Description
The purpose of this paper is to do an empirical analysis assessing whether banks highly
involved into trading activities show specific business model choices. Key factors in the analysis are a
proper measure for trading activities and a consistent classification of banks in terms of business
choices.
Journal of Financial Economic Policy
Banks under X-rays: business model choices and trading
Francesca Campolongo J essica Cariboni Nathalie Ndacyayisenga Andrea Pagano
Article information:
To cite this document:
Francesca Campolongo J essica Cariboni Nathalie Ndacyayisenga Andrea Pagano , (2015),"Banks
under X-rays: business model choices and trading", J ournal of Financial Economic Policy, Vol. 7 Iss 4
pp. 377 - 400
Permanent link to this document:http://dx.doi.org/10.1108/J FEP-12-2014-0081
Downloaded on: 24 January 2016, At: 21:53 (PT)
References: this document contains references to 35 other documents.
To copy this document: [email protected]
The fulltext of this document has been downloaded 43 times since 2015*
Users who downloaded this article also downloaded:
Eric Osei-Assibey, J oseph Kwadwo Asenso, (2015),"Regulatory capital and its effect on credit growth,
non-performing loans and bank efficiency: Evidence from Ghana", J ournal of Financial Economic
Policy, Vol. 7 Iss 4 pp. 401-420http://dx.doi.org/10.1108/J FEP-03-2015-0018
J acob Kleinow, Tobias Nell, (2015),"Determinants of systemically important banks: the case of
Europe", J ournal of Financial Economic Policy, Vol. 7 Iss 4 pp. 446-476http://dx.doi.org/10.1108/
J FEP-07-2015-0042
Muhammad Umar, Gang Sun, (2015),"Country risk, stock prices, and the exchange rate of the
renminbi", J ournal of Financial Economic Policy, Vol. 7 Iss 4 pp. 366-376http://dx.doi.org/10.1108/
J FEP-11-2014-0073
Access to this document was granted through an Emerald subscription provided by emerald-
srm:115632 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald
for Authors service information about how to choose which publication to write for and submission
guidelines are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company
manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as
well as providing an extensive range of online products and additional customer resources and
services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the
Committee on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for
digital archive preservation.
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
*Related content and download information correct at time of
download.
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Banks under X-rays: business
model choices and trading
Francesca Campolongo, Jessica Cariboni,
Nathalie Ndacyayisenga and Andrea Pagano
Institute for Protection and Security of Citizens,
Joint Research Centre - European Commission, Ispra, Italy
Abstract
Purpose – The purpose of this paper is to do an empirical analysis assessing whether banks highly
involved into trading activities show specifc business model choices. Key factors in the analysis are a
proper measure for trading activities and a consistent classifcation of banks in terms of business
choices.
Design/methodology/approach – We investigate three measures for trading activities proposed by
regulators in the context of bank structural reformin Europe. Through robust statistics we identify the
key trading players and classify banks into a limited number of business model clusters, relying on a set
of balance sheet and income statement indicators.
Findings – Using a sample of 100 European banks in 2007-2012, results show that the measures
identify similar, but not identical, sets of banks highly involved into trading. The measure proposed by
the European Commission selects fewer banks and is more consistent over time. The business model
analysis identifes six rather stable clusters, from small-medium retail-focused banks to very large
investment groups. The measures coherently identify as key trading players the largest investment
groups and select very few retailed focused banks. Differences among measures arise for very large
retail-diversifed and medium/large wholesale banks.
Originality/value – These results could feed the debate on which measures for trading regulators
could consider depending on the target of the reformthey would implement. For instance we showthat
the measure proposed by the European Commission selects less well capitalized retail-diversifed banks
compared to the others.
Keywords Banks, Quantitative and mathematical studies
Paper type Research paper
1. Policy context and scope
The recent crisis has forced regulators and researchers to investigate possible initiatives
to enhance the stability of the fnancial sector. These include new capital and liquidity
requirements, the implementation of new resolution regimes, the review of the
functioning of the markets and the risk entailed in certain types of instruments. The
appropriate design of a regulatory framework for tackling the coexistence of trading
and retail activities for systematically important fnancial institutions remains under
JEL classifcation – G01, G21, C38
The authors would like to thank A. Blundell-Wignall and S. Schich from OECD for their
valuable suggestions and comments, which help us to improve our work. The authors are
responsible for any remaining errors.
The content of this article does not refect the offcial opinion of the European Commission.
Responsibility for the information and views expressed therein lies entirely with the author(s).
The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/1757-6385.htm
Banks under
X-rays
377
Received23 December 2014
Revised3 July2015
Accepted6 July2015
Journal of Financial Economic
Policy
Vol. 7 No. 4, 2015
pp. 377-400
©Emerald Group Publishing Limited
1757-6385
DOI 10.1108/JFEP-12-2014-0081
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
discussion [1]. This is recognized as a potential risk for fnancial stability (Gambacorta
and Rixtel, 2013). Stiroh and Rumble (2006) fnd evidence that diversifcation benefts
for USA holding companies can be accompanied by volatility risks. Moreover trading
can be used for risk-shifting and could bring excessive leverage (Boot and Ratnovski,
2012), hence increasing banks’ complexity and systemic risk (Brunnermeier et al., 2012),
associated with implicit and/or explicit State guarantees (Ueda and di Mauro, 2013;
Davies and Tracey, 2012; Brewer III and Jagtiani, 2013; Inci et al., 2011; EC, 2013).
Many countries are taking steps toward a structural reform of their banking sector
via either ring-fencing or separating retail-oriented services from more risky trading
activities (Liikanen et al., 2012; Vickers, 2011; Merkley and Levin, 2011). The
identifcation of the banks that could undergo through this reform is also under
discussion within European institutions.
Such initiatives are nonetheless controversial due to the costs that their
implementation may imply (Viñals, 2013) and diffcult to pursue, due to the unclear
fence between such activities. Moreover a risk of a migration of some activities to
unregulated parts of the system exists (Chow, 2011). Other voices recognize that
effective structural separation measures should be coupled with other macro/micro
regulatory measures (Laeven et al., 2014) and should be designed and tailored heeding
bank business models (Viñals, 2013).
In this context, this paper presents an empirical analysis addressing the
identifcation of European banks that could be proposed for a structural reform and
it examines the link between banks’ involvement in trading and business model
choices. We frst investigate which are the key players in trading, based on some
defnitions as recently discussed in the European Union (EU). Second, bank business
models are characterized via clustering analyses of balance sheet indicators.
Finally, combining the results of the frst two goals, we assess the link between
business model choices and trading. More specifcally, we discuss how trading is
associated with different fnancing and capital-level choices. We aim at giving
indications through the lens of business model choices on how a legislation aiming,
not only at isolating trading activities, but also at reducing banks’ complexity and
systemic risk, could reshape the fnancial sector. All the analyses are based on a
sample of 100 European banks over years 2007 to 2012.
The paper is structured as follows. Section 2 describes the research context. Section
3 describes the steps of our analysis. Section 4 addresses the issue of howtrading can be
measured based on balance sheet information and howbanks highly involved in trading
can be identifed. Section 5 shows the methodology and results for the business model
analysis. Section 6 pulls together all information, and the fnal section presents our
conclusions. Appendix 1 gives details of the banks included in the sample.
2. Research context
The role of trading within the structural separation debate is a rather newtopic, and the
literature is still not well developed. Nevertheless, several scholars address some of the
issues which can be broadly related to our analysis. In the remaining of this section, we
detail our place in the existing literature according to the three main questions the paper
aims to address.
JFEP
7,4
378
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
2.1 Defnitions of trading
Our frst objective is to measure the involvement of European banks into trading
activities. Aproper defnition of trading is rather ambitious per se, due to e.g. the unclear
distinction between proprietary trading and market making, newchallenges brought by
modern fnance and the use of instruments like high-frequency trading (Chordia et al.,
2013). We compare three measures, recently proposed in the context of structural reform
in Europe, which aims to capture the involvement of banks in trading relying only on
balance sheet data. In Liikanen et al. (2012), trading activities refer only to the assets side
and they also include available for sale. Blundell-Wignall and Roulet (2012) propose a
measure of trading based on the Gross Market Value of Derivatives (GMVD), since they
identify GMVDas one of the drivers of banks’ distance to default. Finally, the European
Commission (EC; see EC Annex A8 (2013)) assesses a measure which aims at capturing
gross volumes of trading, and also proposes screening thresholds to identify the most
important players in trading.
2.2 Banks’ business model
Our second goal is to characterize banks’ business models using robust statistical
techniques. We aim to classify banks into a limited number of business model groups
ranging from pure retail to very large investment institutions, using information from
the consolidated balance sheet on their assets and liabilities, capital and leverage levels
and their proftability.
Research on banks’ business model ranges from a general discussion on their trend
(BIS, 2014; ECB, 2010; 2013) to more specifc issues focusing on how business model
information can be used for regulatory purposes (Blundell-Wignall et al., 2013; Laeven
et al., 2014). Later studies (Ayadi et al., 2011; Ayadi et al., 2012; Ayadi and De Groen,
2014) aim to describe, via standard clustering techniques, the business model evolution
of a set of large European banks, which is also the approach we retain in the current
paper.
A different perspective is followed in some econometric studies linking bank
business model and risk. In general terms it has been shown that retail-focused or
diversifed banks are usually safer in terms of distance-to-default (Blundell-Wignall
and Roulet, 2012; Blundell-Wignall et al., 2014), while riskier banks are those relying
on non-interest income and non-deposit funding (Demirguerc-Kunt and Huizinga,
2013) and/or those characterized by low capital, large size, great reliance on
short-term market funding and aggressive credit growth (Altunbas et al., 2011;
Beltratti and Stulz, 2009).
2.3 Trading and business model
As a fnal point, combining the results of the frst two questions, we assess the link
between business model choices and key players in trading. Our working hypothesis is
that fnancing and capital-level choices play a role when structural separation is
addressed. Literature discussing business model within structural separation is very
limited with few important exceptions such as Blundell-Wignall et al. (2014).
2.4 Methodological aspects
As far as analytical aspects are concerned, banking data usually show skewed
distributions and outliers. We employ robust clustering (Garcia-Escudero et al.,
2008) and robust outlier detection (Riani et al., 2012) to properly assess these
379
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
features, which so far have been rarely employed in the analysis of banking data
(Cariboni et al., 2015).
3. Steps of the analysis
(1) Identifying banks highly involved into trading:
• Measuring trading activities. As already mentioned in 2.1, we investigate
three different measures for trading.
• Selecting banks highly involved in trading activities. Through robust
clustering we set thresholds to discriminate banks mostly engaged into
trading. Candidates are those with a high absolute or relative volume of
trading activities. We discuss similarities and differences among the sets of
banks selected using the various measures of trading.
(2) Business model analysis:
• Selecting few variables describing banks business models. We start from a
large number of balance sheet and income statement variables. We restrict
our attention to 14 indicators which present the lower correlation coeffcient.
Through Principal Component Analysis (PCA) we move to the four most
relevant principal components.
• Determining banks business model clusters. Via robust clustering on the
relevant principal components, we identify six business models, ranging
from very small banks with purely retail activities, to very large investment
banks.
(3) Assessing the link between business model choices and trading
Coupling information fromthe previous steps, we map banks highly involved in
trading onto business model clusters.
4. Measuring trading activities in EU banks
4.1 Data and defnitions
To describe trading, data are extracted from SNL database [2] (see Table I).
Using these variables, we consider three measures for trading recently proposed for
the structural reform:
• EC: The measure proposed by the EC in EC Annex A8 (2013). This measure
focuses on the gross volumes of securities held for trading and includes the
liability data to proxy market and counterparty risk:
M
EC
?
TSA ? TSL
2
.
• GMVD: A proxy for the GMVD [3], as described by Blundell-Wignall and Roulet
(2012) because of its role in determining banks default risk, especially if coupled
with wholesale funding and low levels of liquid trading assets:
M
GMVD
? DA ? DL.
JFEP
7,4
380
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
• HLEG: The measure proposed by high level expert group (HLEG), which
includes also assets available for sale:
M
HLEG
? TSA ? TAFS.
Each measure is accompanied by its share with respect to Total Assets (TA).
Our sample contains 100 banks per period (see Annex 1) located in the EU. This
sample is obtained by selecting all EU banks whose average TA over 2007-2012 are
higher than €30 bn. The sample covers 18 EUcountries, with average aggregated TAof
roughly €32 trillion, representing 73 per cent of EU28 TA(Schoenmaker and Peek, 2014).
All EU Globally Systematically Important Banks are included.
4.2 Methodology and results
We focus, for each measure, on four periods of three-year moving average (2007-2010,
2008-2010, 2009-2011 and 2010-2012), leading to a sample containing 400 observations.
Table I.
Classes of assets and
liabilities used for
computing the
measures assessing
trading activities
(source SNL)
Item Label Data defnition
Assets Total Assets TA All assets owned by the company as carried
on the balance sheet
Derivatives Held for Trading
a
DA Derivatives with positive replacement
values not identifed as hedging or
embedded derivatives
Total Assets Held For
Trading
TSA Trading portfolio assets are: assets acquired
principally for the purpose of selling in the
near term, assets that on initial recognition
are part of a portfolio of identifed fnancial
instruments that are managed together and
for which there is evidence of a recent
actual pattern of short-term proft-taking, or
derivative assets
Total Assets Available for
Sale
TAFS Total loans and securities designated as
available for sale; or are not classifed as
loans and receivables, held-to-maturity
investments or fnancial assets at fair value
through proft or loss
Liabilities Derivative Held for Trading DL Derivatives with negative replacement
values not identifed as hedging
instruments
Total Securities Held for
Trading
TSL Trading liabilities that are taken with the
intent on repurchasing in the near term or a
portfolio of managed fnancial instruments
where there is evidence of a recent actual
patter of short-term proft-taking
Notes:
a
A derivative is a fnancial instrument with all of the following three characteristics: its value
changes in response to the change in an underlying variable; it requires no initial net investment or an
initial net investment that is smaller than would be required for other contracts that would be expected
to have a similar response to changes in market factors; it is settled at a future date
381
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
We aimto select, as in Pagano (2013), banks mostly engaged in trading activities. For
each measure M
i
we consider the two dimensional space defned by its standardized
absolute value and its standardized share over TA[4]. We apply trimmed clustering
techniques (so called t-clustering), which allows the creation of ellipsoidal groups that
better ft skewed data [5].
The t-clustering algorithmrequires choosing some parameters, such as the number of
clusters; the restriction factor, which describes the shape of clusters; and the trimming
level, which concerns the share of expected outliers [6].
Clusters are then used to set thresholds according to the following criteria:
• they avoid cutting clusters;
• they are multiple of €10 bn for the volume and 1 per cent for the share to be
applicable in regulation; and
• they are comparable to the ones discussed among regulators.
Figure 1 presents the results of the clustering exercise for the EC measure. The x-axis
shows the absolute value of trading assets defned by EC (in € bn), the y-axis shows its
corresponding share over TA.
The right plot zooms on the region where thresholds are set. If we focus on banks
with a low share of trading, we observe an empty region with respect to volumes (i.e.
around €70 bn for EC and GMVD, and around € 80 bn for HLEG, not shown).
Considering the share of trading assets over TA, setting the thresholds is more
diffcult, due to high density of banks along this axis: t-clustering however helps
identifying a horizontal threshold at 10 per cent for EC. Based on similar exercises run
on the whole sample, we set thresholds for each measure as reported in the right part of
Figure 1.
_Paper_ECClusters
foats here
JFEP
7,4
382
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table II. One can see they are very close to those found in regulation and literature (see
left part of Table II).
We investigate the similarity between measures for a given period and also the
stability of each measure across the years. Table III (top three rows) reports the number
of banks highly involved in trading according to each measure in each period. One can
see that GMVD and HLEG are the most volatile over the years, with a clear increasing
trend for GMVD, while EC tends to select the same number of banks.
To better assess the different behaviors of the measures considered, we confront each
pair of measures by comparing the list of banks selected by both (intersection) versus
the ones selected by only one of the two (bottompart of Table III). Roughly 50-60 per cent
of the banks in the sample are never selected by any measure, while 20-25 per cent are
selected by all. The remaining 20-25 per cent are classifed in different ways by the three
measures. For this latter group and referring to the period 2010-2012, we have that
GMVD selects several fnancial institutions ignored by EC. The different classifcation
of these banks is related to the fact that they hold a small share of securities liabilities for
trading other than derivatives.
Table II.
Measures for trading
activities and
corresponding
thresholds proposed
in the literature or in
regulation, where
available
Measures
Thresholds proposed (or discussed) in
regulation/literature Thresholds via robust clustering
Total volume (bn €) Share (% total assets) Total volume (bn €) Share (% total assets)
EC 70 10 70 10
GMVD n.a. 10 70 10
HLEG 100 15-25 80 20
Table III.
Comparison between
numbers of banks
selected by the
considered metrics
for each moving
average period
EC GMVD HLEG 2007-2009 2008-2010 2009-2011 2010-2012
EC Y – – 28 28 28 30
GMVD – Y – 30 34 38 40
HLEG – – Y 35 33 38 41
EC, versus GMVD N N – 65 61 59 58
N Y – 7 11 13 12
Y N – 5 5 3 2
Y Y – 23 23 25 28
EC versus HLEG N – N 61 61 57 54
N – Y 11 11 15 16
Y – N 4 6 5 5
Y – Y 24 22 23 25
HLEG versus GMVD – N N 61 61 57 54
– N Y 9 7 9 7
– Y N 4 8 9 8
– Y Y 26 26 29 33
EC versus HLEG versus GMVD N N N 58 56 51 50
Y Y Y 22 20 22 25
Note: N and Y indicate banks non-selected or selected
383
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
As quantitative metric of similarity between measures or across periods, we use the
Jaccard index (Real and Vargas, 1996). It is defned as the size of the intersection divided
by the size of the union of the selected sets.
For a given measure and a given period, we defne a binary vector Bwhose entries are
ones if the measure selects the bank in the period, and zero otherwise.
For example, given the measure EC , the Jaccard index for the periods 2007 ?2009 and
2010 ?2012 is:
JIdx
(?
EC, 2007 ? 2009
?
,
?
EC, 2010 ? 2012
?)
?
Size(B
EC, 2007?2009
? B
EC, 2010?2012
)
Size(B
EC, 2007?2009
? B
EC, 2010?2012
)
Results are presented in Figure 2. The top plot compares the measures fxing the time
period, while the bottom plot compares each measure across time. One can see that EC
is the most stable among measures and there is a good agreement between EC and
GMVD, and between GMVD and HLEG.
5. Business model
5.1 Data and defnitions
Our goal is to describe banks business model choices by using information frombalance
sheets and income statements. On the same set of banks, we consider the following 14
fnancial variables describing business model characteristics [7]:
• Asset position: TA, net loans to customers, net loans to banks, assets available for
sale.
• Position in trading activities: Derivatives assets held for trading, net position in
securities held for trading.
Figure 2.
_Paper_Jaccard
foats here
JFEP
7,4
384
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
• Funding strategy: Customer deposits, bank deposits, subordinated debt, total
fnancial liabilities.
• Assets/Liabilities strategy: Loans to deposits ratio, leverage (equity/TA).
• Banks’ performance: Return on assets, trading income over net proft.
5.2 Methodology and results
With respect to the existing literature on banks’ business model classifcation, our
contribution focuses on extracting the main features coming from the data using PCA
and couple it with robust clustering. The most signifcant principal components will be
in fact used as an input for robust statistical clustering. Each cluster obtained from this
procedure will eventually identify a business model.
Our starting data set is a 400 ?14 matrix: each rowis an observation of the indicators
above relative to a specifc bank in a specifc three-year average period.
As often in the case of banking, data present missing values and the distribution of
variables is skewed. Both aspects make delicate the plain use of PCAon banks’ fnancial
data as discussed by Loretan (1997). We thus run PCA on a sub-sample having
eliminated 5 per cent of observations identifed as outliers by robust statistics [8] and
dropped observations with missing value for any of the variables. We standardize
variables and apply PCA on the remaining 329 observations. Figure 3 shows the share
of the variance in the data set explained by the frst ten components (bars) and their
cumulative value (line).
We consider the frst four principal components (PC
i
, i ?1…4) which explain 56 per
cent of variance in the data and we compute the correlation between PC
i
and the original
variables. To impute some missing data, we regress each PC
i
using, as regressor, the
variable with the highest correlation (in our case, they are respectively net loans to
customers, TA, customer deposits and trading income), obtaining a newPCR
i
. We verify
that the correlation does not change signifcantly. This leads to a fnal number of 349
observations (20 outliers and 31 missing observations).
Figure 3.
_PCAVariance_Paper
foats here
385
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
We run robust clustering on the PCR
i
to group banks into business models, with the
following parameters: restriction factor ?5, number of clusters ?6, trimming level ?
5 per cent. To assign a business model to each cluster, we compute some descriptive
statistics focusing mainly on the funding strategy. We present thembelowin ascending
order with respect to the median of net loans to customers:
• Leveraged investment banks (Inv), with a high amount of derivatives held for
trading, both in assets and liabilities. This feature is refected in a relatively high
share of trading income. These banks are very large in asset size [9].
• Wholesale-investment banks (WS-Inv), having a relatively high share of
wholesale funding and showing a stronger orientation toward trading activities
compared to pure wholesale banks. This cluster contains medium-large banks
with the lowest level of capital (in terms of total equity on TA).
• Retail-diversifed banks (Ret-Div), introducing derivative trading in a retail
narrow model. These banks are large and very large.
• Wholesale banks (WS), funding themselves via wholesale markets and having
their assets structure diversifed between retail loans and trading activities [10].
These banks are medium-sized.
• Retail-focused banks (Ret-Foc), focusing on customer deposits and loans. These
banks are medium-sized.
• Small retail banks, well capitalized and focusing on customers’ deposits and
loans.
Figure 4 graphically represents statistics for the Net Loans to Customer Ratio (NLCR)
and Wholesale Ratio (WSR) for 2007-2009 and 2010-2013. Each ellipse denotes a
business model cluster, whose center is the median values of the banks it contains, and
whose diameters are the Median Absolute Deviation (MAD):
MAD(X) ? median
?
X ? median(X)|.
Retail banks (which include Ret-Div, Ret-Foc and Small) have NLCRover 50 per cent and
a small WSR. Wholesale banks (WS and WS-Inv) showthe highest WSRfunding, above
30 per cent. Investment banks contains banks with a rather lowand stable NLCR(below
40 per cent): we will see later that they showthe highest share of derivatives. Comparing
2007-2009 with 2010-2012, WS moves visibly toward the retail clusters. Figure 5 shows
the within-cluster inter-quartile range of selected items from the balance sheet (assets
and liabilities).
Figure 6 presents the median values of capital and leverage. WS-Inv is the cluster
with the highest leverage and poorest capital; as expected, risk weighted assets (RWA)
is higher for retail banks. One can also see the increase of capital ratios from the left to
the right plots.
To assess the quality of the clustering exercise in terms of possible misclassifcation,
we compute the silhouette distance, which measures if a bank is classifed in a certain
cluster while still being close to banks in others: positive values refect a good
classifcation, negative the opposite (Rousseeuw, 1987). Our clusters have positive, high
silhouette values, except for a few WS-Inv, Retail banks and some of the Outliers.
JFEP
7,4
386
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Figure 4.
_Paper_Ellipses
NLC, WS foats here
387
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Figure 5.
_Paper_
AssetsLiabilities
foats here
JFEP
7,4
388
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Figure 6.
_Paper_Capital foats
here
389
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Figure 7 shows the evolution of the size of each cluster over time. Roughly 55 per cent of
the banks are retail-focused or retail-diversifed, while 23-30 per cent are identifed as
wholesale (pure or investment).
Finally we consider the transition matrix (Table IV) which highlights changes in
business model for a given bank between 2007-2009 and 2010-2012. The matrix should
be read by rows: diagonal values give the share of banks remaining in the same cluster
while off-diagonal reports in which cluster(s) banks move to. The classifcation is rather
stable (diagonal values are above 60 per cent). The main change appears on WS, which
lose 24 per cent of its member to Retail in 2010-2012 showing a decrease of wholesale
funding for these banks.
6. Linking key players in trading and business model
This section maps key players in trading into business model clusters. Figure 8 reports
the share of banks identifed as key players in trading by each measure and for each
cluster in terms of TA (top plot) and number (bottom plot).
For instance, GMVDand HLEG select roughly 60 per cent of the banks in WS-Inv in
2007-2009; the two sets are different, since the share of TAof selected banks is larger for
HLEG.
The picture is rather clear for the largest and smallest banks: all measures
consistently identify all investment banks while selecting fewretail-focused institutions
(Small and Ret-Foc). With respect to these Retail clusters, a couple of key aspects are
worth mentioning: EC is more stable in the selection over time, identifying two to three
large banks; HLEG and GMVD are more volatile, they hardly select any banks in
2007-2009, while they add to the selection by EC some small banks in 2010-2012.
The major differences among measures appear in two clusters: retail-diversifed
(Ret-Div) and wholesale-investment (WS-Inv) banks. Focusing on the frst, results show
that GMVD and HLEG recognize three to four banks not selected by EC, with no
peculiar characteristics in terms of our input variables, except their very large size. For
WS-Inv, roughly half of the banks (seven in 2007-2009, eight in 2010-2012) are selected
by all measures. Extra banks selected by GMVD/HLEG, in 2010-2012, are either the
smallest of this cluster or have a particular low share of customer deposits.
Finally the wholesale (WS) cluster contains 8 banks in 2007-2009 and 13 banks in
2010-2012. It is the least stable cluster, with 37 per cent of banks changing classifcation
from 2007-2009 to 2010-2012. Figure 4 showed that this cluster moved toward a more
retail-oriented model, decreasing wholesale funding and slightly increasing
investments into customer loans. EC and HLEG select a single bank in this cluster in
both periods while GMVD appears to be more volatile selecting no bank in 2007-2009
and three banks in 2010-2012.
7. Results and conclusions
This paper has investigated the link between the engagement of European banks in
trading and their business model. Using balance sheet and income statement data (100
banks from 2007 to 2012), we have frst investigated key players in trading activities,
according to three measures discussed by regulators and researchers in the context of
the structural reform of EU banks. Then we have developed a business model analysis
via robust clustering coupled with PCA. Such methodology, not much explored in
literature, leads to classify banks into six business model clusters, which result to be
JFEP
7,4
390
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Figure 7.
_Paper_BanksbyBM
foats here
391
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
rather stable over time. Finally we have mapped information about trading measures on
the business model clusters.
Results show that trading measures behave coherently in selecting banks with
highest and lowest involvement in trading. In fact, the mapping is rather clear for the
cluster of very large investment banks, holding more than 20 per cent of trading assets/
liabilities and having customer loans usually below 45 per cent: these are always
identifed as key players in trading. At the opposite, rare selection is made within small
and well capitalized banks and medium-sized retailed-focused banks, generally
described by a share of customers’ loans over 45 per cent and low reliance on wholesale
funding. These results were somehow expected.
For the other business models, and in particular wholesale-investment and Ret-Div,
the selection depends on the choice of the trading measure. As WS-Inv are concerned,
between 50 and 75 per cent of the banks are recognized as key trading players depending
on the measure. These banks are heavily involved in trading, fnance themselves
through bank deposits and show very high levels of leverage. Our analysis shows that
EC tends to identify fewer banks in this cluster.
Another discrepancy between the measures is observed for some very large banks,
classifed as retail-diversifed: GMVDand HLEGidentify around 85 per cent of banks as
key trading players, while EC only 65 per cent. These banks present in general lower
levels of leverage and better quality of capital, in terms of Tier1, with respect to very
large investment banks. EC selecting fewer banks in this cluster partly answers the
potential critique that trading measures cannot properly heed capital adequacy.
Furthermore, one can see that not only size matters: in particular medium-sized
WS-Inv are selected, while some very large retail-diversifed are not.
Regarding HLEG on Ret-Foc, our analysis confrms the criticism that this measure
might be selecting more retail banks, suggesting that assets available for sale could be
more coupled with retail-oriented activities than trading-oriented ones.
From a policy-makers perspective, our results pose the question whether
business models analysis has a place in the legislative context for structural
separation of trading activities. On one side, mapping trading involvement onto
business models could help regulators in understanding which measure for trading
to choose, depending on their purpose. On the other, business model analysis could
Table IV.
Transition matrix of
banks among
clusters from 2007-
2009 to 2010-2012
(Others category
includes outliers and
NaN)
Inv
(%)
WS-Inv
(%)
Ret-Div
(%)
WS
(%)
Retail
(%)
Small
(%)
Outliers
(%)
NaN
(%)
Total
(%)
Inv 100 100
WS-Inv 87 7 6 100
Ret-Div 100 100
WS 13 63 24 100
Retail 3 3 11 82 3 100
Small 40 60 100
Outliers 17 17 17 50 100
NaN 13 25 38 25 100
Note: Each row reports, for the banks’s classifcation of 2007-2009, in which business model they can
be found in 2010-2012
JFEP
7,4
392
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Figure 8.
_Paper_
SelectedBanksvsBM
foats here
393
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
support regulators in designing alternatives to strict separation, depending on
banks activities at large. In particular one could think of a targeted approach rather
than a one-fts-all: while strict separation could be the way forward for very large
investment banks, regulators could consider enhancing the supervision of
medium-large WS-Inv to better assess their leverage and capital levels and/or
closely monitor Ret-Div, that already show high level and quality of capital.
The duration and extent of discussion of the structural separation proposal within
EUinstitutions confrms the policy relevance of this issue. To conclude, one needs to say
that business model analysis of complex fnancial groups would beneft fromadditional
data such as products strategy, structure and ownership information and geographical
reach, enabling to better assess the regulation impact.
Notes
1. In January 2014 the EC proposed a Regulation on structural measures improving the
resilience of EU credit institutions. In June 2015, the ECOFIN Council agreed its position.http://ec.europa.eu/fnance/bank/structural-reform/index_en.htm
2. The analysis is based on the SNL database since it allows for a detailed disaggregation of
balance sheet items related to assets held for trading. It allows distinguishing derivatives for
trading from derivatives for hedging.
3. GMVD is the cost of replacing all outstanding contracts at current market prices. GMVD is
not readily available in the balance sheet so we propose a proxy based on the sum of
derivatives assets and liabilities.
4. Standardization allows comparing variables with different scales. Given a vector X with
average xˆ and standard deviation ?, the corresponding standardized vector is obtained
as:
X
s
?
X ? xˆI
?
5. Statistical clustering assigns observations into groups. Clusters are built starting from
random centroids and moving their positions so to minimize their dispersion. Outliers can
considerably bias the estimation of the centroids of the clusters, hence affecting the fnal
clustering. For this reason, we opted for robust trimmed clustering t-clust (Garcia-Escudero
et al., 2008), implemented in Matlab in the framework of Forward Search for Data Analysis
(Riani et al., 2012; Cariboni et al., 2015). Its robustness capacity comes fromthe possibilities to
leave a proportion of observations unassigned (trimming) and to allowthe shape of the cluster
to become ellipsoidal.
6. To determine the number of clusters and restriction factor, Bayesian Information Criterion
analysis is employed (Riani et al., 2012). Our results suggest a number of clusters equal to 8
and restriction factor equal to 200.
7. All variables except TA, loan to deposit ratio, return on assets and trading income are
expressed as share of TA.
8. Most of the outliers are banks which received state aid or went through restructuring in the
considered time horizon. Information is available upon request.
9. Our sample does not include any bank with TA lower than € 30 bn. Size buckets are used to
order banks using their TA averaged over 2010-2012: Very large banks (VL): above 500 €bn,
JFEP
7,4
394
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Large banks (L): between 200 and 500 €bn, Medium banks (M): between 100 and 200 €bn and
Small banks (S): between 30 and 100 €bn.
10. In the present work, wholesale funding is estimated as WF ?TFL-CustD-DL-Eqty-SubDebt.
References
Altunbas, Y., Manganelli, S. and Marques-Ibanez, D. (2011), “Bank risk during the fnancial crisis:
do business models matter?”, unpublished manuscript, European Central Bank, available
at: www.ecb.europa.eu/pub/pdf/scpwps/ecbwp1394.pdf
Ayadi, R., Arbak, E. and Groen, W.P.D. (2011), Business Models in European Banking: Apre- and
Post-crisis Screening, September, CEPS Paperbacks.
Ayadi, R., Arbak, E., Groen, W.P.D. and Llewellyn, D.T. (2012), Regulation of European Banks and
Business Models, June, CEPS Paperbacks.
Ayadi, R. and De Groen, W.P. (2014), Banking Business Models Monitor 2014, October, CEPS
Paperbacks, Europe.
Beltratti, A. and Stulz, R. (2009), “Why did some banks perform better during the credit crisis? A
cross-country study of the impact of governance and regulation”, Journal of Financial,
Economics, Vol. 105 No. 1, pp. 1-17.
BIS (2014), VI. The Financial System at a Crossroads, 84th BIS Annual Report, 2013-2014.
Blundell-Wignall, A., Atkinson, P. and Roulet, C. (2013), “Bank business models and the Basel
system: complexity and interconnectedness”, OECD Journal: Financial Market Trends,
Vol. 2013 No. 2, pp. 1-26.
Blundell-Wignall, A., Atkinson, P. and Roulet, C. (2014), “Bank business models and the
separation issue”, OECD Journal: Financial Market Trends, Vol. 2013 No. 2.
Blundell-Wignall, A. and Roulet, C. (2012), “Business models of banks, leverage and the
distance-to-default”, Financial Market Trends, Vol. 2.
Boot, M.A.W. and Ratnovski, L. (2012), Banking and Trading, Andrews McMeel Publishing,
Kansas City, MO.
Brewer E. III and Jagtiani, J. (2013), “How much did banks pay to become too-big-to-fail and to
become systemically Important?”, Journal of Financial Services Research Vol. 43 No. 1,
pp. 1-35.
Brunnermeier, M.K., Dong, G. and Palia, D. (2012), “Banks’ non-interest income and systemic
risk”, Unpublished Working Papers, Princeton University, Princeton.
Cariboni, J., Pagano, A., Perrotta, D. and Torti, F. (2015), “Robust clustering of EU banking data”,
in Morlini I., Vichi M. and Minerva T (Eds), Advances in Statistical Models for Data
Analysis, Springer, New York, NY, pp. 17-25. doi: 10.1007/978-3-319-17377.
Chordia, T., Goyal, A., Lehmann, B.N. and Saar, G. (2013), “High-frequency trading”, Journal of
Financial Markets, Vol. 16 No. 4, pp. 637-645.
Chow, J.T. (2011), “Making banks safer: can Volcker and Vickers do it?”, International Monetary
Fund.
Davies, R. and Tracey, B. (2012), “Too big to be effcient? The impact of too big to fail factors on
scale economies for banks”, Journal of Money, Credit, and Banking, Vol. 46 No. 1,
pp. 219-253.
Demirguerc-Kunt, A. and Huizinga, H. (2013), “Are banks too big to fail or too big to save?
International evidence fromequity prices and CDS spreads”, Journal of Banking &Finance,
Vol. 37 No. 3, pp. 875-894.
395
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
EC (2013), Commission Staff Working Document Impact Assessment, available at:http://eur-lex.
europa.eu/resource.html?uri?cellar:e186dd0b-89b3-11e3-87da-01aa75ed71a1.0001.01/DO
C_1&format?PDF
ECB (2010), “Report on EU banking structures”, September, available at: www.ecb.europa.eu/
pub/pdf/other/eubankingstructures201009en.pdf
ECB (2013), “Banking structures report”, November, available at: www.ecb.europa.eu/pub/pdf/
other/bankingstructuresreport201311en.pdf?5656762fc7710c2ad62b381f432eff9f
Gambacorta, L. and Rixtel, A.A.V. (2013), “Structural bank regulation initiatives: approaches and
implications”, BIS, Working Papers No 412, April.
Garcia-Escudero, L.A., Gordaliza, A., Matrán, C. and Mayo-Iscar, A. (2008), “A general
trimming approach to robust cluster analysis”, The Annals of Statistics, Vol. 36 No. 3,
pp. 1324-1345.
Inci, A. Narain, A., Ilyina, A. and Surti, J. (2011), “The too-important-to-fail conundrum”,
International Monetary Fund, Staff Discussion Notes No. 11/12.
Laeven, M.L., Ratnovski, L. and Tong, H. (2014), Bank Size and Systemic Risk, International
Monetary Fund.
Liikanen, E., Bänziger, H., Campa, J.M., Gallois, L., Goyens, M. Krahnen, J.P., Mazzucchelli, M.,
Sergeant, C., Tuma, Z., Vanhevel, J. and Wijffels, H. (2012), High-level Expert Group on
Reforming the Structure of the EU Banking Sector, 2, Final Report, Brussels.
Loretan, M. (1997), “Generating market risk scenarios using principal components analysis:
methodological and practical considerations”, available at: www.bis.org/publ/ecsc07c.pdf
Pagano, A. (2013), “Trading activities and bank structural separation: possible defnitions and
calibration of screening thresholds”, JRC Scientifc and Policy Report.
Real, R. and Vargas, J.M. (1996), “The probabilistic basis of Jaccard’s index of similarity”,
Systematic Biology, Vol. 45 No. 3, pp. 380-385.
Riani, M., Perrotta, D. and Torti, F. (2012), “FSDA: a matlab toolbox for robust analysis and
interactive data exploration”, Chemometrics and Intelligent Laboratory Systems, Vol. 116,
pp. 17-32.
Rousseeuw, P.J. (1987), “Silhouettes: a graphical aid to the interpretation and validation of cluster
analysis”, Journal of Computational and Applied Mathematics, Vol. 20, pp. 53-65.
Schoenmaker, D. and T., Peek (2014), “The state of the banking sector in Europe”, OECD
Economics Department Working Papers, No. 1102, OECD Publishing, available at: http://
dx.doi.org/10.1787/5k3ttg7n4r32-en
Stiroh, K.J. and Rumble, A. (2006), “The dark side of diversifcation: the case of US fnancial
holding companies”, Journal of Banking & Finance, Vol. 30 No. 8, pp. 2131-2161.
Ueda, K. and di Mauro, B.W. (2013), “Quantifying structural subsidy values for systemically
important fnancial institutions”, Journal of Banking & Finance, Vol. 37 No. 10,
pp. 3830-3842.
Vickers (2011), “The independent commission on banking: the vickers report &the parliamentary
commission on banking standards”, available at:http://researchbriefngs.parliament.uk/R
esearchBriefng/Summary/SN06171
Viñals, J. (2013), “Creating a safer fnancial system: will the Volcker, Vickers, and Liikanen
Structural measures help?”, International Monetary Fund, Staff Discussion Notes No. 13/4.
JFEP
7,4
396
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Appendix. Banks in the sample and their classifcation by cluster type
Table AI.
List of banks in the
sample by cluster
type, classifcation as
of 2010-2012
2010-2012 2007-2009
Classification in Business model
Bank Name | Size
EC GMVD HLEG EC GMVD HLEG
Other business model or size,
only if a change is observed.
Very large and leveraged investment banks Inv 8 8 8 7 7 7
Y Y Y Y Y Y L V s y a l c r a B
Y Y Y Y Y Y L V s a b i r a P P N B
Cr. Agricole Grp VL Y Y Y Y Y Y
Y Y Y Y Y Y L V k n a B e h c s t u e D
Y Y Y Y Y Y L V s g n i d l o H C B S H
Y Y Y L V a e d r o N Missing
Y Y Y Y Y Y L V p u o r G S B R
Y Y Y Y Y Y L V n e G c o S
Medium-large wholesale investment banks
WS-Inv
8 12 10 7 9 9
Y Y Y Y Y Y L V k n a b z r e m m o C
N N L V a i x e D Y Size=L
Y Y Y Y Y Y L B L e h c s i r e y a B
Y Y Y Y Y Y L s u i f l e B
Deutsche Zentral G L Y Y N Y Y N
N L a i x e D Y Y Size=L
L g l H E R o p y H Y Y N N N N
LB Baden Württemberg L Y Y Y Y Y Y
N N L B L / D R O N Y N N Y
Y Y Y L n o g i t r o P Outliers
DekaBank Deutsche GZ M Y Y Y Y Y Y
N M k n a b d r o N H S H Y N WS
La Banque Postale M N N N N N N
N M g l H n i l r e B B L Y Y N Y Y
AXA Bank Europe S Y Y Y Retail
Banque Int à Lux S N Y Y Outliers
Österreichische Volksbanken S N Y N WS
RaifLB Niederösterreich S N N N N N N
Westdeutsche GZ S N N N N N N
Large and very large retail diversified banks
Ret-Div
10 13 13 8 11 13
L V A V B B N Y Y N Y Y
Y Y Y Y Y Y L V r e d n a t n a S
N N L V p r G l e u t u M r C Y N N Y
Y Y Y Y Y Y L V E C P B e p u o r G
Y Y Y Y Y Y L V k n a B G N I
N L V o l o a p n a S a s e t n I Y Y N Y Y
N N Y Y Y L V p u o r G s d y o l L Y
Rabobank Group VL N Y Y N Y Y
Y Y Y Y Y Y L V t i d e r C i n U
Y Y Y Y Y Y L k n a B e k s n a D
N L p r G C B K Y Y Y Y Y
L B E S Y Y N Y Y Y
Y Y Y Y Y Y L t r a h C d n a t S
L k n a b d e w S Y N N Retail
Y Y Y S a c n a b o i d e M Outliers
(continued)
397
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table AI.
Cont (2/3)
2010-2012 2007-2009
Classification in Business model
Bank Name | Size
EC GMVD HLEG EC GMVD HLEG
Other business model or size,
only if a change is observed.
Medium sized wholesale banks WS 1 3 1 1 0 1
Banco Fin. y de Ahorros L N Y g n i s s i m N
Allied Irish Banks, Plc M N N N Retail
N M k n a b d r o N H S H Y N WS-Inv
M A B A L E H Y Y Y Y N Y
l i a t e R N N N M e c e e r G k B l a n o i t a N
l i a t e R N N N M h c i e r r e t s Ö k b Z . f i a R
N N N S k n a B a h p l A Retail
N N N N N N S p r g a t n e g r A
N N N S m u r t s o N e r a M missing
Cyprus Pop Bk Public S N N N Retail
Hypo Alpe-Adria-Bank S N N N N N N
IKB Deutsche Industriebank S N N N Outliers
N N N N N N S S / A k n a B e k s y J
Österreichische Volksbanken S N N N WS-Inv
RaiffLB Oberösterreich S N N N N N N
Sparkasse KölnBonn S N N N Retail
Medium sized retail focused Retail 2 3 6 3 1 1
ABN AMRO Grp L N N N N N N
Monte Paschi di Siena L Y Y Y Y N N
Caja Aho. y Pens Barcel. L N N N N N N
Cassa depositi e prestiti L N N N N N N
L B N D Y N N Size = M
N N N N N N L k n a B p r G e t s r E
Nationwide Bldg Soc. L N N N N N N
N Y N N N N L n e k n a b s l e d n a H
Allied Irish Banks M N N N WS
N N N M l l e d a b a S Size = S
N N N N N N M e r a l o p o P o c n a B
Banco Popular Español M N N N N N N
N N N N N N M d n a l e r I f o k n a B
Caixa Geral de Depósit M N N N N N N
M B N D Y N N Size = L
National Bank of Greece M N N N WS
M t i d e r k l a e R N N N N N N
Raif. ZBk Österreich M N N N WS
M L A A E R S N S N N Y Small
M k n a b d e w S Y N N Ret-Div
N N N N N N M I B U
Aareal Bank S N N N N N N
Alpha Bank S N N N WS
AXA Bank Europe S N N Y WS-Inv
Banca Carige S N N Y Small
Bp lEmilia Romagna S N N N N N N
BP Milano S N N Y N N N
BP Vicenza S N N N N N N
BPI S N N Y N N N
Banco Comercial Português S N N N N N N
Sabadell S N N N Size = M
BAWAG S N Y Y N N N
Bank of Cyprus Public Company S N N N N N N
Bankinter S N N N N N N
Cajas Rurales Unidas S N N N N N N
Clydesdale Bank S N N N N N N
Co-operative Bank S N N N N N N
Cyprus Popular Bank Public S N N N WS
Deutsche Apotheker- und Ärztebank S N N N N N N
Espirito Santo Financial Group S N N N N N N
Liberbank S N N N Missing
Münchener HypothekenbanK S N N N N N N
Permanent TSB S N N N Missing
Piraeus Bank SA S N N N Outliers
Sparkasse KölnBonn S N Y N WS
Volksbanken-Verbund S N N N Missing
Yorkshire Building Society S N N N N N N
(continued)
JFEP
7,4
398
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
About the authors
Francesca Campolongo is the Head of the Economic and Financial Analysis at the European
Commission Joint Research Centre since January 2012. She has been working in the Joint Research
Centre since 1998, where she also obtained a prize as “best young scientist of the year” in 2002.
Currently, she is actively involved in the work of the European Commission to create a safer and
sounder fnancial system and to recover from the economic crisis. In particular, in the last few
years she has contributed to European Commission proposals on: higher capital requirements for
banks, harmonized deposit protection schemes, EUFramework for bank recovery and resolution,
Table AI.
Cont (3/3)
2010-2012 2007-2009
Classification in Business model
Bank Name | Size
EC GMVD HLEG EC GMVD HLEG
Other business model or size,
only if a change is observed.
medium sized wholesale banks
WS
1 3 1 1 0 1
Banco Fin. y de Ahorros L N Y g n i s s i m N
Allied Irish Banks, Plc M N N N Retail
N M k n a b d r o N H S H Y N WS-Inv
M A B A L E H Y Y Y Y N Y
l i a t e R N N N M e c e e r G k B l a n o i t a N
l i a t e R N N N M h c i e r r e t s Ö k b Z . f i a R
N N N S k n a B a h p l A Retail
N N N N N N S p r g a t n e g r A
N N N S m u r t s o N e r a M missing
Cyprus Pop Bk Public S N N N Retail
Hypo Alpe-Adria-Bank S N N N N N N
IKB Deutsche Industriebank S N N N Outliers
N N N N N N S S / A k n a B e k s y J
Österreichische Volksbanken S N N N WS-Inv
RaiffLB Oberösterreich S N N N N N N
Sparkasse KölnBonn S N N N Retail
Small retail
Small
0 0 0 0 0 0
N N N M L A E E R S N S Retail
N N N S e g i r a C a c n a B Retail
N N N N N N S k n a B P T O
N N N N N N S i k s l o P O K P
Volkswagen Bank S N N N N N N
Outliers 1 1 3 1 1 3
Y Y Y M n o g i t r o P WS-Inv
N N S . x u L , E E C B Y N N Y
Banque Int à Lux S N N Y WS-Inv
N N N N N N S t i d e r k F R B
IKB Deutsche Industriebank S N Y N WS
S k n a b n e t n e R N N Y N N N
S a c n a b o i d e M Y N Y Retail-Div
N N N S k n a B s u e a r i P Retail
Missing data 0 0 0 1 1 1
Y Y Y L k n a B a e d r o N Inv
Hamburger Sparkasse N N N N N N
L kred. Baden-Wurttemberg N N N N N N
Banco Fin. y de Ahorros N N N WS
Volksbanken-Verbund N N N Retail
N N N m u r t s o N e r a M WS
N N N k n a b r e b i L Retail
N N N B S T t n e n a m r e P Retail
Notes: The second column describe the size (VL = very large, L = Large, M = Medium-Sized;
S = Small, see footnote 15). Y(N) labels a bank that is selected (not selected) based on the
three-year average value of the metrics, with period specified in the head row; the last columns
specified if a bank has changed cluster moving from the first to the last period and details its
origin in the period 2007-2009. Bolded labels identify cases when not all metrics select the
bank as being above the threshold that identify key player in trading
399
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
and bank structural reform. As background Francesca is a mathematician (she graduated in Pisa
University, Italy) with a PhDin modeling and sensitivity analysis (Griffth University, Australia),
her main research interests focus on fnancial markets modeling and fnancial risk analysis.
Working in support to EU policy makers, she has developed skills in calibration and ex ante
impact assessment of fnancial regulation, with particular attention to the banking sector. She is
the author of ?30 publications in academic journals, co-author of 3 books on sensitivity analysis
published by Wiley and co-author of 2 books on Securitization published by Wiley and Springer.
Jessica Cariboni is an offcer of the Unit Economic and Financial Analysis of the European
Commission Joint Research Centre. Born in 1976, Jessica graduated with honors in Physics at
University of Milan, Italy, in 2000. From March 2001 to December 2002 Jessica was a junior
quantitative analyst in one of the leading Italian companies in fnancial investments, Nextra
Investment Management SGR. In 2007, she completed her PhD in fnancial modeling at the
Department of Mathematics of the Katholieke Universiteit of Leuven. From2003 Jessica has been
working as researcher at the European Commission Joint Research Centre. Her main
responsibility is to develop research and quantitative analyses for impact assessments of new
regulatory proposals in the feld of banking regulation and fnancial stability. Her major felds of
research are fnancial modeling, credit risk modeling and risk analysis. Jessica has over ten
scientifc publications on refereed journals, and she is also co-author of two books on credit
risk-modeling and sensitivity analysis published by Wiley. Jessica Cariboni is the corresponding
author and can be contacted at: [email protected]
Nathalie Ndacyayisenga is a researcher of the Unit Economic and Financial Analysis of the
European Commission Joint Research Centre. Nathalie graduated, major of her promotion, at
Université Paul Sabatier in Toulouse (France) in mathematics and computer sciences. Before
joining the unit, Nathalie has worked for the last decade in analytical and research departments of
large private companies mostly Belgium, Italy, developing technical skills such as accounting,
risk management and currently policy-making processes. Her main domain of expertise
encompasses data mining, prediction modeling and customer base analysis.
Andrea Pagano holds a Ph.D. in mathematics from Brown University. After few years in
academia working on complex geometry, he started to work on environmental modeling at ENEA.
In 2005 he joined the Joint Research Centre (JRC) of the European Commission, where he worked
on Global Sensitivity Analysis. In 2011 he joined the group of Economic and Financial Analyses
where he works as data analyst and modeler.
For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: [email protected]
JFEP
7,4
400
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
doc_901385569.pdf
The purpose of this paper is to do an empirical analysis assessing whether banks highly
involved into trading activities show specific business model choices. Key factors in the analysis are a
proper measure for trading activities and a consistent classification of banks in terms of business
choices.
Journal of Financial Economic Policy
Banks under X-rays: business model choices and trading
Francesca Campolongo J essica Cariboni Nathalie Ndacyayisenga Andrea Pagano
Article information:
To cite this document:
Francesca Campolongo J essica Cariboni Nathalie Ndacyayisenga Andrea Pagano , (2015),"Banks
under X-rays: business model choices and trading", J ournal of Financial Economic Policy, Vol. 7 Iss 4
pp. 377 - 400
Permanent link to this document:http://dx.doi.org/10.1108/J FEP-12-2014-0081
Downloaded on: 24 January 2016, At: 21:53 (PT)
References: this document contains references to 35 other documents.
To copy this document: [email protected]
The fulltext of this document has been downloaded 43 times since 2015*
Users who downloaded this article also downloaded:
Eric Osei-Assibey, J oseph Kwadwo Asenso, (2015),"Regulatory capital and its effect on credit growth,
non-performing loans and bank efficiency: Evidence from Ghana", J ournal of Financial Economic
Policy, Vol. 7 Iss 4 pp. 401-420http://dx.doi.org/10.1108/J FEP-03-2015-0018
J acob Kleinow, Tobias Nell, (2015),"Determinants of systemically important banks: the case of
Europe", J ournal of Financial Economic Policy, Vol. 7 Iss 4 pp. 446-476http://dx.doi.org/10.1108/
J FEP-07-2015-0042
Muhammad Umar, Gang Sun, (2015),"Country risk, stock prices, and the exchange rate of the
renminbi", J ournal of Financial Economic Policy, Vol. 7 Iss 4 pp. 366-376http://dx.doi.org/10.1108/
J FEP-11-2014-0073
Access to this document was granted through an Emerald subscription provided by emerald-
srm:115632 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald
for Authors service information about how to choose which publication to write for and submission
guidelines are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company
manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as
well as providing an extensive range of online products and additional customer resources and
services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the
Committee on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for
digital archive preservation.
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
*Related content and download information correct at time of
download.
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Banks under X-rays: business
model choices and trading
Francesca Campolongo, Jessica Cariboni,
Nathalie Ndacyayisenga and Andrea Pagano
Institute for Protection and Security of Citizens,
Joint Research Centre - European Commission, Ispra, Italy
Abstract
Purpose – The purpose of this paper is to do an empirical analysis assessing whether banks highly
involved into trading activities show specifc business model choices. Key factors in the analysis are a
proper measure for trading activities and a consistent classifcation of banks in terms of business
choices.
Design/methodology/approach – We investigate three measures for trading activities proposed by
regulators in the context of bank structural reformin Europe. Through robust statistics we identify the
key trading players and classify banks into a limited number of business model clusters, relying on a set
of balance sheet and income statement indicators.
Findings – Using a sample of 100 European banks in 2007-2012, results show that the measures
identify similar, but not identical, sets of banks highly involved into trading. The measure proposed by
the European Commission selects fewer banks and is more consistent over time. The business model
analysis identifes six rather stable clusters, from small-medium retail-focused banks to very large
investment groups. The measures coherently identify as key trading players the largest investment
groups and select very few retailed focused banks. Differences among measures arise for very large
retail-diversifed and medium/large wholesale banks.
Originality/value – These results could feed the debate on which measures for trading regulators
could consider depending on the target of the reformthey would implement. For instance we showthat
the measure proposed by the European Commission selects less well capitalized retail-diversifed banks
compared to the others.
Keywords Banks, Quantitative and mathematical studies
Paper type Research paper
1. Policy context and scope
The recent crisis has forced regulators and researchers to investigate possible initiatives
to enhance the stability of the fnancial sector. These include new capital and liquidity
requirements, the implementation of new resolution regimes, the review of the
functioning of the markets and the risk entailed in certain types of instruments. The
appropriate design of a regulatory framework for tackling the coexistence of trading
and retail activities for systematically important fnancial institutions remains under
JEL classifcation – G01, G21, C38
The authors would like to thank A. Blundell-Wignall and S. Schich from OECD for their
valuable suggestions and comments, which help us to improve our work. The authors are
responsible for any remaining errors.
The content of this article does not refect the offcial opinion of the European Commission.
Responsibility for the information and views expressed therein lies entirely with the author(s).
The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/1757-6385.htm
Banks under
X-rays
377
Received23 December 2014
Revised3 July2015
Accepted6 July2015
Journal of Financial Economic
Policy
Vol. 7 No. 4, 2015
pp. 377-400
©Emerald Group Publishing Limited
1757-6385
DOI 10.1108/JFEP-12-2014-0081
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
discussion [1]. This is recognized as a potential risk for fnancial stability (Gambacorta
and Rixtel, 2013). Stiroh and Rumble (2006) fnd evidence that diversifcation benefts
for USA holding companies can be accompanied by volatility risks. Moreover trading
can be used for risk-shifting and could bring excessive leverage (Boot and Ratnovski,
2012), hence increasing banks’ complexity and systemic risk (Brunnermeier et al., 2012),
associated with implicit and/or explicit State guarantees (Ueda and di Mauro, 2013;
Davies and Tracey, 2012; Brewer III and Jagtiani, 2013; Inci et al., 2011; EC, 2013).
Many countries are taking steps toward a structural reform of their banking sector
via either ring-fencing or separating retail-oriented services from more risky trading
activities (Liikanen et al., 2012; Vickers, 2011; Merkley and Levin, 2011). The
identifcation of the banks that could undergo through this reform is also under
discussion within European institutions.
Such initiatives are nonetheless controversial due to the costs that their
implementation may imply (Viñals, 2013) and diffcult to pursue, due to the unclear
fence between such activities. Moreover a risk of a migration of some activities to
unregulated parts of the system exists (Chow, 2011). Other voices recognize that
effective structural separation measures should be coupled with other macro/micro
regulatory measures (Laeven et al., 2014) and should be designed and tailored heeding
bank business models (Viñals, 2013).
In this context, this paper presents an empirical analysis addressing the
identifcation of European banks that could be proposed for a structural reform and
it examines the link between banks’ involvement in trading and business model
choices. We frst investigate which are the key players in trading, based on some
defnitions as recently discussed in the European Union (EU). Second, bank business
models are characterized via clustering analyses of balance sheet indicators.
Finally, combining the results of the frst two goals, we assess the link between
business model choices and trading. More specifcally, we discuss how trading is
associated with different fnancing and capital-level choices. We aim at giving
indications through the lens of business model choices on how a legislation aiming,
not only at isolating trading activities, but also at reducing banks’ complexity and
systemic risk, could reshape the fnancial sector. All the analyses are based on a
sample of 100 European banks over years 2007 to 2012.
The paper is structured as follows. Section 2 describes the research context. Section
3 describes the steps of our analysis. Section 4 addresses the issue of howtrading can be
measured based on balance sheet information and howbanks highly involved in trading
can be identifed. Section 5 shows the methodology and results for the business model
analysis. Section 6 pulls together all information, and the fnal section presents our
conclusions. Appendix 1 gives details of the banks included in the sample.
2. Research context
The role of trading within the structural separation debate is a rather newtopic, and the
literature is still not well developed. Nevertheless, several scholars address some of the
issues which can be broadly related to our analysis. In the remaining of this section, we
detail our place in the existing literature according to the three main questions the paper
aims to address.
JFEP
7,4
378
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
2.1 Defnitions of trading
Our frst objective is to measure the involvement of European banks into trading
activities. Aproper defnition of trading is rather ambitious per se, due to e.g. the unclear
distinction between proprietary trading and market making, newchallenges brought by
modern fnance and the use of instruments like high-frequency trading (Chordia et al.,
2013). We compare three measures, recently proposed in the context of structural reform
in Europe, which aims to capture the involvement of banks in trading relying only on
balance sheet data. In Liikanen et al. (2012), trading activities refer only to the assets side
and they also include available for sale. Blundell-Wignall and Roulet (2012) propose a
measure of trading based on the Gross Market Value of Derivatives (GMVD), since they
identify GMVDas one of the drivers of banks’ distance to default. Finally, the European
Commission (EC; see EC Annex A8 (2013)) assesses a measure which aims at capturing
gross volumes of trading, and also proposes screening thresholds to identify the most
important players in trading.
2.2 Banks’ business model
Our second goal is to characterize banks’ business models using robust statistical
techniques. We aim to classify banks into a limited number of business model groups
ranging from pure retail to very large investment institutions, using information from
the consolidated balance sheet on their assets and liabilities, capital and leverage levels
and their proftability.
Research on banks’ business model ranges from a general discussion on their trend
(BIS, 2014; ECB, 2010; 2013) to more specifc issues focusing on how business model
information can be used for regulatory purposes (Blundell-Wignall et al., 2013; Laeven
et al., 2014). Later studies (Ayadi et al., 2011; Ayadi et al., 2012; Ayadi and De Groen,
2014) aim to describe, via standard clustering techniques, the business model evolution
of a set of large European banks, which is also the approach we retain in the current
paper.
A different perspective is followed in some econometric studies linking bank
business model and risk. In general terms it has been shown that retail-focused or
diversifed banks are usually safer in terms of distance-to-default (Blundell-Wignall
and Roulet, 2012; Blundell-Wignall et al., 2014), while riskier banks are those relying
on non-interest income and non-deposit funding (Demirguerc-Kunt and Huizinga,
2013) and/or those characterized by low capital, large size, great reliance on
short-term market funding and aggressive credit growth (Altunbas et al., 2011;
Beltratti and Stulz, 2009).
2.3 Trading and business model
As a fnal point, combining the results of the frst two questions, we assess the link
between business model choices and key players in trading. Our working hypothesis is
that fnancing and capital-level choices play a role when structural separation is
addressed. Literature discussing business model within structural separation is very
limited with few important exceptions such as Blundell-Wignall et al. (2014).
2.4 Methodological aspects
As far as analytical aspects are concerned, banking data usually show skewed
distributions and outliers. We employ robust clustering (Garcia-Escudero et al.,
2008) and robust outlier detection (Riani et al., 2012) to properly assess these
379
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
features, which so far have been rarely employed in the analysis of banking data
(Cariboni et al., 2015).
3. Steps of the analysis
(1) Identifying banks highly involved into trading:
• Measuring trading activities. As already mentioned in 2.1, we investigate
three different measures for trading.
• Selecting banks highly involved in trading activities. Through robust
clustering we set thresholds to discriminate banks mostly engaged into
trading. Candidates are those with a high absolute or relative volume of
trading activities. We discuss similarities and differences among the sets of
banks selected using the various measures of trading.
(2) Business model analysis:
• Selecting few variables describing banks business models. We start from a
large number of balance sheet and income statement variables. We restrict
our attention to 14 indicators which present the lower correlation coeffcient.
Through Principal Component Analysis (PCA) we move to the four most
relevant principal components.
• Determining banks business model clusters. Via robust clustering on the
relevant principal components, we identify six business models, ranging
from very small banks with purely retail activities, to very large investment
banks.
(3) Assessing the link between business model choices and trading
Coupling information fromthe previous steps, we map banks highly involved in
trading onto business model clusters.
4. Measuring trading activities in EU banks
4.1 Data and defnitions
To describe trading, data are extracted from SNL database [2] (see Table I).
Using these variables, we consider three measures for trading recently proposed for
the structural reform:
• EC: The measure proposed by the EC in EC Annex A8 (2013). This measure
focuses on the gross volumes of securities held for trading and includes the
liability data to proxy market and counterparty risk:
M
EC
?
TSA ? TSL
2
.
• GMVD: A proxy for the GMVD [3], as described by Blundell-Wignall and Roulet
(2012) because of its role in determining banks default risk, especially if coupled
with wholesale funding and low levels of liquid trading assets:
M
GMVD
? DA ? DL.
JFEP
7,4
380
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
• HLEG: The measure proposed by high level expert group (HLEG), which
includes also assets available for sale:
M
HLEG
? TSA ? TAFS.
Each measure is accompanied by its share with respect to Total Assets (TA).
Our sample contains 100 banks per period (see Annex 1) located in the EU. This
sample is obtained by selecting all EU banks whose average TA over 2007-2012 are
higher than €30 bn. The sample covers 18 EUcountries, with average aggregated TAof
roughly €32 trillion, representing 73 per cent of EU28 TA(Schoenmaker and Peek, 2014).
All EU Globally Systematically Important Banks are included.
4.2 Methodology and results
We focus, for each measure, on four periods of three-year moving average (2007-2010,
2008-2010, 2009-2011 and 2010-2012), leading to a sample containing 400 observations.
Table I.
Classes of assets and
liabilities used for
computing the
measures assessing
trading activities
(source SNL)
Item Label Data defnition
Assets Total Assets TA All assets owned by the company as carried
on the balance sheet
Derivatives Held for Trading
a
DA Derivatives with positive replacement
values not identifed as hedging or
embedded derivatives
Total Assets Held For
Trading
TSA Trading portfolio assets are: assets acquired
principally for the purpose of selling in the
near term, assets that on initial recognition
are part of a portfolio of identifed fnancial
instruments that are managed together and
for which there is evidence of a recent
actual pattern of short-term proft-taking, or
derivative assets
Total Assets Available for
Sale
TAFS Total loans and securities designated as
available for sale; or are not classifed as
loans and receivables, held-to-maturity
investments or fnancial assets at fair value
through proft or loss
Liabilities Derivative Held for Trading DL Derivatives with negative replacement
values not identifed as hedging
instruments
Total Securities Held for
Trading
TSL Trading liabilities that are taken with the
intent on repurchasing in the near term or a
portfolio of managed fnancial instruments
where there is evidence of a recent actual
patter of short-term proft-taking
Notes:
a
A derivative is a fnancial instrument with all of the following three characteristics: its value
changes in response to the change in an underlying variable; it requires no initial net investment or an
initial net investment that is smaller than would be required for other contracts that would be expected
to have a similar response to changes in market factors; it is settled at a future date
381
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
We aimto select, as in Pagano (2013), banks mostly engaged in trading activities. For
each measure M
i
we consider the two dimensional space defned by its standardized
absolute value and its standardized share over TA[4]. We apply trimmed clustering
techniques (so called t-clustering), which allows the creation of ellipsoidal groups that
better ft skewed data [5].
The t-clustering algorithmrequires choosing some parameters, such as the number of
clusters; the restriction factor, which describes the shape of clusters; and the trimming
level, which concerns the share of expected outliers [6].
Clusters are then used to set thresholds according to the following criteria:
• they avoid cutting clusters;
• they are multiple of €10 bn for the volume and 1 per cent for the share to be
applicable in regulation; and
• they are comparable to the ones discussed among regulators.
Figure 1 presents the results of the clustering exercise for the EC measure. The x-axis
shows the absolute value of trading assets defned by EC (in € bn), the y-axis shows its
corresponding share over TA.
The right plot zooms on the region where thresholds are set. If we focus on banks
with a low share of trading, we observe an empty region with respect to volumes (i.e.
around €70 bn for EC and GMVD, and around € 80 bn for HLEG, not shown).
Considering the share of trading assets over TA, setting the thresholds is more
diffcult, due to high density of banks along this axis: t-clustering however helps
identifying a horizontal threshold at 10 per cent for EC. Based on similar exercises run
on the whole sample, we set thresholds for each measure as reported in the right part of
Figure 1.
_Paper_ECClusters
foats here
JFEP
7,4
382
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table II. One can see they are very close to those found in regulation and literature (see
left part of Table II).
We investigate the similarity between measures for a given period and also the
stability of each measure across the years. Table III (top three rows) reports the number
of banks highly involved in trading according to each measure in each period. One can
see that GMVD and HLEG are the most volatile over the years, with a clear increasing
trend for GMVD, while EC tends to select the same number of banks.
To better assess the different behaviors of the measures considered, we confront each
pair of measures by comparing the list of banks selected by both (intersection) versus
the ones selected by only one of the two (bottompart of Table III). Roughly 50-60 per cent
of the banks in the sample are never selected by any measure, while 20-25 per cent are
selected by all. The remaining 20-25 per cent are classifed in different ways by the three
measures. For this latter group and referring to the period 2010-2012, we have that
GMVD selects several fnancial institutions ignored by EC. The different classifcation
of these banks is related to the fact that they hold a small share of securities liabilities for
trading other than derivatives.
Table II.
Measures for trading
activities and
corresponding
thresholds proposed
in the literature or in
regulation, where
available
Measures
Thresholds proposed (or discussed) in
regulation/literature Thresholds via robust clustering
Total volume (bn €) Share (% total assets) Total volume (bn €) Share (% total assets)
EC 70 10 70 10
GMVD n.a. 10 70 10
HLEG 100 15-25 80 20
Table III.
Comparison between
numbers of banks
selected by the
considered metrics
for each moving
average period
EC GMVD HLEG 2007-2009 2008-2010 2009-2011 2010-2012
EC Y – – 28 28 28 30
GMVD – Y – 30 34 38 40
HLEG – – Y 35 33 38 41
EC, versus GMVD N N – 65 61 59 58
N Y – 7 11 13 12
Y N – 5 5 3 2
Y Y – 23 23 25 28
EC versus HLEG N – N 61 61 57 54
N – Y 11 11 15 16
Y – N 4 6 5 5
Y – Y 24 22 23 25
HLEG versus GMVD – N N 61 61 57 54
– N Y 9 7 9 7
– Y N 4 8 9 8
– Y Y 26 26 29 33
EC versus HLEG versus GMVD N N N 58 56 51 50
Y Y Y 22 20 22 25
Note: N and Y indicate banks non-selected or selected
383
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
As quantitative metric of similarity between measures or across periods, we use the
Jaccard index (Real and Vargas, 1996). It is defned as the size of the intersection divided
by the size of the union of the selected sets.
For a given measure and a given period, we defne a binary vector Bwhose entries are
ones if the measure selects the bank in the period, and zero otherwise.
For example, given the measure EC , the Jaccard index for the periods 2007 ?2009 and
2010 ?2012 is:
JIdx
(?
EC, 2007 ? 2009
?
,
?
EC, 2010 ? 2012
?)
?
Size(B
EC, 2007?2009
? B
EC, 2010?2012
)
Size(B
EC, 2007?2009
? B
EC, 2010?2012
)
Results are presented in Figure 2. The top plot compares the measures fxing the time
period, while the bottom plot compares each measure across time. One can see that EC
is the most stable among measures and there is a good agreement between EC and
GMVD, and between GMVD and HLEG.
5. Business model
5.1 Data and defnitions
Our goal is to describe banks business model choices by using information frombalance
sheets and income statements. On the same set of banks, we consider the following 14
fnancial variables describing business model characteristics [7]:
• Asset position: TA, net loans to customers, net loans to banks, assets available for
sale.
• Position in trading activities: Derivatives assets held for trading, net position in
securities held for trading.
Figure 2.
_Paper_Jaccard
foats here
JFEP
7,4
384
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
• Funding strategy: Customer deposits, bank deposits, subordinated debt, total
fnancial liabilities.
• Assets/Liabilities strategy: Loans to deposits ratio, leverage (equity/TA).
• Banks’ performance: Return on assets, trading income over net proft.
5.2 Methodology and results
With respect to the existing literature on banks’ business model classifcation, our
contribution focuses on extracting the main features coming from the data using PCA
and couple it with robust clustering. The most signifcant principal components will be
in fact used as an input for robust statistical clustering. Each cluster obtained from this
procedure will eventually identify a business model.
Our starting data set is a 400 ?14 matrix: each rowis an observation of the indicators
above relative to a specifc bank in a specifc three-year average period.
As often in the case of banking, data present missing values and the distribution of
variables is skewed. Both aspects make delicate the plain use of PCAon banks’ fnancial
data as discussed by Loretan (1997). We thus run PCA on a sub-sample having
eliminated 5 per cent of observations identifed as outliers by robust statistics [8] and
dropped observations with missing value for any of the variables. We standardize
variables and apply PCA on the remaining 329 observations. Figure 3 shows the share
of the variance in the data set explained by the frst ten components (bars) and their
cumulative value (line).
We consider the frst four principal components (PC
i
, i ?1…4) which explain 56 per
cent of variance in the data and we compute the correlation between PC
i
and the original
variables. To impute some missing data, we regress each PC
i
using, as regressor, the
variable with the highest correlation (in our case, they are respectively net loans to
customers, TA, customer deposits and trading income), obtaining a newPCR
i
. We verify
that the correlation does not change signifcantly. This leads to a fnal number of 349
observations (20 outliers and 31 missing observations).
Figure 3.
_PCAVariance_Paper
foats here
385
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
We run robust clustering on the PCR
i
to group banks into business models, with the
following parameters: restriction factor ?5, number of clusters ?6, trimming level ?
5 per cent. To assign a business model to each cluster, we compute some descriptive
statistics focusing mainly on the funding strategy. We present thembelowin ascending
order with respect to the median of net loans to customers:
• Leveraged investment banks (Inv), with a high amount of derivatives held for
trading, both in assets and liabilities. This feature is refected in a relatively high
share of trading income. These banks are very large in asset size [9].
• Wholesale-investment banks (WS-Inv), having a relatively high share of
wholesale funding and showing a stronger orientation toward trading activities
compared to pure wholesale banks. This cluster contains medium-large banks
with the lowest level of capital (in terms of total equity on TA).
• Retail-diversifed banks (Ret-Div), introducing derivative trading in a retail
narrow model. These banks are large and very large.
• Wholesale banks (WS), funding themselves via wholesale markets and having
their assets structure diversifed between retail loans and trading activities [10].
These banks are medium-sized.
• Retail-focused banks (Ret-Foc), focusing on customer deposits and loans. These
banks are medium-sized.
• Small retail banks, well capitalized and focusing on customers’ deposits and
loans.
Figure 4 graphically represents statistics for the Net Loans to Customer Ratio (NLCR)
and Wholesale Ratio (WSR) for 2007-2009 and 2010-2013. Each ellipse denotes a
business model cluster, whose center is the median values of the banks it contains, and
whose diameters are the Median Absolute Deviation (MAD):
MAD(X) ? median
?
X ? median(X)|.
Retail banks (which include Ret-Div, Ret-Foc and Small) have NLCRover 50 per cent and
a small WSR. Wholesale banks (WS and WS-Inv) showthe highest WSRfunding, above
30 per cent. Investment banks contains banks with a rather lowand stable NLCR(below
40 per cent): we will see later that they showthe highest share of derivatives. Comparing
2007-2009 with 2010-2012, WS moves visibly toward the retail clusters. Figure 5 shows
the within-cluster inter-quartile range of selected items from the balance sheet (assets
and liabilities).
Figure 6 presents the median values of capital and leverage. WS-Inv is the cluster
with the highest leverage and poorest capital; as expected, risk weighted assets (RWA)
is higher for retail banks. One can also see the increase of capital ratios from the left to
the right plots.
To assess the quality of the clustering exercise in terms of possible misclassifcation,
we compute the silhouette distance, which measures if a bank is classifed in a certain
cluster while still being close to banks in others: positive values refect a good
classifcation, negative the opposite (Rousseeuw, 1987). Our clusters have positive, high
silhouette values, except for a few WS-Inv, Retail banks and some of the Outliers.
JFEP
7,4
386
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Figure 4.
_Paper_Ellipses
NLC, WS foats here
387
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Figure 5.
_Paper_
AssetsLiabilities
foats here
JFEP
7,4
388
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Figure 6.
_Paper_Capital foats
here
389
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Figure 7 shows the evolution of the size of each cluster over time. Roughly 55 per cent of
the banks are retail-focused or retail-diversifed, while 23-30 per cent are identifed as
wholesale (pure or investment).
Finally we consider the transition matrix (Table IV) which highlights changes in
business model for a given bank between 2007-2009 and 2010-2012. The matrix should
be read by rows: diagonal values give the share of banks remaining in the same cluster
while off-diagonal reports in which cluster(s) banks move to. The classifcation is rather
stable (diagonal values are above 60 per cent). The main change appears on WS, which
lose 24 per cent of its member to Retail in 2010-2012 showing a decrease of wholesale
funding for these banks.
6. Linking key players in trading and business model
This section maps key players in trading into business model clusters. Figure 8 reports
the share of banks identifed as key players in trading by each measure and for each
cluster in terms of TA (top plot) and number (bottom plot).
For instance, GMVDand HLEG select roughly 60 per cent of the banks in WS-Inv in
2007-2009; the two sets are different, since the share of TAof selected banks is larger for
HLEG.
The picture is rather clear for the largest and smallest banks: all measures
consistently identify all investment banks while selecting fewretail-focused institutions
(Small and Ret-Foc). With respect to these Retail clusters, a couple of key aspects are
worth mentioning: EC is more stable in the selection over time, identifying two to three
large banks; HLEG and GMVD are more volatile, they hardly select any banks in
2007-2009, while they add to the selection by EC some small banks in 2010-2012.
The major differences among measures appear in two clusters: retail-diversifed
(Ret-Div) and wholesale-investment (WS-Inv) banks. Focusing on the frst, results show
that GMVD and HLEG recognize three to four banks not selected by EC, with no
peculiar characteristics in terms of our input variables, except their very large size. For
WS-Inv, roughly half of the banks (seven in 2007-2009, eight in 2010-2012) are selected
by all measures. Extra banks selected by GMVD/HLEG, in 2010-2012, are either the
smallest of this cluster or have a particular low share of customer deposits.
Finally the wholesale (WS) cluster contains 8 banks in 2007-2009 and 13 banks in
2010-2012. It is the least stable cluster, with 37 per cent of banks changing classifcation
from 2007-2009 to 2010-2012. Figure 4 showed that this cluster moved toward a more
retail-oriented model, decreasing wholesale funding and slightly increasing
investments into customer loans. EC and HLEG select a single bank in this cluster in
both periods while GMVD appears to be more volatile selecting no bank in 2007-2009
and three banks in 2010-2012.
7. Results and conclusions
This paper has investigated the link between the engagement of European banks in
trading and their business model. Using balance sheet and income statement data (100
banks from 2007 to 2012), we have frst investigated key players in trading activities,
according to three measures discussed by regulators and researchers in the context of
the structural reform of EU banks. Then we have developed a business model analysis
via robust clustering coupled with PCA. Such methodology, not much explored in
literature, leads to classify banks into six business model clusters, which result to be
JFEP
7,4
390
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Figure 7.
_Paper_BanksbyBM
foats here
391
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
rather stable over time. Finally we have mapped information about trading measures on
the business model clusters.
Results show that trading measures behave coherently in selecting banks with
highest and lowest involvement in trading. In fact, the mapping is rather clear for the
cluster of very large investment banks, holding more than 20 per cent of trading assets/
liabilities and having customer loans usually below 45 per cent: these are always
identifed as key players in trading. At the opposite, rare selection is made within small
and well capitalized banks and medium-sized retailed-focused banks, generally
described by a share of customers’ loans over 45 per cent and low reliance on wholesale
funding. These results were somehow expected.
For the other business models, and in particular wholesale-investment and Ret-Div,
the selection depends on the choice of the trading measure. As WS-Inv are concerned,
between 50 and 75 per cent of the banks are recognized as key trading players depending
on the measure. These banks are heavily involved in trading, fnance themselves
through bank deposits and show very high levels of leverage. Our analysis shows that
EC tends to identify fewer banks in this cluster.
Another discrepancy between the measures is observed for some very large banks,
classifed as retail-diversifed: GMVDand HLEGidentify around 85 per cent of banks as
key trading players, while EC only 65 per cent. These banks present in general lower
levels of leverage and better quality of capital, in terms of Tier1, with respect to very
large investment banks. EC selecting fewer banks in this cluster partly answers the
potential critique that trading measures cannot properly heed capital adequacy.
Furthermore, one can see that not only size matters: in particular medium-sized
WS-Inv are selected, while some very large retail-diversifed are not.
Regarding HLEG on Ret-Foc, our analysis confrms the criticism that this measure
might be selecting more retail banks, suggesting that assets available for sale could be
more coupled with retail-oriented activities than trading-oriented ones.
From a policy-makers perspective, our results pose the question whether
business models analysis has a place in the legislative context for structural
separation of trading activities. On one side, mapping trading involvement onto
business models could help regulators in understanding which measure for trading
to choose, depending on their purpose. On the other, business model analysis could
Table IV.
Transition matrix of
banks among
clusters from 2007-
2009 to 2010-2012
(Others category
includes outliers and
NaN)
Inv
(%)
WS-Inv
(%)
Ret-Div
(%)
WS
(%)
Retail
(%)
Small
(%)
Outliers
(%)
NaN
(%)
Total
(%)
Inv 100 100
WS-Inv 87 7 6 100
Ret-Div 100 100
WS 13 63 24 100
Retail 3 3 11 82 3 100
Small 40 60 100
Outliers 17 17 17 50 100
NaN 13 25 38 25 100
Note: Each row reports, for the banks’s classifcation of 2007-2009, in which business model they can
be found in 2010-2012
JFEP
7,4
392
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Figure 8.
_Paper_
SelectedBanksvsBM
foats here
393
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
support regulators in designing alternatives to strict separation, depending on
banks activities at large. In particular one could think of a targeted approach rather
than a one-fts-all: while strict separation could be the way forward for very large
investment banks, regulators could consider enhancing the supervision of
medium-large WS-Inv to better assess their leverage and capital levels and/or
closely monitor Ret-Div, that already show high level and quality of capital.
The duration and extent of discussion of the structural separation proposal within
EUinstitutions confrms the policy relevance of this issue. To conclude, one needs to say
that business model analysis of complex fnancial groups would beneft fromadditional
data such as products strategy, structure and ownership information and geographical
reach, enabling to better assess the regulation impact.
Notes
1. In January 2014 the EC proposed a Regulation on structural measures improving the
resilience of EU credit institutions. In June 2015, the ECOFIN Council agreed its position.http://ec.europa.eu/fnance/bank/structural-reform/index_en.htm
2. The analysis is based on the SNL database since it allows for a detailed disaggregation of
balance sheet items related to assets held for trading. It allows distinguishing derivatives for
trading from derivatives for hedging.
3. GMVD is the cost of replacing all outstanding contracts at current market prices. GMVD is
not readily available in the balance sheet so we propose a proxy based on the sum of
derivatives assets and liabilities.
4. Standardization allows comparing variables with different scales. Given a vector X with
average xˆ and standard deviation ?, the corresponding standardized vector is obtained
as:
X
s
?
X ? xˆI
?
5. Statistical clustering assigns observations into groups. Clusters are built starting from
random centroids and moving their positions so to minimize their dispersion. Outliers can
considerably bias the estimation of the centroids of the clusters, hence affecting the fnal
clustering. For this reason, we opted for robust trimmed clustering t-clust (Garcia-Escudero
et al., 2008), implemented in Matlab in the framework of Forward Search for Data Analysis
(Riani et al., 2012; Cariboni et al., 2015). Its robustness capacity comes fromthe possibilities to
leave a proportion of observations unassigned (trimming) and to allowthe shape of the cluster
to become ellipsoidal.
6. To determine the number of clusters and restriction factor, Bayesian Information Criterion
analysis is employed (Riani et al., 2012). Our results suggest a number of clusters equal to 8
and restriction factor equal to 200.
7. All variables except TA, loan to deposit ratio, return on assets and trading income are
expressed as share of TA.
8. Most of the outliers are banks which received state aid or went through restructuring in the
considered time horizon. Information is available upon request.
9. Our sample does not include any bank with TA lower than € 30 bn. Size buckets are used to
order banks using their TA averaged over 2010-2012: Very large banks (VL): above 500 €bn,
JFEP
7,4
394
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Large banks (L): between 200 and 500 €bn, Medium banks (M): between 100 and 200 €bn and
Small banks (S): between 30 and 100 €bn.
10. In the present work, wholesale funding is estimated as WF ?TFL-CustD-DL-Eqty-SubDebt.
References
Altunbas, Y., Manganelli, S. and Marques-Ibanez, D. (2011), “Bank risk during the fnancial crisis:
do business models matter?”, unpublished manuscript, European Central Bank, available
at: www.ecb.europa.eu/pub/pdf/scpwps/ecbwp1394.pdf
Ayadi, R., Arbak, E. and Groen, W.P.D. (2011), Business Models in European Banking: Apre- and
Post-crisis Screening, September, CEPS Paperbacks.
Ayadi, R., Arbak, E., Groen, W.P.D. and Llewellyn, D.T. (2012), Regulation of European Banks and
Business Models, June, CEPS Paperbacks.
Ayadi, R. and De Groen, W.P. (2014), Banking Business Models Monitor 2014, October, CEPS
Paperbacks, Europe.
Beltratti, A. and Stulz, R. (2009), “Why did some banks perform better during the credit crisis? A
cross-country study of the impact of governance and regulation”, Journal of Financial,
Economics, Vol. 105 No. 1, pp. 1-17.
BIS (2014), VI. The Financial System at a Crossroads, 84th BIS Annual Report, 2013-2014.
Blundell-Wignall, A., Atkinson, P. and Roulet, C. (2013), “Bank business models and the Basel
system: complexity and interconnectedness”, OECD Journal: Financial Market Trends,
Vol. 2013 No. 2, pp. 1-26.
Blundell-Wignall, A., Atkinson, P. and Roulet, C. (2014), “Bank business models and the
separation issue”, OECD Journal: Financial Market Trends, Vol. 2013 No. 2.
Blundell-Wignall, A. and Roulet, C. (2012), “Business models of banks, leverage and the
distance-to-default”, Financial Market Trends, Vol. 2.
Boot, M.A.W. and Ratnovski, L. (2012), Banking and Trading, Andrews McMeel Publishing,
Kansas City, MO.
Brewer E. III and Jagtiani, J. (2013), “How much did banks pay to become too-big-to-fail and to
become systemically Important?”, Journal of Financial Services Research Vol. 43 No. 1,
pp. 1-35.
Brunnermeier, M.K., Dong, G. and Palia, D. (2012), “Banks’ non-interest income and systemic
risk”, Unpublished Working Papers, Princeton University, Princeton.
Cariboni, J., Pagano, A., Perrotta, D. and Torti, F. (2015), “Robust clustering of EU banking data”,
in Morlini I., Vichi M. and Minerva T (Eds), Advances in Statistical Models for Data
Analysis, Springer, New York, NY, pp. 17-25. doi: 10.1007/978-3-319-17377.
Chordia, T., Goyal, A., Lehmann, B.N. and Saar, G. (2013), “High-frequency trading”, Journal of
Financial Markets, Vol. 16 No. 4, pp. 637-645.
Chow, J.T. (2011), “Making banks safer: can Volcker and Vickers do it?”, International Monetary
Fund.
Davies, R. and Tracey, B. (2012), “Too big to be effcient? The impact of too big to fail factors on
scale economies for banks”, Journal of Money, Credit, and Banking, Vol. 46 No. 1,
pp. 219-253.
Demirguerc-Kunt, A. and Huizinga, H. (2013), “Are banks too big to fail or too big to save?
International evidence fromequity prices and CDS spreads”, Journal of Banking &Finance,
Vol. 37 No. 3, pp. 875-894.
395
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
EC (2013), Commission Staff Working Document Impact Assessment, available at:http://eur-lex.
europa.eu/resource.html?uri?cellar:e186dd0b-89b3-11e3-87da-01aa75ed71a1.0001.01/DO
C_1&format?PDF
ECB (2010), “Report on EU banking structures”, September, available at: www.ecb.europa.eu/
pub/pdf/other/eubankingstructures201009en.pdf
ECB (2013), “Banking structures report”, November, available at: www.ecb.europa.eu/pub/pdf/
other/bankingstructuresreport201311en.pdf?5656762fc7710c2ad62b381f432eff9f
Gambacorta, L. and Rixtel, A.A.V. (2013), “Structural bank regulation initiatives: approaches and
implications”, BIS, Working Papers No 412, April.
Garcia-Escudero, L.A., Gordaliza, A., Matrán, C. and Mayo-Iscar, A. (2008), “A general
trimming approach to robust cluster analysis”, The Annals of Statistics, Vol. 36 No. 3,
pp. 1324-1345.
Inci, A. Narain, A., Ilyina, A. and Surti, J. (2011), “The too-important-to-fail conundrum”,
International Monetary Fund, Staff Discussion Notes No. 11/12.
Laeven, M.L., Ratnovski, L. and Tong, H. (2014), Bank Size and Systemic Risk, International
Monetary Fund.
Liikanen, E., Bänziger, H., Campa, J.M., Gallois, L., Goyens, M. Krahnen, J.P., Mazzucchelli, M.,
Sergeant, C., Tuma, Z., Vanhevel, J. and Wijffels, H. (2012), High-level Expert Group on
Reforming the Structure of the EU Banking Sector, 2, Final Report, Brussels.
Loretan, M. (1997), “Generating market risk scenarios using principal components analysis:
methodological and practical considerations”, available at: www.bis.org/publ/ecsc07c.pdf
Pagano, A. (2013), “Trading activities and bank structural separation: possible defnitions and
calibration of screening thresholds”, JRC Scientifc and Policy Report.
Real, R. and Vargas, J.M. (1996), “The probabilistic basis of Jaccard’s index of similarity”,
Systematic Biology, Vol. 45 No. 3, pp. 380-385.
Riani, M., Perrotta, D. and Torti, F. (2012), “FSDA: a matlab toolbox for robust analysis and
interactive data exploration”, Chemometrics and Intelligent Laboratory Systems, Vol. 116,
pp. 17-32.
Rousseeuw, P.J. (1987), “Silhouettes: a graphical aid to the interpretation and validation of cluster
analysis”, Journal of Computational and Applied Mathematics, Vol. 20, pp. 53-65.
Schoenmaker, D. and T., Peek (2014), “The state of the banking sector in Europe”, OECD
Economics Department Working Papers, No. 1102, OECD Publishing, available at: http://
dx.doi.org/10.1787/5k3ttg7n4r32-en
Stiroh, K.J. and Rumble, A. (2006), “The dark side of diversifcation: the case of US fnancial
holding companies”, Journal of Banking & Finance, Vol. 30 No. 8, pp. 2131-2161.
Ueda, K. and di Mauro, B.W. (2013), “Quantifying structural subsidy values for systemically
important fnancial institutions”, Journal of Banking & Finance, Vol. 37 No. 10,
pp. 3830-3842.
Vickers (2011), “The independent commission on banking: the vickers report &the parliamentary
commission on banking standards”, available at:http://researchbriefngs.parliament.uk/R
esearchBriefng/Summary/SN06171
Viñals, J. (2013), “Creating a safer fnancial system: will the Volcker, Vickers, and Liikanen
Structural measures help?”, International Monetary Fund, Staff Discussion Notes No. 13/4.
JFEP
7,4
396
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Appendix. Banks in the sample and their classifcation by cluster type
Table AI.
List of banks in the
sample by cluster
type, classifcation as
of 2010-2012
2010-2012 2007-2009
Classification in Business model
Bank Name | Size
EC GMVD HLEG EC GMVD HLEG
Other business model or size,
only if a change is observed.
Very large and leveraged investment banks Inv 8 8 8 7 7 7
Y Y Y Y Y Y L V s y a l c r a B
Y Y Y Y Y Y L V s a b i r a P P N B
Cr. Agricole Grp VL Y Y Y Y Y Y
Y Y Y Y Y Y L V k n a B e h c s t u e D
Y Y Y Y Y Y L V s g n i d l o H C B S H
Y Y Y L V a e d r o N Missing
Y Y Y Y Y Y L V p u o r G S B R
Y Y Y Y Y Y L V n e G c o S
Medium-large wholesale investment banks
WS-Inv
8 12 10 7 9 9
Y Y Y Y Y Y L V k n a b z r e m m o C
N N L V a i x e D Y Size=L
Y Y Y Y Y Y L B L e h c s i r e y a B
Y Y Y Y Y Y L s u i f l e B
Deutsche Zentral G L Y Y N Y Y N
N L a i x e D Y Y Size=L
L g l H E R o p y H Y Y N N N N
LB Baden Württemberg L Y Y Y Y Y Y
N N L B L / D R O N Y N N Y
Y Y Y L n o g i t r o P Outliers
DekaBank Deutsche GZ M Y Y Y Y Y Y
N M k n a b d r o N H S H Y N WS
La Banque Postale M N N N N N N
N M g l H n i l r e B B L Y Y N Y Y
AXA Bank Europe S Y Y Y Retail
Banque Int à Lux S N Y Y Outliers
Österreichische Volksbanken S N Y N WS
RaifLB Niederösterreich S N N N N N N
Westdeutsche GZ S N N N N N N
Large and very large retail diversified banks
Ret-Div
10 13 13 8 11 13
L V A V B B N Y Y N Y Y
Y Y Y Y Y Y L V r e d n a t n a S
N N L V p r G l e u t u M r C Y N N Y
Y Y Y Y Y Y L V E C P B e p u o r G
Y Y Y Y Y Y L V k n a B G N I
N L V o l o a p n a S a s e t n I Y Y N Y Y
N N Y Y Y L V p u o r G s d y o l L Y
Rabobank Group VL N Y Y N Y Y
Y Y Y Y Y Y L V t i d e r C i n U
Y Y Y Y Y Y L k n a B e k s n a D
N L p r G C B K Y Y Y Y Y
L B E S Y Y N Y Y Y
Y Y Y Y Y Y L t r a h C d n a t S
L k n a b d e w S Y N N Retail
Y Y Y S a c n a b o i d e M Outliers
(continued)
397
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table AI.
Cont (2/3)
2010-2012 2007-2009
Classification in Business model
Bank Name | Size
EC GMVD HLEG EC GMVD HLEG
Other business model or size,
only if a change is observed.
Medium sized wholesale banks WS 1 3 1 1 0 1
Banco Fin. y de Ahorros L N Y g n i s s i m N
Allied Irish Banks, Plc M N N N Retail
N M k n a b d r o N H S H Y N WS-Inv
M A B A L E H Y Y Y Y N Y
l i a t e R N N N M e c e e r G k B l a n o i t a N
l i a t e R N N N M h c i e r r e t s Ö k b Z . f i a R
N N N S k n a B a h p l A Retail
N N N N N N S p r g a t n e g r A
N N N S m u r t s o N e r a M missing
Cyprus Pop Bk Public S N N N Retail
Hypo Alpe-Adria-Bank S N N N N N N
IKB Deutsche Industriebank S N N N Outliers
N N N N N N S S / A k n a B e k s y J
Österreichische Volksbanken S N N N WS-Inv
RaiffLB Oberösterreich S N N N N N N
Sparkasse KölnBonn S N N N Retail
Medium sized retail focused Retail 2 3 6 3 1 1
ABN AMRO Grp L N N N N N N
Monte Paschi di Siena L Y Y Y Y N N
Caja Aho. y Pens Barcel. L N N N N N N
Cassa depositi e prestiti L N N N N N N
L B N D Y N N Size = M
N N N N N N L k n a B p r G e t s r E
Nationwide Bldg Soc. L N N N N N N
N Y N N N N L n e k n a b s l e d n a H
Allied Irish Banks M N N N WS
N N N M l l e d a b a S Size = S
N N N N N N M e r a l o p o P o c n a B
Banco Popular Español M N N N N N N
N N N N N N M d n a l e r I f o k n a B
Caixa Geral de Depósit M N N N N N N
M B N D Y N N Size = L
National Bank of Greece M N N N WS
M t i d e r k l a e R N N N N N N
Raif. ZBk Österreich M N N N WS
M L A A E R S N S N N Y Small
M k n a b d e w S Y N N Ret-Div
N N N N N N M I B U
Aareal Bank S N N N N N N
Alpha Bank S N N N WS
AXA Bank Europe S N N Y WS-Inv
Banca Carige S N N Y Small
Bp lEmilia Romagna S N N N N N N
BP Milano S N N Y N N N
BP Vicenza S N N N N N N
BPI S N N Y N N N
Banco Comercial Português S N N N N N N
Sabadell S N N N Size = M
BAWAG S N Y Y N N N
Bank of Cyprus Public Company S N N N N N N
Bankinter S N N N N N N
Cajas Rurales Unidas S N N N N N N
Clydesdale Bank S N N N N N N
Co-operative Bank S N N N N N N
Cyprus Popular Bank Public S N N N WS
Deutsche Apotheker- und Ärztebank S N N N N N N
Espirito Santo Financial Group S N N N N N N
Liberbank S N N N Missing
Münchener HypothekenbanK S N N N N N N
Permanent TSB S N N N Missing
Piraeus Bank SA S N N N Outliers
Sparkasse KölnBonn S N Y N WS
Volksbanken-Verbund S N N N Missing
Yorkshire Building Society S N N N N N N
(continued)
JFEP
7,4
398
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
About the authors
Francesca Campolongo is the Head of the Economic and Financial Analysis at the European
Commission Joint Research Centre since January 2012. She has been working in the Joint Research
Centre since 1998, where she also obtained a prize as “best young scientist of the year” in 2002.
Currently, she is actively involved in the work of the European Commission to create a safer and
sounder fnancial system and to recover from the economic crisis. In particular, in the last few
years she has contributed to European Commission proposals on: higher capital requirements for
banks, harmonized deposit protection schemes, EUFramework for bank recovery and resolution,
Table AI.
Cont (3/3)
2010-2012 2007-2009
Classification in Business model
Bank Name | Size
EC GMVD HLEG EC GMVD HLEG
Other business model or size,
only if a change is observed.
medium sized wholesale banks
WS
1 3 1 1 0 1
Banco Fin. y de Ahorros L N Y g n i s s i m N
Allied Irish Banks, Plc M N N N Retail
N M k n a b d r o N H S H Y N WS-Inv
M A B A L E H Y Y Y Y N Y
l i a t e R N N N M e c e e r G k B l a n o i t a N
l i a t e R N N N M h c i e r r e t s Ö k b Z . f i a R
N N N S k n a B a h p l A Retail
N N N N N N S p r g a t n e g r A
N N N S m u r t s o N e r a M missing
Cyprus Pop Bk Public S N N N Retail
Hypo Alpe-Adria-Bank S N N N N N N
IKB Deutsche Industriebank S N N N Outliers
N N N N N N S S / A k n a B e k s y J
Österreichische Volksbanken S N N N WS-Inv
RaiffLB Oberösterreich S N N N N N N
Sparkasse KölnBonn S N N N Retail
Small retail
Small
0 0 0 0 0 0
N N N M L A E E R S N S Retail
N N N S e g i r a C a c n a B Retail
N N N N N N S k n a B P T O
N N N N N N S i k s l o P O K P
Volkswagen Bank S N N N N N N
Outliers 1 1 3 1 1 3
Y Y Y M n o g i t r o P WS-Inv
N N S . x u L , E E C B Y N N Y
Banque Int à Lux S N N Y WS-Inv
N N N N N N S t i d e r k F R B
IKB Deutsche Industriebank S N Y N WS
S k n a b n e t n e R N N Y N N N
S a c n a b o i d e M Y N Y Retail-Div
N N N S k n a B s u e a r i P Retail
Missing data 0 0 0 1 1 1
Y Y Y L k n a B a e d r o N Inv
Hamburger Sparkasse N N N N N N
L kred. Baden-Wurttemberg N N N N N N
Banco Fin. y de Ahorros N N N WS
Volksbanken-Verbund N N N Retail
N N N m u r t s o N e r a M WS
N N N k n a b r e b i L Retail
N N N B S T t n e n a m r e P Retail
Notes: The second column describe the size (VL = very large, L = Large, M = Medium-Sized;
S = Small, see footnote 15). Y(N) labels a bank that is selected (not selected) based on the
three-year average value of the metrics, with period specified in the head row; the last columns
specified if a bank has changed cluster moving from the first to the last period and details its
origin in the period 2007-2009. Bolded labels identify cases when not all metrics select the
bank as being above the threshold that identify key player in trading
399
Banks under
X-rays
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
and bank structural reform. As background Francesca is a mathematician (she graduated in Pisa
University, Italy) with a PhDin modeling and sensitivity analysis (Griffth University, Australia),
her main research interests focus on fnancial markets modeling and fnancial risk analysis.
Working in support to EU policy makers, she has developed skills in calibration and ex ante
impact assessment of fnancial regulation, with particular attention to the banking sector. She is
the author of ?30 publications in academic journals, co-author of 3 books on sensitivity analysis
published by Wiley and co-author of 2 books on Securitization published by Wiley and Springer.
Jessica Cariboni is an offcer of the Unit Economic and Financial Analysis of the European
Commission Joint Research Centre. Born in 1976, Jessica graduated with honors in Physics at
University of Milan, Italy, in 2000. From March 2001 to December 2002 Jessica was a junior
quantitative analyst in one of the leading Italian companies in fnancial investments, Nextra
Investment Management SGR. In 2007, she completed her PhD in fnancial modeling at the
Department of Mathematics of the Katholieke Universiteit of Leuven. From2003 Jessica has been
working as researcher at the European Commission Joint Research Centre. Her main
responsibility is to develop research and quantitative analyses for impact assessments of new
regulatory proposals in the feld of banking regulation and fnancial stability. Her major felds of
research are fnancial modeling, credit risk modeling and risk analysis. Jessica has over ten
scientifc publications on refereed journals, and she is also co-author of two books on credit
risk-modeling and sensitivity analysis published by Wiley. Jessica Cariboni is the corresponding
author and can be contacted at: [email protected]
Nathalie Ndacyayisenga is a researcher of the Unit Economic and Financial Analysis of the
European Commission Joint Research Centre. Nathalie graduated, major of her promotion, at
Université Paul Sabatier in Toulouse (France) in mathematics and computer sciences. Before
joining the unit, Nathalie has worked for the last decade in analytical and research departments of
large private companies mostly Belgium, Italy, developing technical skills such as accounting,
risk management and currently policy-making processes. Her main domain of expertise
encompasses data mining, prediction modeling and customer base analysis.
Andrea Pagano holds a Ph.D. in mathematics from Brown University. After few years in
academia working on complex geometry, he started to work on environmental modeling at ENEA.
In 2005 he joined the Joint Research Centre (JRC) of the European Commission, where he worked
on Global Sensitivity Analysis. In 2011 he joined the group of Economic and Financial Analyses
where he works as data analyst and modeler.
For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: [email protected]
JFEP
7,4
400
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
5
3
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
doc_901385569.pdf