Description
Brief explanation concerning new firm performance and the replacement of founder ceos jing chen.
New Firm Performance and the
Replacement of Founder-CEOs
Jing Chen*
Copenhagen Business School
Peter Thompson**
Emory University
November 2012
In this paper we study some causes and consequences of founder-CEO replace-
ments among a sample of 4,172 Danish startups created by single founders in
1999 and 2000. In contrast to the extant literature on VC-financed firms, re-
placements among firms in our sample are more likely among the worst- and
best-performing firms, and replacement is not unambiguously associated with
better subsequent performance. Firms that replaced the founder as CEO were
much more likely to fail, but the surviving firms among them grew considerably
faster. We also analyze subsequent earnings and occupational choices of founders
who rescinded operating control of their firms. Although subsequent founder in-
come is increasing in the performance of their firms, those that left good firms
were no more likely than others to found another business. Our results are consis-
tent with the notion that founder-CEO replacement is driven in part by mis-
matches between business quality and founder ability.
JEL Classification Codes: D21, D22, J62, L26
Keywords: Entrepreneurship, business creation, founder turnover, CEO turnover.
* CBS - Department of Innovation and Organizational Economics, Kilevej 14, 2000 Fre-
deriksberg, Denmark. Email: [email protected]. ** Goizueta Business School, Emory Universi-
ty, 1300 Clifton Road, Atlanta, GA 30322, USA. Email: [email protected]. We
appreciate helpful comments from: Christian Catalini, Alberto Galasso, Mirjam Van
Praag, participants in the 2012 Summer Residence Week for Entrepreneurship Econo-
mists at Oxford University, the Roundtable for Engineering Entrepreneurship Research at
Georgia Tech, and in seminars at the University of Amsterdam, the University of Lugano
and Emory University.
1
1. Introduction
How does firm performance influence the likelihood that founder-CEOs of new
businesses cede operating control to a specialist manager? What implications does
the departure of the founder have for subsequent business performance? And
what happens to founders after they have left? This paper reports the results of
an analysis of the causes and consequences of founder-CEO replacement among a
cohort of new businesses founded in Denmark. Our empirical work uses the
matched employer-employee database maintained by Statistics Denmark, and
draws samples from the universe of new businesses created in Denmark during
1999 and 2000. These firms are generally small, they are almost invariably not
financed by venture capital, and they are generally not at risk of an IPO event.
We know surprisingly little about founder turnover
1
among such firms.
A growing literature has provided us with substantial insight into the causes and
consequences of founder-CEO replacement in other types of firms. This literature,
which has focused on high-growth and VC-backed firms
2
, finds that founder-CEO
replacement is most likely to occur after a period of rapid expansion [e.g., Boeck-
er and Karichalil (2002)], or upon attainment of critical milestones such as com-
pletion of product development or securing new rounds of outside funding [Was-
serman (2003)]. More important, perhaps, is that founder-CEO replacement is
typically motivated by an expectation among stakeholders that the firm is likely
in the near future to undergo an episode of transformational growth. Indeed, a
significant portion of the literature explicitly samples on such expectations by
studying only firms that are preparing for an IPO [e.g., Nelson (2003), Jain and
Tabak (2008), Pollack, Fund and Baker (2009)].
The extant research lacks findings that we can confidently extrapolate to the
firms that are the focus of the present paper. First, VC-financed firms appear to
1
We will use the terms founder-CEO replacement and founder turnover interchangeably
to describe an event where the founder of a firm, who had been managing the firm as his
or her primary occupation, rescinds operating control while maintaining ownership rights.
2
Of course, executive turnover in established, typically public, firms has been extensively
studied for many years [e.g., Hofer (1980), Chen and Hambrick (2012)]. There is also an
active literature on succession in family-owned businesses, but the concerns of this litera-
ture are rather different from those in which executive control is passed outside the
founder’s family. See Handler (1994) for a review.
2
exhibit founder-CEO replacements at much higher rates than firms that are not
subject to significant control by outside equity investors [Hellman and Puri
(2002), Boeker and Wiltbank (2005)]. Second, few of the firms in our sample are
ever likely to undergo transformations of the sort that appear to motivate many
founder-CEO substitutions among VC-financed firms. Third, VC-financed firms
are not random draws from the firm population. They are more likely to grow
substantially, they are better financed, and their founders and employees on av-
erage have more human capital. Correlations between variables estimated from a
sample drawn from the upper tail of the quality distribution may not hold in
other segments of the population, especially when there are possibilities of non-
monotonic relationships in the population as a whole.
It is well-known that VC-financed firms account for a tiny fraction of business
launches
3
but they play a much larger role in economic growth, employment crea-
tion and innovation. As a result, academic effort to understand the dynamics of
control in VC-financed firms is appropriately disproportionate to their numbers.
Nonetheless, given their numerical importance, we should also be concerned with
the control dynamics of the typical small business. Patterns of founder turnover
among representative businesses also provide insight into the forces governing
business transfers.
4
They are also key to understanding the expected opportunity
costs of business creation, as well as the dual phenomena of serial and portfolio
entrepreneurs.
In this paper we study founder-CEO replacements among a sample of 4,172 Da-
nish startups created in 1999 and 2000. To ensure that the departure of the
founder indicates a meaningful change in control, the sample consists only of
firms that were created by a single founder, and where the founder worked as his
primary job in the business. The firms and founders we study represent all firms
founded in Denmark that meet these criteria. Only 18 percent of them are classi-
3
For example, in the United States, 626,771 new businesses were created in 2008, only
3,276 of which received VC financing (Bureau of labor Statistics athttp://www.bls.
gov/bdm/entrepreneurship/bdm_chart1.htm; and NCVA (2009).
4
Founders may sell businesses because they are poorly matched to the needs of the busi-
ness, or because they have reasons to cash out their equity position. Looking at founder-
CEO replacements among businesses that are not transferred, allows us to gain insights
about the first of these two motivations.
3
fied as high-tech, while wholesale, retail, construction and manufacturing are
well-represented; this is a much broader sample than has previously been used to
study founder-CEO replacement. Our data are drawn from Denmark’s national
matched employer-employee dataset, so we are also in a unique position to ex-
plore what happens to the founders after they have left the firm. We study in-
stances of founder-CEO replacement and we track the firms and founder until
2005. The data allow us to distinguish business turnover from founder turnover
and, in this paper, we focus on the causes and consequences of events in which
the founder retires from active management but continues to retain controlling
ownership. That is, our focus is on CEO succession in small firms, rather than
business transfers.
5
To guide our empirical work, we develop a simple model of firm creation, surviv-
al, and founder turnover. The goal of founder-CEO replacement among VC-
backed firms is generally held to be resolving mismatches between the skills of
the founder-CEO and the expected demands of the transforming enterprise. We
also develop a framework based on mismatches between business and founder,
although our model focuses on current rather than anticipated mismatches. We
suppose that firm performance is a function of the ability of the founder and the
quality of the business idea, both of which are draws from stable population dis-
tributions. Ability and quality are complements in production so that there can
arise mismatches of sufficient magnitude to justify replacing the founder with an
outside CEO who is better suited to the business.
6
Our theoretical framework motivates us to consider the possibility that founder-
CEO turnover is most likely in the tails of the initial performance distribution,
and that, while good early performance is on average associated with good future
performance, this effect is stronger in firms that had no turnover. What we find
in the data is a little more complex than this. We do find that turnover is more
likely in the tails of the firm performance distribution and, conditional on firm
5
The classic empirical study of small business transfers is Holmes and Schmitz (1995).
6
Our theoretical framework is related to the models of business transfers developed by
Holmes and Schmitz (1990, 1995), and to the model in Braguinsky et al. (2012), which
explores the effect of the cost of evaluating ideas on founder turnover. We shall comment
further on the relationship between our theoretical framework and these models in the
next section.
4
survival, turnover is associated with higher rates of growth. However, turnover is
also associated with higher firm exit rates. We also explore the earnings and oc-
cupational choices of founders who have relinquished day-to-day control of their
firms. Our framework and our empirical results are both consistent with the in-
tuitive notion that founders who were replaced in high-performing businesses
subsequently earn more than those who left low-performing businesses but, less
intuitively perhaps, they are no more likely to create another startup.
2. Theoretical Framework
There are two periods. In the first period, a founder with ability q has an idea of
quality q
1
. Ability is general and determines both the success with which the
founder can manage a firm that implements the idea, and his earnings in alterna-
tive employment. We assume that founder ability is innate, fixed over time and
known to the founder. Idea quality incorporates a variety of factors influencing
performance, including the quality of the underlying idea or technology, its adap-
tation and fit to the market, and the public’s state of knowledge about the firm’s
product. Some of these factors may be regarded as fixed, while others evolve over
time as the business model is refined. Upon entry, the founder knows ?, but he
only knows that q
1
is a random draw from the distribution
1
( | ). F q q The realiza-
tion of q
1
is observed after entry. The business idea is refined and adapted to the
market, yielding a new quality q
2
in the second period, which is drawn from
( )
2 1
| . G q q
In each period, t = 1, 2, output is given by
1
t t
t t
y q
a a
q
-
= , (1)
for some [0, 1].
t
a Î Although, q and q are assumed to be complements in both
periods, we allow for the relative importance of ability and idea quality to vary
from one period to the next.
Profit in period 1 is given by
1 1
1
1 1
q w
a a
p q q
-
= - . (2)
The only cost is the opportunity cost of the founder’s time, which is assumed to
be linearly increasing in ability. For concreteness, we shall suppose this alterna-
5
tive use of the agent's time is in wage work.
At the end of the first period, q
2
is observed and a decision is then made whether
to contract with an outside CEO to replace the founder, to continue without
change of leadership, or to exit. If the outside CEO has ability q
?
, she must be
paid a wage . wq
?
If a replacement CEO is hired, a transition cost, c, must be
paid. This cost reflects both direct costs of any disruption associated with a
change in leadership, as well as any premium necessary to resolve agency prob-
lems arising from the separation of ownership and control [Fama and Jensen
(1983), Jensen and Meckling (1976)].
7
The rewards to recruiting an outside CEO
are that the CEO’s ability can be chosen optimally in light of the realization of
q
2
, and the founder is released to earn wq elsewhere while remaining as the resi-
dual claimant on the firm’s earnings. The founder is replaced by an outsider if
( )
2 2 2 2
1 1
2 2
arg max q w c q w
a a a a
q
q q q q
- -
- - > -
?
? ?
, (3)
and if the term on the left hand side of (3) is positive. The profit-maximizing
ability of the external CEO is given by
( )
2
1/(1 )
2 2
/ w q
a
q a
-
=
?
, (4)
so maximized profits in period 2 are
( )
2 2
1 *
2 2 2
max , q w Aq c
a a
p q q
-
= - - , (5)
where
2 2
/(1 )
2 2
(1 )( / ) . A w
a a
a a
-
= - If
*
2
0, p < the firm exits.
The two terms on the RHS of (5) are plotted separately in Figure 1 for different
values of q
2
while holding q constant. The concave function
2 2
( , ) q p q indicates
profits when the founder continues, and the straight line
2 2
( ) q p? depicts profits
when the founder is replaced. In the case illustrated in the figure, there are four
distinct regions defined by the boundary values 0 :
E
q q q < £ £ if
2
[0, ],
E
q q Î the
firm exits at the end of period 1; if
2
[ , ]
E
q q q Î or if
2
, q q ³
the founder is re-
placed as CEO; and if
2
( , ), q q q Î the founder continues to operate the firm.
7
He (2008), for example, documents that founder CEOs have lower incentive compensa-
tion and lower total compensation than professional CEOs.
6
When
2
[ , ]
E
q q q Î , the founder is replaced because the opportunity cost of his
time does not justify continued commitment to the business model; he is replaced
by an external CEO who is less able than the founder. In contrast, when
2
q q ³
the founder is replaced by an external CEO with higher ability.
Figure 1 also illustrates the effect of an increase in the founder’s ability, from q to
. q
¢
This reduces the profitability of the founder continuing with the firm when
the quality of the idea is low, and increases it when the idea is good. It is easy to
verify that the intersection of
2 2
( , ) q p q and
2 2
( , ) q p q¢ lies between q and , q and
hence that the effect of an increase in q is to raise both q and . q The interval
[ , ]
E
q q will vanish if the founder’s ability declines sufficiently so that .
E
q q <
There are, therefore, two distinct cases. In one case, when founders have limited
ability, the period 2 outcomes fall into three distinct groups, with replacement of
founders occurring only at higher values of q. In the second case, founders are
replaced in firms with business models drawn from the tails of the quality distri-
bution among all surviving firms. Which of these cases arises depends on the
founder’s ability, as well as other parameters of the model. For example, an in-
crease in the transition cost, c, shifts
2 2
( ) q p? down, which may eliminate the in-
terval [ , ];
E
q q
that is, if the increase in c is large enough, replacement of founders
occurs only at high values of q
2
and in each case the replacement has higher abili-
ty than the founder. Clearly, the minimum value of c that eliminates the interval
E
q
q
q
2
q
0
2 2
( ) q p?
2 2
( , ) q p q
c -
wq -
wq¢ -
2 2
( , ) q p q¢
q q
¢
>
FIGURE 1. Second period payoffs; c>0.
7
[ , ]
E
q q
depends on the founder’s ability.
8
Figure 2 provides a more complete depiction of the model’s implications. Firms
that are closed down are those with low-quality ideas. If the founder does not
want to continue to operate the firm, continuation requires a new CEO. Howev-
er, payment of the transition cost cannot be justified in low-quality firms. The
boundary separating firms that are discontinued from those that are continued
with the founder at the helm is given by the pairs
2
{ , } q q satisfying
2
1/(1 )
2
, q w
a
q
-
= showing that low-ability founders may continue to operate a busi-
ness that a high-ability founder would close down. Founder turnover in continu-
ing firms occurs in two distinct groups. First, there is a group of firms with high-
quality ideas but that were founded by agents of sufficiently low ability to merit
replacement with a more able CEO. The boundary between this group and the
group of founders that continue to operate their own firm is defined by the func-
8
When 0, c > it is always the case that . q q < In the limiting case of 0, c = the func-
tion
2 2
( ) q p?
is then tangential to
2 2
( , ) q p q for any value of q. and . q q ? In this case,
every founder is replaced, because it is always efficient to secure a perfect match between
CEO ability and idea quality.
FIGURE 2. Ability, idea quality, and outcomes.
q
2
(log scale)
q
(log scale)
Replace with
less able CEO
Replace with more
able CEO
Founder continues
to operate firm
Firm exits and founder
enters wage employment
2
1/(1 ) 1
A w
a - - -
1
2
q cA
-
=
2
1/(1 )
2
q w
a
q
-
=
( ) q q
( ) q q
8
tion
( ) q q that provides the larger of the solutions to the equation
2 2
1
2 2
. Aq c q w
a a
q q
-
- = - As the transition cost rises, ( ) q q unambiguously shifts
upwards. The second group consists of firms with able founders but whose firms
are of middling quality. The quality of the idea is high enough to prefer continua-
tion over exit, but not high enough to merit continuation of the founder. The re-
placement CEO has lower ability than the founder. The boundary between this
group and those that exit is a horizontal line at
1
2
. q cA
-
= The boundary be-
tween this group and the group of founders that continue to operate their own
firm is given by the function ( ) q q that provides the smaller of the solutions to
2 2
1
2 2
Aq c q w
a a
q q
-
- = - . An increase in the transition cost shifts ( ) q q downwards.
Figure 2 shows that ( ) q q and ( ) q q are both positively sloped, which is consistent
with the discussion around Figure 1.
2.1 Baseline simulations
The empirical implications of our theoretical framework depend critically on the
joint distribution of q
2
and q (and, by virtue of the correlation between q
1
and q
2
,
on the joint distribution of initial idea quality and founder ability). For example,
the implication that founder turnover occurs both when q
2
is high (relative to q)
and when it is low requires that we observe in any sample instances where con-
trasting values of q
2
and q coexist. Similarly, the transition cost cannot be so high
that it eliminates the interval [ , ].
E
q q Because this may not be the case, and de-
spite the simplicity of the theoretical framework, there are no general empirical
predictions about the relationships between performance in the two periods,
founder ability and turnover. We will therefore make use of some numerical simu-
lations to explore the possibilities of the model.
However, it is useful to consider some reasonable ways in which we can limit the
varied possibilities. We do so in two ways. The first imposes a constraint on the
transition cost, c, such that the model predicts an aggregate rate of turnover that
approximates the rate (about 12 percent) that we observe in our data. The
second makes use of a simple free-entry condition to help us constrain the rela-
tionship between ability and the quality of the idea. We want to admit entrepre-
neurs of all abilities. This will be the case if all agents, regardless of ability, are
indifferent between entrepreneurship and wage work in period 1. If the quality of
ideas and agent ability are independent of each other, then high-ability agents,
who bear a higher opportunity cost of becoming an entrepreneur, will find busi-
9
ness creation less attractive than low-ability agents. We will therefore assume
that idea quality and agent ability are positively correlated. For our purposes, it
suffices to suppose that ability and agents care only about expected first-period
returns. In the simulations to follow we therefore impose the constraint
9
1 1
1 1
1
, E q w
a a
q
- -
é ù
=
ê ú
ë û
(6)
which ensures that all agents, regardless of ability, are indifferent between creat-
ing a business and wage employment.
In our baseline model, we assume that in both periods ability and the quality of
the ideas are equally important. That is, we set
1 2
0.5. a a = = We assume that
1
1 a
q
-
has a lognormal distribution, with unit mean and standard deviation of 0.5.
Consistent with (6), we then assume that, for agent i,
1
1
1i
q
a -
is a draw from a log-
normal distribution with mean
1
1
.
i
w
a
q
-
Finally, consistent with the assumption
that
2 1
( | ) G q q is decreasing in q
1
, we assume that
2
1
2i
q
a -
is a draw from a lognor-
mal distribution with mean
1
1
1
.
i
q
a -
The wage, w, is normalized to 1. Finally, we
set 0.35, c = so that our baseline produces an aggregate rate of founder turnover
of about 12 percent, consistent with evidence we shall show later.
Figure 3, which plots 5,000 random draws using the baseline parameters, depicts
outcomes that confirm the boundaries already given in Figure 2. Although found-
ers of exiting firms on average have relatively low ability, this is incidental and
driven by the correlation between idea quality and founder ability. Indeed, once
one conditions on idea quality, low-ability founders are less likely to close down
the firm. The simulations yield observations in the two sets of firms with suffi-
ciently good ideas but poor matches between idea and founder ability to merit
replacement with a more able CEO. Although founder CEO’s of all abilities may
suffer a mismatch sufficient to induce their replacement with an outside CEO, it
is clear that the likelihood of replacement is greater in the tails of the ability dis-
tribution.
9
A forward-looking agent will also consider the option value that entering gives by pro-
viding choices in the second period. In period 2, the founder may choose to remain in
control of the business, replace a new CEO if doing so increases profits, or exit. Even if
expected second-period earnings from continuation are the same as expected first-period
earnings, the replacement and exit options imply for a forward looking agent that
1 1
1
1 1
. E q w
a a
q
- -
<
é ù
ê ú
ë û
10
However, we do not get to observe q and q separately. In the data, we observe
only their combined contribution to revenues, and it is not obvious that we
should also expect to see more frequent founder-CEO replacements in the tails of
the firm performance distribution. Figure 4 therefore plots simulated first- and
second-period revenues of surviving firms by founder status for our baseline simu-
lations. The two distinct groups of founders that experience turnover are also
clearly evident when we look at first-period performance, and they are indeed
drawn more frequently from the tails of the performance distribution. There is a
group with above-average performance in the first-period whose founders are re-
placed with less able CEOs. There is also a group of firms whose founders are
replaced with more able CEOs. This group consists of firms with low-ability
founders, but their poor ability is offset by ideas that in the second-period are
sufficiently high quality to justify payment of the transition cost (firms with the
worst ideas in the second period are closed down). Although these firms have a
mean first-period performance that is not much different from the population av-
0.0001
0.001
0.01
0.1
1
10
100
0.01 0.1 1 10 100
I
d
e
a
Q
u
a
l
i
t
y
i
n
P
e
r
i
o
d
2
,
q
2
Founder ability, q
Replaced with less able CEO Founder remained
Replaced with more able CEO Firm exited
FIGURE 3. Baseline simulations Founder ability versus
second-period idea quality, by outcome. 5000 observations.
Parameters: a
1
=a
2
=0.5, w=1, c=0.35. Each
1
1 a
q
-
is a draw
from the lognormal with mean one,
1
1
1
q
a -
is lognormal with
mean
1
1
,
a
q
-
and each
2
1
2
q
a -
is lognormal with mean
1
1
1
; q
a -
all
three distributions have a standard deviation of 0.5. Founder
turnover occurs in about 12% of the observations.
11
erage, they are overrepresented in the lower tail of the first-period performance
distribution.
A distinctive feature of Figure 4 concerns the effect that CEO turnover has on
the relationship between first-period performance and second-period performance.
Performance among the group of surviving firms with no turnover tends to be
more consistent from one period to the next than in the case of firms that expe-
rienced turnover. Firms in which founders are replaced by less-able CEOs see on
average a decline in their performance, while the arrival of a more capable CEO
induces on average an improvement in performance. Among surviving firms with
no turnover, therefore, the relationship between first- and second-period perfor-
mance is strongly positive. In contrast, the marginal ‘effect’ of first-period per-
formance on second-period performance among firms that experienced turnover is
much weaker in absolute value, and may even be negative.
2.2 Alternative Scenarios
Do these implications of the model persist in the face of significant changes in
parameter values? Figures 5 through 7, which replicate Figure 4 for five alterna-
tive scenarios, suggest that the possibilities depicted so far are, in fact, quite
0.01
0.1
1
10
100
0.01 0.1 1 10 100
S
e
c
o
n
d
p
e
r
i
o
d
p
e
r
f
o
r
m
a
n
c
e
First period performance
FIGURE 4. Baseline simulations: First-period versus second-period
performance among surviving firms, by outcome. 5,000 observations.
Solid line indicates equal performance in both periods.
Replaced with less able CEO Founder remained
Replaced with more able CEO Firm exited
12
0.1
1
10
0.01 0.1 1 10
S
e
c
o
n
d
p
e
r
i
o
d
p
e
r
f
o
r
m
a
n
c
e
First period performance
Replaced with less able CEO Founder remained
Replaced with more able CEO Series4
FIGURE 5. Alternative scenarios I. First-period versus second-period performance
among surviving firms, by outcome. Upper panel: Ability matters more (less) than
idea quality in the first (second) period; a
1
=0.8, a
2
=0.2. Lower panel: Idea quality
matters more (less) than ability in the first (second) period; a
1
=0.2, a
2
=0.3. In
each case c is adjusted to maintain turnover rate of about 12%.
0.01
0.1
1
10
100
1000
10000
100000
0.0001 0.001 0.01 0.1 1 10 100 1000 10000
S
e
c
o
n
d
p
e
r
i
o
d
p
e
r
f
o
r
m
a
n
c
e
First period performance
Replaced with less able CEO Founder remained
Replaced with more able CEO Series4
13
0.1
1
10
100
0.01 0.1 1 10 100
S
e
c
o
n
d
p
e
r
i
o
d
p
e
r
f
o
r
m
a
n
c
e
First period performance
Replaced with less able CEO Founder remained
Replaced with more able CEO Series4
0.1
1
10
100
0.01 0.1 1 10 100
S
e
c
o
n
d
p
e
r
i
o
d
p
e
r
f
o
r
m
a
n
c
e
First period performance
Replaced with less able CEO Founder remained
Replaced with more able CEO Series4
FIGURE 6. Alternative scenarios II. First-period versus second-period performance
among surviving firms, by outcome. Upper panel: First- and second-period idea
qualities more highly correlated (standard deviation of distribution of q
2
reduced to
0.1). Lower panel: First-period idea quality and founder ability more highly corre-
lated (standard deviation of distribution of q
1
reduced to 0.1). In each case c is ad-
justed to maintain turnover rate of about 12%.
14
robust. In Figure 5, the relative importance of idea quality and ability is altered
across the two periods. In the upper panel, ability is most important in the first
period
1 2
( 0.8, 0.2), a a = = while in the lower panel it is more important in the
second
1 2
( 0.2, 0.8). a a = = The results reported in the baseline are not sensitive
to these significant changes in the production technology. Figure 6 sharply alters
the correlations between idea quality and ability. In the upper panel, the stan-
dard deviation of the distribution of
2
q is reduced by 80 percent, to 0.1, while
maintaining its mean equal to the realized value of
1
q ; the lower panel similarly
increases the correlation between q and q
1
by reducing the variance of the distri-
bution of
1
. q
The three groups of surviving firms become even more distinct than
in the baseline, but the original pattern is preserved. Finally, in Figure 7 the cost
of transitioning to a new CEO is reduced sharply, so that the fraction of firms
that experience turnover is increased from about 12 percent to 35 percent. Again,
the baseline results are robust to this change.
10
10
Our data do not afford us an opportunity to explore the effects of changes in the tran-
sition cost. However, Braguinsky et al. (2012) explore the effects of a likely decline in the
0.1
1
10
100
0.01 0.1 1 10 100
S
e
c
o
n
d
p
e
r
i
o
d
p
e
r
f
o
r
m
a
n
c
e
First period performance
Replaced with less able CEO Founder remained
Replaced with more able CEO Series4
FIGURE 7. Alternative scenarios III. First-period versus second-
period performance among surviving firms, by outcome. Transition
cost reduced to c=0.1, increasing founder turnover rate to about
35%.
15
2.3 Related models
There are, of course, different ways one may specify a model of mismatching. In
this subsection we briefly compare our theoretical framework with three related
models. The best known of these is Holmes and Schmitz (1995), in which agents
vary in their ability to produce ideas, although all are equally gifted at managing.
Agents with low innovative ability prefer to buy and manage existing businesses,
while agents with high ability prefer to specialize in business creation (selling or
closing each business they create). The outcomes for founders with intermediate
ability depend on the quality of the business idea: they will close the worst busi-
nesses and manage the best. A key determinant of the boundaries between the
choices is the cost of transferring a business. Figure 8 illustrates. Clearly, only the
better ideas produced by the more able innovators experience founder turnover.
cost of evaluating ideas in Japanese biotech firms, brought about by institutional reform.
They find that a reduction in evaluation cost was associated with a large increase in the
rate of founder turnover.
Transfer business and
start another one
Founder continues
to operate firm
Close business and
find another to
manage.
Close business and
start another one
q
q
FIGURE 8. Holmes and Schmitz (1990)
16
As managerial ability is fixed, the model predicts that business turnover is asso-
ciated with high initial and subsequent firm performance, but it does not predict
any change in performance after turnover.
In Braguinsky et al. (2012), entrepreneurs with known ability found businesses
with unknown quality. As in our model, business quality and founder ability are
complements in production. The quality remains unknown unless the founder
chooses to pay an evaluation cost, c. High-ability agents know that they are at
risk of wasting their time with a poor idea, so they choose to evaluate the idea,
closing down business that are not good enough. Low-ability agents know that
they may not be good enough mangers if their business quality is good; they also
pay the evaluation cost and either close their business or transfer control to a
professional manager. Agents with intermediate ability do not pay the evaluation
cost; they continue to manage a business of unknown quality (see Figure 9). The
model unambiguously predicts that only the better businesses founded by low-
ability managers are at risk of founder turnover, and that turnover elicits an im-
provement in performance. Auxiliary assumptions of the model associate better
initial firm performance (i.e., the combination of business quality and founder
ability) with turnover.
Founder continues
to operate firm
Close business
q
q
Replace founder-
CEO
q
q
Close business
FIGURE 9. Braguinsky et al. (2012)
17
Finally, Holmes and Schmitz (1995) do not have explicit variations in skill. In
this model, output is given by
t t t
y q m = + , where q is business quality and m is
match quality. Both q and m evolve over time, each consisting of permanent and
transitory components. The permanent component of business quality, b, is fixed
when a business is first founded, while the permanent component of match quali-
ty, m, is fixed each time a new manager takes control of the business. Founders
(and subsequent managers) who experience a sufficient decline in match dispose
of their business, either by closing it (if the business quality is sufficiently low) or
by selling it to a new manager (if the quality is sufficiently high). Figure 10 illu-
strates, with a sample path depicting the evolution of business and match quali-
ties. A business begins at point a, and suffers a steady decline in match quality.
At b the business quality is sufficiently high to enable a sale. The new manager
begins with the same business quality but a better match quality, at c, but then
suffers a decline in both business quality and match quality until exit at d.
Holmes and Schmitz show that the boundary between closing and selling a busi-
ness shifts down when the permanent business quality, b, increases. Hence, better
quality businesses are more likely to be sold. However the match quality may
Sell business
Founder continues
to operate firm
Close business
q-b
m-m
a
b
d
c
FIGURE 10. Holmes and Schmitz (1995)
18
be little better after a sale, so despite better average performance after business
turnover, there can be an increased risk of quick failure.
11
It is not possible to re-
late founder ability to the likelihood of business turnover in this model unless we
impose some functional relationship between ability and the pair { , }. m q Doing
so, however, would be unfair to the Holmes and Schmitz model, although it
strikes us that their model could be made consistent with turnover occurring in
the tails of the firm performance distribution.
3. Data and Sample Construction
Our empirical analysis focuses on all new business founded in Denmark in 1999
and 2000. We construct our samples out of three databases maintained by Statis-
tics Denmark: the Entrepreneur Database, the Firm Database, and the Inte-
grated Database for Labor Market Research (IDA).
The Entrepreneur Database is the primary source for identifying new businesses
(partnerships or sole proprietorships) registered in Denmark each year, and pro-
vides unique identifiers for each firm, each plant, and one individual identified in
the registration documents. The Entrepreneur Database spans the period 1996 to
2006. However, accounting information is not collected for all industries until
1999, so this is the earliest cohort we study.
There are several limitations to the Entrepreneur Database. First, since only one
person associated with the new business is recorded, the database provides no
information for identifying possible co-founders. Second, some of those who ap-
pear in the database may not be the “real” founder. For example, there are cases
in which founders used their spouse’s name to register a business. Third, it is un-
clear from the Entrepreneur Database whether or not the identified entrepreneur
actually works at their startup. Although it is possible that entrepreneurs do not
operate businesses themselves, this study focuses on those who were actually
working for their own businesses. In order to identify these entrepreneurs and
possible co-founders, the data obtained from the Entrepreneur Database are
combined with information provided in the Firm Database.
The Firm Database consists of annual employment information on all workers
11
Figure 5 depicts the case where the m is common across matches. If m is lower after
business transfer, the boundaries shift to the right, increasing the risk of business failure.
19
(full-time and part-time) at all firms operating in Denmark. The variable of par-
ticular interest is their position in the firm, through which founders of each new
business are identified in the following way. First, founders are those whose posi-
tions are classified as self-employed, employers, or business tax payers. Second,
for businesses with fewer than four employees, people whose positions are top
managers are also defined as founders.
12
Third, if nobody in a business with fewer
than four employees is identified as a founder based on the previous two criteria,
all employees are considered as cofounders. The fourth group of founders consists
of those who registered the business, were working for the business, but did not
meet the first three criteria for being a founder. Lastly, we identify individuals
who registered the business, but did not appear to be working for the new firms
(we will exclude this last category from the sample).
We link these businesses to the IDA, where more detailed information such as
industry classification and employee demographics is available. The IDA is an
employer-employee matched database, which provides mainly employment infor-
mation at firm, establishment, and employee levels. The IDA enables us to track
over time the turnover of founders, changes in firm status, and occupational
choices of individuals each year until 2008. However, because of incomplete ac-
counting and earnings data in the most recent years, we will follow firms only
through 2004 and individuals through 2005.
Firm identifiers in the IDA are attached to a specific owner, so they change
whenever there is a change of ownership. This design makes it possible to distin-
guish three firm-level events in each year. Survival is assumed if the same firm
identifier from the previous year appears in the current period. An ownership
change occurs if the same establishments of the firm from the previous year are
identified in the current period except that they are associated with a different
firm identifier. A business exit is identified if all the establishments of the firm
can no longer be found in the data.
In our analysis, the original founder of a startup must meet the following criteria:
(1) He or she formed a startup in 1999 or 2000 as a sole proprietor; (2) The per-
son was currently working at the new venture; and (3) The new business could be
identified in the Firm Database and the IDA database, indicating that at least
12
This way to identify founders was first used by Sørensen (2007).
20
one person was working in the business as his or her primary job.
13
Using the
unique person and firm identifiers, we can track these founders’ employment in
the following year to identify whether or not the founder was still working at the
startup, and whether he was still an employer. A founder was staying at the star-
tup if he could be identified in the firm in the following year, whether as an em-
ployee or as an employer. Otherwise, the founder is assumed to have left the orig-
inal firm because of turnover, business exit, or ownership change. In a few cases,
the founder might stay at the startup after a change of ownership, but in each of
these cases a new firm identifier is assigned to the startup.
Table 1 provides tabulations of changes in the status of startups and their found-
ers from the founding year to 2005. Four outcomes are possible in each year: the
firm exits, the firm is acquired or the firm continues, either with or without the
founder remaining in operational control. Of 4,172 firms in the sample, a total of
533 (12.8 percent) of them eventually experienced founder turnover. Turnover
was twice as common as ownership change (6.5 percent), but much less common
than firm exit (50.6 percent). The hazard of turnover declines with firm age, de-
clining from 5.11 percent in the first year, to 2.61 percent at age five.
14
4. Firm Performance and Founder-CEO Replacement
In this section, we evaluate the relationships between startup performance and
subsequent rates of founder turnover, and between founder turnover and subse-
quent firm performance.
4.1 Startup performance and founder-CEO replacement
To assess the effect of initial performance on subsequent rates of founder-CEO
replacement, we remove from the baseline sample 2,388 startups (918 from the
1999 cohort, and 1,470 from the 2000 cohort) that exited or were acquired before
13
This person may be the founder, so our sample includes single-person firms. We have,
however, repeated our analyses after restricting the sample to including firms with at
least two individuals. The results, available from the authors, are very similar to those
reported here.
14
The declining hazard of turnover is consistent with gradual learning about match quali-
ty [Jovanovic (1979)], which we have not attempted to model.
21
TABLE 1
Status Change of Startups by Year
Years after Founding
1 2 3 4 5 6*
Count
Surviving: founder stays 2,996 2,247 1,829 1,535 1,321
Surviving: founder turnover 213 127 82 54 40
Firm exit 898 550 291 193 140
Firm acquired 65 72 45 47 34
Total 4,172 2,996 2,247 1,829 1,535
Hazards (percentage)
Surviving: founder stay 71.81 75.00 81.40 83.93 86.06 87.16
Surviving: founder turnover 5.11 4.24 3.65 2.95 2.61 3.12
Firm exit 21.52 18.36 12.95 10.55 9.12 7.71
Firm acquired 1.56 2.40 2.00 2.57 2.21 2.02
Total 100 100 100 100 100 100
* Percentages for 1999 cohort only.
any occurrence of a founder turnover during the observation window between the
founding year and 2005. Table 2 provides descriptive summary statistics for the
remaining 1,784 startups and their founders. There are no large differences in
means between founders that continued and founders who were replaced. Howev-
er, the latter group are somewhat younger and less likely to have had vocational
training; they are also more likely to come from the tails of the earning distribu-
tion (either less than high school or at least college); they are also somewhat less
likely to have founded businesses in construction, and more likely to be engaged
in high-tech activities.
We first estimate the following logit model,
2
1 1 2 1 , 1 ,
ln ( )
1
t
t t i i t i t
t
p
logsale logsale X Z
p
a b b g d e
- - -
æ ö
÷ ç
÷
ç = + + + + +
÷
ç
÷
÷ ç -
è ø
(7)
22
where p
t
is the probability that the founder leaves a continuing firm in period t,
is a vector of time-invariant variables related to founder demographics (gend-
er, marital status, age, and education) and startup founding conditions (founding
year and industry), and
,
is a vector of potentially time-varying variables,
such as firm age, founder age, marital status, and highest completed education.
Table 2
Descriptive Summary of Subsample 1
All No Turnover Turnover
Age (mean years) 40.07 40.23 37.91
Male (%) 0.76 0.76 0.77
Married (%) 0.64 0.64 0.54
Education (%)
less than high school 16.84 16.55 21.01
exact high school 6.66 6.34 11.07
vocational training 50.91 51.75 39.21
some college 14.72 14.66 15.57
bachelor 1.13 1.02 2.63
Master 7.66 7.66 7.69
Phd 0.69 0.69 0.75
others 1.38 1.33 2.06
Founded in 1999 (%) 42.28 42.69 36.59
Initial sales (1,000 DKK) 2,149 2,135 2,347
Initial no. of full-time employees 1.87 1.85 2.09
Industry (%)
Agriculture 0.29 0.28 0.38
Manufacturing 7.25 7.34 6.00
Construction 17.57 17.86 13.51
Wholesale and retail 37.57 37.55 37.9
Transport, post, telecom 4.89 4.90 4.69
Low-tech businesses 6.93 6.93 6.94
Public Services 0.43 0.42 0.56
Personal Services 6.05 6.26 3.19
Kibs 0.44 0.34 1.88
High-tech 18.24 17.77 24.77
Others 0.34 0.35 0.19
Obs. 7,961 7,428 533
23
The key independent variable is the logarithm of sales in the previous period,
along with its quadratic to allow for the possibility of a non-monotonic effect as
predicted in Section 2.
15
Columns (1) and (2) of Table 3 present the results. In
column (1), we include controls only for firm age and founding year. The nega-
tive coefficient on firm age reflects the declining hazard of founder-CEO replace-
ment already noted in the summary statistics, while the cohort dummy is insigni-
ficant. In column (2) we add some founder characteristics. The probability of
founder-CEO replacement declines with founder age, it is lower for founders with
college education, and it is lower for men and married founders. Both columns
show a U-shaped relationship between the log of sales in the previous period and
the likelihood of founder turnover in the current period, indicating that a founder
is more likely to rescind operating control of the firm if it is located the tails of
the performance distribution among all surviving firms. Columns (3) and (4) re-
place sales in the previous period with sales in the startups’ first year of opera-
tion. The results continue to show the U-shaped relationship, and the same ef-
fects of founder characteristics and firm age on the likelihood of founder-CEO
replacement. The minimum of the U-shaped relationship between prior perfor-
mance and the probability of turnover occurs at revenue levels of DKK580,000 to
DKK800,000, equal to 15 to 37 percent of the sample mean. This is also well
within the sample range. Consistent with the predictions of our simple mismatch-
ing model, therefore, we find that turnover is more common in the tails of the
observed earnings distribution than in the middle.
4.2 Founder-CEO replacement and the subsequent performance of startups
To examine the relationship between founder-CEO replacement and subsequent
firm performance, we use a subset of startups in the baseline sample and limit
the observation window to the interval between the founding year and 2004. The
startups included in this subsample meet two criteria. First, we eliminate firms in
which the founder was replaced after 2002. That is, firms included in the sample
either experienced founder turnover by 2002, or never had founder turnover dur-
ing the observation period. Second, we remove startups that failed or were ac-
quired between the initial founding year and 2002 but prior to any observed
15
For six observations with zero sales, we added one DKK to sales prior to taking the
logarithm.
24
TABLE 3
The Effect of Startup Performance on the Probability of Founder-CEO Replacement
Dept Var: = 1 if founder replaced in year t
Logit Reegression
Using sales in year t-1 Using sales in founding year
(1) (2) (3) (4)
Log of sales -0.940
***
-0.966
***
-1.159
***
-1.270
***
(-5.08) (-5.10) (-3.30) (-3.43)
(Log of sales)
2
0.072
***
0.076
***
0.084
***
0.095
***
(5.71) (5.87) (3.36) (3.61)
Founder characteristics
Founder age
__
-0.017
***
__
-0.019
***
(-3.05) (-3.40)
College educated = 1
__
-0.235
*
__
-0.090
(-1.38) (-0.54)
Male = 1
__
-0.154
__
-0.104
(-1.28) (-0.86)
Married = 1
__
-0.331
***
__ -0.303
***
(-3.25) (-2.98)
Firm characteristics
Firm age -0.366
***
-0.340
***
-0.324
***
-0.288
***
(-10.39) (-9.47) (-9.68) (-8.46)
Cohort dummy (= 1 if
founding year = 2000)
0.130 0.10 3 0.095 0.069
(1.34) (1.04) (0.96) (0.69)
Controls for industry No Yes No Yes
Ave Log Likelihood
-0.239 -0.235 -0.241 -0.237
No. of observations
7,357 7,348 7,305 7,305
Sales in thousand DKK. z-scores are in parentheses. Significance levels: *** 0.01, ** 0.05,
* 0.1.
25
founder turnover event. However, we retain in the subsample startups that expe-
rienced founder turnover and then exited or were acquired. Table 4 illustrates
how we construct the subsample.
Starting with the 1,588 startups in the 1999 cohort, we removed 709 firms that
exited or went through ownership change by the end of 2002 and before any oc-
currence of founder turnover, and 36 firms that experienced founder turnover in
2003 or 2004. For the 2,584 startups of the 2000 cohort, we removed 995 and
then 83 of them for the same reasons. We are left with a sample of 2,349 star-
tups, 843 of which come from the 1999 cohort, and 1,506 from the 2000 cohort.
Table 4
Firms Included in Subsample 2
Turnover by
2002
Turnover after
2002
No Turnover
Exited or acquired by 2002 ? × ×
Exited or acquired after 2002 ? × ?
Survived to 2004 ? × ?
Table 5 summarizes the number of startups in our sample that survived, exited,
or experienced ownership change during the observation period. Among the 843
startups founded in 1999, 142 of them had founder turnover before 2002. Among
them, 50 survived to 2004, 56 exited or changed owners before 2002 but after
founder turnover, and 36 exited after 2002. Among the 1,506 startups founded in
2000, 232 had founder turnover before 2002. Of these, 82 survived the entire ob-
servation period, 70 exited or changed owners before 2002, and 80 exited after
2002.
Table 6 provides the descriptive summary of this subsample (the distributions of
firms in each column across industries is very similar to those shown in Table 2,
and are not reported). The summary statistics show two interesting contrasts be-
tween firms that experienced turnover and those that did not. Turnover by 2002
is strongly associated with lower odds of survival to 2004. Only 35 percent of
firms with turnover survived, compared with 74 percent of firms with no turno-
ver. However, conditional on survival, the growth rate of sales and of employ-
ment is much greater among firm that experienced turnover. For example al-
26
though there is little difference between the two groups in initial sales revenue,
average 2004 sales are 55 percent higher among the turnover group.
Table 5
Survival, Exit, and Ownership Change after Founder Turnover
1999 Cohort
2000 Cohort
Turnover No Turnover
Turnover No Turnover
Total 142 701
232 1,274
Survived 50 545
82 912
Exited before 2002 56 0
70 0
Exited after 2002 36 156
80 362
Table 6
Descriptive Summary of Subsample 2
All No Turnover Turnover
Obs. 2,349 1,975 374
Survival in 2004 (%) 67.65 73.77 35.29
Founded in 1999 (%) 35.89 35.49 37.97
Initial sales (1,000 DKK) 2,008 1,962 2,253
Sales in 2004 (1,000 DKK) 5,168 4,929 7,639
Initial no. of full-time employees 1.68 1.62 2.00
No. of full-time employees in 2004 4.08 3.86 6.39
We estimate the following model
,2004 1 ,0 2
ln( ) ln( )
i i i
sales sales turn a b b = + +
3 ,0 ,2004
ln( )
i i i i
sales turn X b g e + ´ + + , (8)
where turn
i
equals one if the startup experienced founder turnover between its
founding year and 2002, and zero otherwise, and the vector
includes controls
for cohort and industry.
Table 7 presents OLS estimates of (8). In the first three columns, we focus on
27
startups that survived until 2004. Column (1) shows that both initial sales and
founder turnover have a positive impact on startups’ sales in 2004. Interacting
the two factors in column (2), we further find that the marginal effect of founder
turnover on future sales is decreasing with the initial performance of startups (or,
equivalently, that the marginal effect of initial performance is lower for firms that
had turnover). This is consistent with the model in Section 2. However, this
Table 7
Founder-CEO Replacement and Future Performance of Startups
Dept Var: log of sales in 2004
Surviving Firms
OLS
All Firms
Tobit
(1) (2) (3) (4) (5) (6)
Log.sale
0
0.437
***
0.453
***
__ 0.724
***
0.817
***
__
(18.8) (18.6) (5.92) (6.12)
Turnover 0.336
***
1.524
***
__ -5.00
***
-1.032 __
(4.15) (2.80) (-14.4) (-0.45)
Log.sale
0
*turnover __ -0.165
**
__ __ -0.552
*
__
(-2.21) (-1.73)
log.sale
0
percentile:
Brief explanation concerning new firm performance and the replacement of founder ceos jing chen.
New Firm Performance and the
Replacement of Founder-CEOs
Jing Chen*
Copenhagen Business School
Peter Thompson**
Emory University
November 2012
In this paper we study some causes and consequences of founder-CEO replace-
ments among a sample of 4,172 Danish startups created by single founders in
1999 and 2000. In contrast to the extant literature on VC-financed firms, re-
placements among firms in our sample are more likely among the worst- and
best-performing firms, and replacement is not unambiguously associated with
better subsequent performance. Firms that replaced the founder as CEO were
much more likely to fail, but the surviving firms among them grew considerably
faster. We also analyze subsequent earnings and occupational choices of founders
who rescinded operating control of their firms. Although subsequent founder in-
come is increasing in the performance of their firms, those that left good firms
were no more likely than others to found another business. Our results are consis-
tent with the notion that founder-CEO replacement is driven in part by mis-
matches between business quality and founder ability.
JEL Classification Codes: D21, D22, J62, L26
Keywords: Entrepreneurship, business creation, founder turnover, CEO turnover.
* CBS - Department of Innovation and Organizational Economics, Kilevej 14, 2000 Fre-
deriksberg, Denmark. Email: [email protected]. ** Goizueta Business School, Emory Universi-
ty, 1300 Clifton Road, Atlanta, GA 30322, USA. Email: [email protected]. We
appreciate helpful comments from: Christian Catalini, Alberto Galasso, Mirjam Van
Praag, participants in the 2012 Summer Residence Week for Entrepreneurship Econo-
mists at Oxford University, the Roundtable for Engineering Entrepreneurship Research at
Georgia Tech, and in seminars at the University of Amsterdam, the University of Lugano
and Emory University.
1
1. Introduction
How does firm performance influence the likelihood that founder-CEOs of new
businesses cede operating control to a specialist manager? What implications does
the departure of the founder have for subsequent business performance? And
what happens to founders after they have left? This paper reports the results of
an analysis of the causes and consequences of founder-CEO replacement among a
cohort of new businesses founded in Denmark. Our empirical work uses the
matched employer-employee database maintained by Statistics Denmark, and
draws samples from the universe of new businesses created in Denmark during
1999 and 2000. These firms are generally small, they are almost invariably not
financed by venture capital, and they are generally not at risk of an IPO event.
We know surprisingly little about founder turnover
1
among such firms.
A growing literature has provided us with substantial insight into the causes and
consequences of founder-CEO replacement in other types of firms. This literature,
which has focused on high-growth and VC-backed firms
2
, finds that founder-CEO
replacement is most likely to occur after a period of rapid expansion [e.g., Boeck-
er and Karichalil (2002)], or upon attainment of critical milestones such as com-
pletion of product development or securing new rounds of outside funding [Was-
serman (2003)]. More important, perhaps, is that founder-CEO replacement is
typically motivated by an expectation among stakeholders that the firm is likely
in the near future to undergo an episode of transformational growth. Indeed, a
significant portion of the literature explicitly samples on such expectations by
studying only firms that are preparing for an IPO [e.g., Nelson (2003), Jain and
Tabak (2008), Pollack, Fund and Baker (2009)].
The extant research lacks findings that we can confidently extrapolate to the
firms that are the focus of the present paper. First, VC-financed firms appear to
1
We will use the terms founder-CEO replacement and founder turnover interchangeably
to describe an event where the founder of a firm, who had been managing the firm as his
or her primary occupation, rescinds operating control while maintaining ownership rights.
2
Of course, executive turnover in established, typically public, firms has been extensively
studied for many years [e.g., Hofer (1980), Chen and Hambrick (2012)]. There is also an
active literature on succession in family-owned businesses, but the concerns of this litera-
ture are rather different from those in which executive control is passed outside the
founder’s family. See Handler (1994) for a review.
2
exhibit founder-CEO replacements at much higher rates than firms that are not
subject to significant control by outside equity investors [Hellman and Puri
(2002), Boeker and Wiltbank (2005)]. Second, few of the firms in our sample are
ever likely to undergo transformations of the sort that appear to motivate many
founder-CEO substitutions among VC-financed firms. Third, VC-financed firms
are not random draws from the firm population. They are more likely to grow
substantially, they are better financed, and their founders and employees on av-
erage have more human capital. Correlations between variables estimated from a
sample drawn from the upper tail of the quality distribution may not hold in
other segments of the population, especially when there are possibilities of non-
monotonic relationships in the population as a whole.
It is well-known that VC-financed firms account for a tiny fraction of business
launches
3
but they play a much larger role in economic growth, employment crea-
tion and innovation. As a result, academic effort to understand the dynamics of
control in VC-financed firms is appropriately disproportionate to their numbers.
Nonetheless, given their numerical importance, we should also be concerned with
the control dynamics of the typical small business. Patterns of founder turnover
among representative businesses also provide insight into the forces governing
business transfers.
4
They are also key to understanding the expected opportunity
costs of business creation, as well as the dual phenomena of serial and portfolio
entrepreneurs.
In this paper we study founder-CEO replacements among a sample of 4,172 Da-
nish startups created in 1999 and 2000. To ensure that the departure of the
founder indicates a meaningful change in control, the sample consists only of
firms that were created by a single founder, and where the founder worked as his
primary job in the business. The firms and founders we study represent all firms
founded in Denmark that meet these criteria. Only 18 percent of them are classi-
3
For example, in the United States, 626,771 new businesses were created in 2008, only
3,276 of which received VC financing (Bureau of labor Statistics athttp://www.bls.
gov/bdm/entrepreneurship/bdm_chart1.htm; and NCVA (2009).
4
Founders may sell businesses because they are poorly matched to the needs of the busi-
ness, or because they have reasons to cash out their equity position. Looking at founder-
CEO replacements among businesses that are not transferred, allows us to gain insights
about the first of these two motivations.
3
fied as high-tech, while wholesale, retail, construction and manufacturing are
well-represented; this is a much broader sample than has previously been used to
study founder-CEO replacement. Our data are drawn from Denmark’s national
matched employer-employee dataset, so we are also in a unique position to ex-
plore what happens to the founders after they have left the firm. We study in-
stances of founder-CEO replacement and we track the firms and founder until
2005. The data allow us to distinguish business turnover from founder turnover
and, in this paper, we focus on the causes and consequences of events in which
the founder retires from active management but continues to retain controlling
ownership. That is, our focus is on CEO succession in small firms, rather than
business transfers.
5
To guide our empirical work, we develop a simple model of firm creation, surviv-
al, and founder turnover. The goal of founder-CEO replacement among VC-
backed firms is generally held to be resolving mismatches between the skills of
the founder-CEO and the expected demands of the transforming enterprise. We
also develop a framework based on mismatches between business and founder,
although our model focuses on current rather than anticipated mismatches. We
suppose that firm performance is a function of the ability of the founder and the
quality of the business idea, both of which are draws from stable population dis-
tributions. Ability and quality are complements in production so that there can
arise mismatches of sufficient magnitude to justify replacing the founder with an
outside CEO who is better suited to the business.
6
Our theoretical framework motivates us to consider the possibility that founder-
CEO turnover is most likely in the tails of the initial performance distribution,
and that, while good early performance is on average associated with good future
performance, this effect is stronger in firms that had no turnover. What we find
in the data is a little more complex than this. We do find that turnover is more
likely in the tails of the firm performance distribution and, conditional on firm
5
The classic empirical study of small business transfers is Holmes and Schmitz (1995).
6
Our theoretical framework is related to the models of business transfers developed by
Holmes and Schmitz (1990, 1995), and to the model in Braguinsky et al. (2012), which
explores the effect of the cost of evaluating ideas on founder turnover. We shall comment
further on the relationship between our theoretical framework and these models in the
next section.
4
survival, turnover is associated with higher rates of growth. However, turnover is
also associated with higher firm exit rates. We also explore the earnings and oc-
cupational choices of founders who have relinquished day-to-day control of their
firms. Our framework and our empirical results are both consistent with the in-
tuitive notion that founders who were replaced in high-performing businesses
subsequently earn more than those who left low-performing businesses but, less
intuitively perhaps, they are no more likely to create another startup.
2. Theoretical Framework
There are two periods. In the first period, a founder with ability q has an idea of
quality q
1
. Ability is general and determines both the success with which the
founder can manage a firm that implements the idea, and his earnings in alterna-
tive employment. We assume that founder ability is innate, fixed over time and
known to the founder. Idea quality incorporates a variety of factors influencing
performance, including the quality of the underlying idea or technology, its adap-
tation and fit to the market, and the public’s state of knowledge about the firm’s
product. Some of these factors may be regarded as fixed, while others evolve over
time as the business model is refined. Upon entry, the founder knows ?, but he
only knows that q
1
is a random draw from the distribution
1
( | ). F q q The realiza-
tion of q
1
is observed after entry. The business idea is refined and adapted to the
market, yielding a new quality q
2
in the second period, which is drawn from
( )
2 1
| . G q q
In each period, t = 1, 2, output is given by
1
t t
t t
y q
a a
q
-
= , (1)
for some [0, 1].
t
a Î Although, q and q are assumed to be complements in both
periods, we allow for the relative importance of ability and idea quality to vary
from one period to the next.
Profit in period 1 is given by
1 1
1
1 1
q w
a a
p q q
-
= - . (2)
The only cost is the opportunity cost of the founder’s time, which is assumed to
be linearly increasing in ability. For concreteness, we shall suppose this alterna-
5
tive use of the agent's time is in wage work.
At the end of the first period, q
2
is observed and a decision is then made whether
to contract with an outside CEO to replace the founder, to continue without
change of leadership, or to exit. If the outside CEO has ability q
?
, she must be
paid a wage . wq
?
If a replacement CEO is hired, a transition cost, c, must be
paid. This cost reflects both direct costs of any disruption associated with a
change in leadership, as well as any premium necessary to resolve agency prob-
lems arising from the separation of ownership and control [Fama and Jensen
(1983), Jensen and Meckling (1976)].
7
The rewards to recruiting an outside CEO
are that the CEO’s ability can be chosen optimally in light of the realization of
q
2
, and the founder is released to earn wq elsewhere while remaining as the resi-
dual claimant on the firm’s earnings. The founder is replaced by an outsider if
( )
2 2 2 2
1 1
2 2
arg max q w c q w
a a a a
q
q q q q
- -
- - > -
?
? ?
, (3)
and if the term on the left hand side of (3) is positive. The profit-maximizing
ability of the external CEO is given by
( )
2
1/(1 )
2 2
/ w q
a
q a
-
=
?
, (4)
so maximized profits in period 2 are
( )
2 2
1 *
2 2 2
max , q w Aq c
a a
p q q
-
= - - , (5)
where
2 2
/(1 )
2 2
(1 )( / ) . A w
a a
a a
-
= - If
*
2
0, p < the firm exits.
The two terms on the RHS of (5) are plotted separately in Figure 1 for different
values of q
2
while holding q constant. The concave function
2 2
( , ) q p q indicates
profits when the founder continues, and the straight line
2 2
( ) q p? depicts profits
when the founder is replaced. In the case illustrated in the figure, there are four
distinct regions defined by the boundary values 0 :
E
q q q < £ £ if
2
[0, ],
E
q q Î the
firm exits at the end of period 1; if
2
[ , ]
E
q q q Î or if
2
, q q ³
the founder is re-
placed as CEO; and if
2
( , ), q q q Î the founder continues to operate the firm.
7
He (2008), for example, documents that founder CEOs have lower incentive compensa-
tion and lower total compensation than professional CEOs.
6
When
2
[ , ]
E
q q q Î , the founder is replaced because the opportunity cost of his
time does not justify continued commitment to the business model; he is replaced
by an external CEO who is less able than the founder. In contrast, when
2
q q ³
the founder is replaced by an external CEO with higher ability.
Figure 1 also illustrates the effect of an increase in the founder’s ability, from q to
. q
¢
This reduces the profitability of the founder continuing with the firm when
the quality of the idea is low, and increases it when the idea is good. It is easy to
verify that the intersection of
2 2
( , ) q p q and
2 2
( , ) q p q¢ lies between q and , q and
hence that the effect of an increase in q is to raise both q and . q The interval
[ , ]
E
q q will vanish if the founder’s ability declines sufficiently so that .
E
q q <
There are, therefore, two distinct cases. In one case, when founders have limited
ability, the period 2 outcomes fall into three distinct groups, with replacement of
founders occurring only at higher values of q. In the second case, founders are
replaced in firms with business models drawn from the tails of the quality distri-
bution among all surviving firms. Which of these cases arises depends on the
founder’s ability, as well as other parameters of the model. For example, an in-
crease in the transition cost, c, shifts
2 2
( ) q p? down, which may eliminate the in-
terval [ , ];
E
q q
that is, if the increase in c is large enough, replacement of founders
occurs only at high values of q
2
and in each case the replacement has higher abili-
ty than the founder. Clearly, the minimum value of c that eliminates the interval
E
q
q
q
2
q
0
2 2
( ) q p?
2 2
( , ) q p q
c -
wq -
wq¢ -
2 2
( , ) q p q¢
q q
¢
>
FIGURE 1. Second period payoffs; c>0.
7
[ , ]
E
q q
depends on the founder’s ability.
8
Figure 2 provides a more complete depiction of the model’s implications. Firms
that are closed down are those with low-quality ideas. If the founder does not
want to continue to operate the firm, continuation requires a new CEO. Howev-
er, payment of the transition cost cannot be justified in low-quality firms. The
boundary separating firms that are discontinued from those that are continued
with the founder at the helm is given by the pairs
2
{ , } q q satisfying
2
1/(1 )
2
, q w
a
q
-
= showing that low-ability founders may continue to operate a busi-
ness that a high-ability founder would close down. Founder turnover in continu-
ing firms occurs in two distinct groups. First, there is a group of firms with high-
quality ideas but that were founded by agents of sufficiently low ability to merit
replacement with a more able CEO. The boundary between this group and the
group of founders that continue to operate their own firm is defined by the func-
8
When 0, c > it is always the case that . q q < In the limiting case of 0, c = the func-
tion
2 2
( ) q p?
is then tangential to
2 2
( , ) q p q for any value of q. and . q q ? In this case,
every founder is replaced, because it is always efficient to secure a perfect match between
CEO ability and idea quality.
FIGURE 2. Ability, idea quality, and outcomes.
q
2
(log scale)
q
(log scale)
Replace with
less able CEO
Replace with more
able CEO
Founder continues
to operate firm
Firm exits and founder
enters wage employment
2
1/(1 ) 1
A w
a - - -
1
2
q cA
-
=
2
1/(1 )
2
q w
a
q
-
=
( ) q q
( ) q q
8
tion
( ) q q that provides the larger of the solutions to the equation
2 2
1
2 2
. Aq c q w
a a
q q
-
- = - As the transition cost rises, ( ) q q unambiguously shifts
upwards. The second group consists of firms with able founders but whose firms
are of middling quality. The quality of the idea is high enough to prefer continua-
tion over exit, but not high enough to merit continuation of the founder. The re-
placement CEO has lower ability than the founder. The boundary between this
group and those that exit is a horizontal line at
1
2
. q cA
-
= The boundary be-
tween this group and the group of founders that continue to operate their own
firm is given by the function ( ) q q that provides the smaller of the solutions to
2 2
1
2 2
Aq c q w
a a
q q
-
- = - . An increase in the transition cost shifts ( ) q q downwards.
Figure 2 shows that ( ) q q and ( ) q q are both positively sloped, which is consistent
with the discussion around Figure 1.
2.1 Baseline simulations
The empirical implications of our theoretical framework depend critically on the
joint distribution of q
2
and q (and, by virtue of the correlation between q
1
and q
2
,
on the joint distribution of initial idea quality and founder ability). For example,
the implication that founder turnover occurs both when q
2
is high (relative to q)
and when it is low requires that we observe in any sample instances where con-
trasting values of q
2
and q coexist. Similarly, the transition cost cannot be so high
that it eliminates the interval [ , ].
E
q q Because this may not be the case, and de-
spite the simplicity of the theoretical framework, there are no general empirical
predictions about the relationships between performance in the two periods,
founder ability and turnover. We will therefore make use of some numerical simu-
lations to explore the possibilities of the model.
However, it is useful to consider some reasonable ways in which we can limit the
varied possibilities. We do so in two ways. The first imposes a constraint on the
transition cost, c, such that the model predicts an aggregate rate of turnover that
approximates the rate (about 12 percent) that we observe in our data. The
second makes use of a simple free-entry condition to help us constrain the rela-
tionship between ability and the quality of the idea. We want to admit entrepre-
neurs of all abilities. This will be the case if all agents, regardless of ability, are
indifferent between entrepreneurship and wage work in period 1. If the quality of
ideas and agent ability are independent of each other, then high-ability agents,
who bear a higher opportunity cost of becoming an entrepreneur, will find busi-
9
ness creation less attractive than low-ability agents. We will therefore assume
that idea quality and agent ability are positively correlated. For our purposes, it
suffices to suppose that ability and agents care only about expected first-period
returns. In the simulations to follow we therefore impose the constraint
9
1 1
1 1
1
, E q w
a a
q
- -
é ù
=
ê ú
ë û
(6)
which ensures that all agents, regardless of ability, are indifferent between creat-
ing a business and wage employment.
In our baseline model, we assume that in both periods ability and the quality of
the ideas are equally important. That is, we set
1 2
0.5. a a = = We assume that
1
1 a
q
-
has a lognormal distribution, with unit mean and standard deviation of 0.5.
Consistent with (6), we then assume that, for agent i,
1
1
1i
q
a -
is a draw from a log-
normal distribution with mean
1
1
.
i
w
a
q
-
Finally, consistent with the assumption
that
2 1
( | ) G q q is decreasing in q
1
, we assume that
2
1
2i
q
a -
is a draw from a lognor-
mal distribution with mean
1
1
1
.
i
q
a -
The wage, w, is normalized to 1. Finally, we
set 0.35, c = so that our baseline produces an aggregate rate of founder turnover
of about 12 percent, consistent with evidence we shall show later.
Figure 3, which plots 5,000 random draws using the baseline parameters, depicts
outcomes that confirm the boundaries already given in Figure 2. Although found-
ers of exiting firms on average have relatively low ability, this is incidental and
driven by the correlation between idea quality and founder ability. Indeed, once
one conditions on idea quality, low-ability founders are less likely to close down
the firm. The simulations yield observations in the two sets of firms with suffi-
ciently good ideas but poor matches between idea and founder ability to merit
replacement with a more able CEO. Although founder CEO’s of all abilities may
suffer a mismatch sufficient to induce their replacement with an outside CEO, it
is clear that the likelihood of replacement is greater in the tails of the ability dis-
tribution.
9
A forward-looking agent will also consider the option value that entering gives by pro-
viding choices in the second period. In period 2, the founder may choose to remain in
control of the business, replace a new CEO if doing so increases profits, or exit. Even if
expected second-period earnings from continuation are the same as expected first-period
earnings, the replacement and exit options imply for a forward looking agent that
1 1
1
1 1
. E q w
a a
q
- -
<
é ù
ê ú
ë û
10
However, we do not get to observe q and q separately. In the data, we observe
only their combined contribution to revenues, and it is not obvious that we
should also expect to see more frequent founder-CEO replacements in the tails of
the firm performance distribution. Figure 4 therefore plots simulated first- and
second-period revenues of surviving firms by founder status for our baseline simu-
lations. The two distinct groups of founders that experience turnover are also
clearly evident when we look at first-period performance, and they are indeed
drawn more frequently from the tails of the performance distribution. There is a
group with above-average performance in the first-period whose founders are re-
placed with less able CEOs. There is also a group of firms whose founders are
replaced with more able CEOs. This group consists of firms with low-ability
founders, but their poor ability is offset by ideas that in the second-period are
sufficiently high quality to justify payment of the transition cost (firms with the
worst ideas in the second period are closed down). Although these firms have a
mean first-period performance that is not much different from the population av-
0.0001
0.001
0.01
0.1
1
10
100
0.01 0.1 1 10 100
I
d
e
a
Q
u
a
l
i
t
y
i
n
P
e
r
i
o
d
2
,
q
2
Founder ability, q
Replaced with less able CEO Founder remained
Replaced with more able CEO Firm exited
FIGURE 3. Baseline simulations Founder ability versus
second-period idea quality, by outcome. 5000 observations.
Parameters: a
1
=a
2
=0.5, w=1, c=0.35. Each
1
1 a
q
-
is a draw
from the lognormal with mean one,
1
1
1
q
a -
is lognormal with
mean
1
1
,
a
q
-
and each
2
1
2
q
a -
is lognormal with mean
1
1
1
; q
a -
all
three distributions have a standard deviation of 0.5. Founder
turnover occurs in about 12% of the observations.
11
erage, they are overrepresented in the lower tail of the first-period performance
distribution.
A distinctive feature of Figure 4 concerns the effect that CEO turnover has on
the relationship between first-period performance and second-period performance.
Performance among the group of surviving firms with no turnover tends to be
more consistent from one period to the next than in the case of firms that expe-
rienced turnover. Firms in which founders are replaced by less-able CEOs see on
average a decline in their performance, while the arrival of a more capable CEO
induces on average an improvement in performance. Among surviving firms with
no turnover, therefore, the relationship between first- and second-period perfor-
mance is strongly positive. In contrast, the marginal ‘effect’ of first-period per-
formance on second-period performance among firms that experienced turnover is
much weaker in absolute value, and may even be negative.
2.2 Alternative Scenarios
Do these implications of the model persist in the face of significant changes in
parameter values? Figures 5 through 7, which replicate Figure 4 for five alterna-
tive scenarios, suggest that the possibilities depicted so far are, in fact, quite
0.01
0.1
1
10
100
0.01 0.1 1 10 100
S
e
c
o
n
d
p
e
r
i
o
d
p
e
r
f
o
r
m
a
n
c
e
First period performance
FIGURE 4. Baseline simulations: First-period versus second-period
performance among surviving firms, by outcome. 5,000 observations.
Solid line indicates equal performance in both periods.
Replaced with less able CEO Founder remained
Replaced with more able CEO Firm exited
12
0.1
1
10
0.01 0.1 1 10
S
e
c
o
n
d
p
e
r
i
o
d
p
e
r
f
o
r
m
a
n
c
e
First period performance
Replaced with less able CEO Founder remained
Replaced with more able CEO Series4
FIGURE 5. Alternative scenarios I. First-period versus second-period performance
among surviving firms, by outcome. Upper panel: Ability matters more (less) than
idea quality in the first (second) period; a
1
=0.8, a
2
=0.2. Lower panel: Idea quality
matters more (less) than ability in the first (second) period; a
1
=0.2, a
2
=0.3. In
each case c is adjusted to maintain turnover rate of about 12%.
0.01
0.1
1
10
100
1000
10000
100000
0.0001 0.001 0.01 0.1 1 10 100 1000 10000
S
e
c
o
n
d
p
e
r
i
o
d
p
e
r
f
o
r
m
a
n
c
e
First period performance
Replaced with less able CEO Founder remained
Replaced with more able CEO Series4
13
0.1
1
10
100
0.01 0.1 1 10 100
S
e
c
o
n
d
p
e
r
i
o
d
p
e
r
f
o
r
m
a
n
c
e
First period performance
Replaced with less able CEO Founder remained
Replaced with more able CEO Series4
0.1
1
10
100
0.01 0.1 1 10 100
S
e
c
o
n
d
p
e
r
i
o
d
p
e
r
f
o
r
m
a
n
c
e
First period performance
Replaced with less able CEO Founder remained
Replaced with more able CEO Series4
FIGURE 6. Alternative scenarios II. First-period versus second-period performance
among surviving firms, by outcome. Upper panel: First- and second-period idea
qualities more highly correlated (standard deviation of distribution of q
2
reduced to
0.1). Lower panel: First-period idea quality and founder ability more highly corre-
lated (standard deviation of distribution of q
1
reduced to 0.1). In each case c is ad-
justed to maintain turnover rate of about 12%.
14
robust. In Figure 5, the relative importance of idea quality and ability is altered
across the two periods. In the upper panel, ability is most important in the first
period
1 2
( 0.8, 0.2), a a = = while in the lower panel it is more important in the
second
1 2
( 0.2, 0.8). a a = = The results reported in the baseline are not sensitive
to these significant changes in the production technology. Figure 6 sharply alters
the correlations between idea quality and ability. In the upper panel, the stan-
dard deviation of the distribution of
2
q is reduced by 80 percent, to 0.1, while
maintaining its mean equal to the realized value of
1
q ; the lower panel similarly
increases the correlation between q and q
1
by reducing the variance of the distri-
bution of
1
. q
The three groups of surviving firms become even more distinct than
in the baseline, but the original pattern is preserved. Finally, in Figure 7 the cost
of transitioning to a new CEO is reduced sharply, so that the fraction of firms
that experience turnover is increased from about 12 percent to 35 percent. Again,
the baseline results are robust to this change.
10
10
Our data do not afford us an opportunity to explore the effects of changes in the tran-
sition cost. However, Braguinsky et al. (2012) explore the effects of a likely decline in the
0.1
1
10
100
0.01 0.1 1 10 100
S
e
c
o
n
d
p
e
r
i
o
d
p
e
r
f
o
r
m
a
n
c
e
First period performance
Replaced with less able CEO Founder remained
Replaced with more able CEO Series4
FIGURE 7. Alternative scenarios III. First-period versus second-
period performance among surviving firms, by outcome. Transition
cost reduced to c=0.1, increasing founder turnover rate to about
35%.
15
2.3 Related models
There are, of course, different ways one may specify a model of mismatching. In
this subsection we briefly compare our theoretical framework with three related
models. The best known of these is Holmes and Schmitz (1995), in which agents
vary in their ability to produce ideas, although all are equally gifted at managing.
Agents with low innovative ability prefer to buy and manage existing businesses,
while agents with high ability prefer to specialize in business creation (selling or
closing each business they create). The outcomes for founders with intermediate
ability depend on the quality of the business idea: they will close the worst busi-
nesses and manage the best. A key determinant of the boundaries between the
choices is the cost of transferring a business. Figure 8 illustrates. Clearly, only the
better ideas produced by the more able innovators experience founder turnover.
cost of evaluating ideas in Japanese biotech firms, brought about by institutional reform.
They find that a reduction in evaluation cost was associated with a large increase in the
rate of founder turnover.
Transfer business and
start another one
Founder continues
to operate firm
Close business and
find another to
manage.
Close business and
start another one
q
q
FIGURE 8. Holmes and Schmitz (1990)
16
As managerial ability is fixed, the model predicts that business turnover is asso-
ciated with high initial and subsequent firm performance, but it does not predict
any change in performance after turnover.
In Braguinsky et al. (2012), entrepreneurs with known ability found businesses
with unknown quality. As in our model, business quality and founder ability are
complements in production. The quality remains unknown unless the founder
chooses to pay an evaluation cost, c. High-ability agents know that they are at
risk of wasting their time with a poor idea, so they choose to evaluate the idea,
closing down business that are not good enough. Low-ability agents know that
they may not be good enough mangers if their business quality is good; they also
pay the evaluation cost and either close their business or transfer control to a
professional manager. Agents with intermediate ability do not pay the evaluation
cost; they continue to manage a business of unknown quality (see Figure 9). The
model unambiguously predicts that only the better businesses founded by low-
ability managers are at risk of founder turnover, and that turnover elicits an im-
provement in performance. Auxiliary assumptions of the model associate better
initial firm performance (i.e., the combination of business quality and founder
ability) with turnover.
Founder continues
to operate firm
Close business
q
q
Replace founder-
CEO
q
q
Close business
FIGURE 9. Braguinsky et al. (2012)
17
Finally, Holmes and Schmitz (1995) do not have explicit variations in skill. In
this model, output is given by
t t t
y q m = + , where q is business quality and m is
match quality. Both q and m evolve over time, each consisting of permanent and
transitory components. The permanent component of business quality, b, is fixed
when a business is first founded, while the permanent component of match quali-
ty, m, is fixed each time a new manager takes control of the business. Founders
(and subsequent managers) who experience a sufficient decline in match dispose
of their business, either by closing it (if the business quality is sufficiently low) or
by selling it to a new manager (if the quality is sufficiently high). Figure 10 illu-
strates, with a sample path depicting the evolution of business and match quali-
ties. A business begins at point a, and suffers a steady decline in match quality.
At b the business quality is sufficiently high to enable a sale. The new manager
begins with the same business quality but a better match quality, at c, but then
suffers a decline in both business quality and match quality until exit at d.
Holmes and Schmitz show that the boundary between closing and selling a busi-
ness shifts down when the permanent business quality, b, increases. Hence, better
quality businesses are more likely to be sold. However the match quality may
Sell business
Founder continues
to operate firm
Close business
q-b
m-m
a
b
d
c
FIGURE 10. Holmes and Schmitz (1995)
18
be little better after a sale, so despite better average performance after business
turnover, there can be an increased risk of quick failure.
11
It is not possible to re-
late founder ability to the likelihood of business turnover in this model unless we
impose some functional relationship between ability and the pair { , }. m q Doing
so, however, would be unfair to the Holmes and Schmitz model, although it
strikes us that their model could be made consistent with turnover occurring in
the tails of the firm performance distribution.
3. Data and Sample Construction
Our empirical analysis focuses on all new business founded in Denmark in 1999
and 2000. We construct our samples out of three databases maintained by Statis-
tics Denmark: the Entrepreneur Database, the Firm Database, and the Inte-
grated Database for Labor Market Research (IDA).
The Entrepreneur Database is the primary source for identifying new businesses
(partnerships or sole proprietorships) registered in Denmark each year, and pro-
vides unique identifiers for each firm, each plant, and one individual identified in
the registration documents. The Entrepreneur Database spans the period 1996 to
2006. However, accounting information is not collected for all industries until
1999, so this is the earliest cohort we study.
There are several limitations to the Entrepreneur Database. First, since only one
person associated with the new business is recorded, the database provides no
information for identifying possible co-founders. Second, some of those who ap-
pear in the database may not be the “real” founder. For example, there are cases
in which founders used their spouse’s name to register a business. Third, it is un-
clear from the Entrepreneur Database whether or not the identified entrepreneur
actually works at their startup. Although it is possible that entrepreneurs do not
operate businesses themselves, this study focuses on those who were actually
working for their own businesses. In order to identify these entrepreneurs and
possible co-founders, the data obtained from the Entrepreneur Database are
combined with information provided in the Firm Database.
The Firm Database consists of annual employment information on all workers
11
Figure 5 depicts the case where the m is common across matches. If m is lower after
business transfer, the boundaries shift to the right, increasing the risk of business failure.
19
(full-time and part-time) at all firms operating in Denmark. The variable of par-
ticular interest is their position in the firm, through which founders of each new
business are identified in the following way. First, founders are those whose posi-
tions are classified as self-employed, employers, or business tax payers. Second,
for businesses with fewer than four employees, people whose positions are top
managers are also defined as founders.
12
Third, if nobody in a business with fewer
than four employees is identified as a founder based on the previous two criteria,
all employees are considered as cofounders. The fourth group of founders consists
of those who registered the business, were working for the business, but did not
meet the first three criteria for being a founder. Lastly, we identify individuals
who registered the business, but did not appear to be working for the new firms
(we will exclude this last category from the sample).
We link these businesses to the IDA, where more detailed information such as
industry classification and employee demographics is available. The IDA is an
employer-employee matched database, which provides mainly employment infor-
mation at firm, establishment, and employee levels. The IDA enables us to track
over time the turnover of founders, changes in firm status, and occupational
choices of individuals each year until 2008. However, because of incomplete ac-
counting and earnings data in the most recent years, we will follow firms only
through 2004 and individuals through 2005.
Firm identifiers in the IDA are attached to a specific owner, so they change
whenever there is a change of ownership. This design makes it possible to distin-
guish three firm-level events in each year. Survival is assumed if the same firm
identifier from the previous year appears in the current period. An ownership
change occurs if the same establishments of the firm from the previous year are
identified in the current period except that they are associated with a different
firm identifier. A business exit is identified if all the establishments of the firm
can no longer be found in the data.
In our analysis, the original founder of a startup must meet the following criteria:
(1) He or she formed a startup in 1999 or 2000 as a sole proprietor; (2) The per-
son was currently working at the new venture; and (3) The new business could be
identified in the Firm Database and the IDA database, indicating that at least
12
This way to identify founders was first used by Sørensen (2007).
20
one person was working in the business as his or her primary job.
13
Using the
unique person and firm identifiers, we can track these founders’ employment in
the following year to identify whether or not the founder was still working at the
startup, and whether he was still an employer. A founder was staying at the star-
tup if he could be identified in the firm in the following year, whether as an em-
ployee or as an employer. Otherwise, the founder is assumed to have left the orig-
inal firm because of turnover, business exit, or ownership change. In a few cases,
the founder might stay at the startup after a change of ownership, but in each of
these cases a new firm identifier is assigned to the startup.
Table 1 provides tabulations of changes in the status of startups and their found-
ers from the founding year to 2005. Four outcomes are possible in each year: the
firm exits, the firm is acquired or the firm continues, either with or without the
founder remaining in operational control. Of 4,172 firms in the sample, a total of
533 (12.8 percent) of them eventually experienced founder turnover. Turnover
was twice as common as ownership change (6.5 percent), but much less common
than firm exit (50.6 percent). The hazard of turnover declines with firm age, de-
clining from 5.11 percent in the first year, to 2.61 percent at age five.
14
4. Firm Performance and Founder-CEO Replacement
In this section, we evaluate the relationships between startup performance and
subsequent rates of founder turnover, and between founder turnover and subse-
quent firm performance.
4.1 Startup performance and founder-CEO replacement
To assess the effect of initial performance on subsequent rates of founder-CEO
replacement, we remove from the baseline sample 2,388 startups (918 from the
1999 cohort, and 1,470 from the 2000 cohort) that exited or were acquired before
13
This person may be the founder, so our sample includes single-person firms. We have,
however, repeated our analyses after restricting the sample to including firms with at
least two individuals. The results, available from the authors, are very similar to those
reported here.
14
The declining hazard of turnover is consistent with gradual learning about match quali-
ty [Jovanovic (1979)], which we have not attempted to model.
21
TABLE 1
Status Change of Startups by Year
Years after Founding
1 2 3 4 5 6*
Count
Surviving: founder stays 2,996 2,247 1,829 1,535 1,321
Surviving: founder turnover 213 127 82 54 40
Firm exit 898 550 291 193 140
Firm acquired 65 72 45 47 34
Total 4,172 2,996 2,247 1,829 1,535
Hazards (percentage)
Surviving: founder stay 71.81 75.00 81.40 83.93 86.06 87.16
Surviving: founder turnover 5.11 4.24 3.65 2.95 2.61 3.12
Firm exit 21.52 18.36 12.95 10.55 9.12 7.71
Firm acquired 1.56 2.40 2.00 2.57 2.21 2.02
Total 100 100 100 100 100 100
* Percentages for 1999 cohort only.
any occurrence of a founder turnover during the observation window between the
founding year and 2005. Table 2 provides descriptive summary statistics for the
remaining 1,784 startups and their founders. There are no large differences in
means between founders that continued and founders who were replaced. Howev-
er, the latter group are somewhat younger and less likely to have had vocational
training; they are also more likely to come from the tails of the earning distribu-
tion (either less than high school or at least college); they are also somewhat less
likely to have founded businesses in construction, and more likely to be engaged
in high-tech activities.
We first estimate the following logit model,
2
1 1 2 1 , 1 ,
ln ( )
1
t
t t i i t i t
t
p
logsale logsale X Z
p
a b b g d e
- - -
æ ö
÷ ç
÷
ç = + + + + +
÷
ç
÷
÷ ç -
è ø
(7)
22
where p
t
is the probability that the founder leaves a continuing firm in period t,
is a vector of time-invariant variables related to founder demographics (gend-
er, marital status, age, and education) and startup founding conditions (founding
year and industry), and
,
is a vector of potentially time-varying variables,
such as firm age, founder age, marital status, and highest completed education.
Table 2
Descriptive Summary of Subsample 1
All No Turnover Turnover
Age (mean years) 40.07 40.23 37.91
Male (%) 0.76 0.76 0.77
Married (%) 0.64 0.64 0.54
Education (%)
less than high school 16.84 16.55 21.01
exact high school 6.66 6.34 11.07
vocational training 50.91 51.75 39.21
some college 14.72 14.66 15.57
bachelor 1.13 1.02 2.63
Master 7.66 7.66 7.69
Phd 0.69 0.69 0.75
others 1.38 1.33 2.06
Founded in 1999 (%) 42.28 42.69 36.59
Initial sales (1,000 DKK) 2,149 2,135 2,347
Initial no. of full-time employees 1.87 1.85 2.09
Industry (%)
Agriculture 0.29 0.28 0.38
Manufacturing 7.25 7.34 6.00
Construction 17.57 17.86 13.51
Wholesale and retail 37.57 37.55 37.9
Transport, post, telecom 4.89 4.90 4.69
Low-tech businesses 6.93 6.93 6.94
Public Services 0.43 0.42 0.56
Personal Services 6.05 6.26 3.19
Kibs 0.44 0.34 1.88
High-tech 18.24 17.77 24.77
Others 0.34 0.35 0.19
Obs. 7,961 7,428 533
23
The key independent variable is the logarithm of sales in the previous period,
along with its quadratic to allow for the possibility of a non-monotonic effect as
predicted in Section 2.
15
Columns (1) and (2) of Table 3 present the results. In
column (1), we include controls only for firm age and founding year. The nega-
tive coefficient on firm age reflects the declining hazard of founder-CEO replace-
ment already noted in the summary statistics, while the cohort dummy is insigni-
ficant. In column (2) we add some founder characteristics. The probability of
founder-CEO replacement declines with founder age, it is lower for founders with
college education, and it is lower for men and married founders. Both columns
show a U-shaped relationship between the log of sales in the previous period and
the likelihood of founder turnover in the current period, indicating that a founder
is more likely to rescind operating control of the firm if it is located the tails of
the performance distribution among all surviving firms. Columns (3) and (4) re-
place sales in the previous period with sales in the startups’ first year of opera-
tion. The results continue to show the U-shaped relationship, and the same ef-
fects of founder characteristics and firm age on the likelihood of founder-CEO
replacement. The minimum of the U-shaped relationship between prior perfor-
mance and the probability of turnover occurs at revenue levels of DKK580,000 to
DKK800,000, equal to 15 to 37 percent of the sample mean. This is also well
within the sample range. Consistent with the predictions of our simple mismatch-
ing model, therefore, we find that turnover is more common in the tails of the
observed earnings distribution than in the middle.
4.2 Founder-CEO replacement and the subsequent performance of startups
To examine the relationship between founder-CEO replacement and subsequent
firm performance, we use a subset of startups in the baseline sample and limit
the observation window to the interval between the founding year and 2004. The
startups included in this subsample meet two criteria. First, we eliminate firms in
which the founder was replaced after 2002. That is, firms included in the sample
either experienced founder turnover by 2002, or never had founder turnover dur-
ing the observation period. Second, we remove startups that failed or were ac-
quired between the initial founding year and 2002 but prior to any observed
15
For six observations with zero sales, we added one DKK to sales prior to taking the
logarithm.
24
TABLE 3
The Effect of Startup Performance on the Probability of Founder-CEO Replacement
Dept Var: = 1 if founder replaced in year t
Logit Reegression
Using sales in year t-1 Using sales in founding year
(1) (2) (3) (4)
Log of sales -0.940
***
-0.966
***
-1.159
***
-1.270
***
(-5.08) (-5.10) (-3.30) (-3.43)
(Log of sales)
2
0.072
***
0.076
***
0.084
***
0.095
***
(5.71) (5.87) (3.36) (3.61)
Founder characteristics
Founder age
__
-0.017
***
__
-0.019
***
(-3.05) (-3.40)
College educated = 1
__
-0.235
*
__
-0.090
(-1.38) (-0.54)
Male = 1
__
-0.154
__
-0.104
(-1.28) (-0.86)
Married = 1
__
-0.331
***
__ -0.303
***
(-3.25) (-2.98)
Firm characteristics
Firm age -0.366
***
-0.340
***
-0.324
***
-0.288
***
(-10.39) (-9.47) (-9.68) (-8.46)
Cohort dummy (= 1 if
founding year = 2000)
0.130 0.10 3 0.095 0.069
(1.34) (1.04) (0.96) (0.69)
Controls for industry No Yes No Yes
Ave Log Likelihood
-0.239 -0.235 -0.241 -0.237
No. of observations
7,357 7,348 7,305 7,305
Sales in thousand DKK. z-scores are in parentheses. Significance levels: *** 0.01, ** 0.05,
* 0.1.
25
founder turnover event. However, we retain in the subsample startups that expe-
rienced founder turnover and then exited or were acquired. Table 4 illustrates
how we construct the subsample.
Starting with the 1,588 startups in the 1999 cohort, we removed 709 firms that
exited or went through ownership change by the end of 2002 and before any oc-
currence of founder turnover, and 36 firms that experienced founder turnover in
2003 or 2004. For the 2,584 startups of the 2000 cohort, we removed 995 and
then 83 of them for the same reasons. We are left with a sample of 2,349 star-
tups, 843 of which come from the 1999 cohort, and 1,506 from the 2000 cohort.
Table 4
Firms Included in Subsample 2
Turnover by
2002
Turnover after
2002
No Turnover
Exited or acquired by 2002 ? × ×
Exited or acquired after 2002 ? × ?
Survived to 2004 ? × ?
Table 5 summarizes the number of startups in our sample that survived, exited,
or experienced ownership change during the observation period. Among the 843
startups founded in 1999, 142 of them had founder turnover before 2002. Among
them, 50 survived to 2004, 56 exited or changed owners before 2002 but after
founder turnover, and 36 exited after 2002. Among the 1,506 startups founded in
2000, 232 had founder turnover before 2002. Of these, 82 survived the entire ob-
servation period, 70 exited or changed owners before 2002, and 80 exited after
2002.
Table 6 provides the descriptive summary of this subsample (the distributions of
firms in each column across industries is very similar to those shown in Table 2,
and are not reported). The summary statistics show two interesting contrasts be-
tween firms that experienced turnover and those that did not. Turnover by 2002
is strongly associated with lower odds of survival to 2004. Only 35 percent of
firms with turnover survived, compared with 74 percent of firms with no turno-
ver. However, conditional on survival, the growth rate of sales and of employ-
ment is much greater among firm that experienced turnover. For example al-
26
though there is little difference between the two groups in initial sales revenue,
average 2004 sales are 55 percent higher among the turnover group.
Table 5
Survival, Exit, and Ownership Change after Founder Turnover
1999 Cohort
2000 Cohort
Turnover No Turnover
Turnover No Turnover
Total 142 701
232 1,274
Survived 50 545
82 912
Exited before 2002 56 0
70 0
Exited after 2002 36 156
80 362
Table 6
Descriptive Summary of Subsample 2
All No Turnover Turnover
Obs. 2,349 1,975 374
Survival in 2004 (%) 67.65 73.77 35.29
Founded in 1999 (%) 35.89 35.49 37.97
Initial sales (1,000 DKK) 2,008 1,962 2,253
Sales in 2004 (1,000 DKK) 5,168 4,929 7,639
Initial no. of full-time employees 1.68 1.62 2.00
No. of full-time employees in 2004 4.08 3.86 6.39
We estimate the following model
,2004 1 ,0 2
ln( ) ln( )
i i i
sales sales turn a b b = + +
3 ,0 ,2004
ln( )
i i i i
sales turn X b g e + ´ + + , (8)
where turn
i
equals one if the startup experienced founder turnover between its
founding year and 2002, and zero otherwise, and the vector
includes controls
for cohort and industry.
Table 7 presents OLS estimates of (8). In the first three columns, we focus on
27
startups that survived until 2004. Column (1) shows that both initial sales and
founder turnover have a positive impact on startups’ sales in 2004. Interacting
the two factors in column (2), we further find that the marginal effect of founder
turnover on future sales is decreasing with the initial performance of startups (or,
equivalently, that the marginal effect of initial performance is lower for firms that
had turnover). This is consistent with the model in Section 2. However, this
Table 7
Founder-CEO Replacement and Future Performance of Startups
Dept Var: log of sales in 2004
Surviving Firms
OLS
All Firms
Tobit
(1) (2) (3) (4) (5) (6)
Log.sale
0
0.437
***
0.453
***
__ 0.724
***
0.817
***
__
(18.8) (18.6) (5.92) (6.12)
Turnover 0.336
***
1.524
***
__ -5.00
***
-1.032 __
(4.15) (2.80) (-14.4) (-0.45)
Log.sale
0
*turnover __ -0.165
**
__ __ -0.552
*
__
(-2.21) (-1.73)
log.sale
0
percentile: