kennedy 2000, Rollman 2004) and managers of non-bankrupt firms
have repeatedly criticized the destructive implications of Chapter 11 protection. Gary
Kelly, then Chief Financial Officer with Southwest Airlines, for example, notes that “the
length of time an airline can go through bankruptcy protection and offer distressed prices
is very unsettling” (McCafferty 1995). Similarly, Robert Crandall, the former Chief
Executive Officer of American Airlines, argues that “Chapter 11 also undermines
responsible managements. In an intensely competitive industry providing a commodity
product, the ‘dumbest competitor’—unrestrained by fear of failure—sets the standard”
and hence calls for “bankruptcy laws designed to incentivize success and penalize
failure” (Crandall 2005). The criticism of Chapter 11 protection as unfair and destructive
is all but new: a 1989 article published in The Economist discusses the “uses and abuses”
of Chapter 11 and concludes that “what was designed as a shield has become a sword”
(Anonymous 1989).
Most of the previous statements make the explicit or implicit assumption that financially
distressed firms sell at lower prices than their healthier competitors. This contention,
however, has not found consistent theoretical and empirical support.
In the economics stream of research, Borenstein and Rose (1995) find that air fares
slightly decrease prior to bankruptcy filings, but do not further change in the time period
thereafter. Kennedy (2000) and Brander and Lewis (1986) assert that a firm’s financial
15
condition affects its market conduct, and Busse (2002) supports this contention,
indicating that financially distressed firms are more likely to start price wars than their
healthier competitors. The traditional economics literature, however, negates a
relationship between financial condition and firm output market behavior (e.g.
Modigliani and Miller 1958)
6
, and stresses the importance of demand fluctuations in
instigating price reductions.
From a corporate finance perspective, Baker (1973) argues that highly leveraged firms
are more risk-seeking than relatively profitable firms which take some of their “returns in
the form of reduced risk” (Hall and Weiss 1967, p.328). Along the same lines,
Maksimovic and Zechner (1991) suggest that financially distressed firms are more likely
to choose riskier (pricing) strategies. Opler and Titman (1994), in contrast, attribute the
lower performance of troubled firms to the (predatory) aggressiveness of competitors and
the costs of financial distress rather than to the firm’s own pricing behavior.
The strategy literature, finally, has focused the attention on the link between performance
distress and competitive behavior in general. Bowman (1982) contends that troubled
firms may be more risk-assertive (i.e. inclined to compete more aggressively) than
healthy firms, and Miller and Chen (1994) also relate past financial distress to
competitive aggressiveness. Ferrier et al (2002), however, find “that poor-performing
firms were less likely to exhibit aggressive competitive behavior” (p.311) when looking
at the direct relationship between performance distress and competitive aggressiveness. It
6
See also Brander and Lewis (1986) and Kennedy (2000).
16
is noteworthy that none of the studies in the strategy field have examined the link
between financial distress and prices in particular. Rather, competitive behavior has
typically been measured by counting and categorizing competitive actions and reactions
(Chen et al. 1992, Ferrier et al. 2002, Smith et al. 1991, Young et al. 1996).
These examples illustrate the inconclusiveness of prior research and suggest that the link
between financial distress and prices may be more complex (Singh 1986). The general
questions, thus, remain:
? How does a firm’s financial distress impact its pricing behavior?
? What parameters moderate the effect of firm financial distress on the firm’s prices?
As the research results referenced in the preceding paragraphs demonstrate, the answer to
these questions cannot be a straightforward one. There are multiple theoretical
perspectives and contingencies that may partly explain the variability of a troubled firm’s
pricing behavior. Focusing on competitive actions in general, Ferrier et al (2002) have
presented a first attempt to reconcile these conflicting views. They stress the importance
of context-specific contingencies such as industry growth and concentration, as well as
top management team heterogeneity in defining the relationship between performance
distress and competitive behavior. In fact, the strategy literature offers rich insights into
the contingencies that may moderate this relationship. This research builds on the work of
Ferrier et al (2002) in drawing on a broad theoretical basis and proposing a
comprehensive contingency framework that aims at characterizing the relationship
between financial distress and prices, and identifying factors that may affect the
17
magnitude of this relationship. In addition to developing and empirically testing this
contingency framework in the context of the U.S. airline industry, this research extends
the extant body of knowledge in three important respects:
First, price is used as a criterion variable. As mentioned earlier, none of the studies
published in strategic management journals examine the impact of financial distress on
prices. Yet, price is probably the single most important and relevant measure of
competitive behavior: From a consumer perspective, for example, prices are decisive in
determining consumer welfare ? the lower the prices, the greater the consumer surplus.
Consequently, prices are – under the assumption that the products and services offered by
firms are sufficiently homogenous – the primary driver of purchase decisions. From a
firm perspective, price is a key managerial decision variable affecting revenues and a
firm’s bottom line. Low prices may allow a firm to gain market share and obtain an
advantage over competitors, while a differentiation strategy may enable a firm to skim
the market and achieve higher prices (Porter 1980). Moreover, price is of interest from a
public policy point of view. Regulatory government bodies, such as the former Civil
Aeronautics Board (CAB) in the airline industry, and consumer interest groups survey
and screen markets for evidence of predatory pricing and intervene when free market
mechanisms of demand and supply fail to produce satisfactory market outcomes. Using
price as a dependent variable, rather than count and categorical variables such as number
and type of competitive actions, also allows for a more detailed evaluation of the
magnitude of a firm’s reaction to changes in its financial condition.
18
A second contribution lies in examining firm financial distress in more detail than has
been evident in most prior empirical work. While some studies focus on bankruptcy
filings (e.g. Borenstein and Rose 1995), others use measures such as Altman’s Z score
(Altman 1968) to evaluate a firm’s financial situation (e.g. Ferrier et al. 2002). There is,
however, substantial evidence that financial distress may differentially impact firm
behavior before, during, and after a Chapter 11 filing occurs (Borenstein and Rose 1995,
Busse 2002, Kennedy 2000). Therefore, both measures (a Z score-based distress measure
and bankruptcy dummy variables) are included in the empirical analyses to more
precisely sort out the effects of financial distress and bankruptcy per se. Furthermore, a
firm’s financial standing relative to its competitors in the market is considered. In fact,
financial distress in absolute terms may not necessarily imply any pricing actions if
competing firms find themselves in similar financial situations. More specifically, it is
expected that such pricing actions will be more pronounced when a distressed firm’s
financial situation is significantly different from that of its rivals.
Finally, this study is unique with respect to its empirical detail. A panel data set from the
U.S. airline industry is used to investigate the relationship between financial distress and
price. Unlike in many previous studies, the unit of observation in the analyses is a
specific route (i.e. “product”) market rather than a firm year or firm quarter (Busse 2002,
Chattopadhyay et al. 2001, Ferrier et al. 2002). This allows for a much more fine-grained
and statistically robust examination of the hypotheses.
This essay reports a comprehensive effort to understand if, when, and how firm financial
19
distress impacts prices. The empirical results suggest that financially distressed firms
offer lower prices than their healthier competitors, ceteris paribus. The magnitude of the
effect of firm financial distress on prices, however, is shown to decrease with unit
operating costs, increase with firm size, and decrease with firm market shares. The price
effects of financial distress are also stronger in more concentrated markets and when a
firm’s competitors are in significantly different financial situations. The insights provided
by this research will be useful to both firms and policy makers. Distressed firms and their
competitors gain a better understanding of how financial conditions typically impact
pricing decisions and customer demand. Managers of financially distressed firms may
benefit from this knowledge when developing turnaround strategies. Competing (healthy)
firms, on the other hand, can more accurately anticipate distressed firms’ pricing actions
and act accordingly. For policy makers, the findings of this study will help clarify if,
when, and to what extent financial distress and Chapter 11 protection impact sales prices
and the competitive behavior of firms. The findings presented here may help clarify if
current bankruptcy laws serve the purpose they were intended for, and contribute to
maintaining or improving the allocative efficiency of markets.
2.2. Theoretical background and hypothesis development
As briefly outlined above, there are competing perspectives on the relationship between
financial distress and prices. In this section, an overview of these theories from the
strategy, economics and corporate finance fields is provided and hypotheses are derived.
The research hypotheses are developed in two steps: In line with the Structure-Conduct-
20
Performance paradigm (see Figure 1), it is expected that there is a relationship between
financial distress and firm conduct in terms of prices. Several theories which further
support this contention are discussed in Section 2.2.1. Theories that may negate this
relationship are reviewed in Section 2.2.2. A contingency framework is proposed which
suggests that the relationship between firm financial distress and a firm’s pricing
behavior may be moderated by certain firm and structural characteristics (Section 2.2.3).
2.2.1. Financial distress as a driver of competitive pricing behavior
The strategy literature offers two theories, prospect theory
7
and organizational learning
theory that may support a negative relationship between financial distress and a firm’s
prices. Both theories are discussed in turn before empirical evidence and arguments from
standard microeconomic and corporate finance theory are set forth.
Prospect theory posits that decision makers are more risk seeking when facing situations
of likely loss while the inverse is true for decision makers operating in the domain of
profitability (Kahneman and Tversky 1979). Prospect theory can, thus, readily be applied
to evaluate the risk-taking behavior of financially troubled firms: Managers of low-
performing, troubled firms may be risk-assertive in their strategic choices in the
expectation of positive long-term returns to risk (in terms of increased market shares,
revenues, or profits, for example).
7
While prospect theory has its origins in the economics field, its concepts have been widely adopted by
strategic management researchers.
21
There is substantial support for the contention that troubled firms choose riskier strategies
in the strategic management literature (Bowman 1982, Moses 1992, Singh 1986,
Wiseman and Bromiley 1996). Chattopadhyay et al (2001) further investigate firms’
responses to threats such as declining organizational performance by considering
elements such as organizational characteristics and strategic type, and Wiseman and
Gomez-Mejia (1998) examine managerial risk taking across different governance modes.
While extending the basic framework of prospect theory, both papers still support the
hypothesized relationship between a firm’s level of distress and risk seeking behavior.
Authors have, thus, based their arguments on prospect theory when investigating the
relationship between organizational decline and risk taking behavior in general (Bowman
1982, Chattopadhyay et al. 2001, Shoham and Fiegenbaum 2002, Singh 1986, Wiseman
and Gomez-Mejia 1998), organizational adaptation (McKinley 1993) or innovation
(Mone et al. 1998).
With the connection between financial distress and risk taking behavior established, the
relationship between the latter and a firm’s pricing strategy can be characterized as
follows: As noted by Ferrier (2001), pricing actions represent a particular type of
competitive actions which have been associated with organizational risk taking (Ferrier et
al. 2002). Similarly, (Borenstein and Rose 1995) equate bankrupt firms’ “preference for
greater risk” (p.397) to competitive aggressiveness. Moses (1992) further notes that low
price strategies “sacrifice short-run profits in an attempt to establish a market and
generate profits over the long run” (p.40). He concludes that penetration strategies are
high risk strategies because the firm might incur further losses if costs fail to decrease
22
below price levels in the longer term. Pricing actions also entail the risk of imitation or
retaliation by competing firms. LeBlanc (1992), for example, suggests that low-cost
incumbents may choose to price aggressively in response to firms entering their (low-
price) markets. In more general terms, authors have investigated the dynamics of
competitive actions and responses and have found that a firm’s actions drive competitors’
responses (Chen et al. 1992), which in turn, determine the effectiveness and performance
effects of the focal firm’s actions (Chen 1996, Peteraf 1993, Smith et al. 1991). The risk
of choosing low price strategies in a homogenous competitive environment, thus, lies in
the possibility of unbalancing the competitive equilibrium (Xu and Tiong 2001) and the
potential loss resulting from aggressive competitive responses (Young et al. 1996). In
summary, prospect theory supports the argument that financial distress induces firms to
commit to a riskier, more aggressive pricing behavior, i.e. to lower prices.
Organizational learning theory also provides support for a positive relationship
between performance distress and strategic change or competitive aggressiveness (Ferrier
et al. 2002). Lant et al (1992) argue that previously unsuccessful firms undergo a learning
process which may lead to strategic reorientation, and Ferrier (2001) suggests that the
discrepancy between an organization’s goals and its actual performance provides
motivation for future actions and increases the likelihood of strategic change. To the
extent that pricing actions reflect changes in the underlying firm strategy, one may thus
argue that financially distressed firms are more likely to change their prices than are
healthy firms. Ferrier et al (2002), for example, note that “poor performance provides the
firm with strong incentives to aggressively search out new approaches to compete more
23
effectively in the marketplace” (p.304). It is thereby implicit that potential price changes
will typically involve lower prices (see also e.g. Ferrier 2001).
From a microeconomics and corporate finance perspective, Brander and Lewis
(Brander and Lewis 1986) argue that a firm’s “output market behavior will, in general, be
affected by [its] financial structure” (p.957, brackets added). Investigating the linkages
between financial and product markets, they demonstrate that highly leveraged firms will
likely compete more aggressively by increasing their output since riskier strategies with
(potentially) higher returns are more attractive to equity holders as a result of the limited
liability effect of equity financing, than are conservative strategies which primarily
appeal to debt holders. In a similar vein, Maksimovic and Zechner (1991) suggest that
highly leveraged firms choose technologies which are riskier in terms of their expected
cash flows. Hendel (1996) supports this assertion, arguing that “firms under financial
distress use aggressive pricing to generate cash” (p.309) and that prices are a function of
a firm’s liquidity.
A number of authors have empirically examined the relation between firm financial
condition and pricing behavior. Borenstein and Rose (1995) regress the change in prices
on a set of Chapter 11 indicator variables
8
and use a panel dataset from the U.S. airline
industry (1988-1992 data) to estimate their model. They find support for the theoretical
contentions summarized above, indicating that air fares drop by five to six percent in the
months preceding the carrier’s Chapter 11 filing. Kennedy (2000) demonstrates that a
8
As noted by the authors, the effects of many other variables typically included in price estimation
equations are assumed negligible and are excluded from the model specification.
24
distressed firm’s sales revenues and profits (and that of its rivals) decrease prior to
bankruptcy as a result of its altered product market conduct. He analyzes 51 bankruptcy
filings and uses Chapter 11 indicator variables and a small set of market and firm-specific
control variables to predict revenues and profit margins. Analyzing U.S. airline data from
the 1985 to 1992 period, Busse (2002) finds that highly leveraged firms are more likely to
start price wars. Busse also argues that “firms in poor financial condition discount future
revenues more heavily than do financially sound firms” (p.298), thus focusing on
boosting short term sales (by cutting prices, for example).
Taken together, there is theoretical and empirical support for the contention that
financially distressed firms choose riskier strategies and price more aggressively, i.e.
follow a low-price strategy in an effort to gain market shares and boost sales
9
. Hypothesis
1 is stated as follows:
Hypothesis 1: Financial distress negatively impacts prices.
It may also be argued that a firm’s prices will affect firm financial condition. If prices are
consistently below marginal costs, the firm’s financial situation will deteriorate. Prices
above marginal costs, in turn, will positively impact firm financial condition as long as
marginal costs are larger than average costs. The possibility of such reverse causality is
not further explored in this research. Firm financial condition is, of course, a firm-level
phenomenon while prices are market-specific. A firm’s financial distress may, as argued
9
See also Ferrier et al (2002) for a definition of competitive aggressiveness.
25
here, impact a firm’s pricing behavior in all markets, but the sales price in an individual
market will not necessarily affect the firm’s financial standing. In fact, the latter may only
be true if all prices are systematically lower (or higher) than marginal costs. This is,
however, a strong assumption which requires empirical and theoretical substantiation.
Such work is not within the scope of this analysis and is left for future research. This
research uses firm level financial distress to estimate multi-market firms’ market level
prices. It is therefore assumed that problems of endogeneity, caused by reverse causality,
do not arise.
Hypothesis 1 implies that financially distressed carriers may be expected to sell at lower
prices, all else equal. Prior research, however, suggests that the above hypothesized price
effect of firm financial distress may intensify as bankruptcy occurs. As Borenstein and
Rose (1995) and Kennedy (2000) have shown, firms try to prevent insolvency by
generating cash through aggressive competition prior to bankruptcy filings. Once these
firms operate under Chapter 11 protection, however, they benefit from lower operating
costs as debt payments are paused (Barla and Koo 1999, Rose-Green and Dawkins 2002)
to support the restructuring of the firm. This lower cost base may allow bankrupt firms to
charge even lower prices. Moreover, soft demand may force carriers to cut fares once
they operate under bankruptcy protection since the latter signals uncertainty to consumers
(Hofer et al. 2005). Barla and Koo (1999) further suggest that firms “under protection of
Chapter 11 are more likely to adopt short term profit maximization behaviors” which
equate to “prices that are well below long run marginal costs” (p.104) when demand is
low (see also Hofer et al. 2005).
26
In summary, there are three rationales which support the contention that the effect of
financial distress on prices should be stronger during bankruptcy than prior to the Chapter
11 filing (see also Hofer et al. 2005): First, when operating under bankrupt protection,
firms benefit from lower costs and may pass these savings on to consumers in the form of
lower prices. Second, bankrupt firms may experience lower demand due to the
uncertainty concerning the firms’ future operations. Third, bankrupt firms may focus on
short term profit maximization and thus offer lower prices, ceteris paribus. Consequently,
the following hypothesis is suggested:
Hypothesis 2: The negative impact of financial distress on prices is greater during
bankruptcy than prior to the Chapter 11 filing.
As indicated previously, a different set of theories suggest that a firm’s financial distress
may not significantly impact its prices. These perspectives are reviewed below, and
hypotheses that suggest that the relationship between financial distress and prices is
moderated by other factors are formulated.
2.2.2. Conflicting theoretical arguments
In this section, theoretical arguments and empirical results from the industrial
organization economics, game theory and finance literatures that do not provide support
for the financial distress-price relationship are reviewed.
27
The threat-rigidity model has emerged as a counterhypothesis to prospect theory. Staw
et al (1981) argue that individuals, groups, and organizations exhibit restrictive
information processing patterns, centralize control and conserve resources when faced
with threatening situations. These mechanisms result in increased rigidity which reduces
an organization’s ability to change and adapt to its environment (McKinley 1993). As
noted by McKinley (1993) and Mone et al (1998), there is broad empirical support for the
threat-rigidity model: Smart and Vertinsky (1984), for example, find that executives
consult fewer information sources during crises, and Chattopadhyay et al (2001) present
some evidence that organizations respond to control-reducing threats with low risk,
internally directed actions. From this perspective, firms in poor financial conditions may,
thus, be expected to not reduce prices in the short-term, but to behave passively and
conservatively (Ferrier et al. 2002).
Similar rigidity arguments can be found in the industrial organization economics, game
theoretic and finance literature. First, the kinked demand curve theory suggests that
firms in oligopolistic markets with few sellers and rather homogenous products face
highly inelastic demand for price decreases (Waldman and Jensen 2001). Put differently,
firms will refrain from price competition given that their rival firms may be expected to
match these moves, thus offsetting any profit gains (Scherer 1980). This argument is
further supported by game theory: Derfus et al (forthcoming) argue that pricing actions
are negative-sum actions since all competing firms will be worse off after implementing
successive price reductions. Consider, for example, a sequential game between
28
duopolists: Firm Two observes Firm One’s move and subsequently acts in response to
Firm One’s action. Firm One, in turn, observes Firm Two’s action and may choose to
react, etc. (Gibbons 1992). When such moves consist of price reductions, the price may
fall below average cost levels in the course of this competitive interaction of moves and
countermoves (see also Dasgupta and Titman 1998). These theories are, thus, in line with
the imitation/retaliation argument discussed earlier (Busse 2002, Chen 1996, Chen et al.
1992, Peteraf 1993, Smith et al. 1991).
In summary, the threat-rigidity model and arguments from the industrial organization,
game theoretic, and finance literatures suggest that financially distressed firms may
refrain from lowering prices as information processing and decision making processes are
altered in the face of threats or for fear of retaliation.
2.2.3. The contingency approach
This essay attempts to reconcile the apparent theoretical and empirical conflict that has
shaped previous research on the relationship between financial condition and prices. Each
of the groups of theoretical arguments – those supporting and those denying a negative
impact of financial distress on prices – may be valid under specific circumstances. As
will be discussed below, there are a number of contingencies that may impact the
relationship under investigation. Similar to Ferrier et al (2002), a contingency framework
which suggests moderating effects of organizational and market structural characteristics
is developed. This framework aims at defining in what instances the price effects of
29
financial distress are largest.
As shown in Figure 2, two groups of contingencies are hypothesized to impact the
relationship between financial distress and prices are presented: organizational
characteristics and market characteristics. Both groups of variables are discussed in turn,
and hypotheses are formulated.
Organizational characteristics
It is suggested that the relationship between firm financial distress and prices is
moderated by certain organizational characteristics. More specifically, a firm’s operating
costs, its size and market shares are hypothesized to influence the extent to which firm
financial distress impacts the firm’s pricing behavior. The importance of these factors has
been shown in prior research.
Prior research has suggested that a firm’s particular strategic type may impact its
behavior. Chattopadhyay et al (2001), for example, find that a firm’s propensity to
respond to threats with externally as opposed to internally oriented actions is impacted by
its strategic focus. They present empirical support for the contention that firms focusing
on product-market development (prospectors) are more likely to act externally (by
changing prices, for example) since the “effectiveness of a product-market development
strategy depends to a large extent on controlling or modifying the external environment”
(p.940/941). Firms focusing on domain defense (defenders), in turn, “are more likely to
act within themselves to become more efficient through standardizing organizational
30
processes” (p.941). Therefore, a differential impact of financial distress on a firm’s prices
by strategic type is expected, given the firms’ differential inclinations to act externally
versus internally in response to changes in financial situation.
Although there are multiple definitions and classifications of strategic types (see e.g.
Shoham and Fiegenbaum 2002), these can be simplified and synthesized as follows (see
also Chattopadhyay et al. 2001): Defenders are those firms that operate in a stable, well-
defined set of market segments, tend to act conservatively, and are characterized by
deadlocked organizational structures and operating routines. Prospectors, in turn, are
those firms that constantly seek opportunities to expand their business and whose most
distinctive features are their innovativeness and cost-leadership.
In the empirical practice, many operationalizations of strategic types have been
suggested, ranging from simple dichotomies (e.g. Peteraf 1993) to multidimensional
clusters (Smith et al. 1997). There is, however, substantial agreement in the literature that
a firm’s costs are an important differentiator with respect to its strategic type (see the
above definitions of prospectors and defenders). This is particularly true in the U.S.
airline industry: Both the academic and trade presses frequently refer to specific airlines
as either high-cost carriers or low-cost carriers. Peteraf (1993), for example distinguishes
between pre- and post-deregulation air carriers, the former being mostly high-cost firms
10
while the latter are virtually all low-cost airlines. A firm’s strategic type is therefore
identified by means of its operating costs. In fact, an airline’s relative cost
10
Southwest Airlines being a notable exception.
31
(dis)advantage may impact its choice of strategy. Assuming that lower operating costs
also imply higher profit margins, low-cost firms have some financial flexibility to allow
for price reductions and potentially ensuing price wars. Higher operating costs (and lower
profit margins), in turn, would imply that price cuts likely lead to increased operating
losses.
The crucial assumption for this reasoning to be valid is, of course, a negative correlation
between operating costs and profit margins. The empirical analyses will be conducted
using data from the U.S. airline industry
11
. Accordingly, financial data on U.S. airlines
were collected for a total of eight quarterly time periods (1992 and 2002) from the
Bureau of Transportation Statistics. An analysis of these data indicates that the
correlation coefficient between operating costs per available seat-mile and operating
profit per available seat-mile is equal to r = -0.1481 and is statistically significant at the
five percent level (p = 0.0485)
12
. This result provides some support for the contention that
firms with lower operating costs tend to achieve higher profits and may be able to operate
profitably even if prices are cut. Firms with higher costs and lower profit margins, in turn,
do not have this flexibility and may tend to refrain from lowering prices. The negative
effect of financial distress on prices may thus be expected to decrease with the magnitude
of the firm’s operating costs as depicted in Figure 3 below. The coefficient of the
associated interaction term is, thus, expected to be positive.
11
Further information about the data sources and the nature of the data set is provided in Chapter 2.3.
12
This correlation analysis is based on firm-level 228 observations.
32
Figure 3: The moderating effect of operating costs on the distress-price relationship
Accordingly, Hypothesis 3 is proposed as follows:
Hypothesis 3: The negative effect of financial distress on prices decreases with the
magnitude of the firm’s operating costs.
The effect of firm financial distress on prices may also be impacted by the firm’s size.
Commenting on the survivability of large firms, Tiras (2002) notes that creditors have
greater confidence in the turnaround performance of large distressed firms and thus grant
them more favorable loan conditions than to small firms. In a similar vein, Smith and
Graves (2005) suggest that “larger firms are likely to have a higher probability of
survival, as the potential losses to stakeholders are greater. Also, such firms are likely to
have a higher profile and therefore more likely to be kept alive” (p.306).
Financial
distress
Price ($)
a
v
e
r
a
g
e
lo
w
o
p
e
r
a
t
in
g
c
o
s
t
h
ig
h
o
p
e
ra
tin
g
c
o
s
t
positive
interaction
effect
Financial
distress
Price ($)
a
v
e
r
a
g
e
lo
w
o
p
e
r
a
t
in
g
c
o
s
t
h
ig
h
o
p
e
ra
tin
g
c
o
s
t
positive
interaction
effect
33
Looking at bankrupt firms, in particular, prior research suggests that larger bankrupt
firms have a bankruptcy cost advantage due to scale effects in reorganization costs
(Campbell 1996). The costs of bankruptcy consist of both direct and indirect costs, where
the former “include lawyers’ and accountants’ fees, other professional fees, and the value
of managerial time spent in administering the bankruptcy”, and the latter “include lost
sales, lost profits, and possibly the inability of the firm to obtain credit or to issue
securities” (Warner 1977, p.338)
13
. Numerous researchers have attempted to estimate
these costs. Their estimates vary significantly due to differences in cost definitions,
variable measurement, sample composition, and estimation methodology. The estimates
range from an average of 1.3% of the change in firm value during bankruptcy in the
railroad industry (Warner 1977) to 4% of the firm value in the retail business (Altman
1984), and up to 16.35% of the firm value for a cross-section of industries (Branch
2002)
14
.
Many researchers note, however, that there are significant scale economies in bankruptcy
costs: Warner (1977), for example, finds that bankruptcy costs are linearly decreasing
with firm size. His analyses indicate that bankruptcy costs may be as high as about nine
percent of the firm’s market value for firms with a market value of less than 30 million
dollars and as low as two percent for firms with a market value of around 120 million
dollars. Analyzing bankruptcies in the U.S. trucking industry, Guffey and Moore (1991)
also find a significant negative correlation between firm size (as measured by total asset
13
A more detailed discussion of the composition of bankruptcy costs can be found in Guffey and Moore
(1991) and Branch (2002).
14
See also Bradbury and Lloyd (1994) for a summary of prior research estimating bankruptcy costs.
34
values) and bankruptcy costs. Betker (1997), in turn, finds that the relationship between
firm size (total assets) and bankruptcy costs follows an inverted U shape: The direct
effect of assets on bankruptcy costs carries a positive coefficient while the coefficient of
the squared asset value carries a negative coefficient. The observation of the relationship
between firm size and bankruptcy costs has led researchers to conclude that such
bankruptcy costs may significantly impact smaller firms’ decisions while they may not
substantially impact large firms (Bradbury and Lloyd 1994). Consequently, it is expected
that larger bankrupt firms may be able to offer lower prices than smaller firms due to
their bankruptcy cost advantage.
Previous research has also found that larger firms tend to remain in bankruptcy for longer
periods of time and exhibit significantly higher survival rates than smaller firms (Queen
and Roll 1987, Rodgers 2000). The latter observation may be attributed to lower
bankruptcy costs (Campbell 1996), for example. These advantages in terms of credit
conditions, stakeholder confidence, and bankruptcy costs may allow larger distressed
firms to commit to riskier turnaround strategies that involve more aggressive pricing
behaviors. While detrimental in the short term, the latter may drive competitors out of the
market and result in greater long term returns. It is expected that the negative effect of
firm financial distress on prices will be stronger for larger firms. Consequently, the
interaction effect between financial distress and firm size is hypothesized to positively
affect prices, as noted in Hypothesis 4 and illustrated in Figure 4.
35
Hypothesis 4: The negative effect of financial distress on prices increases with firm
size.
Figure 4: The moderating effect of firm size on the distress-price relationship
The magnitude and direction of the effect of firm financial distress on prices may also
depend on the firm’s market share in the particular product (i.e. route) market. In the long
run, greater market shares may result in the achievement of lower marginal costs
through economies of density (Ferrier et al. 2002). Furthermore, high market shares may
be indicative of barriers to entry and mobility that isolate market-leading firms from
intense competition (Caves and Porter 1978, Caves and Ghemawat 1992). From this
perspective, high market shares may be considered a valuable firm resource that allows
for above-normal returns. Consequently, some researchers have argued that firms will
likely try to defend their market power. Busse (2002), for example, presents empirical
evidence that firms are more likely to enter price wars the greater their market shares, and
Financial
distress
Price ($)
a
v
e
r
a
g
e
la
r
g
e
r
f
ir
m
s
s
m
a
lle
r firm
s
negative
interaction
effect
Financial
distress
Price ($)
a
v
e
r
a
g
e
la
r
g
e
r
f
ir
m
s
s
m
a
lle
r firm
s
negative
interaction
effect
36
LeBlanc (1992) argues that firms strive to maintain monopoly profits by implementing
limit or predatory pricing.
These predictions may, however, not hold when explicitly considering distressed firms.
First, note that distressed firms typically focus on short term survival rather than on long
term strategic positioning. While the latter is the ultimate purpose of distressed firms’
turnaround efforts, generating sufficient cash flows is a mandatory obligation these firms
face in the immediate future. In this vein, bankrupt U.S. airlines frequently terminate
unfavorable aircraft leases and collective labor agreements right upon entry into Chapter
11 protection. If liquidity is the prime objective, however, price cuts in an effort to
maintain market shares may prove counter-productive for high market share firms: Any
price reductions will imply lower total revenues since the incremental increase in
customer demand likely will not outweigh the detrimental effect of lower sales prices.
Assuming (quasi-)fixed production costs in the short run, these revenue losses directly
affect the firm’s bottom line. Low-market share firms, in turn, may see a substantial
increase in customer demand when reducing prices. The prospect of increased volume
may, thus, offset the negative effect of lower sales prices. This implies that engaging in
price competition is more appealing to firms with smaller market shares: The potential
market shares to be gained are greater, and any pricing actions hurt the market leading
firm(s) significantly more than the smaller firm. This reasoning reflects the concepts of
Judo economics (Gelman and Salop 1983) and Judo strategy (Yoffie and Kwak 2002),
which essentially posit that a firm’s market shares (and/or size) may constitute a
competitive disadvantage when adequately leveraged against it by smaller firms (in terms
37
of market shares).
Standard microeconomic theory further suggests that firms with greater market shares
possess market power and can charge price premiums (see e.g. Borenstein 1989).
Extending this argument to the present research context, it is expected that distressed
firms with higher market shares have higher degrees of market power and will be
required to compete on prices to a lesser extent than firms with lower market shares and
little market power. Firms with higher market shares may be able to retain greater shares
of market demand due to customer retention instruments such as loyalty programs which
create higher switching costs for consumers. The latter may thus be reluctant to switch to
financially stronger competitors even though they may seem more reliable or offer lower
prices. From this perspective, demand inelasticity confers firms with greater market
shares greater degrees of market power. And such market power, in turn, enables even
distressed firms to maintain higher price levels, ceteris paribus.
In summary, these arguments thus suggest that higher market shares reduce the negative
effect of firm financial distress on prices (see Figure 5), and the associated interaction
effect is expected to be positive. Hypothesis 5 below formally states this contention:
Hypothesis 5: The negative effect of financial distress on prices decreases with the
firm’s market share.
38
Figure 5: The moderating effect of market shares on the distress-price relationship
A second set of contingencies, those relating to market characteristics, are discussed
below.
Market characteristics
Besides organizational characteristics, select market characteristics are hypothesized to
impact a distressed firm’s pricing strategy. Market concentration is one of the most
widely used measures of the competitiveness of markets in the extant literature. While
there are many alternative measures of market structure—the number of sellers in the
market and multi-market contact measures, for example, have been used to characterize
the structure of markets in prior research (e.g. Mazzeo 2002, Scott 1982)—the degree of
market concentration is likely highly correlated with these alternative measures and
Financial
distress
Price ($)
a
v
e
r
a
g
e
lo
w
m
a
r
k
e
t
s
h
a
r
e
h
ig
h
m
a
rk
e
t s
h
a
re
positive
interaction
effect
Financial
distress
Price ($)
a
v
e
r
a
g
e
lo
w
m
a
r
k
e
t
s
h
a
r
e
h
ig
h
m
a
rk
e
t s
h
a
re
positive
interaction
effect
39
appropriately captures the structural characteristics of a market. The second market-
specific factor included in this study is the financial condition of all the firms in a market.
This variable is included to evaluate how a firm’s financial condition differs from the
average distress level of the other firms in the market and how this relative difference
impacts the magnitude of a firm’s pricing actions.
First, market concentration will likely affect a firm’s pricing decision. More
specifically, the expectation of competitive responses and retaliatory moves in highly
concentrated markets impacts a firm’s valuation of the effects of any price changes. The
structure-conduct-performance paradigm posits that industry concentration reduces the
level of competition (Scherer 1980, Waldman and Jensen 2001). Young et al (1996) find
empirical support for this contention, noting that firms in concentrated markets or
industries carry out fewer competitive moves. The underlying assumption of this
reasoning is, however, that the competing firms are similar to one another and that their
products are largely homogeneous. Waldman and Jensen (2001) list a variety of factors
that violate this homogeneity assumption and may hinder effective collusion between
firms in concentrated markets. Cost differences between competing firms, for example,
may negatively affect the ease of collusion.
A deterioration in a firm’s financial position, and bankruptcy in particular, may bring
about such cost differences: firms operating under Chapter 11 protection, in particular,
may pause debt payments and shed financial obligations such as contributions to pension
plans, for example (Rose-Green and Dawkins 2002). This new cost structure may then
40
lead to the firm’s repositioning in the product market. Specifically, a change in a firm’s
operating costs changes the firm’s profit maximization problem, and consequently its
optimal price levels. The interaction of market concentration and financial distress may
therefore lead to a destabilization of collusive arrangements and increase pricing
competitiveness (Barla and Koo 1999). While market concentration is expected to be
positively related to prices, this research contends that this positive relationship will
diminish in magnitude in the light of an aggravation of a firm’s financial condition. Put
differently, the interaction of financial distress and market concentration is expected to
negatively affect prices, ceteris paribus (Hypothesis 6).
Hypothesis 6: The impact of financial distress on prices is greater, the higher the level
of market concentration.
Figure 6 illustrates the differential effect of financial distress on prices as a function of
the degree of market concentration.
41
Figure 6: The moderating effect of market concentration on
the distress-price relationship
A distressed firm’s pricing decisions will, in part, also depend on its competitors’
financial situations. If a firm’s rivals experience similar degrees of distress as the focal
firm does (and assuming that the firms’ products are undifferentiated), then these rivals
may be expected to exhibit comparable or symmetric pricing behaviors. A focal firm’s
price reductions would then be matched by the other firms, and no single firm could gain
a competitive advantage. In fact, game theory suggests that in a perfectly competitive
setting each firm will always have an incentive to slightly undercut its competitor’s
prices, thus eroding profit margins to zero (Gibbons 1992). Financially distressed firms
will, therefore, avoid competing on price when their competitors find themselves in
similar financial conditions. Conversely, Hypothesis 7 is stated as follows:
Financial
distress
Price ($)
a
v
e
r
a
g
e
lo
w
m
a
rk
e
t c
o
n
c
e
n
tra
tio
n
h
ig
h
m
a
r
k
e
t
c
o
n
c
e
n
t
r
a
t
io
n
negative
interaction
effect
42
Hypothesis 7: The greater a firm’s financial distress relative to its competitors, the
lower the firm’s sales prices.
In summarizing, a set of hypotheses on the link between firm financial distress and firm
prices has been formulated based on a variety of theoretical perspectives. Conflicting
viewpoints that may suggest the absence of any significant relationship respectively are
presented, and a contingency framework that more precisely defines for what type of
firms and under what circumstances changes in a firm’s financial situation may indeed
cause changes in the firm’s pricing behavior is proposed. The resulting model is shown in
Figure 7.
Figure 7: Research model
Financial
distress
Market
concentration
Relative fin.
distress
Control variables
Market share
Firm size
Operating costs
Price
M
a
r
k
e
t
(
s
t
r
u
c
t
u
r
a
l
)
c
h
a
r
a
c
t
e
r
i
s
t
i
c
s
F
i
r
m
c
h
a
r
a
c
t
e
r
i
s
t
i
c
s
Financial
distress
Market
concentration
Relative fin.
distress
Control variables
Market share
Firm size
Operating costs
Price
M
a
r
k
e
t
(
s
t
r
u
c
t
u
r
a
l
)
c
h
a
r
a
c
t
e
r
i
s
t
i
c
s
F
i
r
m
c
h
a
r
a
c
t
e
r
i
s
t
i
c
s
43
In the following section, information about the sample data that is used for the empirical
analyses is provided, and measurements of the variables in the research model as well as
methodological issues are discussed.
2.3. Data and methodology
The U.S. airline industry provides the setting for the empirical analyses. This selection is
particularly suitable for a number of reasons. First, the markets are clearly defined (Smith
et al. 1991), and all firms operating in these markets are dominant-business firms (Peteraf
1993), i.e. firm-specific data reflect the firms’ aviation activities and are not diluted by
non-aviation business activities. Second, the U.S. airline industry is highly competitive
and encompasses a large cross-section of routes that differ significantly with respect to
their market characteristics (Peteraf 1993, Smith et al. 1991). Third, the industry has
experienced periods of severe financial distress (Borenstein and Rose 1995), but is
sufficiently heterogeneous with respect to the airlines’ financial conditions. Finally, there
is a wealth of publicly available data on the U.S. airline industry due to the U.S.
Department of Transportation’s reporting requirements
15
.
15
Some sections of this chapter, particularly the sample data and variable descriptions, are similar or equal
to the corresponding sections of a related paper published by Hofer et al (2005).
44
2.3.1. Data sample
Data were collected on the top 1000 U.S. domestic origin and destination route markets
16
,
for all quarters in 1992 and 2002. These years were chosen because the airline industry
experienced serious distress in the early nineties (Barla and Koo 1999) and in the
aftermath of the 9/11 attacks. At the same time, limiting the analyses to two years only
allowed keeping the dataset at a manageable size. The sensitivity of the empirical results
with respect to the selection of these particular time periods is investigated by re-
estimating the regression models using an extended data set that also includes 1997 data.
These results will be discussed in Section 2.4.3.
Quarterly data are used to capture the short-term effects of financial distress and Chapter
11 filings on air fares. The top 1000 route markets cover a wide range of route
characteristics in terms of traffic volume, distance, and intensity of competition. The unit
of observation is a specific carrier’s fare on a particular route market in a given time
period.
The raw data were purchased from Database Products Inc. (DPI), a reseller of the
Department of Transportation’s DB 1A data which contain a 10% sample of all U.S.
domestic origin and destination tickets. DPI downloads the DB 1A data and screens them
for erroneous and redundant data entries. These entries and data points from non-revenue
16
Based on 2002 traffic figures, 48 contiguous states only.
45
transactions
17
are removed from the dataset. The data obtained from DPI thus are filtered
and quality-controlled and provide airline and route specific information on fares,
nonstop and itinerary miles, the number of passengers, and the number of coupons.
Additional air traffic and airline operating and financial data were gathered from the
DOT’s T-1
18
and Form 41
19
databases. Other data sources include the American
Transport Association (ATA; U.S. airline bankruptcy data), the Bureau of Labor
Statistics (BLS; income data and inflation indexes) and the Bureau of Economic Analysis
(BEA; population statistics).
Observations from carriers with less than five percent route market share were deleted
from the data set to keep the data set at a manageable size
20
. Furthermore, a total of 577
observations were excluded because of unidentified carriers
21
, or unavailable airport and
airline-specific data. A total of 23,039 observations were retained for the analyses. Each
observation indicates data for a specific carrier on a specific route market in a specific
time period.
2.3.2. Variables and measurement
This section provides detailed information on the variables used in this research model.
17
E.g. personnel travel and frequent flyer award travel.
18
Table T-1 provides summaries of T-100 data by carrier, aircraft type and service class and includes
information on available seat miles (ASM) and revenue passenger miles (RPM).
19
Form 41 (financial schedule) contains financial information on large U.S. certified air carriers including
data from balance sheets, income statements, and information on cash flows, and aircraft operating
expenses.
20
This is common practice: Borenstein and Rose (1995), for example, exclude all observations of carriers
with less than ten percent route market shares.
21
“XX – unduplicated commuters” and “UK – unknown carrier”.
46
The purpose of this research is to investigate the effect of financial distress on prices.
Consequently, ticket prices (fares) are used as the dependent variable. Among those
factors that may explain and predict variations in ticket prices, firm financial distress is of
particular interest here. Other independent variables include not only the aforementioned
moderating factors—operating costs, firm size, market shares, and market concentration
(see also Figure 7)—but also a set of airline-specific, route-specific, and airport-specific
characteristics that have been shown to impact air fares in prior research (Hofer et al.
2005). The dependent variable is discussed first, followed by the independent variables
of interest. In addition, information on the set of control variables included in the
empirical estimation model is provided. Descriptive statistics and a correlation table are
provided in Section 2.3.3.
2.3.2.1. Dependent variable
Previous studies published in the strategic management literature have measured the
impact of financial condition on firm behavior in terms of the number and type of
competitive actions, response speed and delay, for example (Chen et al. 1992, Ferrier
2001, Ferrier et al. 2002, Smith et al. 1997, Smith et al. 1991, Young et al. 1996). While
price data are commonly used as dependent variables in the economics literature, this is,
to the best of the author’s knowledge, the first study to investigate the impact of financial
distress on prices from a strategic management perspective. More specifically, Fare
kij
is
the average price carrier k charges on the route between airports i and j
22
. All fare values
22
Fare values are averages across all booking classes and do not include taxes and fees.
47
are one-way fares based on roundtrip purchases and are given in real 1992 U.S. dollars
23
.
2.3.2.2. Independent variables
The measurement of financial distress is of particular interest in the context of this study.
Previous studies of financial condition have generally relied on one of two measures.
Ferrier et al (2002) and Chakravarthy (1986), for example, relied on a composite measure
to evaluate a firm’s financial situation. Altman’s (1968) Z score is the most prominent
member of this group of measures and takes into account the firm’s past and present
profitability, its liquidity and its degree of activity. Other researchers have focused on
Chapter 11 filings
24
, the most visible and definite sign of financial distress, to investigate
the effects of firm financial distress (e.g. Borenstein and Rose 1995, Kennedy 2000).
While both measures have their merit, it is important to note that they capture different
aspects of financial distress. Z score-type measures are indicators of a firm’s financial
health (or distress), while Chapter 11 filings refer to a specific point in time at which the
firm is no longer able to meet its debt obligations. The model builds on both of these
indicators and includes four measures of financial distress to more precisely sort out its
effects on firm behavior in terms of pricing:
? Distress
k
is a measure of Airline k’s financial distress. The Distress variable is the
inversion of firms’ Z scores. More specifically, Z’’ scores (Altman 2002) are used, a
revised version of Altman’s original Z score formulation (1968) which is particularly
suitable for firms operating in service industries (such as the airline industry). The
23
All nominal values were converted to real 1992 dollars using the appropriate price indexes published by
the Bureau of Economic Analysis.
24
See Daily (1994) for a comprehensive explanation and discussion of the U.S. Code Chapter 11.
48
more recent Z’’ scores (Altman 2002) are also preferred over the original Z score
formulation (Altman 1968) since it has been shown that “the relation between
financial ratios and financial distress changes over time” (Grice and Ingram 2001)
such that more recent formulations are more reliable and effective in predicting a
firm’s financial distress. Based on discriminant analysis, Altman (2002) developed
the following model to estimate a firm’s financial fitness:
1 2 3 4
'' 6.56* 3.26* 6.72* 1.05* Z X X X X = + + + where X
1
= working capital / total
assets; X
2
= retained earnings / total assets; X
3
= Earnings Before Interests and Taxes
(EBIT) / total assets; X
4
= book value of equity / total liabilities. All airline financial
data needed to compute the Z’’ scores were obtained from the Department of
Transportation’s Form 41 data which are available online on a carrier-time period
basis. High Z’’ scores indicate financial health, while low and negative scores
indicate (serious) financial distress. Specifically, it has been suggested that scores of
2.60 or above indicate financial health, while scores of 1.10 or lower indicate severe
distress. To facilitate the interpretation of the estimation results, the Z scores are
inverted, i.e. ( ) 1 Distress ZScore = ? ? , such that higher (positive) Distress scores
indicate financial distress (see also Ferrier et al. 2002). This variable is used to test
Hypothesis 1. Moreover, the airlines’ Distress scores are used to test the moderating
effects from Hypothesis 3, Hypothesis 4, Hypothesis 5, and Hypothesis 6,
respectively.
? Chpt11Ops
k
is a binary (0/1) variable that identifies those carriers that operate under
Chapter 11 protection (“1”). It thus is an alternate, though rather coarse, measure of a
firm’s financial distress. All bankruptcy data were obtained from WebBRD, a
49
bankruptcy research database that is accessible online athttp://webbrd.com/. This
database is maintained by Professor Lynn M. LoPucki with the University of
California at Los Angeles. This database specifies the dates at which firms (airlines)
entered into and exited from Chapter 11 protection. These data were also double-
checked with the bankruptcy data which are available online at the Air Transport
Association’s website (http://www.airlines.org/econ/). No discrepancies were found.
? Pre4Chpt11
k
, identifies those carriers that will face bankruptcy within the following
four quarters. In the latter case, this binary variable takes on the value of “1”. This
variable is based on the same sources as the Chpt11Ops variable defined above. A
four-quarter period (prior to the Chapter 11 filing) is selected to best capture price
reactions to aggravating financial distress in the time period immediately preceding
bankruptcy.
? Post4Chpt11
k
is similar to the Pre4Chpt11
k
variable, but identifies those carriers that
filed for Chapter 11 within the past four quarters (“1”). This variable is based on the
same sources as the Chpt11Ops variable defined above. The inclusion of the
Pre4Chpt11
k
and Post4Chpt11
k
variables allows capturing the differential impact of
financial distress over time as stated in Hypothesis 2.
? The Chpt11
k
variable is an indicator variable which is equal to “1” if the focal carrier
filed for bankruptcy protection in the current quarter. The number of observations in
which this is the case is small such that this variable is only used for descriptive
purposes (see Figure 9) and is not included in the regression analysis. An overview of
the Chapter 11 dummy variables is provided in Figure 8 below. Note that the
Pre4Chpt11, Chpt11, and Post4Chpt11 variables are mutually exclusive.
50
Figure 8: Overview of Chapter 11 indicator variables
? DistressDiff
kij
is an indicator of an airline’s financial standing relative to its route
competitors. It is based on Altman’s Z’’ score and is computed for each carrier in
each route market for each time period. It is the difference between the focal carrier’s
Z’’ score and the route market share weighted average of its route competitors’ Z’’
scores:
( )
*
competitors competitors
focal
competitors
Distress route shares
DistressDiff Distress
route shares
= ?
?
?
. Higher scores,
thus, indicate that the focal carrier is financially better off relative to its route
competitors and vice versa. The DistressDiff
kij
variable is designed to test Hypothesis
7 which refers to an airline’s financial standing relative to its competitors. This
variable, thus, differs from the Distress and Chpt11 variables in that it indicates an
airline’s relative financial standing, i.e. the focal firm’s Distress relative to the market
share weighted average Distress of its competitors, rather than its absolute financial
distress. Positive DistressDiff values indicate that the airline is financially worse off
than its route competitors, while negative values indicate relative financial wellbeing.
Chapter 11
filing
Pre4Chpt11 Post4Chpt11
Chpt11
Chpt11Ops
time
(quarters)
Chapter 11
filing
Pre4Chpt11 Post4Chpt11
Chpt11
Chpt11Ops
time
(quarters)
51
Further, a set of moderating variables is included as suggested in Hypothesis 3 to
Hypothesis 6. More specifically, it is hypothesized that the impact of financial distress on
prices varies by strategic type/operating costs (Hypothesis 3), firm size (Hypothesis 4),
firm market shares (Hypothesis 5), and market concentration (Hypothesis 6). These
moderating variables are operationalized as follows:
? AirlineCost
k
is an indicator of an airline’s operating efficiency. It is defined by the
ratio of operating expenses to available seat miles (ASM).
? Size
k
indicates the firm’s size in terms of its total assets (measured in 000s of U.S. $).
? RouteShare
kij
measures an airline’s market share on a route market (based on its share
of route passengers).
? RouteHHI
ij
is a measure of route market concentration. It is based on the Herfindahl-
Hirschmann Index (HHI), the sum of the squared market shares of all firms
competing in the route market. The route HHI is computed on an airport-to-airport
basis rather than on a city-to-city basis. This allows capturing airport-specific effects.
These variables are interacted with the Distress variable to estimate their moderating
effects in the relationship between firm financial distress and prices.
2.3.2.3. Control variables
A set of firm and market specific control variables that have been shown to impact prices
in previous research (see e.g. Borenstein 1989) is included in the empirical model. The
firm-specific variables are the following:
? MaxAirportShare
kij
indicates an airline’s market share in the airport market i or j,
52
whichever is highest. The rationale for this approach is that a higher market share at
an airport conveys an airline some degree of market power in that airport market
which may be expected to impact fares in route markets involving that airport
(Borenstein 1989). Higher airport market shares likely imply higher fares.
? Circuity
kij
is another measure of the quality and convenience of carrier k’s service
between airports i and j. Circuity is the ratio of itinerary miles; i.e. the distance
actually flown, and nonstop miles between airport i and j. The higher this ratio; i.e.
the larger the detour, the lower the quality of the transportation service. At the same
time, however, higher circuities mean higher operating costs. The impact on fares is,
thus, undetermined.
? AirlinePass
kij
is the number of passengers carried by airline k between airports i and j
in a given time period. Higher numbers of passengers may be associated with
economies of density, and, thus, lower costs and lower fares. On the other hand, high
traffic volumes reflect high demand levels which may result in high prices.
? Loadfactor
k
is the average fill rate of a carrier k’s passenger aircraft during a given
time period. Note that this variable is not route specific since data were not available
on a route basis. Higher load factors may imply economies of density and utilization,
and fares may be expected to be lower for carriers with high load factors. At the same
time, high load factors may be associated with poorer service quality (e.g. possibly
lower frequency of service, less space for each traveler in a fully booked cabin, less
attentive/personalized cabin service) and lower fares.
The group of market specific control variables consists of the following variables:
53
? Distance
ij
is the distance (in miles) between airports i and j. In general, fares may be
expected to rise as distance increases.
? DistanceSquared
ij
is the square of the Distance
ij
variable. Its inclusion allows for a
nonlinear relationship between distance and fares.
? SlotRoute
ij
is a binary variable that indicates whether one or both airports i, j are slot-
controlled
25
. Such airports are typically highly congested and access is limited. Fares
are therefore expected to be higher on routes to or from these airports.
? MaxAirportHHI
ij
indicates the degree of concentration of an airport market. Rather
than including two values for both airports i and j, only the higher HHI value is
retained in this analysis. The rationale for this approach is that the more concentrated
airport is more likely to be the “bottleneck”, and fares on routes involving this airport
may be expected to be higher than fares on routes between “unconcentrated” airports.
? LCCCompForHCC
ij
is a binary variable. It takes on the value “1” when the carrier in
the observation is a high cost carrier and faces route competition by a low-cost
carrier. While some studies focused on Southwest Airlines only (e.g. Morrison,
2001), others employed a wider definition of low-cost carriers and defined all carriers
that started operations after deregulation as low-cost carriers (e.g. Dresner et al.
1996). In an effort to rigorously define LCCs in this research, financial data on all
airlines included in this analysis were collected. To account for the fact that operating
expenses per available seat mile (ASM) are likely higher for airlines operating
predominantly short haul flights, a carrier’s operating expenses per ASM were
regressed on its average stage length. The error terms, thus, reflect differences in
25
Presently, JFK, LGA, and DCA are the only slot-controlled airports in the U.S.; ORD was slot controlled
until June 2002
54
operating costs that cannot be attributed to differences in average stage length and are
indicators of an airline’s operating efficiency. A ranking of these error terms revealed
consistent patterns across all time periods considered in this research, and twelve
airlines were identified as low-cost carriers (see Appendix 1): Southwest Airlines,
Reno Air, Sun Country Airlines, Spirit Air Lines, JetBlue Airways, Western Pacific
Airlines, Airtran Airways, American Trans Air, Braniff Int'l Airlines, America West
Airlines, Frontier Airlines, Valujet Airlines.
? LCCCompForLCC
ij
is a binary variable which takes on the value “1” when the carrier
in the observation is a low-cost carrier and competes with another low-cost carrier in
the route market. LCCCompForHCC
ij
and LCCCompForLCC
ij
specify the presence
of low-cost carrier competition. These two variables are used to allow for differential
impacts in terms of pricing on LCCs and high cost carriers.
? AltRouteLCC1M
ij
is another dummy variable which indicates if there are one or more
adjacent route markets that are served by one or more low-cost carriers. The inclusion
of this variable builds on the work by Dresner, Lin and Windle (1996) and Morrison
(2001) who analyzed the impact of adjacent route competition on fares. Based on the
population statistics (i.e. PMSA, CMSA, MSA) published by the Bureau of Economic
Analysis (BEA), the following markets have been defined as metropolitan multi-
airport markets in this research: Boston (BOS, MHT, PVD), Chicago (ORD, MDW),
Cleveland (CLE, CAK), Dallas (DAL, DFW), Detroit (DTW, FNT), Houston (HOU,
IAH), Los Angeles (BUR, LAX, LGB, ONT, SNA), Miami (MIA, FLL), New York
(EWR, JFK, LGA, ISP, HPN), Norfolk (ORF, PHF), Philadelphia (PHL, ACY), San
Francisco (OAK, SFO, SJC), Tampa (TPA, PIE), Washington (BWI, DCA, IAD).
55
? Time variables are also included in the analysis to capture macroeconomic changes as
well as seasonal fluctuations (quarter dummies Quarter2, Quarter3, Quarter4) and
general trends over time (year dummy 2002).
? Population
ij
is used as a first-stage instrument and is the product of the metropolitan
area populations around airports i and j:
ij i j
Population Population Population = ? . A
first stage estimation of the AirlinePass variable is required to address the
endogeneity of the Fare and AirlinePass variables (see Section 2.3.4 for further detail
on this endogeneity issue). For econometric reasons, the first-stage estimation of the
endogenous variable (AirlinePass) requires the use of at least one instrumental
variable. Population is one of two instrumental variables used in this research.
? Income
ij
is also used as a first-stage instrument and is the population-weighted
average income in the metropolitan areas around airports i and j:
( ) ( )
( )
i i j j
ij
i j
Income Population Income Population
Income
Population Population
? + ?
=
+
.
2.3.3. Descriptive statistics
Correlations are presented in Table 1 below. Due to the large number of observations,
most correlations are statistically significant at the five percent level. Few correlations
coefficients, however, are larger than 0.50. The market share and market concentration
measures are highly correlated
26
, as are the measures of firm financial distress
27
.
26
Airport market shares (MaxAirportShare) and route market shares (RouteShare), for example, have a
correlation coefficient of 0.72.
27
The correlation coefficient for the Distress and DistressDiff variables is 0.79, for example.
56
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
1 Fare
2 Distance 0.52
3 SlotRoute 0.13 -0.04
4 RouteHHI -0.19 -0.59 0.01
5 MaxAirportHHI 0.15 -0.21 -0.10 0.50
6 RouteShare -0.12 -0.40 0.04 0.52 0.25
7 MaxAirportShare 0.05 -0.34 -0.01 0.44 0.46 0.72
8 Size 0.20 0.11 0.06 -0.06 0.01 -0.01 0.20
9 LCCCompForHCC -0.10 0.19 -0.12 -0.29 -0.19 -0.18 -0.14 0.28
10 LCCCompForLCC -0.27 -0.04 -0.10 -0.04 -0.11 -0.02 -0.06 -0.31 -0.20
11 AltRouteLCC1M -0.06 0.06 0.26 0.02 -0.03 0.04 -0.01 0.02 0.07 0.00
12 Circuity 0.00 0.03 -0.07 -0.06 -0.08 -0.37 -0.28 0.11 0.11 -0.07 -0.14
13 Distress 0.10 0.06 0.06 -0.05 0.01 -0.13 -0.16 -0.37 0.06 -0.13 -0.01 0.00
14 DistressDiff -0.03 0.00 0.00 0.00 0.00 -0.10 -0.13 -0.32 0.12 -0.06 0.00 0.00 0.79
15 Chpt11Ops 0.08 0.04 0.00 -0.04 -0.01 -0.09 -0.10 -0.23 -0.03 -0.03 -0.01 0.00 0.43 0.33
16 Pre4Chpt11 -0.06 0.03 0.02 -0.02 -0.02 -0.03 0.00 0.11 0.14 -0.06 0.07 -0.01 0.18 0.16 -0.08
17 Post4Chpt11 0.03 0.00 0.02 -0.01 0.00 -0.06 -0.08 -0.18 -0.04 0.00 -0.01 -0.01 0.23 0.18 0.55 -0.04
18 Loadfactor -0.17 0.12 0.03 -0.07 -0.07 -0.03 0.00 0.32 0.24 -0.05 0.10 0.05 -0.12 -0.02 -0.18 0.18 -0.14
19 AirlineCost 0.43 0.00 0.09 -0.03 0.06 -0.05 0.06 0.20 0.06 -0.37 -0.10 0.05 0.20 0.06 0.09 0.05 0.07 -0.31
20 AirlinePass -0.32 -0.41 0.12 0.35 0.08 0.71 0.56 -0.05 -0.10 0.09 0.18 -0.47 -0.10 -0.06 -0.10 0.02 -0.07 0.08 -0.17
(correlation coefficients in bold are significant at the 5% level)
Table 1: Correlation matrix (n = 23,039)
57
The change of firms’ Distress during bankruptcy is illustrated in Figure 9: For each
possible state with respect to bankruptcy, the airlines’ unweighted mean Distress scores
are graphed. The averages are computed for all carriers and across all time periods (eight
quarters in 1992 and 2002) included in the dataset. The sample size n indicates the
number of firm-quarter observations the respective statistics are based on. As can be seen
in Figure 9, non-bankrupt (i.e. comparatively healthy) carriers have a mean Distress
score of 0.8, with the minimum and maximum Distress scores being -3.0 and 18.6,
respectively. Financially sound airlines are thereby assumed to have negative Distress
scores while troubled (yet non-bankrupt) carriers have positive Distress scores. Mean
Distress scores for carriers approaching bankruptcy (Pre4Chpt11) are substantially
higher (12.2) and always positive, ranging from 0.4 to 69.5. A Welch-Aspen two-sample
t test for independent groups
28
is performed to evaluate whether these differences are
statistically significant. The test statistic is t = 2.55 with 15.1 degrees of freedom (df).
This result is statistically significant at the five percent level, indicating that firms
approaching bankruptcy have significantly higher distress scores. Carriers’ Distress
scores average 1.7 during the quarter in which the Chapter 11 filing occurs (Chpt11). It
should be noted, however, that this statistic is based on four carrier observations only
29
.
Bankrupt carriers’ Distress scores averaged 7.7 in the four quarters following the entry
into bankruptcy (Post4Chpt11). In this latter case, the Distress scores range from 0.4 to
26.8. Based on these descriptive statistics it is concluded that financially distressed
carriers will always have positive Distress scores. Financially sound airlines, in turn,
have negative Distress scores.
28
Given the difference in sample sizes, unequal variances are assumed.
29
TWA filed for Chapter 11 in the first quarter of 1992, Markair filed in the second quarter of 1992, US
Airways filed in the third quarter of 2002, and United Airlines filed in the fourth quarter of 2002.
58
Figure 9: Distribution of Distress scores prior to and during bankruptcy
Table 2 provides the mean and standard deviation, minimum and maximum values for
some selected variables included in the model. The mean Distress score of 0.59 indicates
that most carriers experienced some level of financial distress in 1992 and 2002
30
. While
there is no direct interpretation for this value, it implies that a significant proportion of
passengers used (severely) troubled carriers. This contention is further substantiated by
the mean of the Chpt11Ops variable (0.11) which suggests that eleven percent of all
passengers traveled with bankrupt airlines. It is further noted that approximately seven
percent of all passengers traveled with near-bankrupt carriers (Pre4Chpt11, 1571
observations), while approximately five percent of all passengers used airlines that filed
30
The mean of the Distress variable is significantly different from 0. A one-sample mean comparison test
yields a test statistic of t = 43.74 which is statistically significant at the one percent level.
18.6
12.2
1.7
7.7
-3.0
26.8
2.8
69.5
0.8 0.4 0.4
0.6
-10
0
10
20
30
40
50
60
70
80
No Chapter 11 Pre4Chpt11 Chpt11 Post4Chpt11
Distress
(n = 138) (n = 16) (n = 4) (n = 9)
Max.
Avg.
Min.
Max.
Avg.
Min.
Max.
Avg.
Min.
Max.
Avg.
Min.
59
for Chapter 11 protection within the past year (Post4Chpt11, 1196 observations).
Moreover, the data set contains 2,414 carrier-route market observations (out of a total of
23,039 observations) with DistressDiff values larger than 2.86 (one standard deviation
above the mean), indicating an airline’s severe financial distress relative to its route
competitors.
Variable Mean Std. dev. Min Max
Fare (1992 U.S. dollars) 114.26 60.61 25.17 1140.19
Distress 0.59 2.28 -3.02 69.53
DistressDiff 0.05 2.81 -31.97 68.59
Chpt11Ops 0.11 0.31 0 1
Pre4Chpt11 0.07 0.25 0 1
Post4Chpt11 0.05 0.21 0 1
AirlineCost ($) 0.08 0.02 0.04 0.24
RouteShare 55.47% 27.28% 5% 100%
RouteHHI 5298.78 2172.02 1259.66 10000
Size (1,000 $) 11,900,000 8,887,998 3,936 29,300,000
AirlinePass 1570.76 2408.01 1 29368
Distance (miles) 950.86 643.09 30 2717
SlotRoute 0.22 0.41 0 1
MaxAirportShare 46.09% 23.89% 0.015% 100%
MaxAirportHHI 3966.56 1885.11 1131.37 10000
Circuity 1.02 0.05 1 2.2
Loadfactor 67.81% 6.18% 35.1% 84.6%
Quarter1 0.23 0.42 0 1
Quarter2 0.26 0.44 0 1
Quarter3 0.27 0.44 0 1
Quarter4 0.25 0.43 0 1
1992 0.42 0.49 0 1
2002 0.58 0.49 0 1
Mean values weighted based on number of airline passengers, except for "AirlinePass"
F
i
n
a
n
c
i
a
l
d
i
s
t
r
e
s
s
M
o
d
e
r
a
t
o
r
s
C
o
n
t
r
o
l
v
a
r
i
a
b
l
e
s
(
s
e
l
e
c
t
e
d
)
Table 2: Descriptive statistics for selected variables (n = 23,039)
60
2.3.4. Empirical methodology
A log-linear price estimation equation forms the basis of the model used in this research.
More specifically, an airline’s fare on a route is modeled as a function of a set of route,
airport, and carrier specific variables, as well as a number of control variables. The
estimation of the model requires the implementation of a two-stage least squares
procedure since AirlinePass
kij
is an endogenous variable; i.e. the number of airline
passengers may impact airfares while at the same time the latter may have an effect on
the number of passengers. In a first stage regression, the number of airline passengers
(AirlinePass) is modeled as a function of all exogenous variables including two
instrumental variables, Income and Population. Fitted values for AirlinePass are then
used to estimate fares (Fare) in the second stage model. The basic estimating model is
defined as follows:
Equation 1: First-stage regression model
lnAirline Passengers = ?
0
+ ?
1
lnDistance + ?
2
(lnDistance)
2
+ ?
3
Slot Route
+ ?
4
lnRoute HHI + ?
5
lnMax Airport HHI + ?
6
Route Share + ?
7
Max Airport Share
+ ?
8
LCC Comp for HCC + ?
9
LCC Comp for LCC + ?
10
Alt Route LCC
+ ?
11
lnCircuity + ?
12
Distress + ?
13
Load Factor + ?
14
lnAirline Cost + ?
15
lnSize
+ ?
16
lnPopulation + ?
17
lnIncome + ?
18
2002 + ??
t
Quarter
t
61
Equation 2: Second-stage regression model
lnFare = ?
0
+ ?
1
lnAirline Passengers (fitted) + ?
2
lnDistance + ?
3
(lnDistance)
2
+ ?
4
Slot Route + ?
5
lnRoute HHI + ?
6
lnMax Airport HHI + ?
7
Route Share
+ ?
8
Max Airport Share + ?
9
LCC Comp for HCC + ?
10
LCC Comp for LCC
+ ?
11
Alt Route LCC + ?
12
lnCircuity + ?
13
Distress + ?
14
Load Factor
+ ?
15
lnAirline Cost + ?
16
lnSize + ?
17
2002 + ??
t
Quarter
t
The OLS assumptions of homoskedasticity and independence are frequently not met
when dealing with cross-sectional time series data (Greene 2003). Therefore, tests to
detect the potential problems of heteroskedasticity and autocorrelation of the error terms
are implemented.
First, the Breusch-Pagan/Cook-Weisberg Lagrange multiplier test (Breusch and Pagan
1979, Cook and Weisberg 1983) uses fitted values of the dependent variable (Fare) to
determine whether the residuals vary with the fitted values of the dependent variable; i.e.
violate the homoskedasticity assumption. This test is implemented after an OLS
regression similar to the second stage model described above (the sole difference being
that the actual values of Airline Passengers are used rather than fitted values; see
Appendix 2). The implementation of this test yields a test statistic of 614.33 which
follows a ?
2
distribution. The null hypothesis of constant variance is clearly rejected with
a significance level of less than one percent.
62
Second, the Wooldridge test for autocorrelation in panel data (Drukker 2003, Wooldridge
2002) suggests the presence of first-order auto-correlation with an F-statistic of
F = 829.37 which is statistically significant at the less than one percent level. Given the
presence of heteroskedasticity, autocorrelation and endogeneity (as discussed previously),
a generalized two-stage least squares (G2SLS) procedure is recommended (Greene
2003).
The generalized least squares procedure is typically implemented in one of two distinct
econometric specifications: fixed effects or random effects
31
. These two specifications
differentially address the heterogeneity of unobserved group and time specific effects,
which in the classical ordinary least squares approach, are subsumed in the error term.
In the fixed effects model, the constant term is adjusted for each group and each time
period such that the regression model becomes '
it i t kit
y x ? ? ? ? ? = + + + + . The first term
on the right-hand side of the equation is the constant term ( ) ? , and the second term
represents the sum of the products of the regressors ( ) x and their respective
coefficients ( ) ? . The third and fourth terms are the group ( )
i
? and time ( )
t
? fixed
effects which effectively adjust the constant term for group and time specific effects. The
last term in the model is the individual error term associated with the kth observation in
group i in time period t ( )
kit
? .
31
The reader is referred to any econometrics textbook for a detailed discussion of the econometric issues
revolving around generalized least squares models and the choice between fixed and random effects
specifications. The overview provided here is based on Greene (2003).
63
The random effects model proposes a different specification of the error term in the
econometric model. In this case, the unobserved individual heterogeneity is assumed
independent of the regressors ( ) x , and the group and time specific adjustments to the
constant term are assumed to be randomly distributed across cross-sectional units and
time. The benefit of the random effects procedure relative to the fixed effects procedure
lies in the preservation of a significant number of degrees of freedom since only two
random variables are needed (random group and time effects) rather than an exhaustive
set of group and time specific dummy variables. If, however, the group and time effects
are correlated with the regressors, the random effects procedure may produce inconsistent
estimates.
To decide whether it is appropriate to use the fixed effects or random effects procedure,
the Hausman specification test (Hausman 1978) is used to test for orthogonality between
the regressors and the random effects. If the null hypothesis of no correlation cannot be
rejected, the random effects model is both consistent and efficient and preferred over the
fixed effects model which, in this case, is inefficient. If, however, the null hypothesis is
rejected, only the fixed effects model is consistent and thus preferred over the random
effects model. The implementation of the Hausman specification test requires the
estimation of the model (Equation 1 and Equation 2) using the fixed effects and random
effects procedures, respectively. The test statistic W is based on the covariance matrix ?
of the difference vector of the respective coefficients [ ] b ? ? and is given by
[ ] [ ]
1
' W b b ? ? ?
?
= ? ? (Hausman 1978). The test produces a
2
? distributed statistic of
64
W = 995.43 which is significant at the less than one percent level. The null hypothesis of
no correlation is therefore clearly rejected, suggesting that the fixed effects model should
be selected.
As noted above, the fixed effects model has the disadvantage of consuming a large
number of degrees of freedom due to the inclusion of group and time specific dummy
variables in the regression analysis. Greene (2003) and Yaffee (2003) therefore suggest
carefully evaluating the benefits of the fixed effects G2SLS procedure relative to the
standard 2SLS procedure. The F test of joint significance of fixed effects (Greene 2003)
evaluates the contribution of the fixed group and time effects to the fit of the model. To
that end, two regression analyses must be performed: The baseline regression which does
not include any fixed effects, and the fixed effects regression. The improvement in the fit
of the model which is achieved by adding fixed effects is measured by the following F
statistic:
( ) ( )
( ) ( )
2 2
2
1
1
fixed effects no effects
fixed effects
no effects
R R n
F
R nT n k
? ?
=
? ? ?
(Greene 2003, Yaffee 2003), where n
is the number of groups, nT is the number of observations, and k is the number of
regressors.
In this study, the cross-section is defined by route-carrier combinations (a total of 4,508
groups), and there are eight distinct time periods (two years with four quarters each). This
implies that 4,514 dummy variables must be added to the baseline 2SLS regression
65
equation
32
. The statistical software package used for this research (Intercooled STATA
8.2) does not support such an operation due to the software’s insufficient matrix size. For
the purpose of this test, the number of dummy variables is therefore reduced by
estimating fixed carrier effects only as opposed to fixed route-carrier effects, thereby
reducing the number of cross-sectional indicators from 4,508 to 30. By constraining
seasonal (quarterly) effects to be constant over time (in 1992 and 2002), the number of
time indicators is reduced to four as specified in Equation 1 and Equation 2. This test is a
highly conservative approximation of the full fixed effects test with 4,514 fixed effects
and therefore provides a lower bound for the joint significance of the fixed effects
33
. The
baseline model (see Appendix 3) yields an R
2
of 0.376, while the reduced fixed effects
model (see Appendix 4) yields an R
2
of 0.512. The resulting F statistic is F = 194.22
which is significant at the less than one percent level. It is therefore concluded that the
fixed effects generalized two-stage least squares procedure is the most appropriate data
analysis technique.
2.4. Empirical results and discussion
The regression results are discussed in two stages: The first-stage regression, in which
AirlinePass is the dependent variable, is discussed, before the second-stage regression
results are presented. The second-stage regression uses Fare as the dependent variable
and tests the hypotheses set forth in this essay.
32
Note that this must be done manually to ensure consistency of the R
2
computation (the computation of
the R
2
statistic differs between the 2SLS and [fixed effect] G2SLS).
33
The breakdown of the 30 carrier indicator variables to 4,508 route-carrier indicator variables will
necessarily result in an increased R
2
statistic.
66
2.4.1. First-stage regression
In the first-stage regression, all independent variables (which are assumed exogenous)
and the Population and Income instruments are used to estimate exogenously determined
fitted values for AirlinePass. Table 3 presents the coefficient estimates of the first stage
regression as specified in Equation 1.
Most of the results displayed in Table 3 are in accordance with prior theoretical reasoning
and empirical research: The relationship between the number of passengers and route
distance is nonlinear as evidenced by the negative coefficient of the Distance variable
and the positive sign of the DistanceSquared coefficient. This suggests that the number of
passengers increases with the distance flown at a rate which increases in route length.
The positive coefficient of the SlotRoute dummy variable is indicative of congestion and
higher passenger volumes on slot-controlled routes. Moreover, greater firm market shares
at the route and airport market levels (RouteShare and MaxAirportShare) imply greater
numbers of passengers. Holding market shares constant, an increase in market
concentration (RouteHHI, MaxAirportHHI) then results in lower passenger numbers (see
also Ravenscraft 1983). Competition in adjacent route markets (AltRouteLCC1M) has a
slight negative effect on the number of passengers, while more circuitous routings
(Circuity) exhibit significantly decreased passenger numbers. Higher load factors
(LoadFactor) are, of course, associated with more passengers, and higher operating costs
(AirlineCost), presumably implying higher prices, negatively affect demand. The
67
coefficient of the Size variable is statistically insignificant, indicating that firm size per se
does not influence passenger demand. The positive coefficients of the quarter dummies
indicate seasonal effects (Quarter2-4), while the time trend variable (2002) carries a
negative, though statistically insignificant coefficient, thus hinting at the downturn in the
airline industry in 2002. The first instrumental variable, Population, carries a positive
coefficient indicating that passenger numbers increase as the potential market volume
increases. The coefficient of the Income variable is statistically insignificant which may
be attributed to its limited variability.
There are three variables with unexpected signs: First, the LCCCompForHCC and
LCCCompForLCC variables both have positive coefficients, indicating that the presence
of a low-cost competitor increases passenger demand for the focal carrier. This result is
most likely due to the focal airline lowering its prices as it faces aggressive competition.
These lower prices then translate into higher passenger demand. The Distress variable
carries a positive and statistically insignificant coefficient which suggests a firm’s
financial distress does not impact passenger demand. A potential explanation may be that
distressed carriers mitigate potentially negative demand effects by charging lower prices
or that passengers have few or no alternative carrier choices.
In summary, it is noted that the first-stage model is highly significant (F = 1,038.3,
significant at the less than one percent level), and that most independent variables are at
least marginally significant with most coefficients having the expected signs. Appendix 5
presents the first stage regression results for all five specifications of the model.
68
Dependent variable:
AirlinePass Coefficient P>|z|
Constant 444.458 0.000
Distance -136.296 0.000
DistanceSquared 10.181 0.000
SlotRoute 0.099 0.001
RouteHHI -0.412 0.000
MaxAirportHHI -0.367 0.000
RouteShare 0.027 0.000
MaxAirportShare 0.001 0.007
LCCCompForHCC 0.227 0.000
LCCCompForLCC 0.238 0.000
AltRouteLCC1M -0.022 0.061
Circuity -2.326 0.000
Distress 0.005 0.190
Loadfactor 0.016 0.000
AirlineCost -0.095 0.023
Size 0.028 0.150
Quarter 2 0.048 0.000
Quarter 3 0.074 0.000
Quarter 4 0.047 0.000
2002 -0.055 0.315
Population 0.579 0.000
Income -0.034 0.785
F 1038.3 0.000
R-squared (within) 0.541
Table 3: First stage G2SLS regression estimates (n = 23,039)
34
2.4.2. Second-stage regression
In this section, the results from five different second-stage regression analyses are
reported. The first and second second-stage analyses test Hypothesis 1 by including the
Distress and Chpt11Ops variables, respectively. The third regression tests the differential
34
The carrier fixed effects are omitted in this table.
69
effect of financial distress over time (Hypothesis 2) by estimating the model with the
Pre4Chpt11 and Post4Chpt11 variables. Hypothesis 1 and Hypothesis 2 are tested
separately to avoid confounding the results by including both the Distress and Chpt11
variables in a single regression (since all bankrupt airlines have high Distress scores).
The fourth second-stage regression model tests the moderating effects of firm costs, firm
market shares, and market concentration (Hypothesis 3 to Hypothesis 6). These
interactions are not included in model 1 to allow for a direct interpretation of the direct
effect of the Distress variable in model 1. As noted by Aiken and West (1991), when
interaction effects are present, a variable’s direct effect cannot be assessed by interpreting
the variable’s coefficient only, but it must be evaluated in conjunction with all its
interactions. The fifth and final model tests the importance of a firm’s relative financial
distress as discussed in Hypothesis 7. Similar to the argumentation above, the
DistressDiff variable is tested separately to avoid confounding the effects of absolute
(Distress) and relative (DistressDiff) financial distress. The second-stage regression
results are presented in Table 4.
Before focusing on the variables of interest in the respective regressions, it is noted that
all second-stage models are highly significant (Wald ?
2
? 23,900,000). Caution must be
used, however, when interpreting the R-squared statistics
35
. In the generalized least
squares (GLS) procedure, the total sums of squares are not broken down as in the
ordinary least squares procedure. The GLS R-squared, therefore, is not bounded between
zero and one and cannot be interpreted as the percentage of variability explained. In
35
The information on the use and meaning of R-squared statistics in GLS regressions was obtained from
the STATA manuals and the STATA website at www.stata.com.
70
addition, there are two sources of variation: within variation and between variation. When
fixed effects models (i.e. within estimators) are used, only the within R-squared should be
used
36
. The R squared for within variation indicates to what extent the model is able to
predict a new observation on one of the subjects already in the study. The R squared for
total variation indicates the quality of predictions relating to a new observation on a new
subject. While all R-squared (within, between, overall) statistics are reported, the reader’s
attention is directed toward the within R-squared measures which range between 0.741
and 0.783 as reported in Table 4.
36
This statistic is obtained by fitting a mean-deviated regression model where all the group effects are
assumed to be fixed. These group effects are subtracted out of the model and no attempt is made to quantify
their overall effect on the fit of the model.
71
Second-stage G2SLS regression Number of obs. 23039 Obs. per group: min. 1
(fixed effects) Number of groups 4508 avg. 5.1
max. 8
Dependent variable:
Fare Coefficient P>|z| Coefficient P>|z| Coefficient P>|z| Coefficient P>|z| Coefficient P>|z|
Constant -231.741 0.000 -237.998 0.000 -247.393 0.000 -245.434 0.000 -268.520 0.000
AirlinePass (fitted) -0.093 0.000 -0.076 0.006 -0.042 0.138 -0.141 0.000 -0.022 0.441
Distance 69.277 0.000 70.705 0.000 73.037 0.000 74.364 0.000 78.903 0.000
DistanceSquared -5.006 0.000 -5.108 0.000 -5.268 0.000 -5.420 0.000 -5.671 0.000
SlotRoute 0.090 0.000 0.096 0.000 0.083 0.000 0.083 0.000 0.088 0.000
RouteHHI -0.009 0.495 -0.003 0.795 0.008 0.553 -0.008 0.518 0.021 0.130
MaxAirportHHI 0.055 0.000 0.067 0.000 0.091 0.000 0.030 0.019 0.105 0.000
RouteShare 0.002 0.010 0.001 0.053 0.001 0.437 0.003 0.000 0.000 0.994
MaxAirportShare 0.002 0.000 0.002 0.000 0.002 0.000 0.002 0.000 0.002 0.000
LCCCompForHCC -0.110 0.000 -0.118 0.000 -0.130 0.000 -0.091 0.000 -0.132 0.000
LCCCompForLCC -0.024 0.034 -0.014 0.242 -0.001 0.914 -0.011 0.311 -0.005 0.684
AltRouteLCC1M -0.027 0.000 -0.026 0.000 -0.027 0.000 -0.021 0.000 -0.028 0.000
Circuity -0.443 0.000 -0.402 0.000 -0.320 0.000 -0.536 0.000 -0.269 0.001
Distress -0.036 0.000 0.496 0.000
Chpt11Ops -0.072 0.000
DistressDiff -0.009 0.000
Pre4Chpt11 -0.007 0.179
Post4Chpt11 -0.042 0.000
Loadfactor -0.015 0.000 -0.014 0.000 -0.014 0.000 -0.013 0.000 -0.015 0.000
AirlineCost 0.224 0.000 0.210 0.000 0.235 0.000 0.223 0.000 0.226 0.000
Size 0.055 0.000 0.119 0.000 0.152 0.000 -0.018 0.061 0.130 0.000
Quarter 2 -0.031 0.000 -0.040 0.000 -0.044 0.000 -0.034 0.000 -0.045 0.000
Quarter 3 -0.026 0.000 -0.034 0.000 -0.048 0.000 -0.028 0.000 -0.044 0.000
Quarter 4 -0.034 0.000 -0.046 0.000 -0.054 0.000 -0.027 0.000 -0.053 0.000
2002 -0.257 0.000 -0.306 0.000 -0.332 0.000 -0.208 0.000 -0.314 0.000
AirlineCost*Distress 0.062 0.000
Size*Distress -0.013 0.000
RouteShare*Distress 0.0002 0.000
RouteHHI*Distress -0.023 0.000
Wald ?
2
26,800,000 25,900,000 24,500,000 28,500,000 23,900,000
Prob > ?
2
0.000 0.000 0.000 0.000 0.000
R-squared: within 0.769 0.761 0.748 0.783 0.741
between 0.080 0.084 0.092 0.041 0.101
overall 0.087 0.091 0.097 0.049 0.104
5 2 4 1 3
Table 4: Second-stage G2SLS regression estimates
72
Turning to the control variables first, it is noted that most coefficient estimates are
consistent across all five second-stage models and are statistically significant at the less
than one percent level: Prices are shown to increase with Distance, but at a decreasing
rate, as evidenced by the negative coefficient of DistanceSquared. As expected, fares
tend to be higher in route markets involving one or two slot-controlled airports
(SlotRoute), and both airport market concentration (MaxAirportHHI) and airport market
shares (MaxAirportShare) are associated with higher fares, ceteris paribus. The presence
of low-cost carrier competition has a strong negative effect on a high cost carrier’s prices
(LCCCompForHCC), as does the presence of low-cost carriers in adjacent route markets
(AltRouteLCC1M). Prices for less convenient connecting traffic are shown to be lower
than for direct service (Circuity), and higher load factors (LoadFactor) – indicative of
economies of density – also tend to result in lower fares. An airline’s operating costs
(AirlineCost) and size (Size), finally, are both shown to positively impact air fares. The
time variables capture both seasonal price fluctuations (Quarter2-4) as well as a clearly
negative time trend (2002).
The following variables have either unexpected or statistically insignificant coefficients:
While the coefficient of the AirlinePass variable is negative as expected in all instances,
it is statistically insignificant in models 3 and 5. There is, nonetheless, at least some
evidence that higher passenger numbers – implying economies of density – result in
lower prices, all else equal. Note that the coefficients of the RouteShare variable, while
positive as expected, are also statistically insignificant in models 3 and 5. The two
variables (AirlinePass and RouteShare) are highly correlated as expected (? = 0.57, see
73
Table 1) with RouteShare being the ratio of AirlinePass and the total number of
passengers in the route market. It is, therefore, likely that multicollinearity between right-
hand side variables cause some degree of variance inflation. The RouteHHI variable
carries a statistically insignificant coefficient in all model specifications, suggesting that
route market concentration does not have a direct effect on prices. Also, the presence of
LCC competitors does not appear to impact other low-cost carriers’ prices as indicated by
the insignificant coefficient estimates of the LCCCompForLCC variable in models 2-5.
Only in the baseline model (1) can the expected negative effect be observed.
The attention is now directed to the variables of interest that test the hypotheses set forth
in this paper.
The negative and significant coefficient of the Distress variable in the first second-stage
regression (? = -0.036, p = 0.000) provides clear support for the contention that greater
levels of financial distress result in lower prices, ceteris paribus (Hypothesis 1). This
result thus confirms the basic finding in the extant literature that financially distressed
firms behave more aggressively in the output market. More specifically, this result
suggests that the reduction of a firm’s Distress score by one unit leads to a price
reduction of 3.6 percent, all else held constant. The second regression presents an
alternative test of Hypothesis 1 using the ChptOps variable. The latter carries a
statistically significant coefficient of -0.072 (p = 0.000) which implies that, on average,
airlines operating under Chapter 11 protection charge about seven percent less than their
non-bankrupt competitors, ceteris paribus. This finding is consistent with the result of the
74
Distress variable and clearly in support of Hypothesis 1.
The third regression presented in Table 4 tests the differential impact of financial distress
prior to and after Chapter 11 filings. The coefficient of the Pre4Chpt11, while negative,
is statistically insignificant (? = -0.007, p = 0.170) which suggests that there are no
significant price changes as an airline approaches bankruptcy. The Post4Chpt11 variable,
however, carries a negative and statistically significant coefficient (? = -0.042,
p = 0.000). This indicates that airlines tend to lower prices upon declaring bankruptcy and
that the effect of firm financial distress on prices is substantially larger (-4.2%) once the
airline operates under bankruptcy protection. This finding supports the contention that
passengers may be reluctant to choose bankrupt carriers given the uncertainty about its
reliability and future operations. This may entice such firms to cut prices in an effort to
stimulate or maintain passenger demand. Moreover, bankrupt carriers may simply pass
some of the cost savings that result from operating under bankruptcy protection
37
on to
consumers. Hypothesis 2 is thus supported.
The fourth column in Table 4 presents a test of the hypothesized interaction effects
(Hypothesis 3 to Hypothesis 6). Hypothesis 3 argues that a firm’s operating costs
positively moderate the relationship between financial distress and prices, meaning that
the effect of firm financial distress on prices will be of lesser magnitude for high-cost
firms than for lower-cost firms (see Figure 3). The rationale for this contention is that
low-cost firms likely have higher profit margins and can more easily (and profitably)
37
Due to paused leasing and debt payments, for example.
75
afford price cuts than high-cost firms. The strategic management literature further argues
that operating costs are a good proxy for a firm’s strategic type: Low-cost firms are often
referred to as prospectors, and high-cost firms have been termed defenders. Prior
research has shown that prospectors tend to act more aggressively (in terms of prices, for
example), while defenders tend to behave more conservatively and focus on internally-
oriented rather than market-oriented actions which involve price and product changes.
The coefficient of the interaction term AirlineCost*Distress is positive and statistically
significant at the less than one percent level (? = 0.062, p = 0.000). As discussed above,
the effect of financial distress on prices is generally negative, implying that distressed
firms sell at lower prices, all else equal. The interaction with operating costs
(AirlineCost) then adds a positive term to the distressed firm’s price, where the value of
this addition increases in the firm’s operating costs. The analyses, thus, present some
evidence for the contention that distressed firms will tend to refrain from competing on
price when their operating costs are higher, as suggested in Hypothesis 3.
It has been suggested in Hypothesis 4 that firm size will increase a distressed firm’s
tendency to compete on price (see Figure 4). More specifically, it has been argued that
larger firms benefit from greater reputation, creditor trust and resource availability which
increase their survivability. Consequently, it is expected that larger distressed firms
leverage their size advantage and do not avoid price competition to the extent smaller,
more fragile airlines do: Larger firms can afford the detrimental short-term effects of
price cuts and may pursue such aggressive pricing strategies in an effort to eliminate
smaller competitors and thus enhance their long-term profitability prospects. The
76
interaction term between Distress and Size is negative and statistically significant (? = -
0.013, p = 0.000). This result thus implies that larger distressed firms will price more
aggressively than smaller distressed firms, all else equal. Consequently, Hypothesis 4 is
supported.
In Hypothesis 5, it was argued that the impact of firm financial distress on prices is
moderated by firm market shares (see Figure 5). Firm with higher market shares may
have higher degrees of market power and therefore experience less pressure to lower
prices in the light of financial distress. In addition, for firms with high market shares, the
potential benefits of cutting prices are limited since the expected gains in terms of market
volume may not offset the losses due to lower sales prices. The coefficient of the
interaction term of the RouteShare and Distress variables is positive and significant
(? = 0.0002, p = 0.000). Higher route market shares, thus, reduce a distressed firm’s
pricing aggressiveness as stated in Hypothesis 5.
As to the moderating effect of (route) market concentration, it was hypothesized that the
interaction of financial distress and route market concentration will positively impact
prices (Hypothesis 6), ceteris paribus (see Figure 6). While high market concentration
per se may facilitate collusive price fixing among firms, deteriorations in a firm’s
financial condition and ensuing changes in that firm’s cost structure may lead to the
breakdown of collusive arrangements with competitors and greater degrees of price
competition. The interaction term of RouteHHI and Distress has a negative and
statistically significant coefficient (? = -0.023, p = 0.000). As stated in Hypothesis 6, this
77
implies that greater levels of distress and market concentration increase a heavily
troubled firm’s tendency to compete aggressively and sell at lower prices (after
controlling for the moderating effects of route market shares).
Hypothesis 7 suggests that the difference between a focal firm’s Distress score and that of
its (route market) competitors affects the focal firm’s prices. This hypothesis is motivated
by the fact that firms that are in similar financial conditions may be expected to behave
symmetrically. In this case, no single firm would benefit from price reductions and
reinforced price competition. It is, therefore, expected that a focal firms pricing actions
will be more pronounced the greater the focal firm’s financial distress relative to its
competitors. To test Hypothesis 7 the coefficient of the DistressDiff variable from the
fourth second-stage regression can be interpreted straightforwardly. The negative and
significant coefficient (? = -0.009, p = 0.000) indicates that a firm’s financial distress
relative to its competitors negatively impacts the focal firm’s prices as stated in
Hypothesis 7
38
.
2.4.3. Second-stage regression: Sensitivity analysis
The results discussed above are based on the analysis of 1992 and 2002 data. These time
periods were chosen since the airline industry experienced substantial financial distress
during those years. To investigate the sensitivity of the results with respect to the
38
Recall that positive DistressDiff values indicate relative financial distress, while negative values indicate
relative financial wellbeing.
78
selection of the time period studied, the regression models are re-estimated using data
from 1992, 1997, and 2002. The addition of 1997 data brings the total number of
observations to 34,097. 1997 data were selected since this year is in the middle of the
1992-2002 time period. Also, the airline industry as a whole performed relatively well
during that year. It is therefore expected that the findings with respect to the effect of
financial distress on prices will be weaker when 1997 data are included in the analyses.
Nonetheless, the empirical results should be consistent with the contentions set forth in
Hypothesis 1 to Hypothesis 7.
Table 5 presents the second-stage regression results which are based on the analysis of
the extended data set (including 1997 data). It is noted that the fit of the regression
models is generally inferior compared to the results presented in Table 4 which were
based on 1992 and 2002 data only. Specifically, the R-squared within statistics shown in
Table 5 suggest that the models explain only about fifty to sixty percent of the variability
as compared to the seventy to eighty percent variability explained for the 1992 and 2002
data (see Table 4). While most variables have statistically significant coefficients with the
expected signs, the Distance and DistanceSquared variables have insignificant coefficient
estimates in all models.
The hypothesis testing results can be summarized as follows:
? Hypothesis 1: The negative coefficient of the Distress variable (? = -0.020,
p = 0.000) in the first regression, provides support for the contention that greater
levels of financial distress result in lower prices. This contention is further
79
corroborated by the negative and significant coefficient of the Chpt11Ops variable
in the second regression (? = -0.047, p = 0.000).
? Hypothesis 2: The differential effect of financial distress over time (prior to versus
during bankruptcy) is empirically examined in the third regression where the
Pre4Chpt11 and Post4Chpt11 variables are included in the model. While the
Pre4Chpt11 variable carries a statistically significant negative coefficient, the
coefficient of the Post4Chpt11 variable is statistically insignificant. This suggests
that, on average, carriers approaching bankruptcy tend to cut prices, while carriers
operating under bankruptcy protection do not cut prices. This finding is contrary
to Hypothesis 2 and inconsistent with the results shown in Table 4. It is noted that
virtually no airline bankruptcies were observed in 1997. As a result, it is not
surprising that adding 1997 data to the regression analysis weakens the robustness
of the regression results with respect to the effect of bankruptcy on prices.
? Hypothesis 3: Hypothesis 3 suggests that a firm’s operating costs positively
moderate the relationship between financial distress and prices. The positive and
significant coefficient of the AirlineCost*Distress interaction effect (? = 0.015,
p = 0.000) confirms this expectation. This finding is consistent with the results
shown in Table 4.
? Hypothesis 4: The distress-price effect was hypothesized to be stronger for larger
firms than for smaller firms. In line with the regression results reported earlier,
this hypothesis is supported even when 1997 data are included: The Size*Distress
interaction effect carries a negative and significant coefficient (? = -0.009,
p = 0.000).
80
? Hypothesis 5: The results shown in Table 5 suggest that the distress-price effect
does not change in magnitude as a firm’s route market share increases. The
RouteShare*Distress interaction effect does not yield a statistically significant
coefficient (? = 0.0000, p = 0.495), whereas this interaction effect was positive
and significant in the analysis of 1992 and 2002 data (see Table 4). Again, the
lack of a significant finding may potentially be attributed to the fact that the
addition of 1997 data tends to dilute statistical effects of financial distress since
the airline industry experienced little distress in that year.
? Hypothesis 6: The RouteHHI*Distress interaction carries the expected negative
coefficient (? = -0.008, p = 0.001), suggesting that the distress-price effect is
greater in more concentrated markets than in less concentrated markets. This
finding is consistent with Hypothesis 6 and the previously reported results (see
Table 4).
?
? Hypothesis 7: A firm’s financial distress relative to its competitors in the route
market is also shown to significantly impact prices (? = -0.003, p = 0.000).
Hypothesis 7 is, thus, supported.
In summary, five out of seven hypotheses are supported when 1997 data are included in
the analyses. The lower model fit statistics and smaller coefficient values, however,
confirm the contention that adding 1997 data—a period of relative financial health in the
airline industry—tends to dilute the results. Nonetheless, the hypothesis testing results are
shown to be largely robust.
81
Second-stage G2SLS regression Number of obs. 34097 Obs. per group: min. 1
(fixed effects) Number of groups 4798 avg. 7.1
max. 12
Dependent variable:
Fare Coefficient P>|z| Coefficient P>|z| Coefficient P>|z| Coefficient P>|z| Coefficient P>|z|
Constant -30.762 0.477 -16.919 0.707 -27.484 0.531 -40.643 0.364 -29.798 0.489
AirlinePass (fitted) -0.505 0.000 -0.544 0.000 -0.515 0.000 -0.550 0.000 -0.493 0.000
Distance 13.024 0.318 8.645 0.525 11.527 0.384 16.686 0.217 12.118 0.351
DistanceSquared -1.025 0.295 -0.690 0.498 -0.895 0.368 -1.315 0.195 -0.936 0.338
SlotRoute 0.188 0.000 0.194 0.000 0.179 0.000 0.183 0.000 0.188 0.000
RouteHHI -0.161 0.000 -0.177 0.000 -0.168 0.000 -0.172 0.000 -0.157 0.000
MaxAirportHHI -0.079 0.000 -0.087 0.000 -0.073 0.000 -0.105 0.000 -0.067 0.000
RouteShare 0.013 0.000 0.014 0.000 0.013 0.000 0.014 0.000 0.013 0.000
MaxAirportShare 0.003 0.000 0.003 0.000 0.003 0.000 0.003 0.000 0.003 0.000
LCCCompForHCC -0.054 0.000 -0.047 0.000 -0.055 0.000 -0.036 0.000 -0.060 0.000
LCCCompForLCC 0.073 0.000 0.087 0.000 0.090 0.000 0.078 0.000 0.084 0.000
AltRouteLCC1M -0.043 0.000 -0.043 0.000 -0.044 0.000 -0.039 0.000 -0.045 0.000
Circuity -1.266 0.000 -1.359 0.000 -1.285 0.000 -1.373 0.000 -1.231 0.000
Distress -0.020 0.000 0.194 0.000
Chpt11Ops -0.047 0.000
DistressDiff -0.003 0.000
Pre4Chpt11 -0.041 0.000
Post4Chpt11 -0.005 0.222
Loadfactor -0.011 0.000 -0.009 0.000 -0.009 0.000 -0.011 0.000 -0.010 0.000
AirlineCost 0.062 0.000 0.048 0.000 0.056 0.000 0.055 0.000 0.052 0.000
Size 0.049 0.000 0.086 0.000 0.098 0.000 -0.010 0.225 0.094 0.000
Quarter 2 0.005 0.126 0.001 0.796 -0.002 0.582 0.006 0.047 -0.002 0.483
Quarter 3 0.027 0.000 0.024 0.000 0.017 0.000 0.029 0.000 0.020 0.000
Quarter 4 -0.018 0.000 -0.018 0.000 -0.025 0.000 -0.010 0.001 -0.023 0.000
2002 -0.157 0.000 -0.183 0.000 -0.188 0.000 -0.115 0.000 -0.192 0.000
AirlineCost*Distress 0.015 0.000
Size*Distress -0.009 0.000
RouteShare*Distress 0.0000 0.495
RouteHHI*Distress -0.008 0.001
Wald ?
2
25,300,000 23,300,000 24,600,000 23,600,000 25,500,000
Prob > ?
2
0.000 0.000 0.000 0.000 0.000
R-squared: within 0.565 0.528 0.552 0.533 0.569
between 0.005 0.001 0.007 0.033 0.010
overall 0.001 0.006 0.015 0.019 0.019
5 2 4 1 3
Table 5: Second-stage G2SLS regression estimates using 1992, 1997, and 2002 data
82
2.5. Summary and discussion
The study’s results are summarized in Table 6 below. Ample support for the theoretical
arguments set forth in this paper is found. The implications of these findings are
discussed in this section, and some limitations and directions for future research are
noted.
The primary objective of this research is to reconcile the extant theoretical conflict
revolving around the impact of firm financial distress on prices. Based on a review of
varied theoretical perspectives and numerous empirical studies, it is suggested that
financial distress is negatively related to prices. It is noted, however, that this may not be
true in all cases. More specifically, it is hypothesized that operating costs, firm size and
market shares, as well as market concentration and a firm’s financial standing relative to
its competitors may impact the magnitude of a troubled firm’s pricing actions. A strategic
contingency framework which incorporates these moderating effects is developed and
tested using a comprehensive panel dataset from the U.S. airline industry.
The empirical results provide clear statistical support for all hypotheses: Firm financial
distress negatively impacts prices, and it is shown that these price effects are greatest for
carriers that operate under bankruptcy protection. The empirical results further suggest
that this is particularly true for firms with lower operating costs and smaller market
shares, and for firms operating in highly concentrated markets. The difference between a
83
focal firm’s financial distress and that of its competitors is also shown to impact the
magnitude of airlines’ pricing actions. All hypothesized direct and moderating effects are
thus supported.
Hypothesized effect
on prices
H
y
p
o
t
h
e
s
i
s
Testing variable Direct
Interaction
w/ Distress
variable Finding
Empirical
support for
hypothesis?
1 Distress – – Yes
2
(Post4Chpt11 –
Pre4Chpt11)
< 0
< 0 Yes
3 AirlineCost + + Yes
4 Size – – Yes
5 RouteShare + + Yes
6 RouteHHI
– – Yes
7 DistressDiff – – Yes
Table 6: Summary of results
This study’s results suggest that passengers traveling on distressed or bankrupt carriers
pay nearly four percent less than other passengers, all else equal. This is, of course, a
desirable outcome from a consumer perspective as a distressed firm’s lower prices
implies increases in consumer welfare. Since bankruptcy only sometimes results in a
firm’s liquidation, there is no indication for longer term negative effects of Chapter 11
protection on consumer welfare through, for example, reduced competition or reduced
service levels.
84
For managers and policy makers, however, this finding may be troubling. Financial
distress appears to negatively affect a firm’s revenue streams by virtue of lower prices
and, in some instances, lower demand
39
and may ultimately reduce the firm’s profitability
(see also Kennedy 2000). Taken together, these findings raise questions about the
rationality of a distressed firm’s pricing behavior and the adequacy of Chapter 11
protection. The results suggest that financial distress is both at the beginning and at the
end of a vicious circle of literally destructive price competition (see also Moulton and
Thomas 1993 for a discussion of the success rates of reorganizations under bankruptcy).
Managers and policy makers try to avoid organizational failure by offering lower prices
and supporting reorganization efforts respectively, but the very opposite effect may be
observed in at least some instances: Financial distress, and Chapter 11 protection in
particular, lead to an increase in the competitive pressures, thus increasing the firm’s
distress and spreading it beyond the firm’s boundaries. While it is not an objective of this
research to make any managerial or public policy prescriptions, the findings presented in
this study may be useful in gaining a greater understanding of the effects of financial
distress on prices by considering the moderating effects of firm and market
characteristics.
The key message of this study is clear: Microeconomic and corporate finance theory
alone cannot fully explain the relationship between a firm’s capital structure and its
output market behavior. The diversity of firms and circumstantial characteristics have to
be considered when investigating the effect of financial distress on prices. Strategic
39
See Table 38: the negative coefficient of the Post4Chpt11 variable implies that bankrupt firms face lower
demand, all else equal.
85
management research offers an array of theoretical approaches to further explore this
issue, and a contingency framework appears to be an appropriate means to do so. In that
vein, the hypotheses reflect and the results present evidence for elements of prospect
theory, organizational learning theory, and strategic groups research, for example. The
author is unaware of any other research that has examined the research question at hand
from a strategic management perspective. By combining multiple theoretical perspectives
and incorporating them in a single, comprehensive contingency framework, the
understanding of the link between firm financial distress and prices is advanced.
Data from the U.S. airline industry are used for the empirical analyses. While this
selection has many desirable qualities in terms of the detail and availability of data, one
must consider the possibility that these findings may not be generalizable to other
industries. The U.S. airline business is particularly competitive and, to some extent, still
marked by the era of regulation
40
. The exploration of the effects of financial distress on
prices in a cross-section of industries is left for future research. Moreover, the DOT
airline data do not contain any information about booking and service classes. As noted
by Lee and Luengo-Prado (2005), the failure to recognize these distinctive attributes of
the tickets purchased is a potentially critical shortfall of any empirical analysis of air
fares.
Research of the impact of financial distress faces a general dilemma: While financial
distress is a firm-level phenomenon, prices are clearly market-specific. In this research,
40
Regulation by the Civil Aeronautics Board ended in 1978, but has shaped the industry in many ways.
Although formally deregulated, regulatory controls (e.g. slot controls, antitrust rulings) continue to impact
the industry.
86
the impact of firm-level financial distress on individual product market prices is
investigated. This approach presents some challenges in that it is more difficult to isolate
statistical effects, and it may be desirable to investigate this research question in the
context of single-market firms. The latter are, however, hard to find nowadays.
On a final note, it should be stated that this study’s results may also depend upon the
measurement of financial distress. This study employed distress measures based on Z
scores and Chapter 11 dummy variables given that they have been widely applied in the
extant literature. The finance literature offers numerous variations of these measures as
well as entirely different ones (see e.g. Gritta 2004 for a comprehensive review of some
of these measures). Future research may explore the sensitivity of the results with respect
the measurement of financial distress.
This research contributes to the literature on the link between firm financial distress and
output market behavior. It is shown that this issue is far from being fully understood and
that strategic management theory offers avenues for further exploration of the impact of
financial distress on prices. This study has made a first step in this direction by estimating
the moderating effect of a number of strategic contingencies on this relationship.
87
3. The effect of firm financial distress on firm inventories: A supply chain
perspective
In this chapter, the effects of financial distress on inventory holdings are discussed and
tested empirically. The structure of this chapter is similar to that of Chapter 2 of this
dissertation: The subject matter of this essay is introduced in Section 3.1. The latter
includes a statement of the research questions and contributions of this research. Section
3.2 presents a review of theories and prior research on the relationship between firm
financial condition and inventories, and a baseline hypothesis is formulated. A supply
chain perspective is discussed in Section 3.3. It is hypothesized that firm power not only
directly impacts firm inventories, but also moderates the effect of financial distress on
inventories. Details about the data sample and the empirical methodology are provided in
Section 3.4. The empirical results are presented and discussed in Section 3.5, and the
study’s findings are summarized in Section 3.6. Managerial implications are discussed,
and suggestions for future research are provided, while the study’s limitations are noted.
3.1. Introduction
Financial considerations play an important role in inventory decision making. The survey
results presented by Osteryoung et al (1986), for example, indicate that 73.5% of all
respondents consider the firm’s cash position, and 57.3% factor in anticipated changes in
interest rates when making inventory decisions. It is intuitively appealing to assume that
firms under financial distress will shed inventories to generate liquidity. For American
88
car manufacturer Chrysler Corp., for example, reducing inventories was a major
component of its turnaround efforts (Stundza and Milligan 2001). Case Corporation, a
U.S. manufacturer of construction and agricultural equipment, also drastically cut
inventories when it restructured its business in the early 1990s (Buxbaum 1995).
While anecdotal evidence suggests that declining firm financial condition implies lower
inventory levels, prior empirical research on this relationship has produced ambiguous
results (see e.g. Corbett et al. 1999, Guariglia 1999). Upon closer examination of
previously published work, which relies exclusively on finance and economic theory, it
becomes clear that the link between firm finances and inventories is not yet well
understood, both in theoretical and empirical terms. Insights from inventory theory and
supply chain research will be useful to better understand this relationship and to improve
upon the specification of empirical estimation models.
A supply chain perspective on the link between financial distress and inventories is of
particular interest in this research. More specifically, this research is concerned with the
effect of a (distressed) firm’s power relative to buyers and suppliers on the firm’s
inventory decisions. In other words, can a distressed firm with greater levels of power
push greater amounts of inventory onto suppliers and buyers? If so, firms may want to
pay more attention to the financial condition of potential supply chain partners and be
aware of the potentially adverse impact of distressed firms’ inventory decisions. The
interplay between financial distress, supply chain power, and inventories remains
unexplored. The following paragraphs summarize the state of knowledge in this area and
89
outline the agenda of this research.
A sizeable literature in the economics and finance fields deals with the effects of financial
parameters on inventories. Some researchers have taken a rather macroeconomic
approach, analyzing the impact of monetary policy on aggregate inventory levels across
industries (see e.g. Corbett et al. 1999). Another set of research papers has investigated
the relationship between firm financial parameters such as bank lending rates or cash
flows and firm inventories (see e.g. Carpenter et al. 1998, Gertler and Gilchrist 1994).
While approaching the phenomenon from different theoretical and methodological
angles, many researchers contend that unfavorable financial conditions are associated
with lower inventory levels across an economy and within firms. The empirical findings,
however, provide only partial support for the researchers’ contentions. Corbett et al
(1999), for example, find that interest rates are a significant predictor of inventory levels
in certain industries only. Similarly, the results presented by Gertler and Gilchrist (1994)
suggest that the coverage ratio, i.e. the ratio of a firm’s cash flow and short term interest
expenses, explains the inventory behavior of small firms but not that of large firms. A
study by Guariglia (1999), finally, establishes a significant relationship between firm
finances and firm inventories during recessionary periods only.
The inconsistency of prior findings may, in part, be explained by differences in variable
measurement, the composition of data samples, estimation techniques, and perhaps most
importantly, variations in model specification. As the relationship between firm financial
factors and firm inventories appears to be more complex than previously assumed,
90
important explanatory variables may have been omitted in past research. In fact, most
published articles in this area rely exclusively on corporate finance and economic theory.
As pointed out by Roumiantsev and Netessine (2007), authors thereby ignore the insights
provided by inventory theory. Classical inventory models suggest that firm inventories
are a function of factors such as average demand, average lead times, holding costs,
demand and lead time variability, for example. Out of these factors, only demand has
been incorporated in the models of the articles referenced in the previous paragraph.
Potential specification problems encountered in prior research may therefore be alleviated
by drawing on inventory theory to a greater extent than has been done before.
Another shortcoming of the extant literature relating firm financial factors to inventories
may be the myopic treatment of inventories as firm decision parameters and the neglect
of the supply chain context in which most firms operate. While a firm’s managers
ultimately decide on the amount of inventory they order and sell, firms typically operate
within the confines of the terms and conditions negotiated with buyers and suppliers.
Some firms, for example, commit to specified service levels and must hold more
inventory to meet these performance targets. In other instances, buyers and sellers closely
cooperate by implementing Vendor-Managed Inventory (VMI) programs, for example.
Under this regime, a firm physically holds inventory that is managed (and possibly
owned) by a supplier until items are used in production or sold. Regardless of the
inventory policy in place, a firm’s bargaining power relative to its buyers and suppliers
will significantly impact the extent to which the firm exerts control over its inventories
(Wallin et al. 2006). Supply chain considerations, thus, may have a substantial impact on
91
a firm’s inventory holdings and on the degree to which a firm’s financial distress affects
inventories. This research adds to prior work in the firm finance-inventory area by
drawing on the supply chain power literature and incorporating associated measures in
the empirical analysis.
In summarizing, this essay theoretically and empirically revisits the link between firm
financial distress and firm inventories. An objective of this research thereby is to gain a
refined understanding of why a firm’s financial situation may have an impact on
inventories. This relationship is also tested empirically. Particular attention is paid to the
specification of the regression equation using not only microeconomic theory, but also
inventory theory, and insights from supply chain research. Also investigated is how the
nature of supply chain relationships, i.e. inter-firm power (im)balances, impact the extent
to which firms can reduce inventory holdings when experiencing financial distress. The
following research questions, thus, emerge:
1. Does a firm’s financial situation have an impact on its inventories after controlling for
other relevant parameters prescribed by inventory theory and supply chain research?
2. How does a firm’s (supply chain) power impact its inventory holdings?
3. Is the magnitude of the presumed effect of financial distress on inventories impacted
by power (im)balances in supply chain relationships?
The contributions of this research are manifold. First, it is shown that firm inventories
should respond to changes in firm financial condition. Prior research has not provided
such theoretical rationales. This is, to the best of the author’s knowledge, the first attempt
92
to investigate the research question at hand from both a microeconomic and an inventory
theory perspective, thus providing a broader, more complete theoretical basis for the
empirical analyses. Second, the impact of supply chain relationship variables on
inventory management is investigated. To date, few researchers have empirically
analyzed how the nature of buyer-supplier power balances impact firm inventory levels
41
.
In this essay, the role of firm power and concentration in both the upstream (supplier) and
downstream (buyer) markets in explaining focal firm inventories is examined. Moreover,
the analysis of the relationship between firm financial distress and inventories is extended
beyond the boundaries of the firm and is approached from a supply chain perspective,
thus more appropriately capturing the external influences on firms’ (inventory) decisions
(Cox et al. 2003, Dobson 2005). More specifically, it is argued that a firm’s buying and
selling power moderates the distress-inventory relationship. This contingency framework
may help reconcile prior findings by defining when and under what conditions the effect
of firm distress on inventories is greatest. This research thus adds to both the inventory
and supply chain literatures by analyzing the relationships between firm financial
distress, supply chain power, and firm inventories.
Besides its academic theoretical appeal, this research also has potentially important
managerial implications for supplier selection. If, for example, distressed firms are shown
to use their power to push inventory ownership to buyers or suppliers, firms may want to
carefully evaluate a potential partner firm’s financial condition and determine how the
41
Amihud and Mendelson (1989) study how firm market power affects firm inventory. They do not,
however, consider a firm’s power over suppliers or market concentration measures. Blazenko and
Vandezande (2003), in turn, investigate the relationship between market concentration and inventories only
and also ignore characteristics of the upstream supplier market.
93
partner’s distress might affect inventory ownership in the supply chain. In addition,
Halley and Nollet (2002) note that supplier selection and supplier development become
increasingly strategic, long-term firm decisions. An investigation of the role and impact
of financial considerations on such decisions, therefore, seems timely and managerially
relevant.
3.2. The financial distress-inventory relationship
In this section, the theoretical bases for a link between firm financial distress and
inventories are reviewed. Most prior research relied on economic theory when
investigating this relationship. This literature and the underlying theoretical rationales are
reviewed below. The second subsection discusses the firm finance-inventory link from an
inventory theory perspective. It is also suggested that inventory theory offers various
determinants of firm inventories that have not been included in prior economics research.
This section concludes with the formulation of a baseline hypothesis.
3.2.1. Economic theory
Within the economics stream of research, three articles, all first published in 1994, merit
particular attention. Gertler and Gilchrist (1994) are among the first to explore the
relationship between monetary policy (interest rates) and firm inventory levels. Kashyap
et al (1994) present a very similar study but use firm liquidity rather than security-market
interest rates as a measure of financial condition. Both papers support the lending view
94
which suggests that a firm’s dependence on external finance drives the strength of the
relationship between firm financial condition and inventories. Carpenter et al (1994),
finally, focus uniquely on the availability of internal finance as a determinant of
inventory (dis)investments and disregard macroeconomic factors such as security-market
interest rates. All three papers are discussed in more detail below.
Gertler and Gilchrist (1994) investigate the relationship between monetary policy and
firm behavior with respect to sales and inventories. The authors present two theoretical
rationales which suggest that tight monetary policy (i.e. an increase in interest rates)
negatively affects firm output and inventories. First, it is noted that rising interest rates
weaken firms’ balance sheet positions by reducing cash flows (net of interest) and
lowering the value of collateral assets. Consequently, borrowers reduce their spending
which implies output and inventory contractions. Second, monetary policy regulates the
pool of funds that is available to bank-dependent borrowers. The effect of monetary
policy on firm behavior is argued to be particularly strong for firms with limited access to
public capital markets. Both rationales, thus, suggest that monetary policy may affect
firm sales and inventories, and that firm financial factors, the access to capital markets in
particular, influence this relationship.
Gertler and Gilchrist (1994) use firm size to approximate a firm’s access to capital
markets and use industry-level time series data disaggregated by firm size classes to
estimate the effects of monetary policy on firm behavior. Descriptive analyses and the
estimation of structural inventory equations with the firm’s coverage ratio (cash flow
95
over total interest payments) as the key independent variable of interest indicate that
small firms’ sales and inventories decline more significantly during and after periods of
tight monetary policy. This result is shown to be significant and quantitatively
meaningful for small firms but not for large firms which supports the contention that tight
monetary policy particularly affects small firms with limited access to public capital
markets.
The work of Kashyap et al (1994) is closely related to that of Gertler and Gilchrist (1994)
and is motivated by the observation that there has been little empirical support for a
relationship between real interest rates and inventory investment. Yet, the observations
that inventory movements explain a substantial portion of the variability in aggregate
output, and that economic downturns typically follow periods of tight credit strongly
suggest such a relationship.
Kashyap et al (1994) attribute the lack of empirical support to measurement
imperfections. More specifically, the authors suggest that measures such as security-
market interest rates do not fully capture firm financial conditions or the cost of external
finance (e.g. bank loans). The latter, however, is argued to have a greater impact on
inventories than security-market interest rates. The authors’ key hypothesis thus states
that firms that depend on external finance should see their inventories fall more sharply
than firms with higher levels of internal funds and better access to public debt markets.
This contention is frequently referred to as the “lending view” in extant research.
96
Kashyap et al (1994) seek to empirically validate their hypothesis by regressing the
change in inventories on a set of firm-level determinants which include most notably the
inventory/sales ratio, the change in sales over the current and preceding years, and a
measure of liquidity (cash and marketable securities over total assets). A series of
different regression analyses using time series data indicate that firm liquidity is
consistently positively and significantly related to changes in inventory. This is, however,
only true for the 1974-75 and 1981-82 time periods when there were substantial liquidity
constraints. Data from 1985-86 are used as a control sample, and for this time period the
coefficient of the “liquidity” variable is statistically insignificant. In summarizing, the
authors thus conclude that financial factors influence inventory movements during tight
money (recessionary) episodes but not otherwise.
Most prior research relating inventory investments to financial parameters focuses on the
effects of monetary policy (e.g. Gertler and Gilchrist 1994) or financial factors such as
commercial paper spread and the mix of bank loans and commercial paper on firm
inventories (e.g. Kashyap et al. 1993). Carpenter, Fazzari and Petersen (1994) build on
this stream of research and add to it on two accounts: First, they focus on the flow of
internal finance as opposed to on monetary policy effects (see Gertler and Gilchrist 1994)
and external (bank) finance (see Kashyap et al. 1994). Moreover, Carpenter et al (1994)
test the importance of financing constraints using high-frequency (i.e. quarterly) panel
data and are thus able to observe short term changes in inventory investment levels. The
perspective of financing constraints, as adopted by Carpenter et al (1994), builds on the
notion that external finance (e.g. loans, bonds, commercial paper) is substantially more
97
expensive than internal finance (e.g. earnings and depreciation flows). The latter, thus, is
the preferred means of financing (inventory) investments.
Internal finance, however, is extremely volatile over the business cycle as it is
immediately affected by a slow-down in sales revenues given fixed or quasi-fixed
production costs in the short-run. As a consequence, comparatively liquid assets with
relatively low adjustment costs, such as inventories, are likely to absorb most of the
internal finance fluctuations of financially constrained firms. Carpenter et al (1994) argue
that this is particularly true for small firms whose access to external finance alternatives
such as corporate bonds and commercial paper is impeded by the lack of publicly
available information and the ensuing information asymmetry, adverse selection and
moral hazard problems. Small firms, the authors suggest, are thus forced to rely on
expensive bank loans as a last recourse to compensate for fluctuations in internal finance.
The effect of internal finance constraints on inventory (dis)investment is therefore
expected to be even greater for small firms than for large firms. The authors further
suggest that the magnitude of this effect depends on the optimality of inventory levels at
the beginning of the period. This contention builds on the idea that the marginal cost of
liquidating inventory stocks increases as current inventory levels (negatively) deviate
from optimal inventory levels.
Carpenter et al (1994) use Compustat data from the U.S. manufacturing industry (1981-
1992) to perform a series of regression analyses. The results generally indicate that the
level of cash flows is positively related to inventory investment, or put differently,
98
internal finance flows account for a significant portion of the variability in inventory
investment. While this is found to be true for both small and large firms, the authors note
that the effect tends to be greater in magnitude for small firms. The general result, thus, is
in line with the authors’ theoretical expectations. It is further noted, that the movements
of cash flows are highly procyclical, which, combined with the identified cash flow-
inventory link, provides a rationale for the high volatility of inventory investment over
the business cycle.
More recent empirical research also finds partial support for the contention that (firm)
financial factors impact firm inventory. Corbett et al (1999), for example, present a study
of UK and Japanese industries. They find that interest rates are significant predictors of
inventory investments in the paper, chemicals, and non-electric machinery industries
(UK), as well as in the Japanese chemicals, steel and iron, and metal manufacturing
industries. A study by Guariglia (1999) of UK manufacturing firms further explores the
effect of financial factors – Guariglia uses the coverage ratio as a measure of a firm’s
financial condition – on inventories. Her findings indicate a significant positive
relationship between coverage ratios and inventory levels during recessions and periods
of tight monetary policy.
In summary, it is noted that researchers in the economics field expect that less favorable
financial conditions will result in lower inventory levels. The significance levels of
empirical findings, however, vary greatly from study to study, depending on the
measures, data sets, and time periods used. It is also noted that researchers have used a
99
broad range of financial variables (interest rates, coverage ratios, and cash levels, for
example). No prior research has attempted to more comprehensively measure the
multifaceted firm financial distress construct and relate the latter to firm inventories. This
essay fills this gap. In addition, it will be argued in this research that previously
unobserved factors may also impact firm inventories and moderate the magnitude of the
financial distress-inventory relationship.
3.2.2. Inventory theory
Firms hold inventory for at least two reasons. First, delivery and production cycles are
typically not perfectly aligned. Natural stocks of raw materials as well as intermediate
and finished products therefore occur at various points throughout the production and
distribution process. These inventories are typically referred to as cycle stocks. Second,
inventories buffer against uncertainty. Specifically, unexpectedly high demand or longer
than usual lead times may lead to costly disruptions in manufacturing and delivery. Safety
stocks are a means of mitigating this risk by holding extra inventory that will be used
only if the need arises.
Determining the magnitude of cycle and safety stocks is a crucial task in inventory
management. While holding inventory is costly due to warehousing and opportunity
costs, not holding inventory may result in substantial stockout costs. The latter can take
the form of backorder (e.g. expediting) or lost sales costs, for example. Inventory theory
has been concerned with developing optimal, i.e. cost-minimizing or profit-maximizing
100
inventory policies. Multiple models have been proposed for different settings and
assumptions (see e.g. Tersine 1994). It is not the focus of this research to provide a
comprehensive review of these models. Rather, two questions are asked. First, which
determinants of inventories are proposed by inventory theory? And second, how may
firm financial distress be related to inventories from an inventory theory perspective? To
address these questions, two widely used and commonly known inventory models, the
r,Q model and the s,S model are briefly reviewed below. Particular attention is paid to the
r,Q model, and most of the subsequent discussion refers to this inventory policy. The
general results relating to the determinants of inventory levels and the relationship
between financial distress and inventory levels do, however, hold for most other
inventory models as well.
The r,Q inventory model is an extension of the well-known and widely used economic
order quantity (EOQ) model which, in its most basic form, balances ordering and
inventory holding costs
42
. First developed by Harris (1913), the EOQ and its variants
have been prominently featured in inventory management research and practice for over
ninety years (see Erlenkotter 1990 for a review of the early history of the EOQ model).
This model’s appeal lies in its relative simplicity and ease of use, as well as in its
robustness (Alstrom 2001). Reuter (1978) surveyed a total of 228 firms in five states in
the U.S. and finds that 75.4% of all respondents use the EOQ on a continuing basis with
an additional 9.6% indicating occasional use of the EOQ. In a study conducted by
McLaughlin et al (1994), 28% out of 236 survey respondents reported using the EOQ. In
42
See any textbook on inventory management for a detailed discussion of the economic order quantity
model and its variants (e.g. Tersine 1994)
101
a more recent survey, Rabinovich and Evers (2002) find that the EOQ is deemed
important in managerial practice and is commonly used by logistics managers to
determine optimal order quantities
43
. Zinn and Charnes (2005) note that quick response
(QR) inventory policies have become increasingly popular in modern inventory
management and therefore analyze the relative merits of the EOQ and QR methods,
respectively. Based on a series of numerical analyses, Zinn and Charnes (2005) conclude
that the EOQ continues to be the preferred inventory policy when order costs are
relatively high
44
. Numerous researchers have conducted sensitivity analyses and have
found that moderate deviations from the EOQ’s assumptions do not have a substantial
impact on order quantities and associated total inventory costs (e.g. Sun and Queyranne
2002). Its popularity, simplicity, and robustness make the EOQ a good starting point for
developing an inventory theory perspective on the financial distress-inventory
relationship.
The classical EOQ is based on the following assumptions: the demand rate is constant,
continuous and known, and lead times are zero. Replenishments are received
instantaneously and all at once, and the cost of placing an order as well as unit holding
costs are constant. The classical EOQ model considers only a single product and assumes
that there are no interactions with other inventory items. Moreover, it is assumed that the
firm has sufficient capital and capacity to purchase the economic order quantity. In the
r,Q model, the rather unrealistic assumptions of constant demand and zero lead times are
43
On a five point scale (1 = unimportant, 5 = very important) 256 survey respondents ascribe an average
weight of 3.27 to the EOQ, and 19.61% report the use of the EOQ for determining finished goods orders.
44
See Zinn and Charnes (2005) for a summary of their study’s results. Table 6 (p.139) identifies the
conditions under which QR and EOQ policies are preferred, respectively.
102
relaxed. With stochastic demand, nonzero but constant lead times, and per-unit backorder
costs, the total inventory cost equation is defined by
( )
2
I
S Q
TC A B E M r H r M
Q
(
= + ? > + + ? (
¸ ¸ (
¸ ¸
, where
S is the expected sales volume
45
over the planning horizon, A is the order cost
46
, H is the unit holding cost, and B is the
unit backorder cost. M is lead time demand (i.e. the sales volume during lead time) and r
is the reorder point which, along with order quantity Q, is the decision variable of
interest. ( ) E M r > is the expected stockout quantity, while
( )
r M ? represents the
average size of the safety stock. Taking the derivative of
I
TC with respect to Q and
setting the expression equal to zero yields the cost minimizing order quantity:
( )
*
2S A B E M r
Q
H
+ ? > (
¸ ¸
= . Similarly, the optimal reorder point
*
r is obtained by
setting the derivative of
I
TC with respect to r to zero. This yields the cost-minimizing
stockout probability ( )
*
HQ
P M r
BS
> =
47
. Under the assumption of normally distributed
demand, the latter value converts to standard normal deviate k , and the reorder point is
defined as
*
LTS S
r M k SL k L ? ? = + ? = + ? where
LTS
? is the standard deviation of lead
time sales, and
S
? is the standard deviation of sales
48
.
The control parameters of the r,Q inventory policy are thus defined by the expected sales
45
The inventory literature commonly uses the term Demand instead of sales volume. It is noted however,
that inventory decisions are made a priori based on forecasts.
46
In a manufacturing context, these order costs may also be thought of as production setup costs.
47
Since Q is a function of r and vice versa, the optimal solutions for these parameters are found by
iteration.
48
Lead times are assumed constant. See Tersine (1994) for more detail.
103
volume, sales variability, lead times, ordering costs, holding costs and backorder costs.
Inventory theory suggests that these parameters appropriately predict a firm’s inventory
decisions and thereby firm inventory levels. Figure 10 provides a graphic illustration of
the r,Q policy.
Figure 10: Illustration of the r,Q policy
The s,S policy is similar to the r,Q inventory policy but may differ from the latter in that
the order quantity is variable when inventories are reviewed periodically only, or when
demand is lumpy, i.e. does not follow a pure Poisson process with unit demand. Each
time the inventory level drops below a threshold level (or reorder point) s, on order of
size (S-s) is placed. For details on the mathematical derivation of the inventory control
parameters (s,S) the interested reader is referred to Denardo (2003), for example. In this
context, it shall suffice to note that the s,S policy is defined by the magnitude of demand,
the item’s unit costs, per-unit holding and backorder costs, as well as ordering costs.
Randomness of demand and lead times can also be incorporated in s,S type inventory
Q*
time
r*
Safety
stock
M
average inventory level
Inventory
Q*
time
r*
Safety
stock
M
average inventory level
Inventory
104
models. The determinants of an s,S inventory control policy, thus, are essentially the
same as the determinants of the r,Q policy.
In summary, inventory theory suggests that firm inventories should be a function of
average demand, demand variability, lead times, ordering costs, holding costs and
backorder costs (or lost sales costs), regardless of the specific inventory control policy in
place.
Next, the potential effects of firm financial distress on inventories are discussed from an
inventory theory perspective. Financial costs are most directly reflected in a firm’s
holding costs. Holding costs include a financial cost component representing the capital
cost of inventories (Followill et al. 1990). While holding costs also comprise a noncapital
carrying charge
49
, Timme (2003) notes that the financial component of holding costs
usually exceeds noncapital carrying charges. In accordance with this contention, the
survey results presented by Fraser and Gaither (1984) suggest that 68% of all firms
approximate inventory carrying costs with borrowing costs. The latter are a function of a
firm’s financial condition (see e.g. Buzacott and Zhang 2004, Wiersema 2005).
Specifically, a deterioration of a firm’s financial condition implies higher borrowing costs
and thereby higher inventory holding costs. Returning to the inventory control parameters
of the r,Q policy, it is evident that higher holding costs entail lower inventory levels, all
49
Noncapital carrying costs comprise the costs of warehousing, obsolescence, pilferage, damage, and
insurance, as well as taxes and administrative charges (see Timme, 2003).
105
else equal
50
. The optimal order quantity
( )
*
2S A B E M r
Q
H
| |
+ ? > (
¸ ¸
|
=
|
\ ¹
is decreasing in
holding costs
*
i.e. 0
Q
H
| | ?
<
|
?
\ ¹
, as is the reorder point: The optimal stockout probability
increases in H ( )
*
HQ
P M r
BS
| |
> =
|
\ ¹
51
. This translates to a lower safety factor k and
consequently to a lower reorder point r
( )
*
LTS S
r M k SL k L ? ? = + ? = + ? . A negative
relationship between firm financial distress and optimal firm inventories can therefore
straightforwardly be established from an inventory theory perspective.
While the author is unaware of any empirical inventory research relating firm financial
condition to inventory levels, there is some analytical research on the relationship
between various financial factors in a broader sense and inventory decisions. For
completeness, a few examples of such research are discussed below.
One literature stream, for example, investigates the effects of trade credits, permissible
payment delays granted by suppliers, on economic order quantities. Haley and Higgins
(1973) analyze the interdependence of inventory decisions and credit terms, and
determine jointly optimal order quantities and payment schedules. Most subsequent
research assumed credit terms as exogenously given (i.e. unilaterally defined by the
supplier) and focused on the effects of trade credits on order quantities. Chapman et al
50
In addition, higher holding costs may imply lower firm output choices and hence lower demand.
51
Note that the optimal stockout probability is also a function of Q, which, in turn, decreases in H .
*
P
therefore increases in H and decreases in H , Overall, the optimal stockout probability increases in H .
106
(1984), for example, conduct an average cost analysis and conclude that trade credit
periods, while significantly impacting total costs, do not affect optimal order quantities.
Chand and Ward (1987), on the contrary, find that order quantities increase as payment
delay times increase. Rachamadugu (1989) reconciles these contradictory findings and
ascribes them to differences in the assumptions and setup of the respective models. In
summary, Rachamadugu’s (1989) analyses corroborate Chand and Ward’s (1987)
intuitively appealing findings, as do the results of a more recent study conducted by
Chang and Teng (2004). For a review of some earlier works on inventory models with
consideration of permissible payment delays the interested reader is referred to Kim and
Chung (1990).
Another stream of research is concerned with the impact of budget constraints of
inventory decisions. Financially distressed firms are likely to operate under budgetary
constraints. Rosenblatt (1981) formulates a constrained inventory optimization problem
with limited budget availability. He uses the Lagrangian procedure to demonstrate the
intuitive result that the optimal order quantity will be restricted to the maximum
affordable level when the budget constraint is tight. A multi-item newsvendor problem
with a budget constraint is analyzed by Moon and Silver (2000). The authors’ attention
focuses on rules for optimally allocating scarce resources to different products. In the
context of this research, however, it is sufficient to note that a restriction on total
expenditure is shown to lead to lower than optimal order quantities and increased overall
costs (Moon and Silver 2000). Abdel-Malek and Montanari (2005) extend Moon and
Silver’s (2000) work by conducting an analysis of the multi-product newsvendor problem
107
with two (generic) constraints. Rustenburg et al (2000) also present a study similar to that
of Moon and Silver (2000) in the context of spare parts logistics, where resupply
decisions for multiple items must be made under limited budgets. One of the basic
finding’s of Rustenburg et al (2000) is that budget constraints result in lower part
availability levels.
Empirical inventory research is challenging from a data collection standpoint and
therefore rather scarce (examples include Ballou 1981, Roumiantsev and Netessine
2007). Inventory theory is, however, indispensable when empirically explaining
inventory levels and analyzing the relationship between firm financial distress and
inventories. This research builds on prior work in the economics area by drawing on
inventory theory to explain this relationship and by incorporating a set of previously
ignored inventory variables in the regression model.
3.2.3. The financial distress-inventory hypothesis
The theoretical link between firm financial distress and inventories has been discussed
from both an economics perspective and an inventory theory perspective in the previous
subsections. Clearly, both theories suggest that greater levels of financial distress (i.e. less
favorable financial conditions) result in lower inventory levels, all else equal. Most prior
research argues that budgetary constraints and increased borrowing costs lead distressed
firms to hold less inventory. Prior research has found some support for this hypothesized
relationship (Carpenter et al. 1998, Carpenter et al. 1994, Gertler and Gilchrist 1994,
108
Kashyap et al. 1994). As Roumiantsev and Netessine (2007) point out, however, this
body of work “might contain biases because many important micro-economic data points
that affect inventories have been left out, including lead times, demand uncertainty,
inventory holding costs, etc.” (p.6). This study reexamines the financial distress-
inventory link while controlling for these inventory determinants.
Besides the previously discussed rationale that financial distress results in budgetary
constraints and increased borrowing costs, it may be argued that managers of distressed
firms have an incentive to liquidate assets (Hofer 1980) such as inventories in an effort to
increase liquidity and improve key firm performance measures such as the Return on
Assets (RoA). In summary, there appears to be clear theoretical and at least some
empirical support for Hypothesis 8:
Hypothesis 8: Holding demand constant, greater levels of financial distress result in
lower inventories.
So far, the focus has been on firm level determinants of inventories. In the next section,
this focus is expanded to include a firm’s supply chain partners. Specifically, the effect of
power on inventories and the financial distress-inventory link is discussed.
3.3. The supply chain perspective
In this section, the relationship between distress and inventories is analyzed from a
109
supply chain perspective. Specifically, the role of power in inter-firm relationships and
firm (inventory) decision making is reviewed in Section 3.3.1. In line with prior research
in the industrial organization economics field, it is suggested that a firm’s power position
impacts its inventory decisions. The second subsection analyzes the moderating role of
inter-firm power in the financial distress-inventory relationship. It is hypothesized that
power determines to what extent financial distress affects firm inventories. The resulting
contingency framework is subsequently tested using U.S. industry data.
3.3.1. Supply chain considerations in inventory decisions
Many parameters influence managerial decision making. While firm-level variables such
as holding and purchasing costs, for example, naturally have a strong impact on
managerial decisions relating to sales prices and inventories, market factors and inter-
firm relationship variables cannot be ignored.
First, competitors’ actions clearly impact a firm’s choices. Researchers from both the
economics and strategy fields have contended that managers must anticipate competitive
reactions and evaluate their implications when deciding on sales prices (see e.g. Chen et
al. 1992, Gibbons 1992). By the same token, firms also compete on inventories. Cachon
(2001), for example, analyzes competitive inventory policies and, for a given set of
assumptions, defines a competitive Nash equilibrium in inventories (see also e.g.
Mahajan and Ryzin 2001). As a consequence, a firm’s inventory decisions are a direct
function of competitors’ inventory choices.
110
Second, firms are typically a part of supply chains that extend across many companies
from raw material suppliers to the end customer. As firm decisions impact the
functioning of the entire supply chain, supply chain firms are necessarily interdependent
(Cox et al. 2001). This interdependence is of particular interest in this research. When the
Case Corporation reduced its inventories as a part of its restructuring efforts, for example,
suppliers had to bear the burden, but were willing to do so to improve customer service
levels (Buxbaum 1995). Chrysler’s aggressive cost-cutting measures implemented in
2000 and 2001, in turn, were considered “acts of war” (p.32) by some suppliers who
agreed to cooperate only because they had little choice (Stundza and Milligan 2001).
These examples illustrate how firms’ (inventory) decision making can be constrained by
cooperative arrangements and coercive pressure exerted by buying and supplying firms.
The extent to which firms are willing or forced to yield to these constraints is a function
of the inter-firm power balance. It is argued in this research that power not only impacts a
firm’s inventory decision but also moderates the link between financial distress and
inventories. Following a brief review of the role of power in inter-firm relationships and
some related literature, the corresponding hypotheses are derived below.
3.3.1.1. Inter-firm relationships: The role of power
There exists a sizeable literature base on the nature, drivers, and consequences of power
in inter-firm relationships. Gaski (1984) provides a review of the early work in this field
111
and, in summarizing, defines power as the “ability to evoke change in another’s
behavior” (p.10). Emerson (1962) relates power to dependence and suggests that Firm
A’s power over Firm B equals Firm B’s dependence on Firm A. The sources of (firm)
power in (inter-firm) relationships were first analyzed by French and Raven (1959).
According to French and Raven (1959) these bases of power include:
? Reward power: A can motivate B by granting rewards;
? Coercive power: A can effectively punish B;
? Legitimate power: A has a legitimate right to prescribe B’s behavior;
? Referent power: A serves as a model to B;
? Expert power: A’s expertise conveys A the power to influence B.
The term power often carries a negative connotation (Hingley 2005). French and Raven’s
power bases, however, suggest that power may be used both collaboratively and
coercively. Along the same lines, Frazier and Antia (1995) suggest distinguishing
between the possession and the application of inter-firm power. Frazier and Antia (1995)
argue that the channel context and the specific inter-firm power constellation drive the
communication style between firms. The latter can be either threatening (as seen above in
the case of Chrysler) or collaborative (as evidenced in the previously mentioned example
of Case Corp.). Firms with some degree of power can, thus, exert either coercive control
or collaborative control to affect other firms’ forced or voluntary behavioral change,
respectively (Frazier and Antia 1995, Hingley 2005).
Cox et al (2003) note that power is an element of every buyer-supplier relationship. The
112
authors suggest that each relationship can be characterized by one of four power
structures: buyer dominance, supplier dominance, buyer-supplier interdependence, and
buyer-supplier independence. Cox (2001) further notes that firms strive to be in a
dominant position over buyers and suppliers so as to extract the maximum amount of
value generated in the supply chain. Cox et al (2001) have coined the term “value
appropriation” to describe this mechanism which can take the form of cost squeezing on
the supply side or high-margin pricing on the sales side, for example. As the previously
cited examples of Chrysler and Case Corp. (Buxbaum 1995, Stundza and Milligan 2001)
have illustrated, firms may also use their dominant power position to shift inventory
ownership to suppliers or buyers. In this vein, Wallin et al (2006) contend that “if a firm
within a specific buyer-supplier relationship were to hold bargaining power, this would
greatly enhance its ability to dictate to and make certain demands of a specific supplier”
(p.59) with respect to the inventory management approach used in the supply chain (see
also Dobson 2005). This research empirically tests the contention that a firm’s power
relative to its suppliers and buyers will impact firm inventory levels, ceteris paribus.
Prior work in this area is reviewed in the following subsection.
3.3.1.2. Supply chain power and inventory decisions
Few researchers have investigated the effect of power on inventories. Blazenko and
Vandezande (2003) ascribe the lack of power-inventory research to the fact that “the
academic literature on inventory focuses on production and procurement as the principal
determinants of […] inventory […] management” (p.256) while “the principal focus of
113
the study of inventory in the economics literature is on the macroeconomic role of
inventory as a stabilizing or destabilizing factor for production in business cycles”
(p.256). Much of the research relating dyadic power, i.e. a firm’s power vis-à-vis another
firm, to inventories remains descriptive in nature and is mostly based on case studies (see
e.g. Dobson 2005). Within the supply chain management literature, articles on power and
its implications are, for the most part, purely conceptual. The author is aware of only two
papers that empirically investigate the effects of power on inventories. Both papers are
housed within the industrial organization economics literature and are discussed in turn.
Amihud and Mendelson (1989) suggest that “a firm with market power will use inventory
as a wedge between the quantity available for sale and the quantity shipped to market”
(pp.269-270). According to Amihud and Mendelson (1989), firms build up inventories
when supply exceeds demand in an effort to maintain higher prices and keep production
at constant levels. With demand greater than supply, in turn, firms deplete inventories to
maximize revenues. A firm’s motivation to use inventories to smooth price fluctuations
thereby increases with the firm’s market power as “greater market power implies a
stronger effect of the firm’s sales quantity on price” (p.270). Amihud and Mendelson
(1989) test the market power-inventory relationship using Compustat data from the U.S.
manufacturing industry. The results suggest that firm market power, measured by either
the Lerner index ([price – marginal cost]/price) or the firm’s market share, positively
affects firm inventories after controlling for firm sales, sales trends, sales variability, and
average industry inventory levels. The authors therefore conclude that “market power has
a sizeable effect on inventory, which has been overlooked so far” (p.275).
114
Blazenko and Vandezande (2003) build their article on the contention that stockout costs
should be represented in inventory estimation models. The authors suggest that greater
levels of competition erode profit margins and thereby reduce the amount of profits
foregone in case of a stockout, while, at the same time, more competition also increases
stockout costs due to the greater availability of alternative sources of supply. According
to Blazenko and Vandezande (2003) the effect of market concentration (an indicator of
the level of power firms possess in a given market) on inventories is ambiguous and
depends on whether the effects of lower foregone profit or increased lost sales costs
prevail. The authors empirically investigate the effect of industry concentration
(measured by the two-firm concentration ratio) on finished goods inventory levels (at the
industry level). The control variables included in the model are, most notably, industry
gross-margins and a set of industry indicator variables. Data from the U.S. manufacturing
industry are used for the empirical analyses. The results suggest that higher industry
concentration levels result in lower inventory levels, ceteris paribus. Blazenko and
Vandezande (2003) conclude that “a less competitive product market reduces the adverse
consequences of stock outs and firms respond by reducing inventories” (p.263).
In summary, Amihud and Mendelson (1989) suggest that greater levels of market power
imply higher inventory levels, while Blazenko and Vandezande (2003) find that
inventories are lower in more concentrated markets (implying more powerful firms). This
conflict may, in part, be explained by different levels of analysis (firm vs. industry), and
differences in measurement. In addition, it is noted that both models fail to include
115
variables prescribed by inventory theory, such as lead times and the cost of holding
inventory, for example. The results may, therefore, be biased (Roumiantsev and
Netessine 2007). Also, neither article attempts to relate focal firm or industry power to
the power levels of buyers and suppliers. Yet, power is dyadic in nature (Cox et al. 2001,
Emerson 1962, Frazier and Antia 1995, Gaski 1984), and a complete evaluation of power
must consider a firm’s power relative to another firm or industry. Prior research has
focused uniquely on downstream power vis-à-vis buyers but has ignored the upstream
supply side. Power, however, is “Janus-faced”, i.e. double-sided (Cox 2001), as firms are
engaged in power relationships with both their buyers and their suppliers (as well as with
their competitors). This research addresses this shortcoming in terms of measurement of
power and proposes a comprehensive set of power measures (see Chapter 3.4) capturing
not only focal firm power, but also power levels in the buying and supplying industries.
3.3.1.3. The power-inventory hypotheses
As outlined previously, prior research on the role of power in supply chain relationships
has suggested that greater levels of power allow firms to obtain more favorable terms and
conditions in negotiations with their buyers and suppliers (Blazenko and Vandezande
2003, Cox 2001, Cox et al. 2001, Wallin et al. 2006). More powerful firms may thus be
able to push the burden of inventory ownership onto buyers and suppliers to a greater
extent than less powerful firms. Hypothesis 9 is therefore proposed as a baseline power-
inventory hypothesis:
116
Hypothesis 9: Greater firm power results in lower inventory levels.
Hypothesis 9 can be refined by distinguishing between firm power relative to suppliers
and buyers, respectively. Accordingly, Hypothesis 10 and Hypothesis 11 are introduced
below.
Wallin et al (2006), for example, argue that a firm with bargaining power may impose
item availability targets on suppliers, thus forcing suppliers to hold larger inventories to
meet these targets while reducing the need to hold inventory at the buying firm (see also
Cox et al. 2001). In addition, a powerful firm may be able to demand inventory
consignments from its suppliers, thus providing the buying firm with improved item
availability without incurring the cost of inventory ownership (Wallin et al. 2006).
Hypothesis 10 therefore suggests that greater power over suppliers implies lower
inventory levels, all else equal.
Hypothesis 10: Greater firm power relative to suppliers results in lower inventory
levels.
Hypothesis 11 mirrors the reasoning underlying Hypothesis 10 and projects it to the
downstream relationship between a firm and its buyers. Accordingly, greater levels of
power over buying firms are expected to be associated with lower inventory levels, all
else equal. While the work of Blazenko and Vandezande (2003) presents some evidence
in support of this contention, the results of the study published by Amihud and
117
Mendelson (1989) appear to contradict this expectation. The latter researchers implicitly
assumed that firms can use inventories to mitigate price fluctuations only if they directly
own these inventories. Powerful firms in supply chains, however, may be able to dictate
the release and buildup of inventories even when these inventories are not under direct
ownership and control. From a supply chain perspective, Hypothesis 11, therefore, does
not necessarily disagree with the arguments and results presented by Amihud and
Mendelson (1989).
Hypothesis 11: Greater firm power relative to buyers results in lower inventory levels.
Besides the direct effect of power on firm inventories, it is also contended that firm
power impacts the extent to which firms can reduce inventories when experiencing
financial distress. These moderating hypotheses are developed below.
3.3.2. Firm power as a moderator of the distress-inventory link
The inconsistency of the results presented by prior research on the link between financial
variables and inventories may be an indication that there are factors that affect the
magnitude and significance of this relationship. Prior research has suggested that firm
size may be such a moderator. This rationale is briefly reviewed below. This essay, in
turn, focuses on the moderating role of firm power. The related reasoning is discussed in
Section 3.3.2.2 and the corresponding hypotheses are formulated.
118
3.3.2.1. Prior research: Firm size as a moderator of the distress-inventory link
As discussed in Section 3.2.1, most studies on the link between financial factors and
inventories have suggested that the magnitude of this relationship may differ by firm size.
Gertler and Gilchrist (1994) and Carpenter et al (1998, 1994), for example, perform
separate regression analyses by firm size classes (small vs. large). These authors find
empirical support for their contention that smaller financially constrained firms
experience stronger inventory contractions due to their limited access to capital markets
and, thus, means of financing inventory investments. Kashyap et al (1994) find that the
effect of firm liquidity on inventories is stronger for firms without bond ratings than for
firms with bond ratings. Since unrated firms typically are smaller firms, their results also
suggest that the effect of financial constraints on inventories differs by firm size.
Firm size may be a proxy for a firm’s power, with larger firms being more powerful than
smaller firms, all else equal. Following this reasoning, the negative effect of firm distress
on inventories (as hypothesized in Hypothesis 8) may be expected to decrease with the
firm’s power. It is noted, however, that the arguments set forth in prior research focus
uniquely on the operating implications of financial constraints, suggesting that firms with
limited resources must reduce inventory investments, particularly when external funds
can be procured at high costs only. This research, in turn, suggests that financially
distressed firms want to reduce inventories and will do so to the largest extent possible. In
other words, reducing inventories is considered desirable as long as potentially negative
119
consequences of inventory cutbacks, such as stockouts and decreases in customer service
levels, can be mitigated by increased buyer and supplier efforts (e.g. in terms of increased
inventory holdings, shorter lead times, etc.). This contention is discussed in more detail in
the following subsection.
3.3.2.2. The power moderator hypotheses
While firms consistently strive to increase efficiency and profitability, these efforts are
reinforced during corporate turnarounds (e.g. Hofer 1980). Tom Sidlik, then Executive
Vice President with Chrysler, for example, indicated that “we’ve accelerated our ongoing
cost-reduction programs so that we can take 15% costs out of the system by the end of
2002.” (Stundza and Milligan 2001, p. 30). Sidlik continued to note that “in the current
business situation, we are counting on our supplier partners to stand with our company
[…] in these difficult times” (p.31). The importance of concessions and support offered
by suppliers during corporate turnarounds is further illustrated by Arogyaswamy and
Yasai-Ardekani (1995) who argue that cutting inventory can only be a successful
turnaround strategy if potentially resulting delivery delays can be mitigated through
suppliers’ or buyers’ increased efforts, for example. Finkin (1985) also notes that during
company turnarounds “[t]erms and conditions of sale are worth fighting over” (p.17) and
that a supplier’s agreement to shorter lead times may help reduce inventory levels.
Clearly, a firm’s bargaining power vis-à-vis its suppliers and buyers will determine to
what extent such concessions will be made. In a similar vein, Hambrick and Schecter
(1983) note that a firm’s power might affect its choice of turnaround strategy as “strong
120
channels of distribution […] could allow [the distressed firm] to solve [the] problems at
less human and organizational cost” (p.234), with loyal and obedient distributors carrying
larger shares of the burden.
These arguments and examples lend support for the contention that distressed firms may
be able to reduce inventories to a greater extent when they have higher degrees of power
relative to their suppliers and buyers. Hypothesis 12 is formulated accordingly:
Hypothesis 12: The effect of firm financial distress on inventories increases with the
firm’s power.
Hypothesis 12 can be specified for a firm’s power relative to buyers and suppliers,
respectively:
Hypothesis 13: The effect of firm financial distress on inventories increases with the
firm’s power relative to suppliers.
Hypothesis 14: The effect of firm financial distress on inventories increases with the
firm’s power relative to buyers.
The moderating effect of firm power on the distress-inventory relationship is graphically
illustrated in Figure 11. On average, a negative relationship between the magnitude of
121
firm financial distress and inventories is expected (Hypothesis 8). This relationship,
however, is hypothesized to be stronger the greater the firm’s power (Hypothesis 12-
Hypothesis 14).
Figure 11: The moderating effect of power on the distress-inventory relationship
An overview of the resulting model is given in Figure 12. In summarizing, a set of
hypotheses on the link between firm financial distress and inventories has been
formulated based on a variety of theoretical perspectives. Particular attention is given to
the role of power as a determinant of firm inventories and as a moderator of the distress-
inventory relationship.
Financial
distress
Inventory
a
v
e
r
a
g
e
m
o
r
e
p
o
w
e
r
f
u
l
f
ir
m
s
le
s
s
p
o
w
e
rfu
l firm
s
negative
interaction
effect
Financial
distress
Inventory
a
v
e
r
a
g
e
m
o
r
e
p
o
w
e
r
f
u
l
f
ir
m
s
le
s
s
p
o
w
e
rfu
l firm
s
negative
interaction
effect
122
Figure 12: Research model
3.4. Data and methodology
The hypotheses set forth in the previous sections are tested using data sets comprising
information on a cross-section of U.S. industries. Details on the data samples,
specification of the model, variable measurement, and data sources are provided in the
following subsections.
Financial Distress Inventory
Power
• Total inventory
• Raw materials
• Finished goods
• Firm power
• Firm power relative to suppliers
• Firm power relative to buyers
Control
variables
Financial Distress Inventory
Power
• Total inventory
• Raw materials
• Finished goods
• Firm power
• Firm power relative to suppliers
• Firm power relative to buyers
Control
variables
123
3.4.1. Sample selection
The empirical tests are conducted using data from U.S. manufacturing firms. Two data
sets from 1997 and the time period from 1998 to 2004, respectively, are used for the
analyses. This subsection provides information on the sample selection criteria.
Industries
Most empirical inventory research has focused on manufacturing industries for the
obvious reason that manufacturing firms are likely to hold substantial inventories
(Carpenter et al. 1998, Carpenter et al. 1994, Corbett et al. 1999, Gertler and Gilchrist
1994, Guariglia and Schiantarelli 1998, Kashyap et al. 1994, Roumiantsev and Netessine
2007). Manufacturing industries are defined by the North American Industry
Classification System (NAICS). The NAICS system has replaced the U.S. Standard
Industrial Classification (SIC) system and all U.S. government agencies commonly report
industry statistics by NAICS codes. NAICS codes have between two and six digits and
are structured hierarchically. The first two digits of a NAICS code designate the
“economic sector” and the third digit identifies the “subsector”. The fourth, fifth, and
sixth digits designate the “industry group”, “NAICS industry”, and “national industry”,
respectively. Manufacturing industries are part of the economic sectors 31-33. All firms
in these sectors for which complete data are available are included in the empirical
analyses.
This study approaches the analysis of inventories from a supply chain perspective and
124
investigates, among other things, the role of a firm’s power on firm inventories and on
the relationship between financial distress and inventories (see Figure 12). Given the
dyadic nature of power, the data set also comprises information on the wholesale and
retail trade industries
52
as these industries likely are manufacturing firms’ principal
suppliers and buyers (besides buyers and suppliers within the manufacturing industries).
Other industries that may potentially buy from or sell to manufacturing industries are not
included in the analysis for the two following reasons:
? Service industries: Service industries
53
are of limited interest in the context of
inventory studies and are not considered in this research (see also Roumiantsev
and Netessine 2007).
? Insufficient data availability: The remaining economic sectors are Agriculture,
Forestry, Fishing and Hunting (11), Mining (21), Utilities (22), Construction (23).
While these industries may potentially be involved in the exchange of goods (i.e.
inventories) with manufacturing firms, these industries must be excluded from the
data analysis due to insufficient data availability at the industry level.
Specifically, industry sales data and industry concentration ratios are available at
highly aggregated levels only and, thus, are not usable in the empirical analyses.
52
NAICS codes 42, 44, and 45.
53
Service industries are found in the following economic sectors (two-digit NAICS codes are given in
parentheses): Transportation and Warehousing (48,49), Information (51), Finance and Insurance (52), Real
Estate, Rental and Leasing (53), Professional, Scientific, and Technical Services (54), Management of
Companies and Enterprises (55), Administrative and Support and Waste Management and Remediation
Services (56), Educational Services (61), Health Care and Social Assistance (62), Arts, Entertainment, and
Recreation (71), Accommodation and Food Services (72), Other Services (except Public Administration)
(81), Public Administration (92).
125
Time periods
This research uses data from 1997 to 2004. This time period is selected for two reasons.
First, consistent data at the industry level are available for this time period only. More
recent data (after 2004) were not available at the time of writing, and older data (prior to
1997) were aggregated differently as the industry classification system was revised in
1997 with the move from the Standard Industry Classification (SIC) system to the
NAICS system. Second, the selection of a relatively recent time period is adequate given
that inventory dynamics may have been significantly different in earlier time periods
prior to the widespread adoption of information systems and Just-In-Time practices, for
example (Roumiantsev and Netessine 2007).
In the U.S., an Economic Census is conducted every five years (years ending with “2”
and “7”). During the time period considered here (1997-2004), Economic Census data
were thus collected in 1997 and 2002. While the 2002 Economic Census data were not
available at the time of writing, data from the 1997 Economic Census could be obtained
from the website of the Bureau of Economic Analysis (BEA). The Economic Census data
provide more detailed industry information, for example on industry sales and
concentration ratios, than the data that are collected by the BEA during years in which no
Economic Census is conducted. Therefore, the empirical analyses are performed using
two different datasets: First, a panel data set is constructed. This panel data set contains
information on a cross-section of U.S. manufacturing industries for the time period from
1998 to 2004. Second, a cross-sectional data set using data from 1997 only is constructed.
126
Using two distinct data sets for the empirical analyses has several advantages. The time
series data set, henceforth denoted “data set I”, is relatively large with multiple
observations per firm. Larger sample sizes generally facilitate the empirical analyses and
typically result in more robust coefficient estimates. The 1997 data set, in turn, provides
more fine-grained industry level data. This data set, henceforth denoted “data set II”, thus
is particularly useful when attempting to evaluate the relative power balances between
industries. In addition, the robustness and validity of the model are underlined if both
data sets produce consistent coefficient estimates.
Frequency
As noted by Carpenter et al (1998, 1994), high-frequency quarterly data may be desirable
for the analysis of firm inventories and financial factors due to their dynamic and volatile
nature. Many firms, however, report only selected parameters on a quarterly basis. Raw
materials and finished goods inventory data, for example, are often available on an
annual basis only. In line with prior research and due to greater data availability, annual
data are used in this study (Guariglia 1999).
3.4.2. Model specification
The purpose of this section is to derive an empirical inventory estimation model which is
grounded in inventory theory and supply chain management research. This research
thereby enhances prior economics research which generally modeled inventories as a
function of (lagged) sales, financial indicators, and a small set of control variables only.
127
According to inventory theory, firm inventory decisions should be a function of order
quantities (Q) and safety stock (SS) (see also Figure 10). The specific magnitude of end-
of-period inventories will then also be a function of sales realization ( )
t
S . In addition, it
is argued in this essay that a firm’s distress and power will affect firm inventories.
Dummy variables to account for inventory accounting differences (LIFO, AvgCost
54
) are
included as well (Carpenter et al. 1994, Gertler and Gilchrist 1994, Kashyap et al. 1994,
Roumiantsev and Netessine 2007). This yields the following inventory model:
(1) ( ) , , , , , ,
t
Inv f Q SS S Distress Power LIFO AvgCost = .
As seen in Chapter 3.2.2, the order quantity Q is a function of expected sales (
t S ), order
costs (A), backorder costs (B) and holding costs (H):
(2)
( )
, , , t Q f S A B H = .
Similarly, safety stocks are shown to be a function of lead times (L), sales (
t S ), sales
variability (
S
? ), and the safety factor k which, in turn, is a function of the optimal
stockout probability ( )
*
t
HQ
P M r
BS
| |
> =
|
\ ¹
(see Chapter 3.2.2). Safety stocks can, thus, be
represented as follows:
(3)
( )
, , , , t
S
SS f L H B S ? = .
The author is unaware of prior empirical inventory research that measured order and
backorder costs. For lack of suitable proxy measures, it is common practice to exclude
order and backorder costs from empirical inventory analyses (see e.g. Lieberman et al.
54
Further detail is provided in Section 3.4.3.2.
128
1999, Roumiantsev and Netessine 2007). While an attempt is made to approximate
order/setup costs (see section 3.4.3.2), backorder costs are not further considered in this
research. Holding costs are a function of the cost of the item that is purchased or
produced and the holding cost rate. Unit cost measures are not readily available due to a
lack of output indicators. Total costs of goods sold, however, are highly correlated with
sales, and are therefore not included in the regression model. A proxy for a firm’s capital
carrying charge will be used to approximate holding costs. Substituting Equations (2),
and (3) in Equation (1) and dropping the above mentioned variables, Equation (1) can be
rewritten as follows:
(4)
( )
, , , , , , , , , t
t S
Inv f S S A H L Distress Power LIFO AvgCost ? = .
Expected sales (
t S ) and realized sales (
t
S ) are naturally highly correlated with
S
t
t t
S S ? = + , where
S
t
? is the forecast error. To avoid excessive multicollinearity
55
,
Equation (4) is therefore restated as follows:
(5)
( )
, , , , , , , , ,
S S
Inv f S A H L Distress Power LIFO AvgCost ? ? = .
The resulting basic empirical estimation equation is defined in Equation (6) below:
(6) Inventory
itf
= ?
0
+ ?
1
SalesForecast
itf
+ ?
2
ForecastError
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
LeadTime
itf
+ ?
7
Distress
itf
+ ?
8
Power
itf
+ ?
9
LIFO
itf
+ ?
10
AvgCost
itf
+ ?
11
Distress
itf
* Power
itf
+ ?
itf
The subscripts i, t, and f designate the industry, time period, and firm respectively.
55
Not only are actual and expected sales highly correlated, but actual sales and the standard deviation of
sales are correlated as well.
129
Inventory
itf
, for example, indicates firm f’s inventory level in time period t, where firm f
operates in industry i. The interaction term (Distress
itf
* Power
itf
) is included to test
Hypothesis 1 Hypothesis 12, Hypothesis 13, and Hypothesis 14.
This inventory model specification differs from prior specifications on multiple accounts:
First, the model presented here controls for important predictors of firm inventories as
prescribed by inventory theory. The author is aware of only one study that controlled for
sales variability and lead times (Roumiantsev and Netessine 2007). The latter study,
however, does not investigate the effect of financial distress on inventories, nor does it
consider the role of firm power in inventory management. Second, measures of a firm’s
buying power and selling power are included. While few prior studies analyzed the
impact of market power on inventories (Amihud and Mendelson 1989, Blazenko and
Vandezande 2003), this is the first study to differentiate between power over buyers and
power over suppliers. In addition, a more comprehensive measure of financial distress is
proposed. Prior research relied on one-dimensional measures such as market interest rates
or firm cash flows to estimate holding costs or a firm’s financial situation. Market interest
rates, however, are poor approximations of holding costs (or firm financial condition, for
that matter), as such measures do not account for the heterogeneity of firms’ borrowing
rates. Measures such as cash flows, in turn, may not comprehensively evaluate firm
financial condition. Consequently, this is—to the best of the author’s knowledge—the
first study to investigate the effects of firm financial distress on inventories from an
inventory theory and supply chain management perspective.
130
3.4.3. Variables and Measurement
Variable measurement has been a major challenge in inventory research and probably is
the most important reason for the scarcity of empirical inventory studies. The measures
used in this research are in part based on the work of Roumiantsev and Netessine (2007)
and Carpenter et al (1994). The dependent and independent variables are discussed in
turn.
3.4.3.1. Dependent variable
The dependent variable in this study is firm-level inventory, i.e. firm f’s inventory in
period t. Specifically, three distinct firm inventory measures are used: total inventory,
raw materials inventory, and finished goods inventory
56
. Several researchers have
previously used these inventory variables in empirical analyses (e.g. Blazenko and
Vandezande 2003, Guariglia 1999). In line with prior research, absolute inventory values
are used (see e.g. Roumiantsev and Netessine 2007). Total inventories of firm f (which is
affiliated with industry i) in time period t are denoted
itf
TotalInv and are measured in
U.S. dollars as reported on the balance sheet (see e.g. Amihud and Mendelson 1989)
57
.
Analogously, raw materials inventories and finished goods inventories are denoted
itf
RawMatInv and
itf
FinGoodsInv , respectively.
56
Work-in-process inventories are reported by few firms only and are therefore not analyzed separately.
57
Roumiantsev and Netessine (2007) note that it is generally not necessary to adjust dollar values in time
series data since inflation has been at very low levels in the United States over the past decade.
131
All inventory data are obtained from Standard & Poor’s Compustat database. Only, firms
with at least $5,000 worth of (total) inventory are included in the dataset to ascertain that
only inventory-carrying firms are analyzed. The regression analyses are performed with
all three inventory measures.
3.4.3.2. Independent variables
The set of independent variables is discussed next. Variables suggested by inventory
theory are discussed first, followed by a review of the measures of firm power. Unless
otherwise stated, all data are obtained from Standard & Poor’s Compustat database.
? SalesForecast
itf
Inventory ordering decisions are made based on expected demand. For each firm and
time period, annual sales are forecast as follows:
( )
1
1 2 t
t
S S g
?
= ? + , where the
average growth rate over the past two years
( )
g is defined as
( ) ( )
2 3 3 1 2 2
2
t t t t t t
S S S S S S
g
? ? ? ? ? ?
? + ?
= . When only incomplete prior sales data are
available or prior sales were impacted by merger and acquisition (M&A) activity, the
average growth rate equals the growth rate for the years for which data are available
and no M&A activity was observed.
? SalesSurprise
itf
While firms make inventory decisions based on expected demand, the magnitude of
inventories at the end of the year is impacted by actual demand. If actual demand
exceeds expected demand, year-end inventory levels should be lower. Conversely,
132
lower than expected sales should result in larger year-end inventories. The
SalesSurprise variable provides some control for the difference between expected
demand (SalesForecast) and realized demand. Following the procedure suggested by
Roumiantsev and Netessine (2007), a binary variable is created. Specifically, this
variable equals “1” if actual demand is greater than expected demand and is
“0”otherwise.
? SalesVariability
itf
SalesVariability is measured as the coefficient of variation of sales and is a proxy for
demand variability. The more variable firm demand, the more inventory a firm will
hold, ceteris paribus. While sales may not be equal to actual demand in case of
stockouts, demand variability is approximated with sales variability. The coefficient
of variation of sales is computed as the ratio of the standard deviation of sales over
the past three periods and the mean of sales over the past three periods:
( )
( )
1, 2, 3
1, 2, 3
Sales i t t t f
itf
Sales i t t t f
CVS
?
µ
? ? ?
? ? ?
= . The SalesVariability variable thus is a standardized
measure of the variability of sales.
? SetupCost
itf
Information on firms’ average cost of setting up production or placing orders is not
readily available. The magnitude of setup costs may, however, be reflected in the
magnitude of firms’ order backlogs. Clearly, there are many reasons why firms
backlog orders: high demand, long lead times, or manufacturing problems are just a
few potential causes of order backlogs. On average, however, larger backlogs may
simply reflect higher order setup costs: Firms may prefer to accumulate orders before
starting production if the cost of setting up production is high. Since the absolute
133
value of backlogged orders is likely highly correlated with sales, the standardized
value of order backlogs is used as a proxy for production setup costs:
itf
itf
itf
OrderBacklog
SetupCost
Sales
= .
? HoldingCost
itf
Inventory holding costs consist of warehousing/handling costs and capital carrying
costs (Timme 2003). While the former component cannot be estimated based on
available accounting information, the latter can be approximated as follows: The
capital cost of holding inventory is a function of the capital interest rate which
represents either the opportunity cost of internally financed (inventory) investments
or the borrowing cost of externally financed (inventory) investments. The firm-
specific interest rate is approximated by dividing the firm’s interest expenses by total
debts:
itf
itf
itf
InterestExpenses
HoldingCost
TotalDebt
= .
? LeadTime
itf
The measure of lead times follows the novel procedure suggested by Roumiantsev
and Netessine (2007). Roumiantsev and Netessine propose the following measure:
365
itf
itf
itf
AP
LeadTime
COGS
?
= where
itf
AP stands for Accounts Payable and
itf
COGS
stands for the Cost of Goods Sold. While not a measure of physical lead times, this
proxy captures the quarterly cash conversion cycle which may, to some extent reflect
physical shipment times. Roumiantsev and Netessine (2007) further justify the use of
this measure as follows:
“[A]ccounts payable are credited, then [the] product is shipped and is typically debited,
134
then it is received and [payment is made]. Hence, financial transactions are correlated with
times of shipment and delivery of inputs and therefore are correlated with the lag a
company has to respond to changing market environment by adjusting inventories.“ (p.13).
Roumiantsev and Netessine (2007) empirically verify “that the lead time proxy is not
dominated by standard payment terms (e.g. 30 or 60 days)” (p.14), and that lead times
are not merely a function of firm power (as measured by the firm’s market share)
58
. In
the data sets used here, the correlation coefficients between LeadTime and the firm
power measures are small and negative ( 0.06
LeadTime MarketShare
r
?
= ? and
0.08
LeadTime IndSalesNet
r
?
= ? ; see Table 10), suggesting that more powerful firms tend to
have shorter permissible payment delays. It may also be argued that longer payment
lead times result from a distressed firm’s inability to pay suppliers (implying inflated
accounts payable). The correlation coefficients between LeadTime and Distress are
positive and statistically significant in both data sets, although limited in magnitude
( 0.19 r ? ; see Table 10). This suggests that the lead time proxy used here may indeed
be a function of, amongst other factors, financial distress. Given the lack of suitable
alternative lead time proxies and the limited size of the distress-lead time correlation,
the LeadTime proxy is included in the subsequent analyses. The lead time proxy
yields the expected positive sign in Roumiantsev and Netessine’s (2007) analyses of
firm inventories.
? Distress
itf
Distress is a measure of firm f’s financial distress. The Distress variable is the
negative value of a firm’s Z score, a measure which was first developed by Altman
58
Firms with greater levels of power may be able to squeeze their suppliers and obtain longer permissible
payment delays, thus increasing accounts payable.
135
(1968). Based on discriminant analysis, Altman (1968) developed the following
model to estimate a firm’s financial fitness:
1 2 3 4 5
0.012* 0.014* 0.033* 0.006* 0.999* Z X X X X X = + + + + ,
where X
1
= working capital / total assets, X
2
= retained earnings / total assets, X
3
=
Earnings Before Interests and Taxes (EBIT) / total assets, X
4
= market value of equity
/ total liabilities, and X
5
= sales / total assets. The information needed to compute the
Z scores is included in the firms’ balance sheets and profit and loss statements. These
data, as well as stock market data are obtained from Standard & Poor’s Compustat
database. High Z scores indicate financial health, while low and negative scores
indicate (serious) financial distress. Specifically, a score of 2.67 or above indicates
financial health, and a score of 1.81 or below suggests (severe) financial distress
(Altman 2002). The Z scores are then rescaled to indicate the level of financial
distress, i.e. ( ) 1
itf
Distress ZScore = ? ? , such that higher (positive) Distress scores
indicate greater financial distress (see also Ferrier et al. 2002). The Distress variable
is included to test the effect of firm financial distress on inventories as hypothesized
in Hypothesis 8.
? DistressDummy
itf
While Distress is a continuous variable, DistressDummy is a binary variable which
indicates whether a carrier is considered financially distressed. Firms are categorized
as distressed and non-distressed based on the above-mentioned cutoff levels
suggested by Altman (1968). Specifically, firms with Z scores of less than 1.81 (i.e.
Distress scores of greater than -1.81) are considered financially distressed
(DistressDummy equals “1”). The sensitivity of the results with respect to the
136
definition of this cutoff value is investigated in Section 3.5. The DistressDummy
variable is used to investigate if distressed firms, on average, hold less inventory
(Hypothesis 8), and to split the data samples into distressed and non-distressed firms
(see Chapter 3.4.6 for further detail).
This study adds to prior research by investigating the effects of firm buying and selling
power on inventories and on the distress-inventory relationship. In the past, researchers
have used relatively simple measures of firm power and have ignored the inherently
dyadic nature of power (Cox 2001, Cox et al. 2001). Amihud and Mendelson (1989), for
example approximate firm power with either the firm’s market share or the firm’s gross
profit margin. Blazenko and Vandezande (2003), in turn, use a market concentration
measure to approximate the average level of power firms possess in a particular industry.
These measures may not fully capture inter-firm power balances. Unlike prior research,
this study uses a set of firm power measures which proxy not only the focal firm’s power,
but also the power levels in the supplying and buying industries. Since a focal firm’s
specific supply chain transaction partners (i.e. buyers and suppliers) cannot be identified
using accounting data, buyer and supplier industry characteristics are used as proxies of
buyer and supplier power. The measures of focal firm, supplier industry and buyer
industry power are presented below.
? IndustrySalesNet
itf
Many prior studies have used market shares
FirmSales
MarketShare
IndustrySales
| |
=
|
\ ¹
to
approximate a firm’s power (e.g. Amihud and Mendelson 1989). The regression
model established in Section 3.4.2, however, already contains the SalesForecast
137
variable and thus controls for the magnitude of firm sales. Including market shares in
the regression model would thus entail two potential problems: First, multicollinearity
problems may arise given the high correlation between sales forecasts and market
shares, thus resulting in inefficient estimates. Second, and perhaps more importantly,
the market share variable might then pick up a size effect (larger firms hold more
inventory) rather than the firm power effect it is intended to measure. This issue is
addressed by transforming the market share variable. Since firm sales are already
controlled for by means of the SalesForecast variable, the size of the firm’s
competitors may indicate the level of power a firm exerts in a market. Specifically,
the IndustrySalesNet variable indicates the sales volume (measured in U.S. $) of all
the other firms in the market (excluding the focal firm). To simplify the interpretation
of the coefficient estimates, the industry sales volume (net of firm sales) is inverted so
as to represent a proxy measure of a firm’s power:
( )
1
itf
it itf
IndustrySalesNet
IndustrySales FirmSales
=
?
. This variable thus indicates the
effect of an increase (decrease) in the sales volume of a firm’s competitors on the
firm’s inventory holdings after controlling for firm sales (~ SalesForecast).
Specifically, a positive coefficient estimate of the IndustrySalesNet variable would
suggest the following: The smaller the firm’s competitors, i.e. the more powerful the
focal firm, the more inventory the (focal) firm will hold. Conversely, a negative
coefficient would confirm the expectation expressed in Hypothesis 9: More powerful
firms, on average, hold less inventory.
For the empirical analyses, industries are defined at the six-digit NAICS level. While
some researchers have computed market shares at the four-digit NAICS level
138
(Amihud and Mendelson 1989), it is believed that the more fine-grained six-digit
NAICS industry data are better suited for the purpose of the analyses. The sensitivity
of the empirical results with respect to the granularity of industry definitions will be
investigated in Section 3.5. Industry sales data are obtained from the website of the
Bureau of Economic Analysis (BEA). These data include the values of exports by
U.S. firms, but do not include imports from foreign firms.
? IndCR4
it
Researchers in the industrial organization economics area have suggested that the
level of market concentration is an indicator of the competitiveness of markets (e.g.
Ravenscraft 1983). Specifically, firms in more concentrated markets are believed to
be more powerful since there are fewer competitors and collusion between firms is
easier to achieve (Waldman and Jensen 2001). While Blazenko and Vandezande
(2003) use two-firm concentration ratios (i.e. the sum of sales of the two largest firms
divided by total sales in the industry) as a measure of market concentration, the
Bureau of Economic Analysis (BEA) publishes 4, 8, 20, and 50 firm industry
concentration ratios. Given its widespread use in the extant literature (see e.g. Pryor
2001, Ravenscraft 1983), the four-firm concentration ratio (CR4) is used here.
Holding all else constant, greater values of the four-firm concentration ratio imply
greater firm power. IndCR4 thus is one of the measures of firm power used to test
Hypothesis 9 to Hypothesis 14 in the analysis of the second data set (part II) in which
generic industrial supply chains are constructed.
As with IndustrySales, industry concentration is measured at the six-digit NAICS
level. Industry concentration data are provided by the Bureau of Economic Analyses
139
and are available in Economic Census years only. The IndCR4 measure is therefore
included in the second data set (part II) only.
? SupplyCR4
(i-1)t
Analogous to the IndCR4 measure, SupplyCR4 is an indicator of the weighted
average concentration levels of those industries that sell to a focal industry. The
Input-Output Tables published by the Bureau of Economic Analysis, illustrate the
flow of goods (and services) between industries. Specifically, the I-O Tables not only
identify those industries that sell goods to another industry, but also indicate the value
of the respective transactions. As a result, the relative importance of supplying
industries to a focal industry can be evaluated and the weighted average four-firm
concentration ratio of the supplying industries can be computed (SupplyCR4) as
illustrated in Figure 13 below. As discussed in Section 3.4.1, this study focuses on
inventory-carrying industries. The average supplying industry concentration
measures, therefore, are based on the four-firm concentration ratios of manufacturing,
wholesale and retail trade industries only. Other (e.g. service) industries are not
considered in the computation of the SupplyCR4 measures. Moreover, only domestic
suppliers are considered when computing the supplying industry concentration ratio;
imports from foreign suppliers are not included.
Holding all else constant, an increase in the SupplyCR4 measure, suggests a relative
decrease in the focal industry’s (and focal firm’s) power. This variable is thus used to
test Hypothesis 10 and Hypothesis 13.
? BuyCR4
(i+1)t
Symmetrical to the SupplyCR4 measure, the BuyCR4 measure is the weighted average
140
of the four-firm concentration ratios of a focal industry’s buying industries (see
Figure 13). Holding all else constant, an increase in the BuyCR4 measure, suggests a
relative decrease in the focal industry’s (and focal firm’s) power. This variable is thus
used to test Hypothesis 11 and Hypothesis 14.
Figure 13: Illustration of the construction of industrial supply chains
There are three widely used methods of inventory accounting: The First In, First Out
(FIFO) method values inventories assuming that items are sold out of inventory in the
same order they were inventoried. Hence, the cost of the most recently added items
determines the value of end-of-period inventories. The Last In, First Out (LIFO) method
values inventories assuming that the most recently inventoried items are sold first.
Consequently, at the end of the accounting period, the oldest items are left over in
inventory. The Average Cost method values inventories at the weighted average cost of
40%
25%
20%
15%
Industry B
Industry C
Industry D
Industry E
40%
25%
20%
15%
Industry B
Industry C
Industry D
Industry E
40%
25%
20%
15%
Industry B
Industry C
Industry D
Industry E
4Supply CR
Industry F
35%
20%
15%
15%
15%
Industry G
Industry H
Industry K
Industry L
4Buy CR
Industry F
35%
20%
15%
15%
15%
Industry G
Industry H
Industry K
Industry L
4Buy CR
Industry A
Industry A
Focal Firm
Industry A
Industry A
Focal Firm
141
all units available for sale during the accounting period. As prices typically change over
time, each inventory accounting method results in different inventory valuations at the
end of the accounting period. Specifically, with generally increasing prices, the use of the
LIFO method will understate the true value of inventories, whereas the FIFO method
more appropriately reflects the value of ending inventories. The average cost procedure
results in inventory values that lie between LIFO and FIFO. To account for these
differences, it is common practice to include an indicator variable which identifies those
firms that use one of the “extreme” accounting methods. Following the example of prior
research (e.g. Blazenko and Vandezande 2003), two binary variables are included in the
model to account for differences in inventory accounting methods:
? LIFO
itf
This indicator variable equals “1” if the firm uses LIFO as the primary inventory
accounting method and equals “0” otherwise (see also Roumiantsev and Netessine
2007).
? AvgCost
itf
This indicator variable equals “1” if the firm uses the average cost method as the
primary inventory accounting method and equals “0” otherwise. (see also
Roumiantsev and Netessine 2007)
3.4.4. Data sources
All firm-level data are obtained from the Compustat database which is maintained by
Standard & Poor’s. This database includes accounting information on publicly traded
142
firms. While the focus on public companies excludes smaller, not publicly traded firms
from the analyses, this selection also ensures that all reported operating and financial data
conform to Generally Accepted Accounting Principles (GAAP) (Roumiantsev and
Netessine 2007). The Compustat database contains firm specific accounting data which
are commonly found in balance sheets and profit and loss statements, including all the
information that is required to construct the firm-specific variables.
Industry level data, most notably industry sales, are obtained from the Bureau of
Economic Analysis. Specifically, the BEA provides annual estimates of total industry
shipments (in U.S. dollars) for U.S. manufacturing industries. At the time of writing, data
were available for the time period from 1998 to 2004
59
. As noted above, the availability
of these data thereby defines the timeframe studied in the first part of the data analysis.
Industry level data for the year 1997 were obtained from the U.S. Census Bureau. This
agency’s website provides access to detailed industry statistics collected through the
Economic Census survey (1997). Total industry sales and industry concentration ratios
were collected from the Economic Census website.
The information relating focal firms to buying and supplying industries is found in the
Input-Output Tables, which are also published by the Bureau of Economic Analysis. The
data in these tables summarize the trade flows between industries. Specifically, the “Use”
tables indicate from which industries firms in a particular industry purchased goods and
59
These data are found in the file “GDPbyInd_SHIP_NAICS.xls” available on the BEA website.
143
services and indicate the respective dollar volumes such that the relative importance of
supplying industries can be evaluated. Conversely, the I-O Tables also identify the
industries that purchase from firms in a focal industry, and the relative importance of
buying industries in terms of the shares of total focal industry sales can be inferred.
The major limitation of using I-O Tables is that these tables focus on U.S. domestic firms
only and disregard foreign buyers and suppliers. As a consequence, the industry power
proxies used here (concentration ratios) may overstate the true power levels in these
industries, particularly when foreign firms hold significant market shares in these
industries. By the same token, industry sales data do not include imports from foreign
firms. This may result in incorrect estimates of the true size of industries and of firms’
market shares. It is believed, however, that the I-O Tables provide at least reasonable
estimates of industry characteristics for the purpose of inter-industry comparisons.
3.4.5. Descriptive statistics
This section provides descriptions of the data samples used in this research. Both data
sets (Part I and Part II) are discussed in turn.
3.4.5.1. Descriptive statistics: Part I
As discussed in Section 3.4.1, data from U.S. manufacturing industries (NAICS 3xxxxx)
for the time period from 1998 to 2004 are used for the empirical analyses. All
144
manufacturing firms for which information on all relevant variables were available for
any or all years in the 1998-2004 time period were included in the data set. The firm
observations in this data set represent about 8.5% of total sales and 9.9% of total
inventory holdings by all publicly traded U.S. manufacturing firms that are included in
the Compustat database.
A two-sample Hotelling T-squared test is implemented to evaluate to what extent the
firms included in this data set differ from those firms for which data are available in
Compustat but which are not included in the analyses due to missing data on one or more
variables. Specifically, the Hotelling test compares these two groups on the following
variables: Total inventories, sales, cost of goods sold, total assets and total debt. The test
yields a test statistic of 2.89. This statistic follows an F distribution and, thus, is
statistically significant at the five percent level. This result suggests that the data sample
used for the empirical analysis differs significantly from the population of firms included
in the Compustat database. Upon closer examination of the data, it becomes apparent
that, on average, the sample firms tend to be smaller (in terms of inventories, sales, costs,
assets, and debt) than those firms that are not included in the data sample (see Appendix
6). The results of the analyses presented here may, therefore, not be generalizable to
firms of all size classes.
The composition of the final data set is shown in Table 7. About forty percent of all firm
observations are in the computer and electronics industry. The second largest industry in
this data set is the machinery industry with 852 observations or about sixteen percent of
145
all observations. While the remainder of the data set comprises firms from a broad array
of manufacturing industries, it cannot be ascertained that the empirical results of this
study will be generalizable to all manufacturing industries, given the dominance of the
computer and electronics, and machinery industries
60
.
NAICS Industry N %
334 Computer and electronics 1983 37.9%
333 Machinery 852 16.3%
336 Transportation equipment 341 6.5%
339 Miscellaneous 275 5.3%
335 Electrical equipment 273 5.2%
332 Fabricated metal 252 4.8%
325 Chemical 203 3.9%
315 Apparel 187 3.6%
331 Primary metal 159 3.0%
316 Leather 137 2.6%
326 Plastics and rubber 136 2.6%
337 Furniture 111 2.1%
313 Textile mills 78 1.5%
327 Nonmetallic mineral 69 1.3%
323 Printing 57 1.1%
321 Wood 44 0.8%
322 Paper 36 0.7%
311 Food 26 0.5%
314 Textile products 11 0.2%
312 Beverage and tobacco 6 0.1%
Total 5236 100%
Table 7: Sample composition (Part I)
Table 8 presents the descriptive statistics of this sample. It is noted that raw materials and
finished goods data were not available for all firms. Hence, the sample size is smaller for
these particular variables. There is substantial variability in all variables. In some
instances, however, the standard deviations are larger than the means suggesting
skewness in the data. Consequently, all inventory variables, as well as SalesForecasts
60
In future research, within-industry analyses could be performed.
146
and DaysPayable (LeadTime) are log-transformed prior to the empirical estimation. It is
also noted that, on average, sales forecasts closely approximate actual sales and that
about 27 percent of all observations are for financially distressed firms.
Variable Mean Std. dev. Min Max N
Inventory Total (million $) 103.9 483.36 0.005 12,207 5236
Inventory RawMat (million $) 24.17 78.74 0.002 1,802 4505
Inventory FinGood (milion $) 43.98 249.23 0.001 7,319 4307
Sales (million $) 835.6 5,164.7 0.05 155,974 5236
Sales Forecast (million $) 847.2 5,149.6 0.01 158,827 5236
SalesSurprise 0.49 0.5 0 1 5236
Coeff. of Variation of Sales 0.20 0.19 0.001 1.73 5236
OrderBacklog/Sales 0.32 1.43 0 90.34 5236
Interest Rate 0.18 0.28 0 1 5236
Days Payable 49.69 50.57 1.32 1,248 5236
Distress -3.79 13.90 -333.2 220.9 5236
Distress Dummy 0.27 0.45 0 1 5236
Market Share (6 dig. NAICS) 0.07 0.18 0.000001 1 5236
LIFO 0.14 0.35 0 1 5236
AvgCost 0.09 0.28 0 1 5236
Table 8: Pooled descriptive statistics (Part I)
Table 9 presents the descriptive statistics for distressed and non-distressed firms
separately. The most striking differences are found in raw materials inventories and days
payable outstanding. Specifically, distressed firms appear to hold less raw materials
inventory and have larger accounts payable than non-distressed firms. The latter
observation can likely be attributed to distressed firms’ lower ability to pay. The former
observation, however, is interesting and lends some support for the contention that
147
distressed firms try to reduce inventories
61
. This is most easily done with raw materials
inventories since extant stock can be reduced by consuming materials while not placing
any new raw materials orders.
Variable Mean Std. dev. Min Max N Mean Std. dev. Min Max N
Inventory Total (million $) 104.8 388.85 0.011 8,349 3804 101.5 672.94 0.005 12,207 1432
Inventory RawMat (million $) 27.32 85.02 0.005 1,802 3278 15.76 57.99 0.002 993 1227
Inventory FinGood (milion $) 42.44 142.07 0.001 2,209 3180 48.35 424.88 0.001 7,319 1127
Sales (million $) 829.4 3,340.5 0.05 58,198 3804 851.8 8,241.6 0.05 155,974 1432
Sales Forecast (million $) 836.8 3,380.2 0.09 68,849 3804 875.0 8,163.7 0.01 158,827 1432
SalesSurprise 0.51 0.5 0 1 3804 0.42 0.5 0 1 1432
Coeff. of Variation of Sales 0.18 0.16 0.001 1.73 3804 0.25 0.23 0.002 1.73 1432
OrderBacklog/Sales 0.30 0.75 0 34.0 3804 0.40 2.46 0 90.34 1432
Interest Rate 0.18 0.30 0 1 3804 0.15 0.22 0 1 1432
Days Payable 41.74 27.02 1.78 630 3804 70.82 82.46 1.32 1,248 1432
Distress -6.57 11.82 -333.2 -1.8 3804 3.60 16.14 -1.8 220.9 1432
Market Share (6 dig. NAICS) 0.08 0.19 0.000001 1 3804 0.04 0.12 0.000002 1 1432
LIFO 0.15 0.36 0 1 3804 0.11 0.31 0 1 1432
AvgCost 0.08 0.28 0 1 3804 0.10 0.30 0 1 1432
Non-distressed firms Distressed firms
Table 9: Descriptive statistics (Part I) – distressed vs. non-distressed firms
A two-sample Hotelling T-squared test is performed to assess whether distressed firms
are statistically significantly different from non-distressed firms based on the variables
listed in Table 9. The test yields a test statistic of F = 58.23 which is statistically
significant at the one percent level, indicating that distressed firms, on average, differ
from non-distressed firms.
61
An alternative explanation may be that suppliers are reluctant to sell to distressed firms, especially when
the latter purchase on credit.
148
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1 Inventory Total (million $)
2 Inventory RawMat (million $) 0.89
3 Inventory FinGood (million $) 0.91 0.74
4 Sales (million $) 0.94 0.85 0.87
5 Sales Forecast (million $) 0.92 0.83 0.85 0.98
6 Sales Surprise 0.06 0.05 0.04 0.08 -0.06
7 Coeff. of Variation of Sales -0.25 -0.19 -0.24 -0.25 -0.23 -0.06
8 OrderBacklog/Sales -0.03 -0.05 -0.05 -0.06 -0.05 -0.01 0.06
9 Interest Rate -0.11 -0.12 -0.11 -0.10 -0.11 0.00 0.05 0.03
10 Days Payable -0.14 -0.09 -0.10 -0.17 -0.17 0.03 0.21 0.01 -0.02
11 Distress -0.14 -0.10 -0.07 -0.15 -0.14 -0.05 0.04 -0.01 -0.02 0.19
12 Distress Dummy -0.26 -0.23 -0.21 -0.28 -0.26 -0.08 0.17 0.03 -0.05 0.28 0.33
13 Market Share (6 dig. NAICS) 0.53 0.45 0.51 0.54 0.53 0.05 -0.12 0.00 -0.04 -0.06 -0.02 -0.11
14 Net Industry Sales (inverted) 0.22 0.20 0.22 0.22 0.21 0.01 -0.10 0.00 -0.03 -0.08 0.03 -0.07 0.75
15 LIFO 0.29 0.24 0.29 0.29 0.29 0.00 -0.18 -0.03 -0.04 -0.15 0.01 -0.06 0.15 0.09
16 AvgCost 0.01 -0.03 0.00 0.01 0.00 0.00 0.00 0.01 0.02 0.02 -0.01 0.02 0.01 -0.03 -0.12
(Values in bold are significant at the 5% level)
Table 10: Pairwise correlations (Part I)
149
Pairwise correlations are displayed in Table 10. Several observations are worth noting:
? As expected, all size variables (inventories, sales, forecasts) are highly and positively
correlated. Market shares are also highly correlated with sales and forecasts (as
discussed previously in Section 3.4.3.2), with correlation coefficients of up to 0.54.
While NetIndustrySales (inverted) are also significantly and positively correlated with
the size variables, the correlation coefficients are much smaller in magnitude (about
0.20 to 0.22).
? The correlation coefficient of 0.98 between actual and forecasted sales is an
indication of the good quality of the sales forecasts.
? In line with the hypotheses presented here, the Distress variable is negatively
correlated with the inventory variables. There are, however, no excessive correlations
between the distress measures and other independent variables.
3.4.5.2. Descriptive statistics: Part II
The second data set (Part II) differs from the first data set (Part I) in that it comprises
observations from a cross-section of U.S. manufacturing firms for the year 1997 only.
The firm observations in this data set represent about 10.7% of total sales and 12.4% of
total inventory holdings by all publicly traded U.S. manufacturing firms that are included
in the Compustat database.
A two-sample Hotelling T-squared test is implemented to investigate potential
150
differences between those firms that are included in the data sample and those firms that
are not included in the empirical analyses due to missing data. The Hotelling test
compares these two groups on the following variables: Total inventories, sales, cost of
goods sold, total assets and total debt. The implementation of this test yields a test
statistic of 4.15 which is statistically significant at the one percent level. On average, the
sampled firms tend to be smaller (in terms of inventories, sales, costs, assets, and debt)
than those firms that are not included in the data sample (see Appendix 6). It is therefore
noted that the results of the analyses presented here may not be generalizable to firms of
all size classes.
The composition of the second data set is very similar to that of the first data set: 446 out
of 755 observations are from firms in the computer and electronics, and machinery
industries (see Table 11). The remainder of the data sample comprises observations of
firms from broad variety of U.S. manufacturing industries.
151
NAICS Industry N %
334 Computer and electronics 291 38.5%
333 Machinery 155 20.5%
335 Electrical equipment 48 6.4%
339 Miscellaneous 47 6.2%
336 Transportation equipment 46 6.1%
332 Fabricated metal 34 4.5%
331 Primary metal 28 3.7%
337 Furniture 21 2.8%
313 Textile mills 16 2.1%
327 Nonmetallic mineral 14 1.9%
325 Chemical 14 1.9%
321 Wood 11 1.5%
323 Printing 8 1.1%
322 Paper 6 0.8%
311 Food 5 0.7%
316 Leather 3 0.4%
326 Plastics and rubber 3 0.4%
315 Apparel 3 0.4%
314 Textile products 2 0.3%
Total 755 100%
Table 11: Sample composition (Part II)
Table 12 presents the descriptive statistics of this sample. The conclusions that can be
drawn upon observing these statistics are consistent with what was noted about the first
data set. There is substantial variability in all variables. The inventory variables,
SalesForecasts and DaysPayable (LeadTime), however, have particularly large standard
deviations relative to the means and are log-transformed. It is further noted that the sales
forecasts are, on average, substantially larger than actual sales. This result is driven by a
relatively small set of observations for which the particular forecasting technique
employed
62
here resulted in substantial overpredictions. The log-transformation of the
SalesForecast variable deemphasizes the impact these outliers have on the regression
62
Forecasts were calculated based on prior year sales which were progressed using the average sales
growth rate over the previous three years (see Section 3.4.3.2 for more detail).
152
estimates, such that the inferior quality of the sales forecasts is not a great concern
63
.
Compared to the first data set, this data sample also contains three new variables:
IndCR4, SupplyCR4, BuyCR4. The data in Table 12 indicate that, on average, the four
largest firms in the focal, supplying and buying industries control between 29 and 38
percent of the market.
Variable Mean Std. dev. Min Max N
Inventory Total (million $) 110.9 623.42 0.036 12,102 755
Inventory RawMat (million $) 20.92 51.38 0 728 678
Inventory FinGood (milion $) 39.02 296.35 0 7,347 656
Sales (million $) 835.2 6,161.3 0.44 154,329 755
Sales Forecast (million $) 1412.4 17,899.8 0.02 465,806 753
SalesSurprise 0.51 0.5 0 1 755
Coeff. of Variation of Sales 0.23 0.22 0.005 1.72 755
OrderBacklog/Sales 0.38 1.42 0 36.99 755
Interest Rate 0.16 0.58 0 11.33 755
Days Payable 38.26 41.79 2.67 736 755
Distress -4.80 9.62 -114.4 50.5 755
Distress Dummy 0.18 0.39 0 1 755
Market Share (6 dig. NAICS) 0.05 0.13 0.00002 1 755
IndCR4 37.74 16.37 4.6 94.5 755
SupplyCR4 28.96 6.71 14.8 83.2 755
BuyCR4 37.38 15.08 6.8 86.9 755
LIFO 0.16 0.36 0 1 755
AvgCost 0.08 0.27 0 1 755
Table 12: Pooled descriptive statistics (Part II)
63
The correlation coefficient between (logged) Sales and (logged) SalesForecasts is r = 0.97 (see Table
36).
153
Table 13 presents the split-sample comparison between distressed and non-distressed
firms. In this sample, distressed firms appear to be larger than non-distressed firms and
therefore tend to hold more inventory. At the same time, distressed firms, on average,
have smaller market shares than non-distressed firms. This may be an indication that the
distressed firms tend to be concentrated in some (larger) industry sectors.
The result of a two-sample Hotelling T-squared test suggests that, overall, distressed
firms are statistically significantly different from non-distressed firms. The test statistic is
F = 8.0767 which is statistically significant at the one percent level.
Variable Mean Std. dev. Min Max N Mean Std. dev. Min Max N
Inventory Total (million $) 101.0 442.36 0.044 8,967 617 155.0 1121.07 0.036 12,102 138
Inventory RawMat (million $) 22.26 50.21 0 728 561 14.49 56.45 0 509 117
Inventory FinGood (milion $) 31.41 84.39 0 1,078 544 76.01 694.06 0 7,347 112
Sales (million $) 712.7 2,790.0 1.52 45,800 617 1382.9 13,174.0 0.44 154,329 138
Sales Forecast (million $) 1425.7 18,863.6 0.11 465,806 617 1351.9 12,692.2 0.02 147,672 136
SalesSurprise 0.53 0.5 0 1 617 0.42 0.5 0 1 138
Coeff. of Variation of Sales 0.22 0.21 0.005 1.46 617 0.26 0.27 0.01 1.72 138
OrderBacklog/Sales 0.33 0.48 0 5.86 617 0.59 3.17 0 36.99 138
Interest Rate 0.15 0.44 0 6.66 617 0.22 0.98 0.01 11.33 138
Days Payable 34.36 36.43 2.67 736 617 55.70 57.18 5.78 438 138
Distress -6.38 9.41 -114.4 -1.8 617 2.27 6.98 -1.8 50.5 138
Market Share (6 dig. NAICS) 0.06 0.14 0.00002 1 617 0.03 0.09 0.00003 1 138
IndCR4 38.03 16.20 4.6 94.5 617 36.45 17.09 4.6 88.3 138
SupplyCR4 29.10 6.81 16.1 83.2 617 28.34 6.25 14.8 54.3 138
BuyCR4 37.78 15.15 6.8 86.9 617 35.60 14.68 6.8 83.2 138
LIFO 0.16 0.37 0 1 617 0.12 0.33 0 1 138
AvgCost 0.07 0.26 0 1 617 0.11 0.31 0 1 138
Distressed firms Non-distressed firms
Table 13: Descriptive statistics (Part II) – distressed vs. non-distressed firms
Pairwise correlations are presented in Table 14. Again, all size variables are highly
correlated, but no excessive correlations between independent variables are found. Given
the relatively small sample size, few correlation coefficients are statistically significant.
154
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1 Inventory Total (million $)
2 Inventory RawMat (million $) 0.92
3 Inventory FinGood (million $) 0.87 0.76
4 Sales (million $) 0.94 0.88 0.84
5 Sales Forecast (million $) 0.91 0.86 0.81 0.97
6 Sales Surprise -0.12 -0.04 -0.10 -0.13 -0.09
7 Coeff. of Variation of Sales 0.05 0.03 0.05 0.05 -0.09 -0.09
8 OrderBacklog/Sales -0.07 -0.10 -0.08 -0.07 -0.08 0.01 -0.01
9 Interest Rate -0.08 -0.07 -0.06 -0.10 -0.11 0.01 0.06 0.00
10 Days Payable -0.14 -0.11 -0.12 -0.22 -0.21 0.27 -0.03 0.02 0.04
11 Distress -0.03 -0.06 0.00 -0.05 -0.05 -0.15 0.02 0.02 0.01 0.16
12 Distress Dummy -0.31 -0.32 -0.25 -0.33 -0.31 0.07 -0.08 0.07 0.05 0.24 0.35
13 Market Share (6 dig. NAICS) 0.55 0.48 0.52 0.58 0.55 -0.08 0.09 0.00 -0.05 -0.09 0.03 -0.09
14 Net Industry Sales (inverted) 0.11 0.08 0.15 0.11 0.08 -0.04 0.05 -0.04 -0.06 -0.11 0.04 -0.02 0.58
15 IndCR4 0.11 0.04 0.02 0.11 0.11 -0.02 0.05 0.07 0.04 0.07 -0.06 -0.04 0.13 -0.19
16 SupplyCR4 0.08 0.07 -0.01 0.07 0.07 -0.01 0.02 0.09 0.02 0.03 0.00 -0.04 -0.01 -0.31 0.51
17 BuyCR4 0.15 0.12 0.02 0.12 0.12 0.00 0.03 0.09 0.02 0.04 -0.03 -0.06 0.05 -0.24 0.55 0.41
18 LIFO 0.29 0.25 0.31 0.30 0.29 -0.19 0.06 -0.05 -0.06 -0.20 0.05 -0.04 0.16 0.09 -0.11 -0.04 -0.14
19 Avg. Cost 0.01 -0.01 0.05 0.00 -0.01 -0.01 -0.05 0.01 0.06 0.04 -0.05 0.05 0.02 -0.01 0.04 0.00 0.05 -0.13
(Values in bold are significant at the 5% level)
Table 14: Pairwise correlations (Part II)
155
3.4.6. Methodology
The empirical methodology is discussed in this section. A series of regression analyses
are performed to test the hypotheses developed in this essay. An overview of these
regressions is presented in the following subsection. Both the panel data set (Part I) and
the cross-sectional data set (Part II) present particular econometric challenges that have to
be considered when choosing an empirical estimation procedure. The methodologies for
the analyses of both data sets are discussed in turn in Subsections 3.4.6.2 and 3.4.6.3.
3.4.6.1. Overview of regression analyses
The hypotheses set forth in this essay are tested by means of a series of regression
analyses. Table 15 provides an overview of the regressions that are performed.
Data Part I Data Part II
Dependent
variable
Baseline Split-sample Distressed firms
with interaction
effect
Baseline Split-sample Distressed firms
with interaction
effect
Total
inventory
R1 R2 R3 R10 R11 R12
Raw
materials
inventory
R4 R5 R6 R13 R14 R15
Finished
goods
inventory
R7 R8 R9 R16 R17 R18
Table 15: Overview of regression analyses
156
As described previously, two separate data sets are used for the empirical analyses. Nine
regressions are performed to analyze each data set (R1-R9 and R10-R18). For the
analysis of each data set, three lines of regressions are required to estimate the model for
three different dependent variables: Total inventory, raw materials inventories, and
finished goods inventories. For each dependent variable, a baseline regression using the
full data set is implemented first. In a second step, the data set is split into distressed and
non-distressed firms, and the regression is implemented for both subsamples separately.
The Distress*Power interaction effect is included in the third regression model which is
implemented using the subsample of distressed firms only and designed to test
Hypothesis 12 to Hypothesis 14.
The regression models are further discussed in the following paragraphs. The models
below show TotalInventory as the dependent variable. The same models are also
analyzed with raw materials inventories and finished goods inventories as the dependent
variables.
The first regression (R1) estimates the baseline model shown below. This regression is
performed using the entire data sample (part I). The measures of financial distress
(DistressDummy) and of firm power (IndSalesNet) are of particular interest. It is expected
that, on average, distressed firms hold less inventory than non-distressed firms.
157
(R1) lnTotalInventory
itf
= ?
0
+ ?
1
lnSalesForecast
itf
+ ?
2
SalesSurprise
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
lnLeadTime
itf
+ ?
7
DistressDummy
itf
+ ?
8
lnIndSalesNet
itf
+ ?
9
LIFO
itf
+ ?
10
AvgCost + ?
itf
The second regression (R2) is nearly identical to R1. This regression however, is
performed for distressed and non-distressed firms separately, using the DistressDummy
variable to split the sample into these groups. The continuous Distress variable then
replaces the DistressDummy variable in the model. It is expected that greater levels of
financial distress result in lower inventory levels for distressed firms
64
.
(R2) lnTotalInventory
itf
= ?
0
+ ?
1
lnSalesForecast
itf
+ ?
2
SalesSurprise
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
lnLeadTime
itf
+ ?
7
Distress
itf
+ ?
8
lnIndSalesNet
itf
+ ?
9
LIFO
itf
+ ?
10
AvgCost + ?
itf
The third regression (R3) is similar to R2 for distressed firms, the only difference being
that the Distress*IndSalesNet interaction term is included to test the contention that the
(negative) effect of financial distress on inventories increases with the firm’s power.
(R3) lnTotalInventory
itf
= ?
0
+ ?
1
lnSalesForecast
itf
+ ?
2
SalesSurprise
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
lnLeadTime
itf
+ ?
7
Distress
itf
+ ?
8
lnIndSalesNet
itf
+ ?
9
LIFO
itf
+ ?
10
AvgCost
+ ?
11
Distress
itf
* lnIndSalesNet
itf
+ ?
itf
64
This study focuses on the analysis of financially distressed firms’ inventories. The effect of financial
health on inventories is not investigated here.
158
The regression models used to analyze the second data sample (part II) are similar to
those described above.
Regression 10 (R10) estimates the baseline model which includes the focal industry’s
four-firm concentration ratio, as well as the weighted average concentration ratios of the
supplying and buying industry in addition to the variables included in R1. The new
variables are added to approximate firms’ supply chain power. R10 is performed using
the entire data sample (part II).
(R10) lnTotalInventory
itf
= ?
0
+ ?
1
lnSalesForecast
itf
+ ?
2
SalesSurprise
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
lnLeadTime
itf
+ ?
7
DistressDummy
itf
+ ?
8
lnIndSalesNet
itf
+ ?
9
IndCR4
itf
+ ?
10
SupplyCR4
itf
+ ?
11
BuyCR4
itf
+ ?
12
LIFO
itf
+ ?
13
AvgCost + ?
itf
Regression 11 (R11) is performed for distressed and non-distressed firms separately,
similar to R2. The continuous Distress variable then replaces the DistressDummy variable
in model R10.
(R11) lnTotalInventory
itf
= ?
0
+ ?
1
lnSalesForecast
itf
+ ?
2
SalesSurprise
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
lnLeadTime
itf
+ ?
7
Distress
itf
+ ?
8
lnIndSalesNet
itf
+ ?
9
IndCR4
itf
+ ?
10
SupplyCR4
itf
+ ?
11
BuyCR4
itf
+ ?
12
LIFO
itf
+ ?
13
AvgCost + ?
itf
159
Building on R11, regression 12 (R12) adds the interaction terms between the Distress and
IndSalesNet, IndCR4, SupplyCR4, and BuyCR4 variables, respectively.
(R12) lnTotalInventory
itf
= ?
0
+ ?
1
lnSalesForecast
itf
+ ?
2
SalesSurprise
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
lnLeadTime
itf
+ ?
7
Distress
itf
+ ?
8
lnIndSalesNet
itf
+ ?
9
IndCR4
itf
+ ?
10
SupplyCR4
itf
+ ?
11
BuyCR4
itf
+ ?
12
Distress
itf
* lnIndSalesNet
itf
+ ?
13
Distress
itf
* IndCR4
itf
+ ?
14
Distress
itf
* SupplyCR4
itf
+ ?
15
Distress
itf
* BuyCR4
itf
+ ?
16
LIFO
itf
+ ?
17
AvgCost + ?
itf
3.4.6.2. Empirical methodology: Part I
As discussed in Chapter 2, the OLS assumptions of homoskedasticity and independence
are frequently not met when dealing with cross-sectional time series data (Greene 2003).
Tests for heteroskedasticity and autocorrelation of the error terms are implemented prior
to selecting the appropriate empirical estimation procedure.
The Breusch-Pagan/Cook-Weisberg Lagrange multiplier test (Breusch and Pagan 1979,
Cook and Weisberg 1983) evaluates the correlation between the residuals of an OLS
regression and the dependent variable (e.g. TotalInventory). If no such correlation is
found, the homoskedasticity assumption is valid and OLS regressions can be assumed to
provide efficient and unbiased estimates. The test is implemented after estimating
160
regression R1 (see Table 15) using the OLS procedure. The resulting test statistic is
844.18 which follows a ?
2
distribution. This result is statistically significant at the less
than one percent level and suggests that the magnitude of the residuals varies with the
levels of the dependent variable (heteroskedasticity).
The Wooldridge test for autocorrelation in panel data (Drukker 2003, Wooldridge 2002)
is implemented to determine if the residuals are serially correlated over time. This test is
particularly suitable for panel data sets since it evaluates serial correlations within panels
only. The test statistic is F = 144.745 and is significant at the one percent level. This
suggests the presence of first-order autocorrelation.
A generalized least squares procedure (GLS) is recommended for the analysis of panel
data with heteroskedastic and serially correlated error terms (Greene 2003). As noted in
Section 2.3.4, the GLS procedure can be implemented with the unobserved cross-
sectional and time effects modeled as either random or fixed effects. The appropriate
procedure (fixed effects or random effects) is determined by implementing the Hausman
specification test (Hausman 1978). This test analyzes whether the error terms are
independent of the independent variables. If that is the case, the random effects procedure
is preferred, while the fixed effects procedure should be selected otherwise. The test
produces a ?
2
distributed statistic of W = 845.51 which is significant at the less than one
percent level. The null hypothesis of no correlation is therefore clearly rejected,
suggesting that the fixed effects model should be selected.
161
The STATA software package is used for the empirical analyses. This software lets users
specify the way in which the first-order autocorrelation of the error terms should be
modeled. The default method is to compute the autocorrelation based on the Durbin-
Watson statistic. This method is applied here, although the results are found to be largely
insensitive to the way in which autocorrelation is computed.
3.4.6.3. Empirical methodology: Part II
The second data set used for the empirical analyses contains firm-level observations from
the year 1997. Given that there is only one observation per firm (rather than a time series
of firm-level observations), serial correlation of the error terms is not a concern with this
data set. Heteroskedasticity may, however, be observed. Therefore, the Breusch-
Pagan/Cook-Weisberg Lagrange multiplier test (Breusch and Pagan 1979, Cook and
Weisberg 1983) is implemented after an OLS regression (R10, see Table 15). The test
statistic is 23.81 with a ?
2
distribution. This result is statistically significant at the one
percent level. This indicates that the magnitude of the residuals varies with the levels of
the dependent variable (heteroskedasticity). This constitutes a violation of the OLS
assumption of homoskedasticity.
Robust estimation techniques provide a mechanism to control the heteroskedasticity of
errors. While the coefficient estimates themselves remain unchanged relative to the
standard OLS estimation procedure, the values of standard errors are adjusted for
correlations across observations. The robust regression procedure in STATA uses Huber-
162
White sandwich estimators to compute robust standard errors (White 1980). The
empirical results are presented and discussed in the following section.
3.5. Empirical results and discussion
The empirical analyses are performed for both data sets (Part I and Part II) separately.
The regression results are discussed in Subsections 3.5.1 and 3.5.2, respectively, and the
empirical support for the hypotheses set forth in this paper is evaluated.
3.5.1. Empirical results: Part I
In this section, the results of the analyses of data set Part I are discussed in four
subsections:
? First, the regression results for TotalInventory as the dependent variable are
presented. Specifically, the baseline regression (R1, see Table 15) and the split-
sample regression (R2) results are reported, and the interaction between distress
and power is evaluated for distressed firms (R3).
? Second, the sensitivity of the regression results (R1) with respect to the definition
of the DistressDummy variable and the granularity of industry definitions (6-digit
NAICS versus 4-digit NAICS) is also evaluated.
? Third, the regression results for RawMatInventory (raw materials inventory) as
the dependent variable are discussed (R4-R6, see Table 15).
? Fourth, the regression results for FinGoodInventory (finished goods inventory) as
163
the dependent variable are discussed (R7-R9, see Table 15).
3.5.1.1. Regression results: Total inventory
As discussed previously, the baseline regression is specified as follows:
(R1) lnTotalInventory
itf
= ?
0
+ ?
1
lnSalesForecast
itf
+ ?
2
SalesSurprise
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
lnLeadTime
itf
+ ?
7
DistressDummy
itf
+ ?
8
lnIndSalesNet
itf
+ ?
9
LIFO
itf
+ ?
10
AvgCost + ?
itf
This model is tested using the panel data set described in Section 3.4.5.1 and the
autoregressive linear regression estimation procedure outlined in Section 3.4.6.2. The
empirical estimation results are presented in Table 16.
The model’s F statistic (F = 135.8) is statistically significant at the one percent level, and
the R-squared within statistic is 0.33 indicating that the model explains about one third of
the variability in the dependent variable. The coefficient estimates are discussed below:
? Forecast: Higher expected demand should result in larger order quantities and
larger average cycle stocks. This expectation is confirmed by the positive and
significant coefficient ( ) 0.361 ? = . Specifically, this result suggests that a one
percent increase in expected demand should result in an increase in total inventory
holdings by 0.361%. It is noted that this result is consistent with Ballou’s (1981)
contention that inventories should increase as the square-root of demand.
? SalesSurprise: Greater than expected demand should result in lower end-of-period
inventory holdings. This variable’s coefficient ( ) 0.202 ? = , however, is positive
164
and significant. One explanation may be that firms build up inventory once it
becomes apparent that demand may exceed expectations.
? Coefficient of Variation of Sales: Greater demand variability should result in
larger safety stocks and, thus, greater inventory levels. The coefficient estimate
( ) 0.027 ? = ? , however, is statistically insignificant.
? OrderBacklog/Sales: The standardized value of order backlogs is used as a proxy
for production setup costs. The higher this cost, the higher production quantities
and average cycle stocks should be. The coefficient estimate ( ) 0.007 ? = is
positive as expected although only marginally significant.
? InterestRate: The InterestRate measure is used as a proxy for inventory carrying
costs. Higher carrying costs should equal lower inventory levels. The coefficient
estimate ( ) 0.038 ? = ? has the expected sign but is statistically insignificant.
? DaysPayable: Days payable outstanding is used as a proxy for lead times. The
longer the lead times, the more inventory firms should hold. This expectation is
confirmed by the positive and significant coefficient estimate ( ) 0.120 ? = .
? DistressDummy: As discussed previously, the DistressDummy variable identifies
those firms that have high Distress scores and thus find themselves in situations
of financial distress. The key contention of this research is that distressed firms
will hold less inventory, all else equal (Hypothesis 8). The coefficient estimate is
negative and statistically significant ( ) 0.065 ? = ? . This result suggests that, on
average, distressed firms hold 6.5 percent less inventory than financially healthier
firms. This finding provides support for Hypothesis 8.
165
? IndSalesNet: This variable measures a firm’s power in a market relative to its
competitors. As stated in Hypothesis 9, more powerful firms are expected to hold
less inventory, ceteris paribus. The coefficient estimate ( ) 0.105 ? = ? provides
strong support for this hypothesis.
? LIFO and AvgCost: The coefficient estimates of both the LIFO and AvgCost
variables are not statistically significantly different from 0. The results therefore
suggest that in this particular data sample, differences in inventory accounting
methods did not significantly affect inventory valuations.
Total Inv Coef. P>t
Constant -0.317 0.000
Forecast 0.361 0.000
SalesSurprise 0.202 0.000
Coeff. of Variation -0.027 0.621
OrderBacklog/Sales 0.007 0.095
InterestRate -0.038 0.188
DaysPayable 0.120 0.000
DistressDummy -0.065 0.003
IndSalesNet -0.105 0.000
LIFO -0.005 0.928
AvgCost 0.016 0.803
Number of obs 3,862
F(10,2758) 135.8
Prob > F 0.000
R-sq. within 0.330
R-sq. between 0.806
R-sq. overall 0.781
Table 16: Regression results: Total inventory (R1)
The regression model discussed above is also implemented for both distressed and non-
distressed firms separately (R2, see Table 15), using the DistressDummy variable to split
the sample into these groups. In addition, the moderating effect of power (IndSalesNet)
166
on the Distress-Inventory relationship is evaluated by estimating the corresponding
interaction effect for distressed firms (R3). Table 17 presents these regression results.
Total Inv Coef. P>t Coef. P>t Coef. P>t
Constant -0.121 0.014 -0.689 0.000 -0.691 0.000
Forecast 0.432 0.000 0.191 0.000 0.187 0.000
SalesSurprise 0.197 0.000 0.164 0.000 0.162 0.000
Coeff. of Variation 0.044 0.509 -0.025 0.850 -0.022 0.868
OrderBacklog/Sales 0.077 0.058 0.005 0.387 0.005 0.383
InterestRate -0.026 0.369 -0.269 0.004 -0.268 0.005
DaysPayable 0.201 0.000 0.014 0.706 0.013 0.724
Distress -0.001 0.344 -0.011 0.000 -0.015 0.272
IndSalesNet -0.032 0.009 -0.197 0.000 -0.199 0.000
LIFO 0.002 0.970 0.010 0.954 0.010 0.955
AvgCost 0.076 0.235 0.073 0.688 0.076 0.678
Distress*IndSalesNet 0.000 0.733
Number of obs 2,701 857 857
F 157.0 19.4 17.6
Prob > F 0.000 0.000 0.000
R-sq. within 0.458 0.289 0.289
R-sq. between 0.863 0.627 0.616
R-sq. overall 0.842 0.604 0.593
Distressed firms Non-distressed firms
w/o interaction with interaction
Table 17: Split-sample regression results: Total inventory (R2, R3)
The leftmost column of Table 17 shows the regression results for non-distressed, i.e.
healthy firms. It is noted that the coefficient estimates are generally consistent with the
results for the entire data sample (Table 16). The key difference is that the model shown
in Table 17 contains the (continuous) Distress variable. It is interesting to note that the
level of Distress (or financial health in this case
65
) does not appear to impact non-
distressed firms’ inventory holdings.
65
Note that non-distressed firms will have low or negative Distress scores, indicating financial health.
167
The right part of Table 17 shows the regression results for distressed firms. Given the
smaller number of observations (n = 857), the model fit is lower than for non-distressed
firms (F = 19.4, 17.6; R-squared = 0.289). The coefficient estimates are, however,
generally consistent with those for the entire sample (Table 16) and those for non-
distressed firms (Table 17, left column). A few results of the analysis of distressed firms
merit further discussion:
? The coefficients of the OrderBacklog/Sales and DaysPayable variables are
statistically insignificant for distressed firms.
? The InterestRate variable, on the other hand, has a statistically significant
negative coefficient ( ) 0.269 ? = ? .
? The Distress variable carries a negative and statistically significant coefficient
( ) 0.011 ? = ? in the model without the interaction effect. This suggests that, for
distressed firms, greater levels of distress result in even lower inventory levels.
This finding provides further support for Hypothesis 8
66
.
? The interaction effect between Distress and IndSalesNet is added to the model in
the rightmost column of Table 17. The corresponding coefficient estimate is close
to zero and does not add any explanatory power to the model. Hypothesis 12 is,
thus, not supported.
In summary, the regression model is of at least reasonable quality and most coefficient
estimates have the expected signs. In particular, the results indicate that distressed firms
66
A squared term of the Distress variable was also tested to investigate if the relationship between financial
distress and inventories is non-linear. While the results are not reported here, it is noted that the squared
Distress variable carries a negative and significant coefficient, suggesting that the magnitude of the effect
of distress on prices increases with the severity of financial distress.
168
hold less inventory than financially healthy firms (Hypothesis 8), and that greater levels
of distress equate to lower inventory levels. The contention that power moderates the
distress-inventory relationship (Hypothesis 12) is not supported.
3.5.1.2. Sensitivity analyses
The sensitivity of the regression results with respect to the definition of the
DistressDummy, IndSalesNet and SalesSurprise variables is assessed in this section. In
addition, it is investigated if the results hold if average inventories rather than end-of-year
inventories are used as dependent variables
The DistressDummy variable indicates whether a firm has a Distress score of greater than
-1.81. This cutoff level, initially proposed by Altman (1968), is, of course, somewhat
arbitrary. The effect of alternative cutoff definitions on the estimation results is
investigated by comparing the regression results for three distinct cutoff levels.
Figure 14: Alternative definitions of distressed and non-distressed firms
0 -1.81 -2.80
Financial
distress
Financial
health
(Altman 1968) (median)
Distress
Dummy
DistressMed
Dummy
DistressNegZ
Dummy
0 -1.81 -2.80
Financial
distress
Financial
health
(Altman 1968) (median)
Distress
Dummy
DistressMed
Dummy
DistressNegZ
Dummy
169
Figure 14 illustrates the cutoff levels that are proposed here. The standard cutoff level of
-1.81 is shown in the middle of the graph. As seen in Table 12, this results in 27 percent
of the firms in the full data sample being classified as financially distressed. An equal
split into relatively distressed and relatively healthy firm is obtained by using the median
Distress value (-2.80) as a cutoff level. The resulting indicator variable is named
DistressMedDummy. Conversely, a stricter definition of financial distress is obtained by
moving the cutoff level to the right. The DistressNegZDummy variable identifies all
severely distressed firms with negative Z scores (i.e. positive Distress scores). With this
cutoff level, about twelve percent of all firms are considered distressed. Table 18
juxtaposes the regression results for all three cutoff definitions.
TotalInventory Coef. P>t Coef. P>t Coef. P>t
Constant -0.317 0.000 -0.312 0.000 -0.314 0.000
Forecast 0.361 0.000 0.360 0.000 0.361 0.000
SalesSurprise 0.202 0.000 0.200 0.000 0.203 0.000
Coeff. of Variation -0.027 0.621 -0.024 0.665 -0.028 0.617
OrderBacklog/Sales 0.007 0.095 0.005 0.203 0.007 0.094
InterestRate -0.038 0.188 -0.035 0.223 -0.036 0.207
DaysPayable 0.120 0.000 0.121 0.000 0.119 0.000
DistressDummy -0.065 0.003 -0.150 0.000 -0.033 0.113
IndSalesNet -0.105 0.000 -0.105 0.000 -0.106 0.000
LIFO -0.005 0.928 -0.005 0.928 -0.007 0.903
AvgCost 0.016 0.803 0.020 0.760 0.013 0.842
Number of obs 3,862 3,862 3,862
F 135.8 139.0 134.5
Prob > F 0.000 0.000 0.000
R-sq. within 0.330 0.335 0.328
R-sq. between 0.806 0.809 0.804
R-sq. overall 0.781 0.784 0.778
DistressMedDummy DistressDummy DistressNegZDummy
Table 18: Sensitivity analysis: Distressed vs. non-distressed firms
The coefficient estimates are robust across all three cases and there are only minimal
170
variations in the overall fit of the model. Specifically, the only notable result is that the
DistressMedDummy variable does not yield a statistically significant coefficient. This is
not surprising considering that the DistressMedDummy variable is based on a very broad
definition of financial distress. The alternative variables (DistressDummy and
DistressNegZDummy) both carry negative and statistically significant coefficients. It is
therefore concluded that the results are largely insensitive to variations in the definition
of financial distress. The DistressDummy variable (as defined initially) is retained for the
remainder of the analyses.
A second sensitivity analysis is performed to evaluate how the regression results change
as the definition of industries changes. Specifically, industries may be defined at different
levels of granularity. While a narrow definition based on six-digit NAICS codes is used
as a default, the model is re-estimated using the broader four-digit NAICS definition.
These definitions affect the magnitude and variability of the IndSalesNet variable.
Table 19 presents the regression results for both the six and four-digit NAICS industry
definitions. While the broader four-digit NAICS definition provides slightly better results
in terms of model fit than the six-digit NAICS definition, the results are robust and
consistent across both regressions. The six-digit NAICS industry definition is retained to
ensure consistency with the granularity of industry definitions in the second part of the
data analysis
67
.
67
The analysis of the second data set (Part II) introduces the concept of supply chain power by adding
weighted average industry concentration levels in the buying and supplying industries. To obtain sufficient
variability in these variables, industries must be defined at the full six-digit NAICS level.
171
TotalInventory Coef. P>t Coef. P>t
Constant -0.317 0.000 -0.986 0.000
Forecast 0.361 0.000 0.272 0.000
SalesSurprise 0.202 0.000 0.159 0.000
Coeff. of Variation -0.027 0.621 -0.076 0.161
OrderBacklog/Sales 0.007 0.095 0.006 0.128
InterestRate -0.038 0.188 -0.043 0.126
DaysPayable 0.120 0.000 0.077 0.000
DistressDummy -0.065 0.003 -0.072 0.001
IndSalesNet -0.105 0.000 -0.204 0.000
LIFO -0.005 0.928 -0.015 0.779
AvgCost 0.016 0.803 -0.003 0.965
Number of obs. 3,862 3,862
F 135.8 151.4
Prob > F 0.000 0.000
R-sq. within 0.330 0.354
R-sq. between 0.806 0.743
R-sq. overall 0.781 0.721
6-digit NAICS 4-digit NAICS
Table 19: Sensitivity analysis: Granularity of industry definitions
Following the example of Roumiantsev and Netessine (2007), the SalesSurprise variable
is included in the regression to capture the effect of unexpectedly large sales on
inventories. Specifically, this indicator variable takes on the value “1” if actual sales
exceed forecasted sales. Alternatively, the actual value of the difference between actual
sales and forecasted sales, i.e. the forecast error (ForecastError), can be included in the
regression model as a more finegrained measure of the magnitude of the deviation of
actual sales from forecasted sales. The regression results using the SalesSurprise and
ForecastError variables, respectively, are compared in Table 20. The model with the
ForecastError variable is of poorer quality than the model with the SalesSurprise
variable. The signs of the coefficient estimates, however, are consistent across both
models. The SalesSurprise variable is retained for all subsequent analyses.
172
TotalInventory Coef. P>t Coef. P>t
Constant -0.317 0.000 -0.082 0.134
Forecast 0.361 0.000 0.234 0.000
SalesSurprise / Error 0.202 0.000 0.000 0.000
Coeff. of Variation -0.027 0.621 0.001 0.985
OrderBacklog/Sales 0.007 0.095 0.006 0.197
InterestRate -0.038 0.188 -0.039 0.186
DaysPayable 0.120 0.000 0.132 0.000
DistressDummy -0.065 0.003 -0.104 0.000
IndSalesNet6D -0.105 0.000 -0.154 0.000
LIFO -0.005 0.928 0.015 0.798
AvgCost 0.016 0.803 0.053 0.427
Number of obs 3,862 3,862
F 135.8 97.3
Prob > F 0.000 0.000
R-sq. within 0.330 0.261
R-sq. between 0.806 0.561
R-sq. overall 0.781 0.536
SalesSurprise ForecastError
Table 20: Sensitivity analysis: SalesSurprise vs. ForecastError
As noted in Section 3.4.3.1, end-of-year inventories as reported in firms’ balance sheets
are the dependent variables used in this research. It may be argued that end-of-year
inventory values are biased estimates of true average inventory levels as firms may
reduce inventory levels toward the end of the year in order to improve key financial and
operating performance indicators. This concern is addressed as follows: For each firm in
the dataset, an average annual inventory value is approximated by averaging the firm’s
inventory levels at the end of the first, second, third, and fourth quarters. As shown in
Table 8, the mean total inventory (end-of-year inventory values) in the panel data set
(Part I) is $103.9 million. The mean average total inventory, in contrast, is $107.78
million. A paired two-sample t test indicates that end-of-year total inventories and
173
average total inventories are statistically significantly different (t = 4.86, p = 0.000). This
result, thus, is consistent with the above mentioned contention that end-of-year inventory
values may be biased proxies for average inventories. A regression analysis with average
total inventories as the dependent variable is performed and compared to the results with
end-of-year total inventories as the dependent variable. This comparison in shown in
Table 21 below.
Coef. P>t Coef. P>t
Constant -0.317 0.000 -0.279 0.000
Forecast 0.361 0.000 0.376 0.000
SalesSurprise 0.202 0.000 0.174 0.000
Coeff. of Variation -0.027 0.621 0.007 0.903
OrderBacklog/Sales 0.007 0.095 0.006 0.159
InterestRate -0.038 0.188 -0.029 0.299
DaysPayable 0.120 0.000 0.060 0.000
DistressDummy -0.065 0.003 -0.033 0.135
IndSalesNet6D -0.105 0.000 -0.119 0.000
LIFO -0.005 0.928 0.022 0.681
AvgCost 0.016 0.803 0.021 0.740
Number of obs 3,862 3,862
F 135.8 135.0
Prob > F 0.000 0.000
R-sq. within 0.330 0.329
R-sq. between 0.806 0.799
R-sq. overall 0.781 0.774
Total Inventory Average Inventory
Table 21: Sensitivity analysis: Measurement of total inventories
The coefficient estimates are generally consistent across both the TotalInventory and
AverageInventory regressions. The significance levels, however, are weaker when
average inventories are used as the dependent variable. It is noted that there is no
indication suggesting that the use of end-of-year inventories results in biased estimation
results. As has been done in prior research (e.g. Carpenter et al 1994, Roumiantsev and
Netessine 2007), end-of-year inventories are, therefore, retained for the empirical
174
analyses.
The regression results for raw materials inventories are discussed next.
3.5.1.3. Regression results: Raw materials inventory
The full data sample, comprising both distressed and non-distressed firms is used to
estimate the regression model using raw materials inventory as the dependent variable
(R4, see Table 15). The results are reported in Table 22.
The estimation results are consistent with the previously presented results for total
inventories. The sales variability and production setup cost proxies (Coeff. of Variation
and OrderBacklog/Sales, respectively), as well as the LIFO and AvgCost control
variables are the only variables that have statistically insignificant coefficient estimates.
All other variables carry significant coefficients with the expected signs. The
DistressDummy variable is of particular interest. The negative coefficient ( ) 0.096 ? = ?
indicates that distressed firms tend to hold less inventory than their healthier counterparts
(Hypothesis 8).
175
Raw Mat Inv Coef. P>t
Constant -0.724 0.000
Forecast 0.330 0.000
SalesSurprise 0.164 0.000
Coeff. of Variation 0.013 0.858
OrderBacklog/Sales -0.001 0.778
InterestRate -0.136 0.000
DaysPayable 0.073 0.001
DistressDummy -0.096 0.002
IndSalesNet -0.063 0.001
LIFO -0.004 0.951
AvgCost -0.090 0.331
Number of obs 3,288
F(10,2758) 49.1
Prob > F 0.000
R-sq. within 0.174
R-sq. between 0.709
R-sq. overall 0.675
Table 22: Regression results: Raw materials inventory (R4)
The split-sample regression results for non-distressed and distressed firms are shown in
Table 23. Focusing on the results for non-distressed firms in the leftmost column first, it
is interesting to note that greater levels of financial health (Distress) do not impact
inventory holdings ( ) 0.000 ? = . In addition, the insignificant coefficient of the
IndSalesNet variable indicates that financially sound firms do not leverage their power to
push inventory ownership up or down the supply chain. The results for distressed firms
(without interaction effect), in contrast, suggest that greater levels of financial distress
(Distress) and greater levels of power (IndSalesNet) result in lower raw materials
inventory holdings. These findings provide support for Hypothesis 8 and Hypothesis 9,
respectively. Moreover, the significant and negative coefficient of the interaction effect
between the Distress and IndSalesNet variables ( ) 0.004 ? = ? suggests that the
magnitude of the effect of distress on raw materials inventories increases with the firm’s
176
power. This result is consistent with Hypothesis 12.
Raw Mat Inv Coef. P>t Coef. P>t Coef. P>t
Constant -0.195 0.004 -0.902 0.000 -0.920 0.000
Forecast 0.393 0.000 0.160 0.006 0.139 0.019
SalesSurprise 0.180 0.000 0.082 0.098 0.071 0.154
Coeff. of Variation 0.126 0.176 -0.029 0.872 -0.012 0.948
OrderBacklog/Sales 0.088 0.189 0.001 0.904 0.001 0.866
InterestRate -0.131 0.001 -0.394 0.001 -0.396 0.001
DaysPayable 0.073 0.006 0.094 0.057 0.091 0.063
Distress 0.000 0.947 -0.013 0.000 -0.054 0.026
IndSalesNet 0.015 0.458 -0.096 0.073 -0.108 0.044
LIFO -0.012 0.867 -0.044 0.837 -0.046 0.831
AvgCost -0.018 0.855 -0.047 0.824 -0.027 0.899
Distress*IndSalesNet -0.004 0.088
Number of obs 2,304 731 731
F 41.1 7.6 7.2
Prob > F 0.000 0.000 0.000
R-sq. within 0.207 0.160 0.166
R-sq. between 0.687 0.591 0.517
R-sq. overall 0.658 0.552 0.481
Non-distressed firms Distressed firms
w/o interaction with interaction
Table 23: Split-sample regression results: Raw materials inventory (R5, R6)
The regression results for finished goods inventories are discussed in the following
subsection.
3.5.1.4. Regression results: Finished goods inventory
In this section, the regression model is estimated using finished goods inventories as the
dependent variable. The results for the full data set, consisting of both distressed and non-
distressed firms, are provided in Table 24.
177
Fin Good Inv Coef. P>t
Constant -0.677 0.000
Forecast 0.357 0.000
SalesSurprise 0.164 0.000
Coeff. of Variation -0.271 0.008
OrderBacklog/Sales 0.004 0.561
InterestRate -0.076 0.151
DaysPayable 0.041 0.172
DistressDummy -0.035 0.402
IndSalesNet -0.055 0.025
LIFO -0.056 0.562
AvgCost 0.285 0.025
Number of obs 3,130
F(10,2758) 29.0
Prob > F 0.000
R-sq. within 0.117
R-sq. between 0.749
R-sq. overall 0.711
Table 24: Regression results: Finished goods inventory (R7)
It is noted that the overall quality of the model is markedly lower for finished goods
inventories, than for total and raw materials inventories. The F statistic is 29.0 and the R-
squared within is only 0.117. Many independent variables, including the DistressDummy
variable, carry statistically insignificant coefficient estimates. While financially distressed
firms, on average, appear to hold less total and raw materials inventory this is not found
to be true for finished goods inventories. Hypothesis 8, thus, is not supported for finished
goods inventories.
The regression results for non-distressed and distressed firms (without and with
interaction effect) are shown in Table 25. Surprisingly, the Distress variable carries a
positive and marginally significant coefficient in the regression analysis of non-distressed
firms (leftmost column). For distressed firms, however, the Distress variable carries the
178
expected negative sign ( ) 0.024 ? = ? . The IndSalesNet variable also has a negative and
significant coefficient ( ) 0.200 ? = ? . The Distress*IndSalesNet interaction effect,
however, is not statistically significant (rightmost column).
FinGoodInv Coef. P>t Coef. P>t Coef. P>t
Constant -0.402 0.001 -1.889 0.000 -1.833 0.000
Forecast 0.472 0.000 0.173 0.062 0.186 0.045
SalesSurprise 0.184 0.000 0.160 0.036 0.167 0.029
Coeff. of Variation -0.216 0.088 -0.409 0.127 -0.443 0.099
OrderBacklog/Sales 0.276 0.009 0.001 0.935 0.001 0.953
InterestRate -0.074 0.183 -0.243 0.217 -0.266 0.178
DaysPayable 0.029 0.450 0.077 0.325 0.077 0.324
Distress 0.003 0.076 -0.024 0.001 0.055 0.340
IndSalesNet 0.022 0.376 -0.200 0.011 -0.192 0.016
LIFO -0.076 0.460 -0.148 0.621 -0.146 0.627
AvgCost 0.361 0.009 0.147 0.652 0.145 0.658
Distress*IndSalesNet 0.009 0.17
Number of obs 2,228 661 661
F 31.8 4.9 4.7
Prob > F 0.000 0.000 0.000
R-sq. within 0.174 0.123 0.128
R-sq. between 0.730 0.434 0.440
R-sq. overall 0.701 0.410 0.419
Non-distressed firms Distressed firms
w/o interaction with interaction
Table 25: Split-sample regression results: Finished goods inventory (R8, R9)
The analysis of finished goods inventories, thus, provides only limited support for the
hypotheses set forth in this paper. It generally seems as though the hypothesized
relationships between financial distress, power and inventories are strongest for total and
raw materials inventories. A summary of the empirical results is presented in Section 3.6.
179
3.5.2. Empirical results: Part II
The second data set (Part II) used for the empirical analyses is described in Section
3.4.5.2. This data set comprises observations from 1997 only, but provides more detailed
industry level statistics. Specifically, focal industry, buying industry, and supplying
industry four-firm concentration ratios are added to the model.
The analysis of the second data set (Part II) is also conducted in three parts: Total
inventories are analyzed, followed by raw materials, and finished goods inventories,
respectively. For each of these dependent variables, three regression analyses are
performed. First, the model is estimated using the full data set (comprising both
financially healthy and financially distressed firms). Then, the model is estimated for
healthy and distressed firms, separately. In a third step, the interaction effects between the
Distress and Power variables are added to the model which is then estimated using the
subsample of financially distressed firms only. The reader is referred to Table 15 for an
overview of the regression analyses.
3.5.2.1. Regression results: Total inventory
The basic regression model used to analyze total inventories is shown below (R10, see
Table 15). This model includes the focal industry’s four-firm concentration ratio, as well
as the weighted average concentration ratios of the supplying and buying industries in
addition to the variables included in R1 (see Table 15). The new variables are added to
180
approximate a firm’s supply chain power.
(R10) lnTotalInventory
itf
= ?
0
+ ?
1
lnSalesForecast
itf
+ ?
2
SalesSurprise
itf
+ ?
3
SalesVariability
itf
+ ?
4
SetupCost
itf
+ ?
5
HoldingCost
itf
+ ?
6
lnLeadTime
itf
+ ?
7
DistressDummy
itf
+ ?
8
lnIndSalesNet
itf
+ ?
9
IndCR4
itf
+ ?
10
SupplyCR4
itf
+ ?
11
BuyCR4
itf
+ ?
12
LIFO
itf
+ ?
13
AvgCost + ?
itf
This model is estimated using an OLS regression procedure with robust standard errors.
Specifically, the Huber-White sandwich estimator of standard errors is used to provide
some control for heteroskedasticity.
The regression results for TotalInventory are shown in Table 26. The model explains
about 86 percent of the variability in total inventories, and the model’s F statistic is
215.33 which is statistically significant at the less than one percent level.
While some variables have statistically insignificant or unexpected coefficients
(SalesSurprise, Coefficient of Variation, OrderBacklog/Sales, InteresRate), many
variables have significant coefficients with the expected signs. Specifically, total
inventories are shown to increase with forecasted sales (? = 0.868) and days payable
outstanding (the lead time proxy, ? = 0.247). The coefficient of the DistressDummy
variable is negative (? = -0.182), thus supporting Hypothesis 8 which states that
financially distressed firms should hold less inventory than their healthier counterparts.
181
Total Inv Coef. P>t
Constant -1.982 0.000
Forecast 0.868 0.000
SalesSurprise 0.451 0.000
Coeff. of Variation -0.335 0.168
OrderBacklog/Sales 0.006 0.729
InterestRate 0.036 0.230
DaysPayable 0.247 0.000
DistressDummy -0.182 0.023
IndSalesNet 0.052 0.005
IndCR4 -0.003 0.091
SupplyCR4 0.004 0.349
BuyCR4 0.007 0.002
LIFO 0.185 0.013
AvgCost 0.204 0.107
N 753
F( 13, 739) 215.33
Prob > F 0.000
R-squared 0.863
Table 26: Regression results: Total inventory (R10)
Some of the coefficient estimates of the power variables also have the expected signs:
Greater levels of (focal) industry concentration, suggesting greater firm power, are shown
to be associated with lower inventory levels (? = -0.003, Hypothesis 9). In addition, the
buying industry power (BuyCR4) has a positive and statistically significant coefficient (?
= 0.007). An increase in the buying industry’s concentration level (holding focal industry
concentration levels constant), thus, implies that firms will hold more inventory. This
finding supports Hypothesis 11. The SupplyCR4 variable, however, does not have a
statistically significant coefficient, and the coefficient of the IndSalesNet variable, while
significant, does not have the expected sign.
In summary, the analysis of the second data set (Part II) is generally consistent with the
results from the analysis of the first data set (Part I, see Table 16).
182
Table 27 shows the split-sample regressions (non-distressed and distressed firms) and the
interaction effects between Distress and the Power variables are included in the
regression analysis of distressed firms in the rightmost column.
Despite the relatively small sample sizes all models explain 85 percent of the variability
in the dependent variable. Many of the coefficient estimates, however, are statistically
insignificant which may be a function of the small number of observations, especially in
the case of distressed firms (n = 136).
Focusing on the results for non-distressed firms first (Table 27, leftmost column), it is
noted that the statistically significant coefficient estimates are consistent with the results
shown in Table 26. The only unexpected result is the positive and significant coefficient
of the Distress variable (? = 0.006). The other coefficients, including those of the four-
firm concentration ratio variables are statistically insignificant.
183
Total Inv Coef. P>t Coef. P>t Coef. P>t
Constant -1.909 0.000 -2.780 0.001 -2.715 0.002
Forecast 0.857 0.000 0.829 0.000 0.831 0.000
SalesSurprise 0.384 0.000 0.594 0.000 0.562 0.000
Coeff. of Variation -0.306 0.308 -0.207 0.659 -0.247 0.613
OrderBacklog/Sales 0.165 0.014 0.365 0.002 -0.016 0.129
InterestRate 0.016 0.733 -0.014 0.151 0.071 0.381
DaysPayable 0.263 0.000 0.069 0.368 0.356 0.003
Distress 0.006 0.026 -0.036 0.015 -0.038 0.799
IndSalesNet 0.054 0.008 0.591 0.039 0.008 0.917
IndCR4 -0.001 0.654 0.143 0.710 -0.010 0.114
SupplyCR4 0.006 0.152 0.010 0.897 -0.009 0.587
BuyCR4 0.002 0.283 -0.012 0.042 0.019 0.007
LIFO 0.136 0.042 -0.006 0.685 0.569 0.050
AvgCost 0.182 0.060 0.020 0.004 0.116 0.771
Distress * IndShipValueNet -0.014 0.273
Distress * IndCR4 -0.002 0.066
Distress * SupplyCR4 -0.004 0.450
Distress * BuyCR4 0.001 0.464
Number of obs 617 136 136
F 248.02 105.47 90.11
Prob > F 0.000 0.000 0.000
R-squared 0.859 0.851 0.857
Non-distressed firms Distressed firms
w/o interaction with interaction
Table 27: Split-sample regression results: Total inventory (R11, R12)
The results for distressed firms (second column in Table 27) closely resemble the results
for non-distressed firms. The significance are, however, generally lower due to the small
sample size (n = 136). The Distress variable carries a negative and significant coefficient
(? = -0.036) which is in line with Hypothesis 8. This hypothesis suggests that distressed
firms hold less inventory than healthier firms.
The only power measure with a statistically significant coefficient is BuyCR4. The
coefficient is negative (? = -0.012) which suggests that greater buying industry power
results in lower focal firm inventory holding. This finding is surprising and inconsistent
with Hypothesis 11 and the results for the entire data sample (see Table 26).
184
The only distress-power interaction effect that is significant is that of the IndCR4 variable
(? = -0.002, see rightmost column in Table 27). This result indicates that the negative
effect of (focal) industry concentration—a proxy for a firm’s power—on the firm’s
inventory holdings is greater the more distressed the firm is. This finding lends some
support to Hypothesis 12. There are, however no statistically significant interaction
effects between Distress and the IndSalesNet, BuyCR4, and SupplyCR4 variables.
3.5.2.2. Regression results: Raw materials inventory
The key results for the analysis of raw materials inventories are similar to those for total
inventories. Table 28 presents the regression results for the entire sample of non-
distressed and distressed firms (n = 676).
Again, distressed firms are shown to hold less raw materials inventory than healthy firms
(Distress, ? = -0.206), thus confirming Hypothesis 8. It is also interesting to note that all
power variables have statistically significant coefficients:
? IndSalesNet: Greater levels of firm power, as approximated by the IndSalesNet
variable, are shown to be associated with greater inventory holdings (? = 0.094).
The same unexpected result was found in the analysis of total inventories.
? IndCR4: The negative coefficient of the IndCR4 variable (? = -0.007), in turn, is
consistent Hypothesis 9 and suggests that greater power, as approximated by focal
industry concentration levels, should result in lower inventory holdings.
185
? BuyCR4 and SupplyCR4: Both the BuyCR4 and SupplyCR4 variables have
positive and significant coefficients (? = 0.009 for BuyCR4 and ? = 0.013 for
SupplyCR4). These results support Hypothesis 11 and Hypothesis 10, suggesting
that, while holding focal industry power levels constant, greater buying and
supplying industry power levels result in larger inventory holdings.
Raw Mat Inv Coef. P>t
Constant -2.515 0.000
Forecast 0.786 0.000
SalesSurprise 0.363 0.000
Coeff. of Variation -0.106 0.648
OrderBacklog/Sales -0.027 0.014
InterestRate 0.075 0.000
DaysPayable 0.263 0.000
DistressDummy -0.206 0.034
IndSalesNet 0.094 0.000
IndCR4 -0.007 0.008
SupplyCR4 0.013 0.005
BuyCR4 0.009 0.001
LIFO 0.081 0.384
AvgCost 0.122 0.374
N 676
F( 13, 662) 141.38
Prob > F 0.000
R-squared 0.769
Table 28: Regression results: Raw materials inventory (R13)
The results for non-distressed firms only (see Table 29) are largely consistent with the
results for the entire data set. The analysis of distressed firms (second and third columns
in Table 29), in turn, yields only few statistically significant coefficient estimates.
Specifically, the coefficient of the Distress variable (? = -0.034) is not statistically
significant in these regressions, and the only power variables with significant coefficients
are IndCR4 (? = - 0.017) and BuyCR4 (? = 0.015). The Distress*IndCR4 interaction
186
effect also carries a negative and statistically significant coefficient estimate (? = -0.003).
This finding suggests that financial distress only affects firms’ raw materials inventories
when these firms operate in highly concentrated industries, i.e. when firms possess some
degree of market power. This finding thus is consistent with Hypothesis 12. There is,
however, no support for the contention that the distress-inventory effect increases with
the level of buying industry or supplying industry concentration (Hypothesis 13 and
Hypothesis 14, respectively).
Raw Mat Inv Coef. P>t Coef. P>t Coef. P>t
Constant -1.952 0.000 -6.013 0.000 -5.839 0.000
Forecast 0.759 0.000 0.844 0.000 0.807 0.000
SalesSurprise 0.309 0.000 0.662 0.000 0.595 0.001
Coeff. of Variation 0.056 0.845 -0.260 0.551 -0.260 0.552
OrderBacklog/Sales -0.033 0.805 -0.027 0.001 -0.033 0.000
InterestRate 0.093 0.002 0.052 0.377 0.037 0.558
DaysPayable 0.213 0.006 0.642 0.000 0.597 0.000
Distress 0.003 0.436 -0.034 0.140 -0.109 0.557
IndSalesNet 0.123 0.000 -0.066 0.481 -0.094 0.313
IndCR4 -0.005 0.114 -0.017 0.002 -0.011 0.082
SupplyCR4 0.014 0.006 0.025 0.140 0.010 0.621
BuyCR4 0.007 0.017 0.015 0.028 0.019 0.011
LIFO 0.073 0.445 0.277 0.397 0.310 0.347
AvgCost 0.108 0.425 0.203 0.613 0.160 0.701
Distress * IndShipValueNet -0.006 0.746
Distress * IndCR4 -0.003 0.002
Distress * SupplyCR4 0.004 0.576
Distress * BuyCR4 0.000 0.922
Number of obs 561 115 115
F 87.19 136.42 107.48
Prob > F 0.000 0.000 0.000
R-squared 0.731 0.825 0.842
Non-distressed firms Distressed firms
w/o interaction with interaction
Table 29: Split-sample regression results: Raw materials inventory (R14, R15)
187
3.5.2.3. Regression results: Finished goods inventory
The results for finished goods inventories are shown in Table 30 and in Table 31.
Unfortunately, the estimation results are of generally poorer quality than the results for
total and raw materials inventories. None of the hypotheses set forth in this essay are
empirically supported for finished goods inventories and few variables carry statistically
significant coefficients. The lack of significant findings may be attributable to, among
other factors, the particularly small sample sizes.
Fin Good Inv Coef. P>t
Constant -2.885 0.000
Forecast 0.882 0.000
SalesSurprise 0.464 0.000
Coeff. of Variation -0.504 0.129
OrderBacklog/Sales 0.005 0.926
InterestRate 0.127 0.191
DaysPayable 0.302 0.001
DistressDummy 0.090 0.500
IndSalesNet 0.065 0.044
IndCR4 -0.005 0.209
SupplyCR4 -0.004 0.562
BuyCR4 -0.004 0.352
LIFO 0.465 0.001
AvgCost 0.440 0.052
N 654
F( 13, 640) 81.28
Prob > F 0.000
R-squared 0.701
Table 30: Regression results: Finished goods inventory (R16)
188
Fin Good Inv Coef. P>t Coef. P>t Coef. P>t
Constant -3.231 0.000 -1.578 0.246 -1.393 0.360
Forecast 0.885 0.000 0.847 0.000 0.817 0.000
SalesSurprise 0.445 0.000 0.588 0.028 0.556 0.045
Coeff. of Variation -0.772 0.061 0.270 0.700 0.308 0.656
OrderBacklog/Sales -0.645 0.000 0.067 0.000 0.059 0.000
InterestRate 0.088 0.338 -0.099 0.874 -0.139 0.795
DaysPayable 0.388 0.000 0.120 0.547 0.101 0.628
Distress 0.003 0.419 0.015 0.664 0.135 0.630
IndSalesNet 0.057 0.090 0.098 0.491 0.093 0.551
IndCR4 -0.001 0.723 -0.012 0.197 -0.007 0.514
SupplyCR4 -0.001 0.868 -0.012 0.646 -0.025 0.378
BuyCR4 -0.002 0.586 -0.001 0.896 0.002 0.841
LIFO 0.412 0.003 0.631 0.240 0.595 0.266
AvgCost 0.519 0.013 0.415 0.464 0.430 0.468
Distress * IndShipValueNet -0.006 0.841
Distress * IndCR4 -0.001 0.508
Distress * SupplyCR4 -0.001 0.891
Distress * BuyCR4 -0.003 0.282
Number of obs 544 110 110
F 107.14 15.12 14.7
Prob > F 0.000 0.000 0.000
R-squared 0.717 0.672 0.686
Non-distressed firms Distressed firms
w/o interaction with interaction
Table 31: Split-sample regression results: Finished goods inventory (R17, R18)
3.6. Summary and discussion
This paper develops a comprehensive theoretical perspective of the firm distress-
inventory relationship, drawing on theories and prior research from the economics,
inventory theory, and supply chain management fields. Previously, researchers generally
ignored the role of firm financial distress when investigating inventories. This study
contends that financial distress plays a significant role in inventory management and that
a firm’s power relative to its buyers and suppliers will impact the magnitude of this
distress-inventory effect. Specifically, the hypotheses set forth in this essay contend that
189
? financially distressed firms hold less inventory than healthier firms
(Hypothesis 8),
? more powerful firms hold less inventory than less powerful firms (Hypothesis 9),
? greater power relative to suppliers results in lower inventory holdings
(Hypothesis 10),
? greater power relative to buyers results in lower inventory holdings
(Hypothesis 11),
? the effect of financial distress on inventories increases with the firm’s power
(Hypothesis 12),
? greater power relative to suppliers increases the magnitude of the distress-
inventory effect (Hypothesis 13),
? greater power relative to buyers increases the magnitude of the distress-inventory
effect (Hypothesis 14).
The results of the empirical analyses of total, raw materials and finished goods
inventories are summarized in Table 32 and Table 33. Table 32 shows the hypothesis
testing results obtained from the analyses of both distressed and non-distressed firms.
Table 33, in turn, focuses on the results for distressed firms only.
190
Data Part I
(1998-2004)
Data Part II
(1997)
H
y
p
o
t
h
e
s
i
s
Testing variable(s)
E
x
p
e
c
t
a
t
i
o
n
T R F T R F
8 DistressDummy – – – 0 – – 0
9 IndSalesNet – – – – + + +
9 IndCR4 – – – 0
10 SupplyCR4 + 0 + 0
11 BuyCR4 +
+ + 0
T = Total inventory, R = Raw materials inventory, F = Finished goods inventory
0 = statistically insignificant result
Table 32: Summary of results for entire data set
Focusing on the results for the entire data sets (both Part I and Part II, see Table 32) first,
it is evident that there is strong support for the contention that distressed firms, on
average, hold less inventory than financially healthy firms (Hypothesis 8). This finding,
however, is not confirmed for finished goods inventories. This is not surprising since raw
materials inventories can easily be reduced by consuming extant stock without reordering
further supplies.
The hypothesis that greater levels of power should be associated with lower inventory
levels (Hypothesis 9) also finds some empirical support. As shown in Table 32, the
IndSalesNet variable carries the expected negative coefficients in the analysis of the panel
data set (Part I), thus indicating that inventories decrease as firm power (as measured by
the IndSalesNet variable) increases. The analysis of 1997 data (Part II), in turn,
consistently yields (unexpected) positive coefficients for the IndSalesNet variable. At the
191
same time, however, the industry concentration variable (IndCR4), carries negative
coefficients (for total and raw materials inventories). While this latter finding is
consistent with Hypothesis 9, the positive coefficients of the IndSalesNet variable are not
consistent with Hypothesis 9. The multicollinearity between the power variables
(IndSalesNet, IndCR4, SupplyCR4, BuyCR4) may partly explain this contradictory result.
Hypothesis 10 and Hypothesis 11 suggest that greater levels of power over suppliers and
buyers, respectively, should result in lower inventory holdings. Table 32 shows positive
coefficients for SupplyCR4 (raw materials inventory only) and BuyCR4 (total and raw
materials inventory). These results imply that greater supplying and buying industry
concentration levels—i.e. lower focal firm power when focal industry concentration
levels are held constant—equate to greater firm inventory holdings. This is, to the best of
the author’s knowledge, the first study to present empirical evidence for the contention
that inter-firm power balances in the supply chain affect the location and ownership of
inventories in supply chains. The results also indicate that power levels affect raw
materials inventories to a much greater extent than finished goods inventories which do
not appear to be impacted by supply chain power.
The results of the analysis of distressed firms only are summarized in Table 33. The
negative coefficients of the Distress variable further support Hypothesis 8. This result
suggests that the magnitude of financial distress impacts the magnitude of the distressed
firm’s inventory reductions.
192
Data Part I
(1998-2004)
Data Part II
(1997)
H
y
p
o
t
h
e
s
i
s
Testing variable(s)
E
x
p
e
c
t
a
t
i
o
n
T R F T R F
8 Distress – – – – – 0 0
9 IndSalesNet – – – – + 0 0
9 IndCR4 – 0 – 0
10 SupplyCR4 + 0 0 0
11 BuyCR4 +
– + 0
12 Distress*IndSalesNet – 0 – 0 0 0 0
12 Distress*IndCR4 – – – 0
13 Distress*SupplyCR4 + 0 0 0
14 Distress*BuyCR4 +
0 0 0
T = Total inventory, R = Raw materials inventory, F = Finished goods inventory
0 = statistically insignificant result
Table 33: Summary of results for distressed firms
The results for the power-inventory hypothesis (Hypothesis 9) are mixed. In the analysis
of the panel data set (Part I), the IndSalesNet variable carries negative coefficients as
expected, suggesting that more powerful distressed firms tend to hold less inventory. As
seen in Table 32, however, the analysis of the second data set yields unexpected (or
insignificant) coefficient estimates for the IndSalesNet variable. The four-firm
concentration ratio (IndCR4), an alternative proxy for power, is shown to significantly
impact distressed firms’ inventory holdings only in the case of raw materials inventories.
Specifically, distressed firms in more concentrated industries are found to hold less raw
materials inventory, all else equal.
193
The supplying and buying industry concentration variables (SupplyCR4 and BuyCR4)
mostly carry insignificant or unexpected coefficients. Only the BuyCR4 variable has a
positive and significant coefficient in the raw materials inventory regression. This result
suggests that distressed firms facing more powerful buyers may be forced to hold greater
raw materials inventories and provides some support for Hypothesis 11.
There is, finally, only scant evidence that distressed firms reduce inventories to a greater
extent when they are more powerful. Only in three instances did the interaction effects
between Distress and the power variables have the expected negative coefficient
estimates. In the first part of the data analysis (Part I), the effect of financial distress on
raw materials inventories is shown to increase with the firm’s power (IndSalesNet). The
same result is obtained in the second part of the data analysis (Part II) when power is
approximated with the industry concentration ratio (IndCR4). These findings provide
some evidence in support of Hypothesis 12. There is, however, no support for the
contention that the distress-inventory effect depends on the levels of supplying and
buying industry power (Hypothesis 13 and Hypothesis 14).
In summary, many of the hypotheses set forth in this study are empirically supported. It is
shown that a firm’s financial condition significantly impacts a firm’s inventory decisions.
Moreover, it is shown that power balances in supply chains may impact the distribution
of inventory ownership in supply chains. At the same time, the data provide only limited
evidence for the contention that power moderates the distress-inventory relationship.
194
This research contributes to the extant literature on multiple accounts: Different
theoretical perspectives are synthesized to investigate the financial distress-inventory
relationship. Specifically, insights from inventory theory and supply chain management
research are used to improve upon the specification of empirical estimation models
presented in prior economics research. Novel proxies for variables such as order and
holding costs are proposed to overcome measurement problems that have previously
hindered empirical inventory research.
The analyses presented in this essay not only refine the extant knowledge of the distress-
inventory relationship but also provide new insights on inventory management issues in a
supply chain context. Specifically, this is, to the best of the author’s knowledge, the first
study to empirically explore the role of inter-firm power in inventory management. In
addition, the moderating role of power in the financial distress-inventory relationship is
investigated.
Understanding the effect of a firm’s financial condition on its inventory decisions may
also have important managerial implications in terms of supplier selection, for example.
Managers should be aware of how a supply chain partner’s distress and power may affect
inventory ownership in the supply chain. While this research does not evaluate how
financial distress and power imbalances in supply chains affect overall supply chain
performance, it is conceivable that the shifting of inventory ownership from the
distressed firm to suppliers and buyers may reduce a supply chain’s effectiveness in
terms of, for example, service levels and responsiveness. The investigation of these
195
questions is left for future research.
This research investigates if and how financial distress affects firm inventories and it is
shown that distressed firms tend to reduce inventory holdings. Future research may also
investigate if and when cutting inventories is a viable turnaround strategy.
As noted previously, this study adds to the small, emerging body of empirical inventory
research. While efforts have been made to overcome the difficulties of data collection and
variable measurement that are associated with doing research in this field, the work
presented here must be considered exploratory. Secondary accounting data from public
firms only were used for the empirical analyses. The generalizability of the results to the
entire population of manufacturing firms, both public and private, can not be ascertained.
In addition, buyer and supplier power levels could only be approximated using rather
crude measures such as buying industry and supplying industry concentration ratios. The
computation of these ratios relies on the Input-Output Tables and industry concentration
data published by the Bureau of Economic Analysis. As noted previously, the omission
of international firms in the construction of these data may lead to a misrepresentation of
inter-industry power constellations. Moreover, these industry power levels are only rough
approximations of firm power levels. Future research may use qualitative methods and
different data collection techniques, such as dyadic surveys, for example, to further
investigate how power affects supply chain inventories and how it moderates the distress-
inventory relationship.
196
4. Firm decision making under financial distress: Summary and outlook
The structure-conduct-performance (SCP) paradigm is a theoretical framework that is
widely used in the industrial organization and strategic management literatures. The basic
tenet of the SCP paradigm is that the structure of markets influences firms’ conduct, and
the latter then is a determinant of firm and market performance. In addition, it is also
recognized that feedback mechanisms may exist within this framework (Waldman and
Jensen 2001, see also Figure 1). The performance observed in a market, for example, may
attract new entrants, thus changing the market structure. Similarly, firms may change
their conduct in the light of poor past performance. This dissertation is concerned with
this particular feedback mechanism: How does a firm’s financial distress affect its
conduct in terms of sales prices and inventories. While the potential existence of such
relationships has been recognized previously, the author is unaware of any study that has
systematically investigated the nature of these causal links. This dissertation addresses
this gap in the literature by investigating the following two research questions:
? Does financial distress have an impact on prices and inventories, after controlling
for other relevant parameters?
? If so, how can these effects be characterized, i.e. what factors influence the
magnitude of the distress-price and distress-inventory relationships?
These questions are investigated through analyses of prices in the U.S. airline industry
and inventories in U.S. manufacturing industries. Upon reviewing the literature, two sets
197
of hypotheses relating financial distress to prices and inventories, respectively, are
formulated. These hypotheses reconcile the conflicts revolving around prior
conceptualizations of the distress-price, and distress-inventory links. More precisely, it is
suggested that firm-specific and structural contingencies moderate these relationships. As
a consequence, it is implied that financial distress may have a strong influence on prices
and inventories in some instances, but not in others.
Large-scale empirical analyses are conducted to test the hypotheses set forth in this
research. Data from the U.S. airline industry are used to investigate how financial distress
affects prices. The results present substantial evidence in support of the hypotheses.
Financial distress is found to be negatively related to air fares, with the magnitude of this
relationship depending on the distressed firm’s operating costs, market shares, and size.
In addition, the degree of market concentration and the competitors’ financial situations
are shown to impact the distress-price relationship.
As to the effect of financial distress on inventories, data from the U.S. manufacturing
industry are used for the empirical tests. It is shown that greater degrees of financial
distress will result in lower inventory levels, ceteris paribus. In some instances, this
effect is found to be stronger the greater the distressed firm’s power over its buyers and
suppliers.
Both the price and inventory studies thus suggest the following:
? Firm financial distress is an important determinant of a firm’s actions.
198
? The nature of the distress-conduct relationship is further impacted by market
structural and firm characteristics. Specifically such factors impact the occurrence
and magnitude of the effect of distress on firm conduct parameters.
This research thus helps refine and enhance researchers’ understanding of the relationship
between structure, conduct and performance. While only one particular feedback
mechanism within the structure-conduct-performance paradigm—the effect of distress on
firm conduct—is investigated here, it is expected that there exist further, previously
unexplored links between structural, conduct, and performance parameters. The analyses
of such relationships are suggested for future research.
Managers may benefit from this work through an enhanced understanding of how firms’
financial conditions may impact (competing) firms’ behavior. Competitors of distressed
firms, for example, may be able to better anticipate distressed firms’ competitive moves,
and as a consequence, may be in a better position to implement preemptive measures or
respond to distressed firms’ actions. Moreover, the findings of this work may be of
interest to cooperation partners of distressed firms. Specifically, managers may want to
understand how a distressed firm’s actions may ultimately impact the cooperating firm, in
terms of service levels, costs, or required inventory holdings, for example. While the
findings presented here do not provide direct evidence for the implications of a firm’s
distress on its cooperation partners, there are some indications that a distressed firm’s
supply chain partners will be affected by the changes in the distressed firm’s conduct. It
is suggested that future research further explore these issues.
199
This dissertation research, thus, enhances researchers’ and managers’ understanding of
how firm financial distress affects prices and inventories. Following these descriptive
causal analyses, a normative approach to the research question at hand is suggested for
future research. Specifically, the following questions could be addressed:
? Is the cutting of prices or the reduction of inventories a viable turnaround strategy,
i.e. do distressed firms that lower prices or reduce inventories exhibit greater
turnaround performance?
? In what specific instances is price or inventory cutting advisable? Are there
certain organizational or situational characteristics that influence the extent to
which lower prices or inventories result in distressed firms’ performance
improvements?
? How does distress affect firm and supply chain operating performance? Are there
any effects in customer service levels or purchasing lead times, for example?
These and more questions may be of great interest to both the academic and practitioner
communities. While these issues are not within the scope of this dissertation, the work
presented here provides a solid basis for further investigations of the managerial
implications and consequences of firm financial distress.
This dissertation empirically investigates the relationship between financial distress and
select firm conduct parameters using secondary data from the U.S. airline and
manufacturing industries. The use of secondary data is desirable in that advanced
statistical methods can be utilized to analyze large data sets and obtain robust and
200
generalizable estimation results. At the same time, however, the use of secondary data
often requires researchers to approximate variables for which no direct measures are
available or to omit explanatory factors from empirical models altogether should data not
be available. In this dissertation, variables such as order costs and sourcing lead times, for
example, could not be measured directly but could only be approximated. In addition,
some data sources present inherent deficiencies and limitations. The data from the Input-
Output tables, which are used to construct industrial supply chains for the analyses of the
distress-inventory relationship, for example, do not include information on trade flows
involving foreign buyers and suppliers. While a research design based on the analysis of
secondary data is deemed suitable for an initial study of the distress-conduct link, future
research could employ qualitative methods such as case studies, for example, to gain an
in-depth understanding of the managerial decision processes that are triggered by the
deterioration of a firm’s financial condition. At the same time, insights could be gained
into when and why specific turnaround strategies and actions result in performance
improvements.
201
Appendix 1
Table 34 below presents the residuals of the regression of airlines’ operating expenses per
available seat-mile (ASM) on average stage length. This regression was performed to
evaluate U.S. carriers’ operating costs after controlling for differences in the airlines’
average stage length. Negative residuals indicate relative cost advantages, while positive
residuals suggest relative cost disadvantages. Based on the results displayed in Table 34,
twelve carriers were identified as low-cost carriers (LCC). While the cut-off between
low-cost and high-cost carriers (HCC) is arbitrary, it is noted that there is a sizeable
difference in the magnitude of the residuals between the highest-cost LCC (Valujet
Airlines: -0.142), and the lowest-cost HCC (Carnival Airlines: -0.105). In the empirical
analyses, the top twelve airlines in Table 34 are, therefore, identified as LCCs.
202
Carrier code Carrier name Ranked residuals LCC indicator
WN Southwest Airlines, Co. -0.371 1
QQ Reno Air, Inc. -0.303 1
SY Sun Country Airlines -0.268 1
NK Spirit Air Lines -0.237 1
B6 Jetblue Airways -0.231 1
W7 Western Pacific Airlines -0.223 1
FL Airtran Airways Corporation -0.216 1
TZ American Trans Air, Inc. -0.214 1
BE Braniff Int'l Airlines, Inc -0.172 1
HP America West Airlines, Inc. -0.157 1
F9 Frontier Airlines, Inc. -0.142 1
J7 Valujet Airlines, Inc. -0.142 1
KW Carnival Air Lines, Inc. -0.105 0
AS Alaska Airlines, Inc. -0.099 0
NJ Vanguard Airlines, Inc. -0.081 0
KP Kiwi International -0.073 0
N7 National Airlines -0.051 0
TW Trans World Airlines, Inc. -0.039 0
XJ Mesaba Airlines -0.036 0
NW Northwest Airlines, Inc. -0.022 0
BF Markair, Inc. -0.010 0
WV Air South, Inc. -0.002 0
HQ Business Express -0.001 0
DL Delta Air Lines, Inc. 0.012 0
CO Continental Air Lines, Inc. 0.018 0
EV Atlantic Southeast Airlines 0.031 0
JI Midway Airlines, Inc. 0.086 0
YV Mesa Airlines, Inc. 0.090 0
ZN Key Airlines, Inc. 0.090 0
OE Westair Airlines 0.093 0
AQ Aloha Airlines, Inc. 0.112 0
RU Continental Express Airline 0.125 0
FF Tower Air, Inc. 0.132 0
AA American Airlines, Inc. 0.140 0
ZW Air Wisconsin Airlines Corp 0.166 0
UA United Air Lines, Inc. 0.170 0
US US Airways, Inc. 0.171 0
YX Midwest Express Airlines 0.179 0
HA Hawaiian Airlines, Inc. 0.195 0
TB USAir Shuttle 0.263 0
PA Pan American World Airways 1.243 0
Table 34: Ranked residuals of regression of OpEx/ASM on avg. stage length (n=41)
203
Appendix 2
Table 35 presents the OLS regression estimates of the empirical model presented in
Chapter 2 of this dissertation. These basic regression results are used solely to investigate
the presence of heteroskedasticity. This is done by means of the Breusch-Pagan/Cook-
Weisberg Lagrange multiplier test (Breusch and Pagan 1979, Cook and Weisberg 1983).
The test result suggests the presence of heteroskedasticity and motivates the choice of a
generalized least squares procedure for the empirical analyses.
Source SS df MS Number of obs 23039
Model 3522.49 20 176.12 F( 21, 23017) 2754.5
Residual 1471.77 23018 0.06 Prob > F 0.000
Total 4994.26 23038 0.22 R-squared 0.705
Adj R-squared 0.705
Root MSE 0.253
Fare Coefficient Std. error P>|t|
Constant 1.473 0.171 0.000
AirlinePass -0.048 0.002 0.000
Distance 0.713 0.048 0.000
DistanceSquared -0.023 0.004 0.000
SlotRoute 0.134 0.005 0.000
RouteHHI -0.005 0.006 0.416
MaxAirportHHI 0.107 0.005 0.000
RouteShare 0.001 0.000 0.000
MaxAirportShare 0.004 0.000 0.000
LCCCompForHCC -0.148 0.004 0.000
LCCCompForLCC -0.103 0.008 0.000
AltRouteLCC1M -0.016 0.004 0.000
Circuity -0.162 0.024 0.000
Distress 0.005 0.001 0.000
Loadfactor -0.004 0.000 0.000
AirlineCost 0.462 0.013 0.000
Size 0.034 0.002 0.000
Quarter 2 -0.086 0.005 0.000
Quarter 3 -0.120 0.006 0.000
Quarter 4 -0.060 0.005 0.000
2002 -0.260 0.006 0.000
Table 35: OLS regression estimates (n = 23,039)
204
Appendix 3
The regression results shown in Table 36 and Table 37 are used to evaluate the benefit of
adding fixed effects to the regression model. This benefit is measured by means of an F
statistic as proposed by (Greene 2003). The test returns a statistically significant F value,
suggesting that a fixed effects model should be used for the empirical analyses.
Source SS df MS Number of obs 23039
Model 1878.66 16 117.42 F( 17, 23021) 1582.87
Residual 3115.60 23022 0.14 Prob > F 0.000
Total 4994.26 23038 0.22 R-squared 0.3762
Adj R-squared 0.3757
Root MSE 0.3679
Fare Coefficient Std. error P>|t|
AirlinePass (fitted) 0.272 0.012 0.000
Distance -0.033 0.074 0.654
DistanceSquared 0.043 0.006 0.000
SlotRoute 0.051 0.008 0.000
RouteHHI 0.031 0.008 0.000
MaxAirportHHI 0.254 0.008 0.000
RouteShare -0.006 0.000 0.000
MaxAirportShare 0.000 0.000 0.103
LCCCompForHCC -0.223 0.006 0.000
LCCCompForLCC -0.158 0.011 0.000
AltRouteLCC1M -0.136 0.007 0.000
Circuity 1.136 0.057 0.000
Distress -0.001 0.001 0.277
Loadfactor -0.022 0.001 0.000
AirlineCost 0.960 0.016 0.000
Size 0.030 0.003 0.000
Constant 2.536 0.242 0.000
Table 36: 2SLS regression estimates without fixed effects (n = 23,039)
205
The Tables in Appendix 4 and Appendix 5 provide further details of the empirical
estimation results.
Table 37 (Appendix 4) presents the second-stage estimation results of the regression of air
fares on the set of independent variables as specified in Sections 2.3.2.2 and 2.3.2.3. In
addition, air carrier fixed effects are included in this regression to evaluate the contribution
of these fixed firm effects to the explanatory power of the model. This analysis is needed to
determine the appropriate econometric estimation technique as discussed in Section 2.3.4.
Table 38 (Appendix 5) presents the first-stage estimation results for all five empirical
models. This table, thus, is an extension of the baseline first-stage estimation results shown
in Table 3. It is noted that the estimation results are generally consistent across all five
models such that the discussion of the baseline first-stage results (see Section 2.4.1) also
apply to the results shown in Table 38.
206
Appendix 4
Source SS df MS Number of obs 23039
Model 2558.23 50 51.16 F( 51, 22987) 715.34
Residual 2436.04 22988 0.11 Prob > F 0.000
Total 4994.26 23038 0.22 R-squared 0.5122
Adj R-squared 0.5112
Root MSE 0.3255
Fare Coefficient Std. error P>|t|
AirlinePass (fitted) 0.247 0.011 0.000
Distance -0.033 0.066 0.618
DistanceSquared 0.040 0.005 0.000
SlotRoute 0.034 0.007 0.000
RouteHHI 0.069 0.008 0.000
MaxAirportHHI 0.192 0.007 0.000
RouteShare -0.006 0.000 0.000
MaxAirportShare 0.000 0.000 0.353
LCCCompForHCC -0.194 0.006 0.000
LCCCompForLCC -0.106 0.010 0.000
AltRouteLCC1M -0.118 0.006 0.000
Circuity 0.960 0.050 0.000
Distress -0.040 0.003 0.000
Loadfactor -0.022 0.001 0.000
AirlineCost 0.300 0.041 0.000
Size 0.090 0.017 0.000
Constant 0.527 0.376 0.161
Quarter2 -0.045 0.007 0.000
Quarter3 -0.048 0.009 0.000
Quarter4 -0.048 0.006 0.000
2002 -0.301 0.020 0.000
aq 0.097 0.143 0.499
as 0.002 0.050 0.961
b6 0.217 0.074 0.003
be 0.414 0.131 0.002
co 0.243 0.024 0.000
dl 0.060 0.010 0.000
f9 0.007 0.084 0.929
ff 0.190 0.182 0.295
fl 0.048 0.075 0.525
hp 0.172 0.050 0.001
hq 0.006 0.136 0.965
ji 0.794 0.121 0.000
kp -0.074 0.206 0.721
kw 0.656 0.165 0.000
n7 0.629 0.104 0.000
nj 1.345 0.116 0.000
nk 0.098 0.094 0.295
nw 0.133 0.015 0.000
oe 0.287 0.133 0.030
qq 0.287 0.146 0.048
sy -0.012 0.152 0.935
tb -0.479 0.140 0.001
tw 0.300 0.032 0.000
tz 0.162 0.068 0.017
ua 0.184 0.009 0.000
us 0.074 0.019 0.000
wn -0.393 0.037 0.000
yx 0.400 0.091 0.000
zn 3.164 0.374 0.000
zw 0.191 0.163 0.242
Table 37: 2SLS regression estimates with fixed effects (n = 23,039)
207
Appendix 5
First-stage G2SLS regression Number of obs. 23039 Obs. per group: min. 1
(fixed effects) Number of groups 4508 avg. 5.1
max. 8
Dependent variable:
AirlinePass Coefficient P>|t| Coefficient P>|t| Coefficient P>|t| Coefficient P>|t| Coefficient P>|t|
Constant 444.458 0.000 440.911 0.000 446.816 0.000 454.492 0.000 477.332 0.000
Distance -136.296 0.000 -134.880 0.000 -136.560 0.000 -139.647 0.000 -145.605 0.000
DistanceSquared 10.181 0.000 10.067 0.000 10.186 0.000 10.444 0.000 10.821 0.000
SlotRoute 0.099 0.001 0.101 0.001 0.103 0.000 0.104 0.000 0.085 0.004
RouteHHI -0.412 0.000 -0.411 0.000 -0.411 0.000 -0.420 0.000 -0.421 0.000
MaxAirportHHI -0.367 0.000 -0.373 0.000 -0.375 0.000 -0.366 0.000 -0.375 0.000
RouteShare 0.027 0.000 0.027 0.000 0.027 0.000 0.027 0.000 0.027 0.000
MaxAirportShare 0.001 0.007 0.001 0.005 0.001 0.010 0.001 0.004 0.001 0.013
LCCCompForHCC 0.227 0.000 0.229 0.000 0.229 0.000 0.223 0.000 0.221 0.000
LCCCompForLCC 0.238 0.000 0.226 0.000 0.224 0.000 0.236 0.000 0.249 0.000
AltRouteLCC1M -0.022 0.061 -0.021 0.068 -0.021 0.065 -0.022 0.055 -0.019 0.094
Circuity -2.326 0.000 -2.327 0.000 -2.326 0.000 -2.344 0.000 -2.335 0.000
Distress 0.005 0.190 -0.138 0.038
Chpt11Ops -0.015 0.141
DistressDiff 0.018 0.000
Pre4Chpt11 0.018 0.107
Post4Chpt11 -0.046 0.001
Loadfactor 0.016 0.000 0.015 0.000 0.015 0.000 0.015 0.000 0.017 0.000
AirlineCost -0.095 0.023 -0.097 0.021 -0.089 0.034 -0.083 0.053 -0.091 0.029
Size 0.028 0.150 0.008 0.666 0.015 0.370 0.042 0.059 0.065 0.000
Quarter 2 0.048 0.000 0.051 0.000 0.053 0.000 0.051 0.000 0.044 0.000
Quarter 3 0.074 0.000 0.079 0.000 0.079 0.000 0.076 0.000 0.062 0.000
Quarter 4 0.047 0.000 0.051 0.000 0.053 0.000 0.049 0.000 0.044 0.000
2002 -0.055 0.315 -0.036 0.509 -0.038 0.485 -0.057 0.306 -0.108 0.046
Population 0.579 0.000 0.567 0.000 0.568 0.000 0.588 0.000 0.577 0.000
Income -0.034 0.785 -0.035 0.780 -0.048 0.702 -0.033 0.791 -0.014 0.912
AirlineCost*Distress -0.025 0.020
Size*Distress -0.002 0.458
RouteShare*Distress 0.000 0.185
RouteHHI*Distress 0.013 0.033
F 1038.3 1038.3 992.3 873.7 1051.3
Prob > F 0.000 0.000 0.000 0.000 0.000
R-squared: within 0.541 0.541 0.541 0.541 0.544
between 0.006 0.007 0.008 0.005 0.009
overall 0.006 0.007 0.007 0.004 0.010
1 3 4 5 2
Table 38: First-stage G2SLS regression estimates (n = 23,039)
208
Appendix 6
Table 39 presents the means of select variables for those firms that are included in the
statistical analyses and those firms for which data are available in the Compustat database
but which have been deleted from the data sample due to missing data on one or more
variables. As discussed in Sections 3.4.5.1 and 3.4.5.2, the sampled firms, on average, tend
to be smaller than those firms firms that are not included in the data sample. A Hotelling T-
squared test confirms that the two groups are statistically significantly different in both
time periods studied (1997 and 1998-2004).
Variable Sample mean Compustat Sample mean Compustat
Total Inventory (million $) 110.9 190.5 103.9 186.2
Sales (million $) 835.2 1698.7 835.6 1788.2
COGS (million $) 626.0 1162.5 537.4 1217.0
Total Debt (million $) 148.0 488.4 158.1 574.9
Total Assets (million $) 760.1 1792.9 733.8 2023.6
1997 1998-2004
Table 39: Mean comparisons between sampled firms and Compustat population
209
Bibliography
Abdel-Malek, Layek L, and Roberto Montanari. "On the Multi-Product Newsboy Problem
with Two Constraints." Computers & Operations Research 32, no. 8 (2005): 2095.
Aiken, Leona S, and Stephen G West. Multiple Regression: Testing and Interpreting
Interactions. Newbury Park, CA: Sage Publications, 1991.
Alstrom, Poul. "Numerical Computation of Inventory Policies, Based on the Eoq/Ox Value
for Order-Point Systems." International Journal of Production Economics 71, no.
1-3 (2001): 235.
Altman, Edward. Bankruptcy, Credit Risk, and High Yield Junk Bonds. Malden, MA:
Blackwell Publishers, 2002.
Altman, Edward I. "Financial Ratios, Discriminant Analysis and the Prediction of
Corporate Bankruptcy." The Journal of Finance 23, no. 4 (1968): 589-609.
———. "A Further Empirical Investigation of the Bankruptcy Cost Question." The Journal
of Finance 39, no. 4 (1984): 1067.
Amihud, Yakov, and Haim Mendelson. "Inventory Behaviour and Market Power: An
Empirical Investigation." International Journal of Industrial Organization 7, no. 2
(1989): 269.
Anonymous. "American Bankruptcy: The Uses and Abuses of Chapter 11." The Economist
310, no. 7594 (1989): 72.
———. "America's Airlines, Flying on Empty." Economist.com / Global Agenda (2005):
1.
Arogyaswamy, Kamala, Vincent L III Barker, and Masoud Yasai-Ardekani. "Firm
Turnarounds: An Integrative Two-Stage Model." The Journal of Management
Studies 32, no. 4 (1995): 493.
Bain, Joe S. Barriers to New Competition. Cambridge, MA: Harvard University Press,
1956.
Baker, Samuel A. "Risk, Leverage and Profitability - an Industry Analysis." The Review of
Economics and Statistics 55, no. 4 (1973): 503.
Ballou, Ronald H. "Estimating and Auditing Aggregate Inventory Levels at Multiple
Stocking Points." Journal of Operations Management 1, no. 3 (1981): 143.
210
Barla, Philippe, and Bonchun Koo. "Bankruptcy Protection and Pricing Strategies in the Us
Airline Industry." Transportation Research. Part E, Logistics & Transportation
Review 35 (1999): 101.
Betker, Brian L. "The Administrative Costs of Debt Restructurings: Some Recent
Evidence." Financial Management 26, no. 4 (1997): 56.
Blazenko, George W, and Kirk Vandezande. "Corporate Holding of Finished Goods
Inventories." Journal of Economics and Business 55, no. 3 (2003): 255.
Borenstein, Severin. "Hubs and High Fares: Dominance and Market Power in the U.S."
The Rand Journal of Economics 20, no. 3 (1989): 344.
Borenstein, Severin, and Nancy L Rose. "Bankruptcy and Pricing Behavior in U.S. Airline
Markets." The American Economic Review 85, no. 2 (1995): 397.
Bowman, Edward H. "Risk Seeking by Troubled Firms." Sloan Management Review (pre-
1986) 23, no. 4 (1982): 33.
Bradbury, Michael E, and Suzanne Lloyd. "An Estimate of the Direct Costs of Bankruptcy
in New Zealand." Asia Pacific Journal of Management 11, no. 1 (1994): 103.
Branch, Ben. "The Costs of Bankruptcy: A Review." International Review of Financial
Analysis 11, no. 1 (2002): 39.
Brander, James A., and Tracy R. Lewis. "Oligopoly and Financial Structure: The Limited
Liability Effect." The American Economic Review 76, no. 5 (1986): 956.
Breusch, T S, and A R Pagan. "A Simple Test for Heteroskedasticity and Random
Coefficient Variation." Econometrica 47, no. 5 (1979): 1287.
Busse, Maghan. "Firm Financial Condition and Airline Price Wars." The Rand Journal of
Economics 33, no. 2 (2002): 298.
Buxbaum, Peter A. "A Case of Corporate Turnaround." Distribution 94, no. 1 (1995): 30.
Buzacott, John A, and Rachel Q Zhang. "Inventory Management with Asset-Based
Financing." Management Science 50, no. 9 (2004): 1274.
Cachon, Gerard P. "Stock Wars: Inventory Competition in a Two-Echelon Supply Chain
with Multiple Retailers." Operations Research 49, no. 5 (2001): 658.
Campbell, Steven V. "Predicting Bankruptcy Reorganization for Closely Held Firms."
Accounting Horizons 10, no. 3 (1996): 12.
Carey, Susan, and Melanie Trottman. "Airlines, Battered by Losses, Look Set for Rebound;
Falling Costs, Fewer Seats May Be Ticket to Recovery for Some Major Carriers."
Wall Street Journal, Nov 26 2005, A.1.
211
Carpenter, Robert E, Steven M Fazzari, and Bruce C Petersen. "Financing Constraints and
Inventory Investment: A Comparative Study with High-Frequency Panel Data." The
Review of Economics and Statistics 80, no. 4 (1998): 513.
Carpenter, Robert E, Steven M Fazzari, Bruce C Petersen, Benjamin M Friedman, and Anil
K Kashyap. "Inventory Investment, Internal-Finance Fluctuations, and the Business
Cycle; Comments and Discussion." Brookings Papers on Economic Activity, no. 2
(1994): 75.
Caves, R. E., and M. E. Porter. "Market Structure, Oligopoly, and Stability of Market
Shares." The Journal of Industrial Economics 26, no. 4 (1978): 289.
Caves, Richard E., and Pankaj Ghemawat. "Identifying Mobility Barriers." Strategic
Management Journal 13, no. 1 (1992): 1.
Chakravarthy, Balaji S. "Measuring Strategic Performance." Strategic Management
Journal 7, no. 5 (1986): 437.
Chand, Suresh, and James Ward. "A Note on "Economic Order Quantity" under Conditions
of Permissible Delay in Payments"." Journal of the Operational Research Society
38, no. 1 (1987): 83-84.
Chang, Chun-Tao, and Jinn-Tsair Teng. "Retailer's Optimal Ordering Policy under Supplier
Credits." Mathematical Methods of Operations Research 60, no. 3 (2004): 471.
Chapman, C. B., S. C. Ward, Dale F. Cooper, and M. J. Page. "Credit Policy and Inventory
Control." The Journal of the Operational Research Society 35, no. 12 (1984): 1055.
Chattopadhyay, Prithviraj, William H Glick, and George P Huber. "Organizational Actions
in Response to Threats and Opportunities." Academy of Management Journal 44,
no. 5 (2001): 937.
Chen, Ming-Jer. "Competitor Analysis and Interfirm Rivalry: Toward a Theoretical
Integration." Academy of Management. The Academy of Management Review 21,
no. 1 (1996): 100.
Chen, Ming-Jer, Ken G. Smith, and Curtis M. Grimm. "Action Characteristics as Predictors
of Competitive Responses." Management Science 38, no. 3 (1992): 439.
Cook, R. Dennis, and Sanford Weisberg. "Diagnostics for Heteroskedasticity in
Regression." Biometrika 70, no. 1 (1983): 1-10.
Corbett, Jenny, Donald Hay, and Helen Louri. "A Financial Portfolio Approach to
Inventory Behavior: Japan and the Uk." International Journal of Production
Economics 59, no. 1-3 (1999): 43.
212
Cox, Andrew. "Understanding Buyer and Supplier Power: A Framework for Procurement
and Supply Competence." Journal of Supply Chain Management 37, no. 2 (2001):
8.
Cox, Andrew, Chris Lonsdale, Glyn Watson, and Hong Qiao. "Supplier Relationship
Management: A Framework for Understanding Managerial Capacity and
Constraints." European Business Journal 15, no. 3 (2003): 135.
Cox, Andrew, Joe Sanderson, and Glyn Watson. "Supply Chains and Power Regimes:
Toward an Analytic Framework for Managing Extended Networks of Buyer and
Supplier Relationships." Journal of Supply Chain Management 37, no. 2 (2001): 28.
Crandall, Robert. "Coffee, Tea ... Or Bankruptcy?" Wall Street Journal, Sep 16 2005, A.14.
Dasgupta, Sudipto, and Sheridan Titman. "Pricing Strategy and Financial Policy." The
Review of Financial Studies 11, no. 4 (1998): 705.
Denardo, Eric V. Dynamic Programming: Models and Applications. first ed. Mineola, New
York: Dover Publications Inc., 2003.
Derfus, Pamela J., Curtis M. Grimm, Ken G. Smith, and Patrick G. Maggitti. "The Red
Queen Effect: Competitive Action and Performance." Academy of Management
Journal (forthcoming).
Dobson, Paul W. "Exploiting Buyer Power: Lessons from the British Grocery Trade."
Antitrust Law Journal 72, no. 2 (2005): 529.
Dresner, Martin, Jiun-Sheng Chris Lin, and Robert Windle. "The Impact of Low-Cost
Carriers on Airport and Route Competition." Journal of Transport Economics and
Policy 30, no. 3 (1996): 309.
Drukker, D.M. "Testing for Serial Correlation in Linear Panel-Data Models." Stata Journal
3 (2003): 168-77.
Emerson, Richard M. "Power-Dependence Relations." American Sociological Review 27,
no. 1 (1962): 31-41.
Erlenkotter, Donald. "Ford Whitman Harris and the Economic Order Quantity Model."
Operations Research 38, no. 6 (1990): 937.
Ferrier, Walter J. "Navigating the Competitive Landscape: The Drivers and Consequences
of Competitive Aggressiveness." Academy of Management Journal 44, no. 4
(2001): 858.
Ferrier, Walter J, Cormac Mac Fhionnlaoich, Ken G Smith, and Curtis M Grimm. "The
Impact of Performance Distress on Aggressive Competitive Behavior: A
Reconciliation of Conflicting Views." Managerial and Decision Economics 23, no.
4,5 (2002): 301.
213
Finkin, Eugene F. "Company Turnaround." The Journal of Business Strategy 5, no. 4
(1985): 14.
Followill, Richard A., Michael Schellenger, and Patrick H. Marchand. "Economic Order
Quantities, Volume Discounts, and Wealth Maximization." The Financial Review
25, no. 1 (1990): 143.
Fraser, Donald R., and Norman Gaither. "Estimating Financing Costs in Managing
Inventory Positions at Large Corporations: Practice and Implications." Managerial
Finance 10, no. 2 (1984): 21.
Frazier, Gray L, and Kersi D Antia. "Exchange Relationships and Interfirm Power in
Channels of Distribution." Academy of Marketing Science. Journal 23, no. 4
(1995): 321.
French, John R. P., and Bertram Raven. The Bases of Social Power. Edited by Dorwin
Cartwright. Ann Arbor: University of Michigan Press, 1959.
Gaski, John F. "The Theory of Power and Conflict in Channels of Distribution." Journal of
Marketing (pre-1986) 48, no. 3 (1984): 9.
Gelman, Judith R., and Steven C. Salop. "Judo Economics: Capacity Limitation and
Coupon Competition." The Rand Journal of Economics 14, no. 2 (1983): 315.
Gertler, Mark, and Simon Gilchrist. "Monetary Policy, Business Cycles, and the Behavior
of Small Manufacturing Firms." The Quarterly Journal of Economics 109, no. 2
(1994): 309.
Gibbons, Robert. Game Theory for Applied Economists. First ed. Princeton, NJ: Princeton
University Press, 1992.
Greene, William H. Econometric Analysis. Fifth ed. Delhi/India: Pearson Education Inc.,
2003.
Grice, John Stephen, and Robert W. Ingram. "Tests of the Generalizability of Altman's
Bankruptcy Prediction Model." Journal of Business Research 54 (2001): 53-61.
Gritta, Richard D. "Air Carrier Financial Condition: A Review of Models for Measuring
the Financial Fitness of the U.S. Airline Industry." Proceedings of the 2004
Conference of the Air Transport Research Society (2004).
Guariglia, Alessandra. "The Effects of Financial Constraints on Inventory Investment:
Evidence from a Panel of Uk Firms." Economica 66, no. 261 (1999): 43.
Guariglia, Alessandra, and Fabio Schiantarelli. "Production Smoothing, Firms'
Heterogeneity, and Financial Constraints: Evidence from a Panel of Uk Firms."
Oxford Economic Papers 50, no. 1 (1998): 63.
214
Guffey, Daryl M., and William T. Moore. "Direct Bankruptcy Costs: Evidence from the
Trucking Industry." The Financial Review 26, no. 2 (1991): 223.
Haley, Charles W, and Robert C Higgins. "Inventory Policy and Trade Credit Financing."
Management Science (pre-1986) 20, no. 4 (1973): 464.
Hall, M., and L. Weiss. "Firm Size and Profitability." Review of Economics and Statistics
49 (1967): 319-31.
Halley, Alain, and Jean Nollet. "The Supply Chain: The Weak Link for Some Preferred
Suppliers?" Journal of Supply Chain Management 38, no. 3 (2002): 39.
Hambrick, Donald C., and Steven M. Schecter. "Turnaround Strategies for Mature
Industrial-Product Business Units." Academy of Management Journal 26, no. 2
(1983): 231.
Harris, F.W. "How Many Parts to Make at Once." Factory, The Magazine of Management
10 (1913): 135-36, 52.
Hausman, J A. "Specification Tests in Econometrics." Econometrica (pre-1986) 46, no. 6
(1978): 1251.
Hendel, Igal. "Competition under Financial Distress." The Journal of Industrial Economics
44, no. 3 (1996): 309.
Hingley, M K. "Power to All Our Friends? Living with Imbalance in Supplier-Retailer
Relationships." Industrial Marketing Management 34, no. 8 (2005): 848.
Hingley, Martin K. "Power Imbalanced Relationships: Cases from Uk Fresh Food Supply."
International Journal of Retail & Distribution Management 33, no. 8/9 (2005): 551.
Hitt, Michael A., Brian K. Boyd, and Dan Li. "The State of Strategic Management
Research and Vision of the Future." In Research Methodology in Strategy and
Management, 1, edited by David J. Ketchen and Donald D. Bergh, 1-32: JAI Press,
2004.
Hofer, Charles W. "Toward a Contingency Theory of Business Strategy." Academy of
Management Journal (pre-1986) 18, no. 4 (1975): 784.
Hofer, Charles W. "Turnaround Strategies." The Journal of Business Strategy 1, no. 1
(1980): 19.
Hofer, Christian, Martin Dresner, and Robert Windle. "Financial Distress and Us Airline
Fares." Journal of Transport Economics and Policy 39, no. 3 (2005): 323.
Kahneman, Daniel, and Amos Tversky. "Prospect Theory: An Analysis of Decision under
Risk." Econometrica (pre-1986) 47, no. 2 (1979): 263.
215
Kashyap, Anil K, Owen A Lamont, and Jeremy C Stein. "Credit Conditions and the
Cyclical Behavior of Inventories." The Quarterly Journal of Economics 109, no. 3
(1994): 565.
Kashyap, Anil K, Jeremy C Stein, and David W Wilcox. "Monetary Policy and Credit
Conditions: Evidence from the Composition of External Finance." The American
Economic Review 83, no. 1 (1993): 78.
Kennedy, Robert E. "The Effect of Bankruptcy Filings on Rivals' Operating Performance:
Evidence from 51 Large Bankruptcies." International Journal of the Economics of
Business 7, no. 1 (2000): 5.
Kim, Yong H., and Kee H. Chung. "An Integrated Evaluation of Investment in Inventory
and Credit: A Cash Flow Approach." Journal of Business Finance & Accounting
17, no. 3 (1990): 381.
Lant, Theresa K., Frances J. Milliken, and Bipin Batra. "The Role of Managerial Learning
and Interpretation in Strategic Persistence and Reorientation: An Empirical
Exploration." Strategic Management Journal 13, no. 8 (1992): 585.
LeBlanc, Greg. "Signalling Strength: Limit Pricing and Predatory Pricing." The Rand
Journal of Economics 23, no. 4 (1992): 493.
Lee, Darin, and María José Luengo-Prado. "The Impact of Passenger Mix on Reported
"Hub Premiums" in the U.S. Airline Industry." Southern Economic Journal 72, no.
2 (2005): 372.
Lieberman, Marvin B, Susan Helper, and Lieven Demeester. "The Empirical Determinants
of Inventory Levels in High-Volume Manufacturing." Production and Operations
Management 8, no. 1 (1999): 44.
Longenecker, Justin G, and Charles D Pringle. "The Illusion of Contingency Theory as a
General Theory." Academy of Management. The Academy of Management Review
(pre-1986) 3, no. 000003 (1978): 679.
Luthans, Fred, and Todd I Stewart. "A General Contingency Theory of Management."
Academy of Management. The Academy of Management Review 2, no. 2 (1977):
181.
Mahajan, Siddharth, and Garrett van Ryzin. "Inventory Competition under Dynamic
Consumer Choice." Operations Research 49, no. 5 (2001): 646.
Maksimovic, Vojislav, and Josef Zechner. "Debt, Agency Costs, and Industry
Equilibrium." The Journal of Finance 46, no. 5 (1991): 1619.
Mason, Edward S. "The Current State of the Monopoly Problem in the United States."
Harvard Law Review 62, no. June 1949 (1949): 1265-85.
216
———. "Price and Production Policies of Large-Scale Enterprise." American Economic
Review 29, no. March 1939 (1939): 61-74.
Mazzeo, M.J. "Product Choice and Oligopoly Market Structure." The Rand Journal of
Economics 33, no. 2 (2002): 221-42.
McCafferty, Joseph. "Is Bankruptcy an Unfair Advantage?" CFO 11, no. 6 (1995): 28.
McCracken, Jeffrey. "Auto Maker Recovery Bid Intensifies; Ford to Accelerate, Deepen Its
Restructuring Efforts; Gm Cites Rosier Fiscal View." Wall Street Journal, Aug 10
2006, A.2.
McKinley, William. "Organizational Decline and Adaptation: Theoretical Controversies."
Organization Science 4, no. 1 (1993): 1-9.
McLaughlin, Curtis P, Gyula Vastag, and D Clay Whybark. "Statistical Inventory Control
in Theory and Practice." International Journal of Production Economics 35, no. 1-3
(1994): 161.
Miller, Danny, and Ming-Jer Chen. "Sources and Consequences of Competitive Inertia: A
Study of the U.S. Airline Industry." Administrative Science Quarterly 39, no. 1
(1994): 1.
Modigliani, Franco, and Merton H. Miller. "The Cost of Capital, Corporation Finance and
the Theory of Investment." The American Economic Review 48, no. 3 (1958): 261-
97.
Mone, Mark A, William McKinley, and Vincent L Barker III. "Organizational Decline and
Innovation: A Contingency Framework." Academy of Management. The Academy
of Management Review 23, no. 1 (1998): 115.
Moon, I, and E A Silver. "The Multi-Item Newsvendor Problem with a Budget Constraint
and Fixed Ordering Costs." The Journal of the Operational Research Society 51,
no. 5 (2000): 602.
Morrison, Steven A. "Actual, Adjacent, and Potential Competition." Journal of Transport
Economics and Policy 35, no. 2 (2001): 239.
Moses, O Douglas. "Organizational Slack and Risk-Taking Behaviour: Tests of Product
Pricing Strategy." Journal of Organizational Change Management 5, no. 3 (1992):
38.
Moulton, Wilbur N, and Howard Thomas. "Bankruptcy as a Deliberate Strategy:
Theoretical Considerations and Empirical Evidence." Strategic Management
Journal 14, no. 2 (1993): 125.
Opler, Tim C, and Sheridan Titman. "Financial Distress and Corporate Performance." The
Journal of Finance 49, no. 3 (1994): 1015.
217
Osteryoung, Jerome S., Eldon Nosari, Daniel E. McCarty, and Walter J. Reinhart. "Use of
the EOQ Model for Inventory Analysis." Production and Inventory Management
27, no. 3 (1986): 39.
Peteraf, Margaret A. "Intra-Industry Structure and the Response toward Rivals."
Managerial and Decision Economics (1986-1998) 14, no. 6 (1993): 519.
Porter, Michael. Competitive Strategy. New York, NY: Free Press, 1980.
Pryor, Frederic L. "New Trends in U.S. Industrial Concentration." Review of Industrial
Organization 18, no. 3 (2001): 301.
Queen, Maggie, and Richard Roll. "Firm Mortality: Using Market Indicators to Predict
Survival." Financial Analysts Journal 43, no. 3 (1987): 9.
Rabinovich, Elliot, and Philip T Evers. "Enterprise-Wide Adoption Patterns of Inventory
Management Practices and Information Systems." Transportation Research. Part E,
Logistics & Transportation Review 38E, no. 6 (2002): 389.
Rachamadugu, Ram. "Effect of Delayed Payments (Trade Credit) on Order Quantities."
Journal of the Operational Research Society 40, no. 9 (1989): 805-13.
Ravenscraft, David J. "Structure-Profit Relationships at the Line of Business and Industry
Level." The Review of Economics and Statistics 65, no. 1 (1983): 22.
Reuter, Vincent G. "Are We Using the Inventory Management Tools?" Industrial
Management 20, no. 3 (1978): 26.
Rodgers, Kimberly Jo Guthrie. "Chapter 11 Reorganization Policy and the Economic
Viability of Filing Firms." 3017860, Purdue University, 2000.
Rollman, Daniel P. "Flying Low: Chapter 11's Contribution to the Self-Destructive Nature
of the Airline Industry Economics." Emory Bankruptcy Developments Journal 21,
no. 1 (2004): 381.
Rose-Green, Ena, and Mark Dawkins. "Strategic Bankruptcies and Price Reactions to
Bankruptcy Filings." Journal of Business Finance & Accounting 29, no. 9/10
(2002): 1319.
Rosenblatt, Meir J. "Multi-Item Inventory System with Budgetary Constraint: A
Comparison between the Lagrangian and the Fixed Cycle Approach." International
Journal of Production Research 19, no. 4 (1981): 331-39.
Roumiantsev, Serguei, and Serguei Netessine. "What Can Be Learned from Classical
Inventory Models: A Cross-Industry Empirical Investigation." Manufacturing &
Service Operations Management forthcoming (2007).
218
Rustenburg, W D, G J van Houtum, and W H M Zijm. "Spare Parts Management for
Technical Systems: Resupply of Spare Parts under Limited Budgets." IIE
Transactions 32, no. 10 (2000): 1013.
Scherer, F M. Industrial Market Structure and Economic Performance. Second ed.
Chicago: Rand McNally College Publishing Company, 1980.
Schoonhoven, Claudia Bird. "Problems with Contingency Theory: Testing Assumptions
Hidden within the Language of Contingency ''Theory''." Administrative Science
Quarterly 26, no. 3 (1981): 349.
Scott, John T. "Multimarket Contact and Economic Performance." Review of Economics
and Statistics 64 (1982): 368-75.
Shoham, Aviv, and Avi Fiegenbaum. "Competitive Determinants of Organizational Risk-
Taking Attitude: The Role of Strategic Reference Points." Management Decision
40, no. 1/2 (2002): 127.
Singh, Jitendra V. "Performance, Slack, and Risk Taking in Organizational Decision
Making." Academy of Management Journal 29, no. 3 (1986): 562.
Smart, Carolyne, and Ilan Vertinsky. "Strategy and the Environment: A Study of Corporate
Responses to Crises." Strategic Management Journal 5, no. 3 (1984): 199.
Smith, Ken G, Curtis M Grimm, Stefan Wally, and Greg Young. "Strategic Groups and
Rivalrous Firm Behavior: Towards a Reconciliation." Strategic Management
Journal 18, no. 2 (1997): 149.
Smith, Ken G., Curtis M. Grimm, Martin J. Gannon, and Ming-Jer Chen. "Organizational
Information Processing, Competitive Responses, and Performance in the U.S.
Domestic Airline Industry." Academy of Management Journal 34, no. 1 (1991): 60.
Smith, Malcolm, and Christopher Graves. "Corporate Turnaround and Financial Distress."
Managerial Auditing Journal 20, no. 3 (2005): 304.
Spanos, Y., G. Zaralis, and S.Lioukas. "Strategy and Industry Effects on Profitability:
Evidence from Greece." Strategic Management Journal 25, no. 2 (2004): 139-66.
Staw, Barry M., Lance E. Sandelands, and Jane E. Dutton. "Threat-Rigidity Effects in
Organizational Behavior: A Multilevel Analysis." Administrative Science Quarterly
26, no. 4 (1981): 501.
Stundza, Tom, and Brian Milligan. "How Chrysler Will Cut Costs." Purchasing 130, no. 3
(2001): 30.
Sudarsanam, Sudi, and Jim Lai. "Corporate Financial Distress and Turnaround Strategies:
An Empirical Analysis." British Journal of Management 12, no. 3 (2001): 183.
219
Sun, Daning, and Maurice Queyranne. "Production and Inventory Model Using Net Present
Value." Operations Research 50, no. 3 (2002): 528.
Tersine, Richard J. Principles of Inventory and Materials Management. Fourth ed. Upper
Saddle River, NJ: Prentice-Hall, 1994.
Timme, Stephen G. "The Real Cost of Holding Inventory." Supply Chain Management
Review 7, no. 4 (2003): 30.
Tiras, Samuel L. "Which Firms Are Likely to Survive Chapter 11?" Commercial Lending
Review 17, no. 6 (2002): 18.
Trottman, Melanie. "Airline Bankruptcy Filings Put Continental, American at Risk." Wall
Street Journal, Sep 22 2005, B.4.
Waldman, Don E, and Elizabeth J Jensen. Industrial Organization: Theory and Practice.
Second ed: Addison Wesley Longman, 2001.
Wallin, Cynthia, M. Johnny Rungtusanatham, and Elliot Rabinovich. "What Is the "Right"
Inventory Management Approach for a Purchased Item?" International Journal of
Operations & Production Management 26, no. 1 (2006): 50.
Warner, Jerold B. "Bankruptcy Costs - Some Evidence." The Journal of Finance 32, no. 2
(1977): 337.
White, Halbert. "A Heteroskedasticity-Consistent Covariance Matrix Estimator and a
Direct Test for Heteroskedasticity." Econometrica 48, no. 4 (1980): 817-38.
Wiersema, William H. "Financing the Unfinanceable." Electrical Apparatus 58, no. 5
(2005): 42.
Wiseman, Robert M, and Philip Bromiley. "Toward a Model of Risk in Declining
Organizations: An Empirical Examination of Risk, Performance and Decline."
Organization Science 7, no. 5 (1996): 524.
Wiseman, Robert M, and Luis R Gomez-Mejia. "A Behavioral Agency Model of
Managerial Risk Taking." Academy of Management. The Academy of Management
Review 23, no. 1 (1998): 133.
Wooldridge, J M. Econometric Analysis of Cross Section and Panel Data. Cambridge,
MA: MIT Press, 2002.
Xu, Tianji, and Robert L K Tiong. "Risk Assessment on Contractors' Pricing Strategies."
Construction Management and Economics 19, no. 1 (2001): 77.
Yaffee, Robert. "A Primer for Panel Data Analysis." New York University, NY:http://www.nyu.edu/its/pubs/connect/fall03/yaffee_primer.htm, 2003.
220
Yoffie, David B, and Mary Kwak. "Judo Strategy: 10 Techniques for Beating a Stronger
Opponent." Business Strategy Review 13, no. 1 (2002): 20.
Young, Greg, Ken G Smith, and Curtis M Grimm. ""Austrian" and Industrial Organization
Perspectives on Firm-Level Competitive Activity and Performance." Organization
Science 7, no. 3 (1996): 243.
Zinn, Walter, and John M Charnes. "A Comparison of the Economic Order Quantity and
Quick Response Inventory Replenishment Methods." Journal of Business Logistics
26, no. 2 (2005): 119.
doc_294730646.pdf