The Anatomy of Bank Diversification

Ralf Elsas*, Andreas Hackethal†, Markus Holzhäuser‡

March 2009

Abstract

We use panel data from nine countries over the period 1996 to 2003 to test how revenue diversi-
fication in conjunction with increasing bank size affects bank value. Using a comprehensive
framework for bank performance measurement, we find robust evidence against a conglomerate
discount, unlike studies concerned with industrial firms. Rather, revenue diversification in-
creases bank profitability and is associated with higher market valuation. This performance
effect does not depend on whether diversification was achieved through organic growth or
through M&A activity. We further demonstrate that previous results in the literature on the im-
pact of diversification on bank value presumably differ due to the way diversification is meas-
ured, and the negligence of an indirect positive value effect via bank profitability.

Keywords: Bank Diversification, Conglomerate Discount, Universal Banking

JEL Classification: G24, G21, G34

*
LMU Munich - Institute for Finance & Banking, Ludwigstr. 28 RG/V, 80539 Munich, Germany. Email:
[email protected], phone: +49 (89) 2180 2579, fax: +49 (89) 2180 3607
†
Johann Wolfgang Goethe-University & E-Finance Lab, Grüneburgplatz 1, 60323 Frankfurt/Main, Germany. Email:
[email protected], phone: +49 (69) 798 33700, fax: +49 (69) 798 33530.
‡
Johann Wolfgang Goethe-University & E-Finance Lab, Grüneburgplatz 1, 60323 Frankfurt/Main, Germany. Email:
[email protected].
1. Introduction
While many non-financial firms around the world have been striving for corporate business focus
over the last two decades, financial services firms and especially banks have instead increased
business diversification. This paper analyzes empirically whether or not the trend towards diversi-
fication in global banking has been in the interests of bank shareholders and it aims to shed light
on the channels through which diversification might affect value creation in banking.

Our study documents that average diversification levels of the world’s largest banks were almost
one third higher in 2003 than they had been in 1996. Commercial banks typically increased diver-
sification by moving into fee-based businesses. Banks with already strong fee-based revenues ex-
panded into trading activities. Yet other banks diversified revenues by underwriting insurance con-
tracts.

The disparate diversification trends between non-financial firms and banks raises the question of
whether banking institutions are prone to make the same mistakes that many non-financial firms
made during the conglomeration wave of the sixties and nineties of the last century. Alternatively,
the banking business is possibly truly special in the sense that a broadening of business scope cre-
ates value for bank shareholders.

Such value creation could come from two broad sources. One is bank-specific economies of scope.
Unlike most firms from other industries, banks often entertain long-term contractual relationships
with their customers. Over time banks can gather extensive customer information and reuse that
information not only in the business area where the information was originally gathered but also in
other non-related business areas. Moreover, banks that operate with high operational leverage (i.e.,
a high ratio of fixed costs to variable costs) might find that diversification into related businesses
awards them with a cost advantage over specialized competitors. For example, selling life insur-
ance through the existing retail bank branch network might result in cost economies of scope. If
such economies of scope truly exist in banking, diversification will tend to have positive effects on
aggregate welfare and financial system stability.

The second reason why business diversification might be in the interest of bank shareholders has
to do with the current state of the financial services industry. Mainly due to technological progress
and deregulation, the financial services industry has been undergoing dramatic change over the last
two decades. Boot (2003) argues that banks have extended their business scope mainly as a strate-
gic response to this business uncertainty. Banks have been investing into diverse business areas

1
early on to acquire the skills needed to make efficient production decisions and to reap profits
when a particular business area eventually turns out to flourish. Given the real option character of
skill building investments, wealth implications for shareholders of diversifying financial institu-
tions are a function of the degree of strategic uncertainty, the effectiveness of early skill building
or skill reusing, and ultimately the ability of an institution to create shareholder value from the
opportunities at hand. This foot-in-the-door strategy cannot turn out to be successful for all the
banks that have embarked on it. Rather one would expect that some institutions will indeed de-
velop into broad powerhouses while others will have to absorb the losses from a failed market en-
try. If the observed trend towards more diversification was largely driven by such a foot-in-the-
door strategy, aggregate effects on welfare and stability are ambiguous at best and bank sharehold-
ers as well as regulators should watch diversification trends very carefully.

This paper attempts to provide a fresh view on the direct and indirect effects of revenue diversifi-
cation on equity market value by disentangling the aggregate diversification effect typically meas-
ured by other studies. To this end we measure the effects of diversification on financial indicators
from three tiers of a comprehensive framework for bank valuation. We also account for potential
interactions between diversification, growth and vertical integration by explicitly controlling for
simultaneous changes along these two other dimensions of banking firm boundaries.

Contrary to related studies, our sample covers large banks from nine developed countries with well
functioning financial systems. This allows studying banks that face hardly any restrictions regard-
ing diversification levels. We thus complement other studies, that either consider only U.S. banks
(which may be special due to the Glass/Steagall-Act, that was effective over most of the available
history of firm-level data in the U.S.) or that cover banks from developed countries together with
banks from developing countries which are very heterogeneous with regard to regulation, public
ownership and influence, or market structure.

Our econometric framework allows discriminating between diversification effects from opera-
tional economies of scope and from a foot-in-the-door-strategy strategy.

Finally, we conduct several robustness tests of our results, considering economic significance,
sample splits, and three different exercises to control for potential endogeneity between bank value
and diversification.

Our main finding is that revenue diversification enhances contemporaneous bank profitability via
higher margins from non-interest businesses and lower cost income ratios. Higher contemporane-
ous bank profitability translates into higher market valuations, implying that diversification has an

2
indirect positive impact on bank value and giving rise to a conglomerate premium in banking. This
result is in line with the economies of scope explanation for diversification. Once we control for
the indirect effect of via profitability we find no direct effect of diversification on market valua-
tions. We therefore largely rule out the foot-in-the-door-strategy being the primary reason for bank
diversification.

The paper unfolds as follows. The next section provides a quick review of the relevant empirical
literature, in particular delineating how our analysis differentiates from and complements a recent
study by Laeven/Levine (2007). Section 3 presents the multi-tier framework for measuring diversi-
fication effects. Our data are presented in section 4. In section 5 we present and discuss our em-
pirical results. Section 6 provides extensive robustness tests regarding measurement issues, en-
dogeneity, and selectivity. Section 7 concludes.

2. Literature Review
There is a large body of literature on the costs and benefits of diversification. Among the identified
benefits are economies of scope (e.g., Chandler 1977; Teece 1982), an improved resource alloca-
tion through internal capital markets (e.g., Williamson 1975; Stein 1997), a potentially lower tax
burden due to higher financial leverage (e.g., Lewellen 1971) and the ability to use firm-specific
resources to extend a competitive advantage from one market to another (e.g., Wernerfelt & Mont-
gomery 1988 and Bodnar et al. 1997). These benefits have to be traded off against the costs asso-
ciated with diversification. Cost may stem from agency problems afflicting diversifying invest-
ments (e.g., Jensen 1986; Meyer et al. 1992), inefficient internal resource allocation due to a mal-
functioning of internal capital markets (e.g., Lamont 1997; Scharfstein 1998; Rajan et al. 2000),
informational asymmetries between head office and divisional managers (e.g., Harris et al. 1992),
and increased incentives for rent-seeking behaviour by managers (e.g., Scharfenstein & Stein
2000).

There is abundant empirical evidence for US industrial firms that the cost of diversification out-
weighs its benefits from a shareholder’s perspective. The by now classic studies by Lang & Stulz
1994 and Berger & Ofek 1995 report that diversified US firms trade at an 8% and 13% to 15%
equity discount, respectively, as compared to their specialized peers. However, more recent work
has cast some doubt on the general existence of a diversification discount. It has been argued that
measurement problems (e.g., Whited 2001), data problems (e.g., Harris 1998; Villalonga 2004),
selection biases in terms of firms, observation period or country (Graham et al. 2002; Lins & Ser-

3
vaes 1999) and failure to account for the endogeneity of the diversification decision (Campa &
Kedia 2002) have driven the results of earlier studies. While the academic debate has not reached a
final consensus, a common broad picture emerges according to which value generating diversifica-
tion is rather the exception than the rule in most industries.

The empirical literature on the merits of bank diversification has largely focused on the question of
whether the repeal of the Glass-Steagall Act allowed US commercial banks to reduce business risk
by diversifying into non-traditional financial services. The potential to reduce earnings volatility
was found for combinations of earnings streams from banking and insurance activities (Boyd et al.
1993; Laderman 1999; Lown et al. 2000) but was hardly found at all for the combination of earn-
ings streams from interest-based banking activities and fee-based securities activities (Allen &
Jagtiani 2000; Estrella 2001).

Stiroh & Rumble 2003 measure the effect of diversification on the risk-adjusted profitability of US
financial holding companies for the period 1997–2002. They find that revenue diversification to-
wards fee income reduced risk-adjusted returns because over their observation period, fee-based
activities were more volatile but not necessarily more profitable than traditional interest earning
activities. The lack of evidence for positive diversification effects on profitability is echoed by
event studies on diversifying bank mergers (e.g., DeLong 2001) and by the abundant empirical
literature that applies frontier efficiency analysis to examine the productive efficiency of banks. In
their extensive survey article Berger & Humphrey 1997 report that there is a lack of strong evi-
dence in favour of or against the joint provision of different financial services.

Laeven & Levine 2007 apply a modification of the Lang & Stulz 1994 ‘chop shop’ method to
measure diversification effects on bank market valuation. They compare the market-to-book ratio
of a diversified financial institution with that of a corresponding portfolio of selected banks spe-
cializing in either interest-based or fee-based business. They assume a linear relationship between
diversification and market value and find that diversification reduces market-to-book ratios by up
to ten percentage points.

Our approach differs from Laeven/Levine 2007 along four important dimensions. Firstly we utilize
several alternative measures for bank diversification. One of these measures is arguably more
granular than the type used in Laeven/Levin 2007. This allows us to investigate the robustness of
results with regards to measurement of diversification.

Secondly, we include additional explanatory variables into our regressions that help to avoid po-
tential omitted variables biases. Most importantly we include contemporaneous bank profitability

4
as a regressor when estimating diversification effects on market values. This allows us to discrimi-
nate between indirect, operational diversification effects from economies of scope and direct ef-
fects from a foot-in-the-door strategy. In addition, we control for mergers and acquisitions activity
and for differences in the vertical integration of our sample banks.

Thirdly we adopt a comprehensive framework of regression models (see next section) that permits
us to investigate along which channels diversification affects bank profitability and bank valuation.

Finally, we use a more homogenous sample of exchange-listed banks from well-developed coun-
tries rather than a sample that also comprises banks from developing countries.

Similarly to Laeven/Levine (2007), however, we spend considerable effort on examining robust-

ness to measurement issues and potential endogeneity problems of the econometric specification.

3. Methodology
To investigate diversification effects on both a bank’s market valuation and its financial perform-
ance one needs a consistent bank valuation framework that ties market values to observable finan-
cial indicators. This section presents the simple framework that underlies our empirical analysis.

The fundamental value (FV) of a bank’s equity equals the present value (PV) of future cash flows
to shareholders (CF). Under clean surplus accounting the present value of cash flows is equal to
the book value of invested shareholders’ capital (IC) plus the present value of future economic
value creation, as measured by residual income (RI). The present value of future residual incomes
is equivalent to the net present value of all current and future projects of a given bank.

∞ ∞
FVt = ∑ PV (CFT ) = ICt +
T =t +1
∑ PV ( RI
T =t +1
T )
(1)

Residual income in period t is defined as excess net operating profits after tax (NOPAT) over a
capital charge for the capital invested by shareholders. The capital charge equals invested capital
(IC) times cost of equity (CoE).

RIT = NOPATT − ( ICT * CoET ) (2)

The right-hand side of equation (2) can also be expressed as the spread between the return and the
cost of equity where NOPAT/IC corresponds to the after tax return on equity (ROE)

5
 NOPATT
RI T = ICT ⋅ ( − CoET )
ICT (3)

Inserting (3) into (1) yields:

∞
NOPATT
FVt = ICt + ∑ PV ( IC
T =t +1
T ⋅(
ICT
− CoET ))
, (4)

The fundamental value of a bank’s equity can therefore be expressed as a function of today’s
nominal value of invested capital, the expected growth path of invested capital and the expected
development of spread. Given that investors use information on past growth spread to forecast fu-
ture performance, fundamental values will be a function of a vector of past growth and spread and
a vector of parameters that investors believe to be indicative of the future development of eco-
nomic value creation.

FVt = f ( ICt , ICt −1 ,..., Spread t , Spread t −1 ,..., X t ) (5)

If we assume that market value is a (stochastic) function of fundamental value and if we further
assume that book value (BV) is a good proxy for invested capital, we can express the ratio of mar-
ket value to book value (MTB) as a function of the arguments of f (.) and substitute IC with BV.

MVt g ( FVt , ε )
MTBt = = = h( BVt , BVt −1 ,..., Spread t , Spread t −1 ,..., X t )
BVt BVt (6)

Equation (6) implies that there are two channels through which diversification can affect the rela-
tive market value of banks. Firstly, diversification could affect current spreads due to economies
(or diseconomies) of scope. Secondly, diversification could affect the expected future development
of bank growth and spreads. Such effects can be positive if diversification is expected to provide
the foot-into-the-door to exploit upcoming business opportunities or they can be negative if diver-
sification is expected to increase business complexity and bureaucracy. In both cases the vector X
should contain a measure for diversification. The existence of multiple, possibly countervailing
effects might give rise to a non-linear relationship between diversification and MTB and possibly
to an optimal degree of diversification. Our estimation model for market-to-book (MTB) in equa-
tion (7) below therefore also contains a quadratic term for diversification (DIV).

6
 MTBi ,t = a + b ⋅ log(BV ) i ,t + c ⋅ BVgrowthi ,t + d ⋅ SPREADi ,t
(7)
+ x1 ⋅ DIVi ,t + x2 ⋅ DIVi ,t + ϕ '⋅ X i ,t + ε i ,t
2

X contains further possible determinants of future performance, such as the degree of vertical inte-
gration.

Note that coefficients x1 and x2 measure the effect of current diversification on investor expecta-
tions regarding future growth and spread. If they are found to be statistically insignificant, this
casts doubt on the foot-in-the-door channel of diversification. Insignificant coefficients on diversi-
fication in our MTB-model, however, do not rule out the economies-of-scope channel of diversifi-
cation because such diversification effects will be captured by the variables spread and growth.

In order to investigate the economies-of-scope channel we estimate diversification effects on the

components of spread.. Equation (8) decomposes spread into the corporate tax rate (T), the cost of
equity and most importantly into four components of pre-tax operating ROE,:

At TNORt TOE t LLPt

SPREADt = ⋅( − − ) ⋅ (1 − T ) − CoE t
BVt At At At
14 444 4244 444 3
Pr etax Operating Income
, (8)

where A denotes total assets, TNOR denotes total net operating revenues, TOE denotes total oper-
ating expenses, LLP denotes loan loss provisions, and T is the corporate tax rate.

The ratio of total net operating revenues over assets can be further decomposed into four asset
margins: net interest revenues over assets, net fee revenues over assets, net trading revenues over
assets and net other operating revenues over assets. Figure 2 provides a graphical representation of
equation (8) and a numerical example for Deutsche Bank AG in 2003.

In the majority of the empirical analysis, our regressions focus on the market-to-book ratio or the
spread as the dependent variable, to test for the direct and indirect (i.e. via the spread) impact of
diversification on bank value. To trace the performance impact of diversification through the op-
erational channel, we will also report regressions, that have either leverage (assets over book eq-
uity), one of the four asset margins, expenses over assets, loan loss provisions over assets, or costs
of equity as the dependent variable.

The general structure of these regressions is shown in equation (9), where Y denotes any of the
aforementioned dependent variables:

7
 Yi ,t = a + z1 ⋅ DIVi ,t + z2 ⋅ DIVi ,t + ϕ '⋅ X i ,t + ε i ,t
2
, (9)

where X denotes a matrix of control variables for vertical integration, growth, size, systematic risk,
business focus (non-interest revenues over interest revenues) and possibly interaction terms. We
take the panel nature of our data into account by using fixed-effects regressions, and control for
any time variation or macro-factors by including a set of year dummy variables.

The coefficients z1 and z2 of equation (9) will indicate whether there exists a relationship between
diversification and the individual components of the spread. Comparisons across spread compo-
nent models will allow us to scrutinize how exactly diversification affects performance. An aggre-
gation of diversification effects across spread-component models according to equation (8) will
allow us to check for consistency of our estimates. The aggregate effects should be broadly in line
with the total effects from regressing SPREAD directly on diversification.

4. Data
We obtain financial accounting data from the Bankscope database. The database covers 2,072
banks and bank holding companies from the USA, Canada, Australia, UK, Germany, France,
Spain, Italy, and Switzerland. Bankscope covers virtually all existing large banks but coverage for
small banks has been quite heterogeneous across countries and over time. In order to mitigate the
selection bias from that heterogeneous coverage, we required that the total assets of sample banks
exceeded 1bn USD in at least one of the nine years of our observation period. The remaining 1,378
banks from Bankscope were matched with DataStream to obtain banks with available share price
information. The final sample contains 380 listed banks with a total of 1,917 bank years for the
years 1996 to 2003. The number of sample banks was not stable over the years due to mergers and
acquisitions. 65% of the sample banks are from the US, 3% are from Canada, 29% are from
Europe and 3% from Australia. As discussed in detail as a robustness test in Section 6.3, the large
sample concentration on banks from the U.S. does not affect our results qualitatively.

The CAPM beta of the banks was estimated based on matched DataStream price series. M&A
growth numbers were calculated based on matched M&A transaction data from Thomson Finan-
cial’s SDC database (see section 4.2 for more details).

8
4.1. Definition of Diversification
Unlike other studies on non-financial industries we cannot measure bank diversification based on
SIC codes and segmental accounting data. SIC-code classification for banks is not granular enough
and is not consistent across countries. Moreover, segmental reporting is not consistent across
banks and across time.

Instead, we use an adjusted Herfindahl-Hirshman index to measure revenue diversification. Vari-

ous authors have applied a closely related approach (see, e.g., Comment & Jarrell 1995; Desai &
Jain 1999; Acharya et al. 2002; Stiroh & Rumble 2003; Stiroh 2004). We report extensive robust-
ness tests for this choice of the diversification measure in Section 5, however.

Equation (10) shows the construction of our diversification index.

⎛ ⎛ INT ⎞ 2 ⎛ COM ⎞ 2 ⎛ TRAD ⎞ 2 ⎛ OTH ⎞ 2 ⎞

DIV = 1 − ⎜ ⎜ + + + ⎟
⎜ ⎝ TOR ⎟⎠ ⎜⎝ TOR ⎟⎠ ⎜⎝ TOR ⎟⎠ ⎜⎝ TOR ⎟⎠ ⎟
⎝ ⎠ (10)

INT denotes gross interest revenue, COM denotes net commission revenue, TRAD denotes net
trading revenue and OTH denotes all other net revenue. TOR denotes total operating revenue and
is equal to the sum of the absolute values of INT, COM, TRAD and OTH.4

We use gross interest revenue so that our diversification measure is not unduly distorted by the
profitability of the bank’s interest business. Unfortunately, Bankscope does not consistently report
gross numbers for the other revenue categories. For those banks, for which non-interest gross
revenues are reported we find that direct (and presumably mostly external) expenses for trading
and fee-based activities are in the range of 5% to 15% of respective gross revenues. We conclude
that the distortion from using non-interest net figures is not material.

Note that, because we are using gross interest revenue and the absolute values for the other three
revenue streams, total operating revenue (TOR) in equation (10) is different from total net operat-
ing revenue (TNOR) in equation (8). In equation (8) we are interested in the profitability of the
four business areas and not, as is the case in equation (10), in the relative significance of the reve-
nues streams.

4
Negative net revenue values would lead to negative shares for some revenue streams and shares greater than one for
other revenue streams. As a consequence, DIV would be strongly affected by business unit performance and could take
on values far greater than 0.75.

9
We subtract the sum of squared revenue shares from unity so that DIV increases in the degree of
revenue diversification. By definition DIV can take on values between zero (the bank is fully spe-
cialized in one business area) and 0.75 (the bank generates a fully balanced revenue mix from all
four business areas).

Figure 1 depicts average revenue shares and average diversification levels across all sample banks.
Diversification increased from below 30% in 1996 to over 38% in 2003. Average diversification
levels of US banks increased from 29.8% to 38.3% with the highest increase in the years after the
abolishment of the Glass-Steagall Act in 1999 (see Table 3). Figure 1 also indicates that the overall
increase in diversification is not merely caused by a general decline in the revenue share of gross
interest income. Rather, average revenues shares have remained fairly constant since 1998. There-
fore some banks must have diversified from interest-related businesses into e.g. fee-based business
while others must have taken the opposite route.

4.2. Definition of Other Variables

The level of revenue diversification refers to the horizontal boundaries of a banking firm. Man-
agement decisions to alter horizontal boundaries are often intertwined with decisions regarding the
level of vertical integration and the overall size and growth of the institution. For example, if busi-
ness complexity is a positive function of both the level of horizontal diversification and the level of
vertical integration and if business complexity bears negatively on performance, then bank man-
agement has to trade off horizontal diversification against vertical integration and possibly also
against size.

To account for possible interaction effects between the three dimensions of the boundaries of
banking firms we need to introduce control variables for size, growth, and vertical integration into
our regression analysis.

Following standard definitions, we measure size by the natural logarithm of year-end total assets.
Growth is defined based on the annual percentage change in equity book values. We matched
Bankscope data with M&A transaction data from Thomson Financial’s M&A database SDC to
break down total book equity growth into an M&A growth component and a residual organic
growth component. Because Thomson Financial only reports transactions at market prices we first
had to estimate book values for the acquired equity stakes. For that purpose we divided equity
market values by the average market-to-book ratio of all banks from the same country as the target
in the year of the transaction. We then added book values of all acquisitions and divestitures un-
dertaken by the same bank in a given year and divided that sum by the start-of-year equity book

10
value of that bank to arrive at an estimate of its M&A growth in that year. We identified 892 deals
that contributed on average 3.8% to a bank’s book equity growth per annum. Organic growth is
defined as total growth rate minus M&A growth rate and amounts to 14.4% per annum for the av-
erage bank in our sample.

The measurement of vertical integration is based on the following idea. We consider a bank as
highly vertically integrated if most of its output is generated through the employment of a bank’s
own resources. A bank is considered to have a low degree of vertical integration if it relies largely
on outside resources and services to generate output. Following Tucker & Wilder 1977 we use
gross revenues minus pretax profit as a proxy for output volume 5 and the sum of interest expenses,
labour expenses, fixed charges and loan loss provisions as a proxy for internal input volume. Ver-
tical integration is defined as input from internal resources over profitability adjusted output. The
sample average for vertical integration measured this way declined from 79.4% in 1996 to 73.9%
in 2003.

The definitions of the remaining variables are standard and are reported in Table 1. Table 2 shows
descriptive statistics for the 380 sample banks. Data series are winsorized at the 0.5% and 99.5%
quantiles to mitigate the impact of outlier observations on regression estimates whilst keeping the
information that an extreme realisation of the variable has been observed.

5. Results

5.1. Diversification and Value Creation

We start by estimating the direct impact of diversification on market-to-book ratios. Control vari-
ables include the level of vertical integration, organic growth and M&A growth, bank size, the
ratio of non-interest income over gross interest income and year dummies. The results for a fixed-
effects model with bank-individual effects and year dummies (not shown) are presented in Table 4,
column 2. This is our baseline model, which we will put to several robustness checks in what fol-
lows.

5
As mentioned above, Bankscope reports gross figures for interest revenues but only net figures for the other revenue
types. Therefore we had to approximate gross revenues from fee-based businesses, trading, and other business based on
their corresponding net revenue figures.

11
The coefficients on both the linear and the quadratic term for revenue diversification are statisti-
cally insignificant. Revenue diversification does therefore not affect market valuations directly.
This casts doubt on the foot-in-the-door explanation for diversification because market participants
do not seem to factor in diversification when forecasting future growth and spread.

As expected, contemporaneous spread has a statistically significant and strong positive effect on a
bank’s relative market valuation. Apparently, investors rely to a considerable extent on current
value creation when forecasting future value creation. Organic or M&A growth do not systemati-
cally affect bank valuations, while the size control is negative and statistically significant.

The regression estimates in columns 3-5 of Table 4 confirm the results from the baseline model. In
column 3, we drop squared explanatory variables. In column 4 we repeat the estimation of the
baseline model but change the measure of diversification. The corresponding Herfindahl-index
now consists only of two blocks – interest income and non-interest income, thus aggregating fee-
based, trading and commission income into one component. In both robustness exercises results
remain unchanged, diversification does not affect market valuation, but the spread as a measure for
value creation is highly positive and statistically significant.

In column 5 of Table 4, we complement the regression model by controlling for extreme changes
in diversification levels. We introduce a dummy variable DIVERSIFY that is equal to one if the
change in diversification is in the highest decile for a given year. Similarly, we use another dummy
variable SPECIALIZE which is equal to one if the change in diversification is in the lowest decile
for a given year. We interact these dummies with our measure of diversification to test whether (at
least) major changes in diversification affect market values directly. The coefficients of both inter-
action terms turn out to be insignificant, casting further doubt on the foot-in-the-door explanation.
We provide further robustness tests of this result in Section 6.

The non-existence of a direct diversification effect on bank value does not rule out that diversifica-
tion affects bank value indirectly through the spread (i.e. current bank profitability). Table 5 shows
regression estimates of banks’ spread (i.e. the excess of earnings over cost of capital as a proxy for
periodic value creation). Here, we also repeat the robustness tests documented in Table 4. Hence,
the first column of coefficients shows our baseline model, the second excludes squared terms, the
third relies on the Herfindahl-index of diversification based on interest vs. non-interest income,
and the fourth repeats the analysis of major changes in banks’ degree of diversification.

12
As can be seen from all specifications in Table 5, there is a strong positive effect of diversification
on spread.6 This finding is consistent with the existence of economies of scope in banking to be
exploited through revenue diversification. Given that bank value increases in the contemporaneous
spread (see Table 4), the positive diversification effect on spread is also consistent with a positive,
yet indirect diversification effect on bank value.

As an aside the spread regressions in Table 5 also indicate that growth through M&A activities
reduce spread while organic growth leaves spread unaffected. In addition, reducing vertical inte-
gration (e.g. through outsourcing) is generally not associated with an increase in profitability.

5.2. Decomposition of Determinants

In Table 6, we analyse the determinants of the spread-components to achieve a more detailed
analysis of how diversification bears on bank profitability, relying on the next layer of the spread
decomposition shown in Figure 2.7 Note that all models control for a general trend towards more
non-interest revenue by including the variable non-interest income over interest income. Also in all
models we include squared terms of the diversification index based on four income types.The ra-
tionale for including squared terms is to allow for non-linearity in diversification effects. As it
turns out, non-linearity is only an issue for two spread components, namely net interest margin and
cost of equity. The significant coefficients for the linear and the quadratic diversification variables
carry the same sign in all other eight models.

All three non-interest income margins appear to be positive and linear functions of diversification,
indicating that banks benefit from revenue economies of scope (see models 2–4). Expenses per
asset dollar are also increasing in the degree of diversification (see model 5), indicating that banks
do not benefit from cost economies of scope. However, this result does not hold anymore if one
uses the cost-income ratio to measure cost efficiency. Column 6 of Table 6 shows that more diver-
sified banks incur lower expenses per dollar in total revenues. A possible explanation for the dis-
crepancy in both measures is that leverage declines as a consequence of higher diversification (see
column 7). Expenses over assets may thus increase primarily because diversified banks operate
with less assets per dollar in equity and also with less assets per dollar in operating expenses.

6
Note that the regressions do not include a measure for the riskiness of banks since the cost of capital are already sub-
tracted from the dependent variable.
7
While not shown here we also estimated extended models that feature interaction terms of DIV with the variables for
vertical integration and growth. The statistical and economic relevance of diversification effects remained unchanged.

13
Loan loss provisions are only very weakly affected by diversification. The eighth model in Table 6
predicts that expected loan losses are slightly higher for diversified banks than for focused banks.
This result does not change if we use total revenues instead of total assets as the denominator of
the dependent variable.

Model 9 shows that diversification is associated with smaller current equity growth. In the last sec-
tion we reported that there is no systematic relationship between current equity growth and market
valuation. We can therefore largely rule out that diversification reduces market valuations through
a deterioration of growth rates.

The estimated effects of diversification on net interest margins and on the cost of equity are not
monotonous across the support of the diversification variable. They can be best analyzed by means
of numerical examples. The same is true for the aggregate effect of diversification on spread-
components. The next section presents numerical examples for the median bank in the sample and
three specific bank types, also to examine the economic significance of these effects.

5.3. Economic Significance

Columns 2 and 3 of Table 8 report sample median values for the eight spread-components. We
define the median bank from our sample as a bank for which spread-components take on the val-
ues shown in column 2. Column 4 reports the estimated change for each of the spread-components
if the median bank augments diversification by 10 percentage points (roughly two thirds of the
sample standard deviation for DIV). Estimated changes in single spread-components are computed
based on the corresponding coefficients on DIV and squared DIV in Table 6.8 Aggregate changes
on spread are computed based on estimated individual changes in conjunction with equation (8).
To verify results on aggregate spread effects we also computed diversification effects on SPREAD
based on the coefficients from the regression shown in column 2 of Table 5.

Our model predicts that if the median bank increased its diversification level from 32% to 42%, its
total revenue margin would ceteris paribus increase by 62 basis points. The slight 10 basis point
drop in net interest margin would be overcompensated by a combined 72 basis point increase in
fee, trading and other income margins.

8
We used both significant and insignificant coefficients. To check whether the results in Table 6 are driven by this pro-
cedure we dropped all insignificant variables from the models in Table 5, re-estimated the reduced models and used
only significant coefficients for the numerical examples. Estimated effects on spread remained largely unchanged.

14
The model also predicts that total operating expenses over assets would have gone up by 22 basis
points and that loan loss provisions over assets would have added another 2 basis points to total
expenses over assets. Both leverage and cost of equity are estimated to decline slightly as a result
of more diversification.

These effects are commensurate with the existence of non-trivial revenue and cost economies of
scope in banking. As banks extend their business scope, they might find it easier to (cross-) sell
more products to the same customers. In some cases non-interest business can, however, slightly
cannibalize a bank’s conventional interest-related business, as indicated by the negative diversifi-
cation effect on the interest margin.

When measured over total revenues, expenses decline in response to diversification (see last row
in Table 8). This implies that diversified banks need ceteris paribus fewer inputs to generate the
same revenue volume as focused banks. Possible reasons for this result are that revenue diversifi-
cation permits banks to use some of their resources (e.g., branches, IT systems, brands) more pro-
ductively.

Table 8 also shows that diversification effects on current spread are quite sizable. Our model for
the median bank predicts that individual diversification effects on spread-components add up to a
total effect of 4.0%. This estimate is comparable to the 3.2% increase in spread that is implied by
the model in column 2 of Table 5. We conclude that diversification has increased value creation in
banking over the past years.

If current spread is a positive function of diversification and if market-to-book is a positive func-

tion of current spread, then market-to-book should be a positive function of diversification. If
spread increased by four percentage points (e.g., induced by diversification), the model in Table 4
predicts that market-to-book grows by 22 basis points. Hypothetically, the median bank could have
increased its market-to-book ratio from 1.73 to 1.95 through a 10 percentage point increase in di-
versification.

To see whether our results also hold for specific bank types other than the median bank, we select
retail banks, investment banks and universal banks from our sample (refer to Table 7 for selection
criteria), compute the group medians for the variables and then estimate the impact of diversifica-
tion on spread-components and spread. Table 8 documents that economies of scope seem to exist
for all three bank types and that diversification effects are somewhat stronger for universal banks
and somewhat weaker for retail banks than for the median bank. Moreover, the results shown in
Table 8 suggest that aggregate diversification effects are extremely powerful for investment banks.

15
6. Further Robustness Tests

6.1. Measuring Diversification and Market Valuation

Our first main empirical result, that diversification has no direct effect on bank value, is not in line
with Laeven/Levine 2007 who report a negative (direct) effect. There are four potential reasons for
this discrepancy:

a) Different measures of diversification: Laeven/Levine 2007 use the right hand side of equation
(11) as their measure of revenue diversification, and an equivalent measure with loans and non-
loan assets to measure asset diversification.

net interest income − other operating income

Diversification( LL) = 1 −
total operating income
(11)

Table 9 shows the correlation coefficient between the Laeven/Levine 2007 measure and our in-
come-based Herfindahl-measure to be only 0.65. Moreover, asset-based diversification meas-
ures differ from all income measures substantially, e.g. the correlation with our diversification
measure is -0.185.
b) Different measures of bank value: Laeven/Levine 2007 use a so-called activity adjusted excess
market-to-book ratio as the dependent variable in their main regressions. This ratio is obtained
by subtracting a hypothetical market-to-book ratio from the observed market-to-book ratio for a
given bank. The hypothetical ratio is a weighted average of contemporaneous market-to–book
ratios of two groups of banks that either specialize in interest-generating businesses or that spe-
cialize in fee-generating businesses.
c) Control variables: we include additional explanatory variables when estimating the direct diver-
sification effect on bank value to avoid omitted variables biases
d) Sample composition

In this section we investigate to what extent a) to c) are responsible for the discrepancy in empiri-
cal result. Section 6.3 will discuss the likely impact of d). In what follows we repeat our major
regressions from Table 4 but use alternative diversification measures (including (11)) and an activ-
ity adjusted excess market-to-book ratio. Table 10 shows estimates from fixed-effects regressions
designed to compare our estimation results with those by Laeven/Levine 2007.

The second column of Table 10 illustrates that we can replicate the Laeven/Levine 2007 result of a
negative diversification effect on bank value for our sample if we use their diversification and
16
market value measure, and do not use further control variables. The R2-statistic of this model is
much lower than those reported for the other specifications, pointing to a potential misspecification
of this reduced model. In fact, if we introduce more controls and in particular spread as explana-
tory variables, the negative revenue diversification effect almost vanishes, with a p-value of 9.9%.
It vanishes completely if we use our Herfindahl diversification measure instead of the simple
measure from (11). The fifth column shows the effect from asset diversification on excess value if
we control for spread. This effect is ambiguous because the linear and quadratic coefficients are
both significant but carry different signs.

These results indicate that the discrepancy between the results in this paper and those in
Laeven/Levin (2007) are unlikely to be due to how bank value is measured (reason b), partly due
to how diversification is measured (reason a), but largely due to the inclusion of spread as a control
variable (reason c).

It is not clear a priori, which diversification measure is better suited. We would argue that meas-
ures based on more granular information will also better reflect different bank structures. Given
equation (6), it is at least clear, however, that contemporaneous profitability needs be included as a
control variable based on theoretical grounds.

6.2. Endogeneity
Our second main result, that diversification improves bank profitability and therefore has an indi-
rect positive effect on bank value, rests on the assumption that diversification affects profitability
or value, but not vice versa. Moreover, we implicitly assume that it is not the very same character-
istics that determine bank values and diversification levels. If these assumptions do not hold, our
estimation results and inferences could be biased by problems of endogeneity or selectivity. To
control for these issues, we take three countermeasures.

The first countermeasure follows automatically from the fact that our regression results in Section
4 are all based on panel regressions that include bank and time fixed effects. The fixed effects
model does not require independence between the regressors and individual error terms. This con-
trols simultaneously for all unobserved heterogeneity with respect to time-invariant, firm specific
characteristics (like the country of the registered seat of the bank, legal form, etc) and firm-
invariant characteristics that vary over time (like the interest level, other macro-variables, or inter-
national regulations like Basle I).

Second, we use instrumental variable regressions to explicitly control for the endogeneity between
bank value (or spread) and diversification. It is of course challenging to identify instrumental vari-

17
ables that are correlated with one endogenous variable (diversification) but not the other (market
value). A common procedure in econometrics is to exploit the panel nature of the data and use
lagged variables as instruments. These variables are not fully exogenous but predetermined (be-
cause lagged diversification is set before the actual market value is determined), a characteristic
that is for example systematically exploited in the GMM-framework, when estimating dynamic
panel models (see Arellano/Bond 1991), or the Fama/MacBeth-approach in asset pricing tests.

Table 11 shows estimation results when using lagged diversification as the instrumental variable
for current diversification. In the first step, we simply base the fixed-effects regression on lagged
diversification as regressor, to test whether the implicit timing assumption of our baseline regres-
sion in Table 4 affects estimation results. Then, we apply two-stage least squares (2SLS) with the
fixed-effects estimator, use lagged diversification as the instrument for current diversification and
repeat our central regression with either the market-to-book ratio or the spread as dependent vari-
able.9

As can be seen from Table 11 in columns 2 and 4, the coefficient on the lagged diversification
variable (and the instrumented counterpart) is statistically not different from zero in both regres-
sions of the market-to-book ratio on diversification and other variables. The estimates shown in
column 6 of the table illustrate that this result remains robust, when using the activity-adjusted
excess market-to-book ratio suggested by Laeven/Levine 2007 as the dependent variable. Columns
3 and 5 of Table 11 show that diversification affects the banks’ spread significantly positive even
when controlling for endogeneity. As a consequence, instrumental variable regressions confirm
our earlier results.

As a third exercise, we follow Laeven/Levine 2007 and model the impact of diversification on the
market-to-book ratio and the determinants of a bank’s degree of diversification simultaneously.
To this end, we estimate a Heckman (1979) treatment effects model by maximum likelihood, with
standard errors corrected for clustering of observations for each bank. This explicitly controls for
selectivity, i.e. the potential problem that the same characteristics which affect the decision to di-
versify affect a bank’s market value. The model consists of one equation for the determinants of
the market-to-book ratio, where a dummy variable indicates whether a firm is diversified. The

9
Recall that we do not need to include additional risk measures in the regression model of the spread because spread is
defined as ROE minus cost of capital, which is based on the estimate of the bank’s systematic risk (beta).

18
dummy equals one, if the Herfindahl index of diversification (based on interest versus non-interest
bank income) exceeds 40%, which resembles the 80% quantile of the empirical distribution.

The model comprises a simultaneous probit estimation, where the dummy for a diversified bank is
explained by variables exogenous to the market value. As can be seen from our estimations in Ta-
ble 4, bank market value does not depend on the growth characteristics. However, it is obvious that
M&A growth and organic growth are two important strategic instruments for banks to manage
their level of diversification. Accordingly, we can use growth through acquisitions and organic
growth as exogenous instruments. We also include the cost-income-ratio as an explanatory vari-
able for the diversification status, because increasing diversification might be driven by a bank’s
desire to use potential synergies through economies of scale and scope more efficiently. The cost-
income ratio is a standard bank-efficiency measure in the literature.

Table 12 shows the results of these selectivity estimations. In column 2, we present estimation re-
sults based on the market-to-book ratio as in Table 4, while in column 3 we show estimation re-
sults when firm value is measured by the activity-based excess value as suggested by
Laeven/Levine 2007, similar to the robustness exercise of Table 10, column 2. The coefficient es-
timates clearly support our main results – diversification does not affect market values but the
spread does. Hence, our results are robust to selectivity as well.

6.3. Subsamples and Extended Sample

So far, our analyses were based on a sample of large, exchange-listed banks from nine countries.
More than two thirds of the sample banks are domiciled in the U.S. To examine whether this sam-
ple concentration affects our results, we repeat our estimations for the non-U.S. subsample, com-
prising 133 listed banks.

As can be seen in Table 13, our major empirical findings remain qualitatively unchanged. Column
2 of Table 13 shows estimation results for the non-U.S. sample using our baseline specification.
The diversification measures are insignificant, while the spread affects the market-to-book-ration
positively. The estimation results shown in column 3 illustrate in turn, that the spread positively
depends on diversification even in the non-U.S. subsample. Column 4 of Table 13 shows, further
shows, that the insignificance of diversification and the positive impact from the spread prevail, if
on uses the activity adjusted excess Q measure.

19
In the last two columns of Table 13, we report in addition estimates for the baseline model (see
Table 4, column 3), where we include an interaction term between diversification and a dummy
variable INTERNATIONAL, which takes the value of one if international operations contribute at
least 20% to total revenues of banks. This serves as an additional test for rsik heterogeneity in our
sample. However, neither is the coefficient on the interaction term significant in the market-to-
book regression nor in the spread regression, and all other coefficients remain qualitatively unaf-
fected. Thus, our evidence is robust against this additional test for bank risk as well.

Finally, although not reported in this paper for brevity, further robustness test show that the posi-
tive dependence of the return on equity (i.e. the principle spread component) is robust when in-
cluding non-listed banks in the sample. This adds further 438 non-listed banks with assets greater
than 1bn USD from the same nine countries. The extended sample comprises 52% US banks, 2%
Canadian banks, 36% European banks (8% from the UK) and 2% Australian banks. Again, the
regressions results for ROE-components remain qualitatively the same.10

7. Conclusions
This paper is motivated by the observation that over the last decade and all around the world, many
banks have diversified while non-financial firms have specialized their operations. We hypothesize
that either economies of scope are greater for banks than for non-financial firms, or that the high
degree of uncertainty in the banking industry increased the attractiveness of foot-in-the-door diver-
sification strategies.

The paper provides strong evidence that diversification improves bank profitability. This result is
shown to be very robust for alternative measures of diversification, for alternative subsamples, and
also after controlling for endogeneity and selectivity. Based on this result, we believe that econo-
mies of scope (at least among related activities) are indeed pronounced in banking.

The evidence presented in this paper is not consistent with the foot-in-the-door explanation sug-
gested by Boot 2003. When controlling for contemporaneous profitability, bank market value is
not directly affected by diversification. Both results contradict the findings by Laeven/Levine
2007, who find a negative direct diversification effect on bank value. We show that the discrep-

10
Results are available from the authors upon request.

20
ancy in results is mostly driven by the fact that the authors use different measures for bank value
and revenue diversification, and do not use a profitability control in their major regressions.

Given that diversification benefits are tied more to the way financial services are produced and less
so to the particular challenges the banking industry has been facing over the last decade, it seems
likely that banks will also benefit from diversification in the future. And given the fact that a ma-
jority of banks still operate at medium levels of diversification, we assume that the industry wide
trend towards revenue diversification will continue over the next years.

Finally, on a methodological note, our paper suggests that simple measures of diversification like
non-interest revenue over total revenue might in fact be overly simplistic to capture all diversifica-
tion effects. The same is probably true for econometric models that focus on the direct diversifica-
tion effects on market valuations. Our paper shows that diversification effects do not typically ma-
terialize through such direct effects but rather through indirect effects on current operating per-
formance and is thereby able to reveal a conglomeration premium in banking.

21
8. References
Acharya, V.V., Hasan, I., Saunders, A. 2002. Should banks be diversified? Evidence from individual bank loan portfo-
lios.
Allen, L., Jagtiani, J. 2000. The risk effects of combining banking, securities, and insurance activities. Journal of Eco-
nomics and Business 52, 485-497.
Berger, A.N., Humphrey, D.B. 1997. Efficiency of financial institutions: International survey and directions for future
research. European Journal of Operational Research 98, 175-212.
Berger, P., Ofek, E. 1995. Diversification´s effect on firm value. Journal of Financial Economics 37, 39-65.
Bodnar, G.M., Tang, C., Weintrop, J. 1997. Both Sides of Corporate Diversification: The Value Impacts of Geographic
and Industrial Diversification. NBER Working Paper
Boot, A.W.A. 2003. Restructuring in the Banking Industry with Implications for Europe. 2003 EIB Conference on Eco-
nomics and Banking.
Boyd, J.H., Graham, S.L., Hewitt, R.S. 1993. Bank holding company mergers with nonbank financial firms: Effects on
the risk of failure. Journal of Banking & Finance 17, 43–63.
Campa, J.M., Kedia, S. 2002. Explaining the diversification discount. The Journal of Finance 57, 1731-1761.
Chandler, A.D. 1977. The Visible Hand: The Managerial Revolution in American Business. Belknap Press, Cambridge,
MA.
Comment, R., Jarrell, G.A. 1995. Corporate Focus and Stock Returns. Journal of Financial Economics 37, 67-87.
DeLong, G.L. 2001. Stockholder gains from focusing versus diversifying bank mergers. Journal of Financial Economics
59, 221-252.
Desai, H., Jain, P.C. 1999. Firm performance and focus: Long-run stock market performance following spinoffs. Journal
of Financial Economics 54, 75-101.
Estrella, A. 2001. Mixing and matching: Prospective financial sector mergers and market valuation. Journal of Banking
and Finance 25, 2367-2392.
Graham, J.R., Lemmon, M.L., Wolf, J. 2002. Does corporate diversification destroy value? The Journal of Finance 57,
695-720.
Harris, M., Kriebel, C., Raviv, A. 1992. Asymmetric Information, Incentives, and Intrafirm Resource Allocation. Man-
agement Science 28, 604-620.
Harris, M.S. 1998. The association between competition and managers’ business segment reporting decisions. Journal of
Accounting Research 36, 111–128.
Holzhäuser, M. 2005. Long-term performance effects of bank diversification. E-Finance Lab Working Paper Series.
Jensen, M.C. 1986. Agency Costs of Free Cash Flow, Corporate Finance, and Takeovers. American Economic Review
76, 323-329.
Laderman, E.S. 1999. The potential diversification and failure reduction benefits of bank expansion into nonbanking
activities. Federal Reserve Bank of San Francisco Working Paper Series.
Laeven, L., Levine, R. 2007. Is there a Diversification Discount in Financial Conglomerates? Journal of Fianncial Eco-
nomics 85, 331-367.
Lamont, O. 1997. Cashflow and investment: Evidence from internal capital markets. The Journal of Finance 52, 83-109.
Lang, L.H.P., Stulz, R.M. 1994. Tobin’s q, Corporate Diversification, and Firm Performance. Journal of Political Econ-
omy 102, 1248-1280.
Lewellen, W.G. 1971. A Pure Financial Rationale for the Conglomerate Merger. The Journal of Finance 26, 521–537.
Lins, K., Servaes, H. 1999. International Evidence on the Value of Corporate Diversification. The Journal of Finance 54,
2215-2239.

22
Lown, C.S., Osler, C.L., Strahan, P.E., Sufi, S. 2000. The changing landscape of the financial services industry: What lies
ahead? Federal Reserve Bank of New York, Conference on Specialization, Diversification and the Structure of the
Financial System: The Impact of Technological Change and Regulatory Reform.
Meyer, M., Milgrom, P., Roberts, J. 1992. Organizational Prospects, Influence Costs, and Ownership Changes. Journal of
Economics and Management Strategy 1, 9–35.
Rajan, R., Servaes, H., Zingales, L. 2000. The Cost of Diversity: The Diversification Discount and Inefficient Investment.
The Journal of Finance 55, 35-80.
Scharfstein, D., Stein, J.C. 2000. Herd Behavior and Investment: Reply. American Economic Review 90, 705-706.
Scharfstein, D. 1998. The dark side of internal capital markets II: Evidence from diversified conglomerates. NBER
Working Paper Series.
Stein, J.C. 1997. Internal capital markets and the competition for corporate resources. The Journal of Finance 52, 111-
134.
Stiroh, K.J. 2004. Do Community Banks Benefit from Diversification? Journal of Financial Services Research 25, 135-
160.
Stiroh, K.J., Rumble, A. 2003. The Darkside of Diversification: The Case of U.S. Financial Holding Companies. Work-
ing Paper.
Teece, D.J. 1982. Towards an economic theory of the multiproduct firm. Journal of Economic Behaviour and Organiza-
tion 3, 39-63.
Tucker, I.B., Wilder, R.P. 1977. Trends in Vertical Integration in the U.S. Manufacturing Sector. The Journal of Indus-
trial Economics 26, 81-94.
Villalonga, B. 2004. Does Diversification Cause the Diversification Discount? Financial Management 33, 5-27.
Wernerfelt, B., Montgomery, C.A. 1988. Tobin’s q and the Importance of Focus in Firm Performance. American Eco-
nomic Review 78, 246-250.
Whited, T.M. 2001. Is It Inefficient Investment that Causes the Diversification Discount? The Journal of Finance Vol. 56,
1667-1691.
Williamson, O.E. 1975. Markets and Hierarchies: Analysis and Antitrust Implications. Free Press, New York.

23
Figure 1: Revenue Composition and Diversification by Year

The figure shows the mean proportions of the different revenue streams as a percentage of total gross operat-
ing revenues. Operating revenue is the sum of the four revenue streams. In addition the graph presents the
annual means of diversification levels for the 380 sample banks.

100%

90%

80%

70%
Other Income
60%
Trading Income
50% Fee Income
Interest Income
40%
Diversification
30%

20%

10%

0%
1996 1997 1998 1999 2000 2001 2002 2003

24
Figure 2: Spread Decomposition

The figure shows a graphical illustration of equation (8) in the text, where the spread (defined as return on
equity after tax minus cost of capital of equity) is decomposed in its definitory components. It also depicts
information from Deutsche Bank AG annual reports, to provide a numerical example. Assets and equity
values are averages of the start- and end-of-year balance sheet numbers for the fiscal year 2003. Cost of
Equity are calculated assuming a beta-factor of 0.9, a risk-free rate of 5%, and a market risk premium of 5%.

Pretax ROE Op. Income/ Op. Revenues /

Assets Assets
8.14% 0.29% 2.87%
x x -

Spread 1-tax rate Assets / Expenses /

Equity Assets
-5.75% 50.47% 27.91 2.40%
x- -

Cost of Loan Loss

Equity Prov. / Assets
9.50% 0.18%

Deutsche Bank AG (in thousand $)

Assets (2002/03) $ 905,071,626.00
Equity (2002/03) $ 32,433,438.50

Op. Revenues (2003) $ 25,986,082.00

Expenses (2003) $ 21,728,532.00
Loan Loss Prov. (2003) $ 1,616,631.00

25
Table 1: Definition of Variables

Diversification Measures
Diversification Herfindahl-Hirshman index based on types of reve-
nues, see equation (10)
Diversify Dummy, equal to one if a bank was in the highest
decile of yearly changes in diversification.
Specialize Dummy, equal to one if a bank was in the lowest dec-
ile of yearly changes in diversification.
Diversification (Interest) Diversification measure using interest versus non-
interest income (Herfindahl-Hirshman index)
Diversification (LL) Diversification measure similar to Laeven/Levine
2007, see equation (11).
Diversification (Loans) Asset-based diversification measure (Herfindahl-
Hirshman index) using loans versus other assets.
Return on equity (Net revenues – operating expenses – loan loss provi-
sions)*effective tax rate / (0.5*beginning-of-period
book equity + 0.5*end-of-period book equity)
Spread Return on equity – cost of equity
Cost of equity 0.05+0.05*Beta
Market-to-book Market capitalization on Dec 31st/end-of-year book
equity
Beta Calculated based on 250 days of data on return indi-
ces for bank and for MSCI World market index (data
from DataStream).
Vertical integration (Staff expenses + interest expenses + loan loss provi-
sions) / (gross operating revenues – pretax operating
income)
Equity growth (end-of-period book equity / beginning-of-period
book equity) – 1
M&A growth (Aggregate deal value from acquisitions and divesti-
tures per bank) / (beginning-of-period book equity) –
1
Organic growth (1 + equity growth) / (1 + M&A growth) – 1

26
Table 2: Descriptive Statistics

Obs Mean Std.Dev. Min Max

Bank Structure
Diversification 1917 32.7% 14.1% 0.0% 70.6%
Vertical integration 1917 77.7% 9.4% 36.4% 110.7%
Total Assets (in USDbn) 1917 61.6 145.3 0.1 1264.0
Growth
Growth due to M&A buys 1917 3.8% 11.9% 0.0% 93.1%
Organic Growth 1917 14.4% 27.4% -97.1% 263.6%
Shareholder value
ROE 1917 17.5% 10.0% -35.8% 56.1%
Spread 1917 3.8% 6.7% -34.6% 23.0%
Market-to-book 1917 1.995 1.439 0.075 11.351
Revenue composition
Interest income 1917 62.7% 21.8% -76.9% 102.8%
Fee income 1917 15.2% 15.5% -4.0% 91.1%
Trading income 1917 3.9% 9.6% -15.9% 133.3%
Other operating income 1917 18.2% 19.3% -3.4% 193.0%
Operations
Cost/income 1917 62.8% 14.5% 19.6% 153.4%
Expenses/assets 1917 3.3% 2.6% 0.2% 23.4%
Non-interest income/interest income (gross) 1917 65.1% 230.5% -1.1% 2709.3%
Loan loss provisions/assets 1917 0.31%
0.00031% 0.40%
0.00040% -0.28%
-0.00028% 3.14%
0.00314%
Risk
Assets/equity 1917 13.438 6.176 1.502 62.106
Beta 1917 0.644 0.475 -0.634 1.964

27
Table 3: Development of Diversification

Summary statistic of the development of the diversification measure in the US compared to Europe from 1996–2003.
1996 1997 1998 1999 2000 2001 2002 2003
USA 29.8% 30.3% 31.7% 31.3% 27.5% 30.3% 34.3% 38.3%
Europe 28.2% 28.8% 31.9% 34.6% 34.8% 33.3% 33.7% 37.2%
Total sample 29.3% 29.9% 31.8% 32.5% 29.6% 31.2% 34.3% 38.0%

28
Table 4: Market-to-Book Ratio Regressions

The table reports the results for fixed-effects regressions with the market-to-book ratio of equity (MTB) as
dependent variable. DIVERSIFY is a dummy variable, which is equal to one if a bank was in the highest
decile of yearly changes in banks’ diversification, while SPECIALIZE is a dummy variable that a bank was
in the lowest decile. Diversification is measured as one minus a Herfindahl-index, based on interest, fee,
commission and trading income. Alternatively, for Diversification (Interest) the Herfindahl-index is based
on interest vs. non-interest-income only. *,**,*** denote significance at the 10%, 5% and 1%-level, re-
spectively. All models include bank individual effects and year dummies (not shown). p-values in parenthe-
ses. The number of observations is 1917, comprising 380 banks over the period 1996-2003.

MTB / in- MTB / income MTB / diver- MTB / major

come diversi- diversification sification changes in
fication based on in- diversification
Baseline terest vs. non-
specification interest in-
come
Diversification -0.83 -0.09 --- -0.08
(0.311) (0.885) (0.897)
Diversification2 1.01 --- --- ---
(0.417)
Diversification (Interest) --- --- -0.18 ---
(0.921)
Diversification2 (Interest) --- --- 0.74 --
(0.813)
Vertical Integration -4.55 -0.51 -3.38 -3.41
(0.128) (0.583) (0.500) (0.499)
Vertical Intergration2 2.551 --- 1.94 1.89
(0.171) (0.543) (0.555)
Spread 5.81*** 5.73*** 5.69*** 5.70***
(0.000) (0.000) (0.000) (0.000)
M&A growth -0.22 -0.23 -0.22 -0.22
(0.218) (0.202) (0.214) (0.232)
Organic growth 0.09 0.08 0.08 0.09
(0.277) (0.483) (0.467) (0.422)
Interest / Non-interest -0.12*** -0.09*** -0.09*** -0.09***
Income (gross) (0.000) (0.002) (0.003) (0.001)
Size (log(assets)) -1.47*** -1.46*** -1.46*** -1.46***
(0.000) (0.000) (0.000) (0.000)
Diversification x DI- --- --- --- -0.09
VERSIFY (0.748)
Diversification x SPE- --- --- --- -0.35
CIALIZE (0.142)
R2 0.318 0.315 0.316 0.317

29
Table 5: Spread-Regressions

The table reports the results for fixed-effects regressions with SPREAD as dependent variable. DIVERSIFY
is a dummy variable, which is equal to one if a bank was in the highest decile of yearly changes in banks’
diversification, while SPECIALIZE is a dummy variable that a bank was in the lowest decile. Diversification
is measured as one minus a Herfindahl-index, based on interest, fee, commission, and trading income. Alter-
natively, for Diversification (Interest) the Herfindahl-index is based on interest vs. non-interest-income only.
*,**,*** denote significance at the 10%, 5% and 1%-level, respectively. All models include bank individual
effects and year dummies (not shown). p-values in parentheses. The number of observations is 1917, com-
prising 380 banks over the period 1996-2003.

SPREAD / SPREAD / in- SPREAD / SPREAD /

income diver- come diversifi- diversification major
sification cation based on in- changes in
Baseline terest vs. non- diversification
specification interest in-
come
Diversification 0.10** 0.27*** --- 0.26***
(0.023) (0.000) (0.000)
Diversification2 0.297*** --- --- ---
(0.000)
Diversification (Interest) --- --- 0.28*** ---
(0.000)
Diversification2 (Interest) --- --- -0.00 --
(0.960)
Vertical Integration 1.09*** 0.37*** 1.02** 0.94**
(0.000) (0.000) (0.023) (0.015)
Vertical Intergration2 -0.43*** --- -0.43 -0.38
(0.000) (0.145) (0.148)
M&A growth -0.04*** -0.04*** -0.04*** -0.04***
(0.000) (0.000) (0.000) (0.000)
Organic growth -0.00 0.00 -0.00 -0.00
(0.349) (0.951) (0.849) (0.736)
Interest / Non-interest 0.06*** 0.04* 0.05** 0.01**
Income (gross) (0.000) (0.077) (0.016) (0.011)
Size (log(assets)) -0.01** -0.01* -0.01** -0.02***
(0.040) (0.06) (0.049) (0.002)
Diversification x DI- --- --- --- -0.05***
VERSIFY (0.000)
Diversification x SPE- --- --- --- -0.03*
CIALIZE (0.050)
R2 0.245 0.203 0.209 0.219

30
Table 6: Spread-Components Regressions

The table reports the results from fixed effects regression models. Dependent variables are in column headers. *,**,*** denote significance at the 10%, 5% and 1%-
level, respectively. The full models include year dummies (not reported).

Net interest income / as- Fee income / assets Trading income / assets Other income / assets
sets (1) (2) (3) (4)
Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
Diversification -0.056*** 0.000 0.015** 0.010 0.006* 0.054 0.029*** 0.000
Diversification2 0.062*** 0.000 0.013 0.124 0.022*** 0.000 -0.004 0.647
Vertical Integration 0.037*** 0.002 -0.028 0.177 0.020* 0.076 -0.139*** 0.000
2
Vertical Intergration -0.029*** 0.000 0.011 0.409 -0.008 0.273 0.081*** 0.000
M&A growth -0.002*** 0.009 0.000 0.900 0.000 0.998 0.001 0.613
Organic growth -0.001*** 0.000 0.001* 0.097 0.001* 0.064 0.000 0.460
Size (log(assets)) -0.002*** 0.000 -0.004*** 0.000 -0.001* 0.014 -0.004*** 0.000
Interest / Non-interest Income -0.001*** 0.000 -0.001*** 0.000 0.001** 0.000 0.004*** 0.000
(gross)
Beta 0.000 0.133 0.000 0.531 0.000*** 0.205 0.000 0.456
Constant 0.055*** 0.000 0.056*** 0.000 -0.003 0.644 0.099*** 0.000
N 1917 1917 1917 1917
Groups 380 380 380 380
2
R 0.346 0.131 0.127 0.309
F-Test 47.34*** 0.000 13.47*** 0.000 13.03*** 0.000 39.91*** 0.000
Table 6: Spread-Components Regressions (cont.)

The table reports the results from fixed effects regression models. Dependent variables are in column headers. *, **, *** denote significance at the 10%, 5% and 1%-
level, respectively. The full models include year dummies (not shown here).
Loan loss
Expenses/assets (5) Cost/income ratio (6) Assets/equity (7) Cost of equity (9) Equity growth (10)
provisions/assets (8)
coefficient p-value coefficient p-value coefficient p-value coefficient p-value coefficient p-value coefficient p-value
Diversification -0.010 0.137 -0.152 ** 0.021 -0.839 0.686 -0.002 0.235 0.022 * 0.090 -2.774 *** 0.003
(Diversification)^2 0.044 *** 0.000 -0.123 0.219 -5.965 * 0.059 0.006 * 0.075 -0.059 *** 0.003 0.212 0.663
Vertical Integration -0.261 *** 0.000 -2.618 *** 0.000 -9.598 0.201 -0.005 0.551 -0.015 0.746 -1.723 0.160
(Vertical Integration )^2 0.127 *** 0.000 1.237 *** 0.000 9.214 * 0.052 0.008 * 0.090 0.012 0.673 0.305 0.649
Spread
Growth due to M&A buys 0.004 *** 0.005 0.080 *** 0.000 -1.833 *** 0.000 0.000 0.486 -0.004 0.158
Organic Growth 0.000 0.496 0.016 ** 0.012 -1.574 *** 0.000 0.000 0.178 -0.003 ** 0.021
Log(assets) -0.011 *** 0.000 -0.019 ** 0.016 1.516 *** 0.000 0.000 0.759 0.008 *** 0.000 0.327 *** 0.000
Non-interest income/interest income (gross) 0.001 *** 0.000 -0.010 *** 0.000 -0.239 *** 0.002 0.000 0.216 0.000 0.763 0.010 0.324
Beta 0.000 0.918 0.006 0.330 0.101 0.623 0.000 0.977 -0.074 *** 0.006
Constant 0.259 *** 0.000 2.045 *** 0.000 3.459 0.349 0.000 0.951 0.016 0.477 -1.056 * 0.097

N 1917 1917 1917 1917 1917 1917

Groups 380 380 380 380 380 380
R2 0.3379 0.2269 0.1337 0.0911 0.1215 0.1268
F-Test 45.64 *** 0.000 26.24 *** 0.000 14.67 *** 0.000 8.96 *** 0.000 13.15 *** 0.000 13.80 *** 0.000
Table 7: Categorization of Banks

This table reports the criteria for categorizing sample banks into subgroups, used in Table 8.
Investment bank Universal bank Retail bank
Deposits/assets <40% >60% >70%
Loans/assets >50%
Gross interest income/assets <75% <80% >80%*

* Additionally: Trading income/operating income <5%

33
Table 8: Hypothetical Changes in Operating Performance Given an Increase in Diversification

The table reports the estimated change in spread-components, spread and market-to-book ratio if diversification level is increased by 10 percentage points. Changes
for spread-components are calculated based on the corresponding coefficients in Table 6 and median values per bank type. The aggregate effect on spread is computed
based on estimated changes in the spread-components and based on equation (8). The estimated effect on spread is computed based on the coefficients in column (1)
of Table 5. The effect on market-to-book is computed by entering the aggregate effect on spread in the model from Table 4. Bank type definitions are reported in Ta-
ble 8.
Median Bank Retail Bank Investment Bank Universal Bank
Hypothetical Hypothetical Hypothetical Hypothetical
Median value change in ppt Median value change in ppt Median value change in ppt Median value change in ppt
Diversification 32.07% 10.00% 24.29% 10.00% 53.16% 10.00% 39.89% 10.00%
Expenses/assets 2.75% 0.22% 2.58% 0.16% 2.81% 0.41% 3.28% 0.29%
LLP/assets 0.22% 0.02% 0.19% 0.01% 0.04% 0.04% 0.28% 0.03%
Assets/equity 11.97 -0.53 11.53 -0.43 20.47 -0.78 11.43 -0.62
Net interest income/assets 3.00% -0.10% 3.42% -0.20% 0.78% 0.16% 3.09% -0.01%
Fee income/assets 0.45% 0.24% 0.35% 0.22% 0.69% 0.30% 0.59% 0.27%
Trading income/assets 0.04% 0.22% 0.02% 0.19% 0.70% 0.31% 0.06% 0.26%
Other operating income/assets 0.59% 0.26% 0.49% 0.26% 1.68% 0.24% 1.29% 0.25%
Cost of Equity 7.91% -0.22% 7.15% -0.13% 12.25% -0.47% 8.49% -0.31%
Aggregate Effect on Spread 5.3% 4.0% 10.2% 3.0% 8.2% 10.7% 8.4% 4.2%
Estimated Effect on Spread 4.3% 3.2% 4.8% 2.7% 2.7% 4.4% 4.9% 3.6%
Price-to-book-ratio (indirect) 1.73 0.22 1.75 0.15 1.95 0.66 1.83 0.25

Memo item:
Cost/income ratio 62.06% -2.43% 59.55% -2.24% 72.01% -2.95% 63.21% -2.63%

34
Table 9: Correlation of Alternative Diversification Measures

This table reports correlation coefficients between different measures for the degree of bank diversifica-
tion. The diversification measure by Laeven/Levine 2007 is calculated as Diversification(LL)=
net interest income − other operating income
1− .
total operating income

Income- Interest- Diversifica- Asset-Div.

Div. Div. tion (LL)
Income-Diversification
(Herfindahl-index based on interest,
1
fee, trading, and commission in-
come)
Interest-Diversification (Herfin-
dahl-index based on interest vs. non- 0.96 1
interest-income)
Diversification (LL)
0.65 0.69 1
(Laeven/Levine 2007-measure)
Asset-Diversification
(Herfindahl-Index based on loans -0.16 -0.08 0.04 1
vs. other assets)
Table 10: Robustness Regressions: Diversification and Market Value measured according
to Laeven/Levine 2007
The table reports the results of fixed-effects regressions for alternative diversification measures. Diversi-
fication is one minus a Herfindahl-index, based on interest, fee, commission and trading income. Diversi-
fication (LL) is the measure for revenue diversification used by Laeven/Levine 2007. Diversification
(loans) is one minus a Herfindahlindex based on loan and non-loan shares in total assets. Excess MTB is
the so called activity adjusted excess market to book ratio from Laeven/Levine 2007. *,**,*** denote
significance at the 10%, 5% and 1%-level, respectively. All models include bank individual effects and
year dummies (not shown). p-values are in parentheses. The observation period is 1996-2003.

Excess MTB Excess MTB Excess MTB Excess MTB

Diversification(LL) -0.98*** -0.92* --- ---
(0.000) (0.099)
Diversification(LL)2 --- -0.05 --- ---
(0.927)
Diversification ( loans) --- --- -10.28** ---
(0.05)
Diversification ( loans) 2 --- --- 11.85* ---
(0.062)
Diversification --- --- --- 0.97
(0.427)
Diversification2 --- --- --- -1.35
(0.434)
Vertical Integration --- 0.57 -0.43 1.26
(0.907) (0.927) (0.700)
Vertical Intergration2 --- -0.19 -0.66 -0.43
(0.949) (0.824) (0.841)

Spread --- 5.18*** 5.04*** 5.09***

(0.000) (0.000) (0.000)
M&A growth --- -0.23 -0.15 -0.22
(0.197) (0.432) (0.221)
Organic growth --- 0.02 0.05 0.02
(0.807) (0.639) (0.767)
Interest / Non-interest --- --- --- ---
Income (gross)
Size (log(assets)) -1.47*** -1.41*** -1.51*** -1.40***
(0.000) (0.000) (0.000) (0.000)
R2 0.182 0.258 0.273 0.251

36
Table 11: Endogeneity Tests

The table shows regression results controlling for endogeneity between the degree of diversification and
banks’ market values as robustness tests of our results in Table 4. For a definition of the variables, see
Table 1. Estimator denotes the applied estimation technique, where FE is fixed effects and 2SLS is two-
stage least squares. The first row denotes the dependent variable, where Activity Adjusted Excess Q is
the market value measure suggested by Laeven/Levine (2007). *, **, *** denote significance at the 1%,
5%, and 10%-level, respectively.

Market-to- Spread Market-to- Spread Activity

Book Book Adjusted
Excess
Q
Estimator FE FE FE, 2SLS FE, 2SLS FE,
2SLS
Diversification (Lag 1 / 0.02 0.19*** 0.04 0.37*** 0.92
Instrumented) (0.970) (0.000) (0.970) (0.000) (0.286)
Vertical Integration -3.81 1.03*** -3.82 0.89*** 0.05
(0.272) (0.000) (0.270) (0.000) (0.989)
Vertical Intergration2 2.22 -0.45*** 2.24 -0.28** 0.65
(0.317) (0.000) (0.321) (0.02) (0.773)
Spread 5.71*** --- 5.70*** --- 4.83***
(0.000) (0.000) (0.000)
M&A growth -0.24 -0.04*** -0.23 -0.03*** -0.21
(0.197) (0.000) (0.199) (0.001) (0.236)
Organic growth 0.10 0.00 0.10 0.00 0.04
(0.204) (0.390) (0.206) (0.837) (0.618)

Interest / Non-interest -0.09*** 0.00 -0.09*** -0.01*** ---

Income (gross) (0.001) (0.174) (0.007) (0.000)
Size (log(assets)) -1.46*** -0.01*** -1.46*** -0.02*** -1.51***
(0.000) (0.008) (0.000) (0.004) (0.000)
R2 0.321 0.067 0.321 0.216 0.253

37
Table 12: Selectivity Tests

The table shows regression results for the regression of banks’ market values on a set of explanatory
variables, including a proxy for the degree of diversification as robustness tests of our results in Table 4.
For a definition of the variables, see Table 1. Estimator denotes the applied estimation technique, where
Treatment indicates the Heckman (1979) treatment effects model, and Cluster indicates that standard
errors are corrected for bank-individual clustering. The first row denotes the dependent variable, where
the Activity Adjusted Excess Q is the market value measure suggested by Laeven/Levine (2007). *, **,
*** denote significance at the 1%, 5%, and 10%-level, respectively.

Market-to-Book Activity Adjusted Excess

Q
Estimator Treatment, Clustering Treatment, Clustering
Diversified (Dummy) 0.20 0.02
(0.364) (0.938)
Vertical Integration 1.94 5.53
(0.724) (0.989)
Vertical Intergration2 -3.25 -5.51
(0.347) (0.102)
Spread 8.04*** 7.54***
(0.000) (0.000)
Interest / Non-interest -0.00 ---
Income (gross) (0.0.916)
Size (log(assets)) 0.05** 0.04
(0.032) (0.136)
Simultaneous Probit Estimation (Diversified as Dependant)
M&A growth -0.58* -0.58*
(0.067) (0.065)
Organic growth 0.37*** 0.37
(0.001) (0.001)
Cost-Income-Ratio 1.68*** 1.69***
(0.000) (0.000)
Wald-Test of Inde- (0.109) (0.069)*
pendent Equations

38
Table 13: Subsample-Regressions

The table shows regression results for the non-U.S. subsample of our data and the baseline model aug-
mented by the interaction of diversification and the dummy variable INTERNATIONAL, which takes
the value of 1 if international operations contribute at least 20% to total revenues, and zero else. The
model specifications are similar to Table 4, including bank individual fixed effects and year dummies
(coefficients not reported). For a definition of the variables, see Table 1. The first row denotes the de-
pendent variable, where Activity Adjusted Excess Q is the market value measure suggested by
Laeven/Levine (2007). *, **, *** denote significance at the 1%, 5%, and 10%-level, respectively.

Market-to- Spread Activity Market-to- Spread

Book Adjusted Book
Excess Q
Subsample Not USA Not USA Not USA Intern. Intern.
Diversification 0.30 0.28* 2.22 -0.49 0.28***
(0.879) (0.086) (0.232) (0.555) (0.000)
Diversification2 -0.83 -0.03 -3.07 --- ---
(0.756) (0.897) (0.232)
Diversification x In- --- --- --- 1.91 -0.05
ternational (0.139) (0.563)
Vertical Integration -8.99** 1.92*** -1.70 -0.60 0.37***
(0.031) (0.000) (0.659) (0.58) (0.000)
Vertical Intergration2 5.09** -0.96*** 1.23 --- ---
(0.041) (0.000) (0.600)
Spread 4.38*** --- 3.54*** 5.76*** ---
(0.000) (0.000) (0.000)
M&A growth -0.75** -0.01*** -0.71 -0.22 -0.38***
(0.027) (0.750) (0.03) (0.146) (0.000)
Organic growth -0.05 0.02* -0.12 0.07 0.00
(0.609) (0.056) (0.763) (0.495) (0.937)

Interest / Non-interest -0.04 0.01*** --- -0.07** 0.00

Income (gross) (0.211) (0.003) (0.025) (0.250)
Size (log(assets)) 0.03 -0.03** 0.04 -1.46*** -0.01
(0.847) (0.044) (0.763) (0.000) (0.133)
#obs (banks) 668 (127) 668 (127) 668 (127) 1917 (380) 1917 (380)
R2 0.234 0.275 0.188 0.317 0.204

39