international diversification and risk of multinational banks: evidence from the pre-crisis period

7
ICT reuse in socio-economic enterprises F.O. Ongondo a,, I.D. Williams a , J. Dietrich b , C. Carroll a a Centre for Environmental Sciences, Faculty of Engineering and the Environment, Lanchester Building, University of Southampton, University Rd., Highfield, Southampton, Hampshire SO17 1BJ, UK b Technische Universität Berlin, Centre for Scientific Continuing Education and Cooperation, Cooperation and Consulting for Environmental Questions (kubus) FH10-1, Fraunhoferstraße 33-36, 10587 Berlin, Germany article info Article history: Received 27 March 2013 Accepted 23 August 2013 Available online 14 September 2013 Keywords: WEEE Value-conservation ICT Reuse Resource efficiency Reuse networks abstract In Europe, socio-economic enterprises such as charities, voluntary organisations and not-for-profit companies are involved in the repair, refurbishment and reuse of various products. This paper character- ises and analyses the operations of socio-economic enterprises that are involved in the reuse of Informa- tion and Communication Technology (ICT) equipment. Using findings from a survey, the paper specifically analyses the reuse activities of socio-economic enterprises in the UK from which Europe-wide conclusions are drawn. The amount of ICT products handled by the reuse organisations is quantified and potential barriers and opportunities to their operations are analysed. By-products from reuse activities are discussed and recommendations to improve reuse activities are provided. The most common ICT products dealt with by socio-economic enterprises are computers and related equipment. In the UK in 2010, an estimated 143,750 appliances were reused. However, due to limitations in data, it is difficult to compare this number to the amount of new appliances that entered the UK market or the amount of waste electrical and electronic equipment generated in the same period. Difficulties in marketing prod- ucts and numerous legislative requirements are the most common barriers to reuse operations. Despite various constraints, it is clear that organisations involved in reuse of ICT could contribute significantly to resource efficiency and a circular economy. It is suggested that clustering of their operations into ‘‘reuse parks’’ would enhance both their profile and their products. Reuse parks would also improve consumer confidence in and subsequently sales of the products. Further, it is advocated that industrial networking opportunities for the exchange of by-products resulting from the organisations’ activities should be investigated. The findings make two significant contributions to the current literature. One, they provide a detailed insight into the reuse operations of socio-economic enterprises. Previously unavailable data has been presented and analysed. Secondly, new evidence about the by-products/materials resulting from socio-economic enterprises’ reuse activities has been obtained. These contributions add substan- tially to our understanding of the important role of reuse organisations. Ó 2013 Elsevier Ltd. All rights reserved. 1. Introduction In the new global economy, managing waste electrical and elec- tronic equipment (WEEE) has become a central issue (Oguchi et al., 2011; Ongondo et al., 2011; Schluep et al., 2009). This fastest grow- ing waste stream has become a priority due to quantities of WEEE generated, potential environmental and health impacts, ethical issues relating to disposal or recycling of WEEE in developing countries, and increasingly, the amount of resources needed to manufacture electrical and electronic equipment (EEE). Demand for various scarce raw materials for the manufacture of electronics has forced countries to rethink their strategies for managing WEEE (Ongondo et al., 2011). Recent developments in the supply of critical raw materials (such as rare earth metals) have led to renewed interests in resource efficiency (BIS, 2012; European Commission, 2010; Humphries, 2012): The European Union (EU) has introduced a flagship initiative, the EU 2020 strategy 1 which champions a shift towards a resource-efficient, low-carbon economy for sustainable growth; The USA has recently invested $120 million to set up a new research centre to develop new methods of producing rare earth metals 2 ; and In the UK, a ‘‘Resource Security Action Plan’’ has been estab- lished (BIS, 2012). 0956-053X/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.wasman.2013.08.020 Corresponding author. Tel.: +44 (0)2380 592317; fax: +44 (0)2380 678606. E-mail address: [email protected] (F.O. Ongondo). 1 http://ec.europa.eu/europe2020. 2 http://energy.gov/articles/ames-laboratory-lead-new-research-effort-address- shortages-rare-earth-and-other-critical. Waste Management 33 (2013) 2600–2606 Contents lists available at ScienceDirect Waste Management journal homepage: www.elsevier.com/locate/wasman

Upload: alberto-franco

Post on 23-Dec-2016

222 views

Category:

Documents


0 download

TRANSCRIPT

IE

Ma

b

c

a

ARRAA

JGGFFL

KBRMEM

1

mtiafiitb2

ST

c

h1

Journal of Financial Stability 13 (2014) 30–43

Contents lists available at ScienceDirect

Journal of Financial Stability

journal homepage: www.elsevier.com/locate/jfstabil

nternational diversification and risk of multinational banks:vidence from the pre-crisis period

.A. Gulamhussena, Carlos Pinheirob, Alberto Franco Pozzoloc,∗

ISCTE Business School, Instituto Universitário de Lisboa, PortugalCaixa Geral de Depósitos, PortugalUniversità degli Studi del Molise, Italy

r t i c l e i n f o

rticle history:eceived 20 March 2013eceived in revised form 20 March 2013ccepted 27 February 2014vailable online 12 March 2014

EL classification:2132233622

a b s t r a c t

The recent financial crisis has clearly shown that the relationship between bank internationalizationand risk is complex. Multinational banks can benefit from portfolio diversification, reducing their overallriskiness, but this effect can be offset by incentives going in the opposite direction, leading them to takeon excessive risks. Since both effects are grounded on solid theoretical arguments, the answer of whatis the actual relationship between bank internationalization and risk is left to the empirical analysis. Inthis paper, we study such relationship in the period leading to the financial crisis of 2007–2008. For asample of 384 listed banks from 56 countries, we calculate two measures of risk for the period from 2001to 2007 – the expected default frequency (EDF), a market-based and forward-looking indicator, and theZ-score, a balance-sheet-based and backward-looking measure – and relate them to the degree of banks’internationalization. We find robust evidence that international diversification increases bank risk.

© 2014 Elsevier B.V. All rights reserved.

eywords:anksiskultinational banking

conomic integration

sttsnehoM

arket structure

. Introduction

Scholars have traditionally viewed bank internationalizationore favorably than regulators and policymakers, on the grounds

hat opening the banking market to foreign players will resultn larger economies of scope and scale, increased competition,nd better risk diversification. However, recent research has con-rmed part of the fears of policymakers, showing that bank

nternationalization may induce multinational players to hinderhe development of local banks, cherry-pick the best clientele and

ypass local regulations (Detragiache et al., 2008; Ongena et al.,013).

∗ Corresponding author at: Università degli Studi del Molise, Dipartimento dicienze Economiche Gestionali e Sociali, Via De Sanctis, 86100 Campobasso, Italy.el.: +39 0874404338; fax: +39 087498043.

E-mail addresses: [email protected] (M.A. Gulamhussen),[email protected] (C. Pinheiro), [email protected] (A.F. Pozzolo).

obmSco

w

ttp://dx.doi.org/10.1016/j.jfs.2014.02.007572-3089/© 2014 Elsevier B.V. All rights reserved.

At first sight, one may classify this debate as a standard discus-ion between those in favor of free markets and those who believehat certain economic sectors, such as the financial markets, needo be strictly regulated. But the recent financial crisis has clearlyhown that market forces, especially in the financial sector, areot always capable of driving the economic system to the first bestquilibrium. The call for stricter regulation of financial activitiesas been strong, with particular attention being paid to the rolef the so-called Global Systemically Important Banks (G-SIBs).1

any scholars have argued that G-SIBs consist almost exclusivelyf large multinational banks, a view that is largely shared alsoy regulators and policy makers. In the indicator-based measure-ent approach proposed by the Basel Committee on Banking

upervision to assess whether a bank is a G-SIB, the first categoryonsidered is the degree of cross-jurisdictional activity, with thebjective “to capture the global footprint of banks” (BCBS, 2011,

1 See for example the discussion in Avgouleas et al. (2011) on the role of livingills and burden sharing among G-SIBs in the G20s.

of Fin

pdoctbatbwr

n2irepiictqtiasdsG

ntfif

wbttcStP

o3g4r

s

2(itdstoCo

2

a(r(giidlpg

ltvcsmtbnadttagffitsrccfwtb(adfIcttm

M.A. Gulamhussen et al. / Journal

. 5). According to this view, multinational banks are too risky. Theirefault is likely to generate substantial spillover effects to the restf the system, because they are large and operate in many differentountries (Huang et al., 2011), and their complex corporate struc-ures may trigger perverse incentives and excessive risk-takingehavior. The higher complexity and excessive agency problemsssociated with multinational banks would therefore outweighhe benefits of diversifying the idiosyncratic risks.2 However, sinceoth of these theses are grounded on solid theoretical arguments,e believe that whether multinational banks are more or less

isky than domestic institutions is ultimately an empirical issue.Although a vast number of studies have analyzed the determi-

ants of bank risk-taking (Boyd et al., 2006; Vander Vennet et al.,004), the relationship between international diversification and

ndividual bank risk-taking has received much less attention. Theelevance of the issue is confirmed by a recent paper by Bergert al. (2013) that nonetheless focus only on U.S. banks. In thisaper, we contribute to filling this gap in the literature by study-

ng the relationship between international diversification and riskn a sample of 384 listed banks headquartered in 56 differentountries around the world. We consider two measures of riskhat are widely used in the literature: the expected default fre-uency (EDF), a market-based and forward-looking measure, andhe Z-score, a backward-looking measure based on balance-sheetnformation. Our sample includes both internationally diversifiednd purely domestic banks (the latter representing 23.4% of ourample). We cover the 7 years before the financial crisis, a perioduring which the riskiness of banks and financial markets increasedubstantially, eventually leading to the worst collapse since thereat Depression.3

Our results show that internationally diversified banks are sig-ificantly riskier than domestically oriented banks, consistent withhe recent findings of Berger et al. (2013). These findings are con-rmed adopting different estimation techniques and controlling

or potential endogeneity of the internationalization choice.Our work is related to two major streams of the literature. First,

e contribute to the analysis of the determinants of bank risk takingy considering a specific type of corporate structure, the multina-ional bank, which is most likely to be affected by all of the problemshat standard agency theories have shown to be major causes oforporate risk taking (Jensen and Meckling, 1976; John et al., 2008).econd, we contribute more specifically to the literature analyzinghe characteristics of international banks (Buch and DeLong, 2009;ozzolo, 2009).

The rest of the paper is organized as follows. Section 2 relatesur research to the previous literature on bank risk taking. Section

describes our empirical strategy and the measures of risk andeographic diversification used in the empirical analyses. Section

presents the sources of our information. Section 5 presents theesults. Section 6 concludes this paper.

2 See also Battiston et al. (2012) for an additional channel through which diver-ification may increase rather than decrease stability.

3 Increased bank risk-taking has been recognized as one of the major causes of the007–2008 financial crisis (Gorton and Metrick, 2012; Lo, 2012), but Berger et al.2013), studying US international banks, find that the relationship between banknternationalization and risk taking is not altered during financial crises, althoughhere is some evidence that it becomes stronger during market crises and weakeruring banking crises. We leave to future research adopting an empirical frameworkpecifically aimed at this objective to directly assess the impact of internationaliza-ion on the increase in bank riskiness that led to the 2007–2008 financial crisis andn the following choices made by international banks. The recent paper of Allen andarletti (2013) also begged for a new theoretical redesign to address the preventionf financial crisis.

cttgsi

eArg

ancial Stability 13 (2014) 30–43 31

. Related literature

The benefits from the diversification of idiosyncratic risks aremong the best understood concepts in the economic literatureCochrane, 2001): according to portfolio theory, diversification caneduce the effect of idiosyncratic shocks and thus overall riskLewellen, 1971; Markowitz, 1959). From this perspective, theeographic diversification of banks should dampen the effects ofdiosyncratic shocks and, in this way, reduce their overall risk-ness. Although the potential gains from international portfolioiversification are still an object of current research in the finance

iterature,4 Buch et al. (2010) recently showed that bank assetortfolios exhibit a significant home bias. According to this view,eographic diversification should reduce aggregate bank risk.5

However, multinational banks typically have access to a mucharger set of strategies to increase their risky activities than domes-ic banks. Additionally, these activities may be hidden from theiew of local regulators. Incentive problems lie at the root of thesehoices. Although international diversification may prove to be auboptimal decision once one accounts for the costs of increasedanagement complexity, insiders may still support the acquisi-

ion of foreign participations if in this way they can obtain privateenefits (Jensen, 1986; Jensen and Meckling, 1976). In their semi-al paper, Jensen and Meckling (1976) contend that agency costsrising from the conflicting interests of managers and sharehol-ers should have a negative effect in risk-taking. For example,he increased asset liquidity associated with international opera-ions might provide bank managers with more possibilities to tradegainst the bank’s interest (Myers and Rajan, 1998). In this case,eographic diversification may cause an increase in bank risk. Inact, the relevance of incentive issues in bank risk-taking was con-rmed in a recent study by Laeven and Levine (2009), who showhat control problems have a first-order impact on corporate deci-ions and that banks with less dispersed shareholders are generallyiskier. An important related issue is that international diversifi-ation could lead to more risk simply because it requires a moreomplex and opaque organization, that is in turn more likely to suf-er from agency costs. These agency problems become more acuteith higher bank complexity that may increase bank risk. Since

here is convincing empirical evidence of a negative relationshipetween Tobin’s Q and agency problems in the banking industryDemsetz et al., 1996, 1997; Klein and Saidenberg, 2010), we expect

negative sign of the coefficient estimate of Tobin’s Q when theependent variable is bank risk measured by the expected defaultrequency EDF and the symmetrical of Z-score, as we detail later on.n addition to agency problems, a number of specific features of theulture, the institutional environment and the market structure ofhe foreign country may increase the riskiness of banks’ interna-ional activities, consistent with what Berger et al. (2013) call thearket risk hypothesis.

As we argued above, determining the net effect of the pros andons of diversification is mainly an empirical issue. However, also inhis case the results of the literature studying the effects of differentypes of diversification are rather mixed. Some authors show that

eographic diversification increases bank risk. Hughes et al. (1996),tudying the effects of U.S. branching deregulation, show that anncrease in the number of U.S. States in which a bank holding

4 See Karolyi and Stulz (2002) and Stiroh (2009) for a survey of this literature.5 This is consistent with what Berger et al. (2013) define as diversification hypoth-

sis, according to which international banks should have lower risk. See alsockermann (2008) for both a practitioner’s description of the background of theecent financial crisis and most importantly the view that banks with a greatereographic footprint fared better.

3 of Fin

cn(hbAtidprdtcbbas

bpesdti

3

3

im

R

wtcesdtr(irabds

3

u

oibt

(

foa

raV(sb

Zcadfoe

(esbei

3

dpiwHta2tmbad

m

I

wheioa

2 M.A. Gulamhussen et al. / Journal

ompany operates increased insolvency risk, whereas a rise in theumber of branches per se had the opposite effect. De Nicolò et al.2004) find that large conglomerate corporations did not exhibitigher levels of risk-taking behavior than average banks in 1995,ut did so in 2000. Other studies reached the opposite conclusion.nalyzing US mergers and acquisitions, Zhang (1995) found

hat geographical diversification leads to lower risk by reducingncome variability. Deng et al. (2007) show that banks that areomestically diversified on both the assets and the liabilities sidesay lower bond spreads, which provide indirect evidence of lowerisk. Similarly, Deng and Elyasiani (2008) find that geographicallyiversified banks have lower stock price variability. With regardo international diversification, Amihud et al. (2002) show thatross-border mergers and acquisitions (M&As) have no effects onidders’ systematic risk levels, although this result is questionedy Focarelli et al. (2008), who find instead that bidders experience

reduction in their beta (i.e., the correlation of their returns withtock market returns).6

In a recent paper, Berger et al. (2013) study the relationshipetween bank internationalization and risk-taking on a large sam-le of US banks between 1989 and 2010. Following the previousmpirical literature, they also measure bank risk by their Z-score,howing that U.S. international banks are overall riskier than theiromestic counterparts. These findings can be seen as complemen-ary to the analysis presented below, that is based instead on annternational sample of large banks.

. Empirical strategy

.1. Econometric model

Our test of whether a bank’s risk is a function of its degree ofnternational diversification is based on the following empirical

odel:

iskjt = + International diversificationjt + �controlsjt + εjt,

(1)

here the measures of risk and international diversification refero bank j at time t; the controls include time-varying bank-specificharacteristics as well as time and country dummies; and εjt is anrror term. To account for the large values of the coefficients ofkewness and courtosis of our dependent variables, we trim ourata by excluding observations below the 1st percentile and abovehe 99th percentile, and we estimate the model with standard OLS,obust regression techniques, that give lower weigh to outliersLi, 1985), and quantile regressions evaluated at the median, thatdentify the regression plane minimizing the sum of the absoluteesiduals (Cameron and Trivedi, 2009). In robustness tests we uselternative econometric specifications and alternative measures ofank-risk and international diversification. In the following, weiscuss in detail our measures of risk taking and international diver-ification as well as the controls introduced in our empirical model.

.2. Measures of risk

The empirical literature has proposed a large number of meas-res of bank risk.7 In our analysis, we use a market-based and

6 Our research is also related to the analyses of the risk effects of the diversificationf banking activities: Baele et al. (2007) show that a larger share of noninterestncome is associated with higher systematic risk, which is measured by the marketeta, and Demirgüc -Kunt and Huizinga (2010) also confirm this result with regardo the Z-score.

7 For example, see Berger and De Young (1997), Williams (2004), Garlappi et al.2006), Buch and DeLong (2009), Laeven and Levine (2009), Altunbas et al. (2010),

ddd

C(

r

aa

ancial Stability 13 (2014) 30–43

orward-looking index, the expected default frequency (EDF) basedn Black and Scholes (1973) and Merton (1974), and an accountingnd backward-looking indicator, the Z-score.8

More precisely, as our first measure we take the natural loga-ithm of the expected default frequency, that is the bank’s 5-yearhead cumulative EDF provided by Moody’s KMV, based on theasicek–Kealhofer model, as explained in detail by Kealhofer

2003). The forward-looking feature endows EDF with earlyignaling properties, as compared to other measures of bank risk,oth traditional and market-based (Saldias, 2013).

Following Laeven and Levine (2009), our second measure, the-score, is computed as (CAR + EQT)/�(ROA), where CAR is theapital-to-total-assets ratio, EQT is the equity-to-total-assets ratio,nd �(ROA) is the standard deviation of the return on assets (ROA)uring our sample period.9 Under the assumption that profitsollow a normal distribution, the Z-score measures the numberf standard deviations that a bank’s ROA has to drop below itsxpected value before the equity is entirely depleted.

Because the Z-score is a negative function of the risk of defaulti.e., banks with a higher Z-score are less likely to default), in ourstimates we use the opposite of the Z-score, which we call Z′-core, so that a positive sign of its coefficient implies an increase inank risk, similar to the measure based on EDF. Similar to Bergert al. (2013), we have also analyzed the relationship between banknternationalization and each one of the components of the Z-score.

.3. Measures of international diversification

Most conventional empirical studies measure internationaliversification by considering the number of subsidiaries in a cor-oration or the number of countries in which the corporation

s present. Some studies simply use a binary variable indicatinghether the corporation is present in a given foreign country.owever, these measures do not precisely assess the level and

he intensity of banks’ international diversification and do notccount for its steep trend before the financial crisis (Pozzolo,009). Therefore, following Gulamhussen et al. (2010), we computehree alternative measures of international diversification. Each

easure positions the banks over a continuum, with the loweround corresponding to purely non-diversified (domestic) banksnd the upper bound corresponding to the most internationallyiversified banks.

Our first measure is a proxy for international reach and is for-ally defined as follows:

nternational reachj,t = nj,t

nmax,t(2)

here nj,t is the number of foreign countries in which the bank jas a subsidiary in year t, and nmax,t is the maximum number of for-ign countries in which the most diversified bank has subsidiariesn year t. Clearly, international reach is a stock variable, continu-us, and bounded between 0 and 1. Purely domestic banks assume

value of 0, and values close to 1 indicate more internationally

ispersed banks. This index normalizes the measure of geographiciversification by accounting for the yearly variation of the mostiversified banks.

hiaramonte and Casu (2010), Fiordelisi et al. (2010) and De Haan and Poghosyan2011).

8 Bank specific Z-scores are commonly used as a measure of individual banks’iskiness (see, e.g., Demirguüc -Kunt and Detragiache, 2011).

9 We have checked that our results are confirmed also using the standard devi-tion of ROA calculated over the previous 3 years; results are available from theuthors upon request.

of Fin

ad2tdw

I

ct

mttetcobhrso

aoacfc

I

wv

saro

3

ncscpiapn

(

fiata

pittttdiusatttfinntdawbwcaioci(fomcrpiec

3

tbew

4

4

M.A. Gulamhussen et al. / Journal

Our second measure, international share, computes the share ofssets on a country-by-country basis by considering a bank’s assetispersion across its subsidiaries (similar to Buch and Lipponer,007). We compute the difference between the total assets andhe foreign assets of the bank’s subsidiaries, and then scale thisifference by the total assets of all bank’s subsidiaries. Formally,e calculate the following index:

nternational sharej,t = foreign subsisiaries assetsj,t

total subsidiaries assetsj,t(3)

International share is also bounded between 0 and 1, with valueslose to 0 indicating low geographic diversification and values closeo 1 indicating high geographic diversification.

Compared with international reach, international share conveysore information, in that it considers the incidence of foreign par-

icipations in the aggregate activities of a banking group. In fact,he international reach of a bank that has 4 foreign subsidiaries,ach of which accounts for just 2% of its total assets, is identical tohat of a group that spreads its activities equally across differentountries and that has 5 subsidiaries, each of which represents 20%f the total assets; on the contrary, the international share of the twoanks is, respectively, 0.08 and 0.80. However, international shareas the drawback that a bank with just one foreign subsidiary thatepresents 50% of its activities is identical to a bank with foreignubsidiaries in 10 different countries, each of which represents 5%f this bank’s total assets.

To account for both the number of foreign countries in which bank is present and the weight of each activity, we calculatedur third measure of diversification, international concentration, as

transformed Hirsch–Herfindhal Index (Mercieca et al., 2007). Weompute this index based on the total assets of a banking group’soreign participations in each subsidiary in the various foreignountries.10 Formally, we define the index as follows:

nternational concentration=1 −nj∑

j=1

(subsidiaries assetsj

total subsidiaries assets

)2

(4)

International concentration is also bounded between 0 and 1,ith values close to 0 indicating low geographic diversification and

alues close to 1 indicating internationally dispersed banks.The three measures described above provide a different per-

pective on the degree and mode of internationalization of a bank,s it is made clear also by their positive but imperfect degree of cor-elation (between 0.45 and 0.76). For this reason, we will presentur results using all three measures.

.4. Other bank characteristics

Bank riskiness is related to other characteristics than just inter-ational diversification. For example, during the recent financialrisis, scholars have forcefully argued that large banks had exces-ively high risk attitudes because they discounted the fact that, inases of distress, the government would have bailed them out usingublic money (i.e., they were too big to fail). At the same time, it

s well known that larger banks are more likely to be internation-lly diversified. Because we are interested in measuring only theartial correlation between international diversification and risk,eglecting to control for size might introduce a bias in favor of

10 Similar concentration measures can be found in the works of Acharya et al.2006) and Stiroh and Rumble (2006).

fi

T

u

ancial Stability 13 (2014) 30–43 33

nding a positive relationship just because larger banks are riskiernd more international at the same time. As argued above, an addi-ional issue relates to the link between international diversificationnd organizational costs.

To account for these and other bank characteristics that mightotentially bias our results, we include a number of time vary-

ng bank-specific controls in our specification. First, we considerhe logarithm of total assets as a measure of bank size. In additiono the too-big-to-fail argument, the asset and loan portfolios andhe activities of larger banks are typically far more diversified thanhose of smaller institutions. This difference obviously impacts theegree of risk that banks take on, independently of their degree of

nternational diversification. In a set of robustness checks availablepon request we also included total income that measures bankize taking indirectly into account also the role of off-balance sheetctivities. Second, we control for the share of retail deposits overotal liabilities, because the funding composition is likely to affecthe banks’ lending and investing strategies. Third, we account forhe role of organizational and operational complexity. For largernancial institutions, as it is the case of the vast majority of inter-ational banks, the danger of becoming a large number of ‘silos’ oneext to the other instead of a single institution achieving beneficialeamwork and collaboration is indeed stronger than for smalleromestic banks. Also, the misalignment between insiders (man-gers) and outsiders (shareholders) is an unavoidable issue. Sincee are interested in measuring the effect of internationalization on

ank risk abstracting from these internal organizational problems,e account for complexity and agency problems introducing three

ontrol measures: the ratio of the number of employees over totalssets (Kaen and Baumann, 2003; Becker-Blease et al., 2010), anndex of income diversity (Laeven and Levine, 2007), and a proxyf Tobin’s Q, obtained as the ratio of the sum of the market value ofommon stocks, the book value of preferential shares and minoritynterests, and the book value of debt to the book value of total assetsLindenberg and Ross, 1981; Sweeney et al., 2001).11 The rationaleor these measures is that banks with a larger number of employeesver total assets and more diversified revenue sources are typicallyore complex. At the same time, since banks with a higher fran-

hise value suffer less from agency problems and therefore are lessisky (Demsetz et al., 1996, 1997; Klein and Saidenberg, 2010), weroxy agency costs with our measure of franchise value and use

t to control for the role of the higher agency costs. Therefore, wexpect that lower values of the Tobin’s Q ratio, that imply higheromplexity, are associated with higher risk.12

.5. Country controls

Country characteristics, such as the strictness of the regula-ory environment, are most likely to have a strong impact onank strategies and risk. To control for these possibly confoundingffects on the relationship between internationalization and risk,e include country dummies in all our specifications.

. Data and summary statistics

.1. Sources

We focus on commercial banks, a homogeneous group ofnancial institutions that has been found to have compelling

11 We are grateful to an anonymous reviewer for the suggestion to include alsoobin’s Q as a measure of complexity.12 Unfortunately, from our data it was not possible to calculate more precise meas-res of organizational complexity such as those in Laeven and Levine (2008).

34 M.A. Gulamhussen et al. / Journal of Financial Stability 13 (2014) 30–43

Table 1Summary statistics.Expected default frequency (EDF) is the logarithm of measure at the 5-year horizon; the transformed Z′-score is the symmetric value of (CAR + EQT)/�(ROA), where CAR isthe capital asset ratio, EQT is the equity-to-assets ratio, and �(ROA) is the volatility of returns. International reach is the ratio of nj,t to nmax,t , where njt is the number of foreigncountries in which bank j has a subsidiary in year t, and nmax,t is the maximum number of foreign countries in which the most diversified bank has subsidiaries in year t.International share is the geographic dispersion of subsidiaries (geographic share), which is estimated by foreign subsidiaries assets/total subsidiaries assets. Internationalconcentration is the transformed Hirsch–Herfindhal index (HHI) (1 −

∑j(subsidiaryj assets/total subsidiaries assets)2. Employees to assets are the logarithm of the bank’s

number of employees divided by its assets. Income diversity is computed as 1 − |(net interest income − other operating income)/(total operating income)|. Share of diversifiedbanks is the share of geographically diversified banks in the country. Tobin’s Q is the ratio of the sum of the market value of common stocks, the book value of preferentialshares and minority interests, and the book value of debt to the book value of total assets. Dummy SP takes the value 1 for banks included in the Standard & Poor’s 500financial index.

Variable Mean Coefficient of variation 1st percentile 99th percentile

Expected default frequency (EDF) −0.90 −1.36 −3.79 1.95Z′-score −3.73 −1.08 −19.43 1.84International reach (n/nmax) 0.07 2.27 0.00 0.81International share 0.17 1.80 0.00 1.00International concentration (HHI) 0.17 1.47 0.00 0.80Log of assets 6.91 0.13 5.14 9.12Deposits to liabilities 0.88 0.19 0.18 0.99Employees to assets −13.34 −0.20 −17.80 −8.19Income diversity 0.63 0.41 0.00 0.99Share of diversified banks 0.74 0.31 0.00 1.00

1205

r2eodotwsCmbvummab

yoda1

som

4

i

BEIMQLr

iisp((iNBmstB(i(sPNblS

eacifc

Tobin’s Q 1.06 0.Dummy SP 0.10 3.

easons to internationalize their activities (Barba Navaretti et al.,010; Focarelli and Pozzolo, 2005). To assemble our data, we firstxtracted the yearly account and market data from 2001 to 2007n all of the listed commercial banks on Bankscope, a commercialatabase produced by Bureau van Dijk, with total assets in excessf US$ 100 million. We excluded smaller banks that may face addi-ional challenges and costs in diversifying across borders comparedith large banks. We also excluded banks headquartered in off-

hore centers, such as Bermuda, Gibraltar, the Virgin Islands or theayman Islands, because they typically have less standard businessodels. We retrieved some missing information from Worldscope

y Thomson Financial, Datastream by Thomson Reuters and indi-idual bank websites, integrating in this way our initial data set. Wendertook a painstaking effort to clean and complement the infor-ation downloaded from Bankscope and to avoid incongruent andissing data on crucial account and market variables. Our data-

ssembling exercise yielded an initial sample of 577 commercialanks and 4039 bank-year observations.

By matching our initial 577 publicly traded banks with theearly data on the banks’ subsidiaries, we obtained a final samplef 384 banks headquartered in 56 countries for which time-varyingata on domestic and foreign subsidiaries were available.13 Japannd the U.S. have the largest number of banks in our sample, with7.0% and 9.4% of the total number of banks, respectively.

Data on EDF are from Moody’s KMV. When merging our balanceheet information with the risk measures, the number of banks inur sample falls to slightly less than 250 banks, depending on theeasure of risk that we used.

.2. Summary statistics

Table 1 presents the summary statistics of the dependent andndependent variables in our empirical model. The measures of

13 The 56 countries in our sample are the following: Australia, Bangladesh,elgium, Brazil, Canada, China, Colombia, Croatia, Czech Republic, Denmark, Egypt,stonia, Finland, France, Germany, Greece, Hong Kong, India, Indonesia, Ireland,srael, Italy, Japan, Jordan, Kenya, Rep. of Korea, Kuwait, Lebanon, Lithuania,

alaysia, the Netherlands, Oman, Pakistan, Peru, Philippines, Poland, Portugal,atar, Romania, Saudi Arabia, Singapore, Slovakia, Slovenia, South Africa, Spain, Srianka, Sweden, Switzerland, Taiwan, Thailand, Tunisia, Turkey, United Arab Emi-ates, United Kingdom, United States, and Venezuela.

itttasitatil

0.89 1.570.00 1.00

nternational diversification and of risk show substantial variabil-ty. The more internationally diversified commercial banks in ourample exhibit an international reach in excess of 0.75. Exam-les of these banks include ABN Amro (Netherlands), BNP ParibasFrance), Société Générale (France), Citibank (U.S.) and the HSBCU.K.). Purely domestic banks, that have no foreign subsidiaries,nclude instead 1st Source Bank (U.S.), Citizens Bank (U.S.), Cityational Bank (U.S.), Banca Italalease (Italy), Canadian Westernank (Canada), Howa Bank (Japan), and Daishi Bank (Japan). Wheneasuring international diversification in terms of international

hare (i.e., the weight of foreign subsidiaries’ assets in subsidiaries’otal assets), the most internationally dispersed banks are Deutscheank (Germany), Unicredit (Italy) and the Royal Bank of ScotlandU.K.). According to the modified Hirsch–Hirfindhal Index of banknternational concentration, the most diversified banks are BBVASpain), ING (Netherlands) and the National Bank (Greece). Ourample shows high variability also in terms of bank size: BNParibas (France), Deutsche Bank (Germany), HSBC (U.K.), ING (Theetherlands), Santander (Spain) and UBS (Switzerland), the largestanks in terms of total assets, are about two order of magnitudes

arger than small financial intermediaries such as Citizens Bank,unwest Bank, First California Bank (U.S.) and Howa Bank (Japan).

Bank riskiness also shows high variability. The logarithm of thexpected default frequency (EDF), the forward looking measuredopted in our empirical model, ranges from −3.79 at the 1st per-entile to 1.95 at the 99th percentile, and its coefficient of variations −1.36. The Z′-score, our balance sheet based measure, also rangesrom −19.43 at the 1st percentile to 1.84 at the 99th, and it has aoefficient of variation of −1.08.

Table 2 presents the pairwise correlations among the variablesncluded in the empirical model. Our two measures of riskiness,he EDF and the Z′-score, have a positive correlation of 0.33, statis-ically significant at the 1% level. As already mentioned above, alsohe three measures of international diversification are positivelynd statistically significantly correlated: international reach has aample correlation of 0.45 with international share and 0.58 withnternational concentration, while the correlation between the lat-er two measures is 0.76. The sample correlation between bank size

nd risk is negative and statistically significant only with respecto the forward looking measure, the EDF, while it is insignificantn the case of the Z′-score, that is a backward looking measure. Aarger share of funding by means of deposits is instead associated

M.A. Gulamhussen et al. / Journal of Financial Stability 13 (2014) 30–43 35

Table 2Correlation matrix.Expected default frequency (EDF) is the logarithm of measure at the 5-year horizon; the transformed Z′-score is the symmetric value of (CAR + EQT)/�(ROA), where CAR is thecapital asset ratio, EQT is the equity-to-assets ratio, and �(ROA) is the volatility of returns. International reach is the ratio of nj,t to nmax,t , where njt is the number of foreigncountries in which bank j has a subsidiary in year t, and nmax,t is the maximum number of foreign countries in which the most diversified bank has subsidiaries in year t.International share is the geographic dispersion of subsidiaries (geographic share), which is estimated by foreign subsidiaries assets/total subsidiaries assets. Internationalconcentration is the transformed Hirsch–Herfindhal index (HHI): 1 −

∑j(subsidiaryj assets/total subsidiaries assets)2. Employees to assets are the logarithm of the bank’s

number of employees divided by its assets. Income diversity is computed as 1 − |(net interest income − |other operating income)/(total operating income)|. Share of diversifiedbanks is the share of geographically diversified banks in the country. Tobin’s Q is the ratio of the sum of the market value of common stocks, the book value of preferentialshares and minority interests, and the book value of debt to the book value of total assets. Dummy SP takes the value 1 for banks included in the Standard & Poor’s 500financial index. Numbers in bold denote significance at the 1% level.

1 2 3 4 5 6 7 8 9 10 11

1 EDF 12 Z′-score 0.33 13 International reach (n/nmax) −0.14 −0.16 14 International share −0.14 −0.08 0.45 15 International concentration (HHI) −0.15 −0.14 0.58 0.76 16 Log of assets −0.18 0.00 0.58 0.41 0.47 17 Deposits to liabilities 0.14 0.05 −0.46 −0.23 −0.27 −0.19 18 Employees to assets −0.09 −0.06 0.29 0.38 0.47 0.12 −0.19 19 Income diversity −0.27 −0.17 0.03 0.04 0.04 0.05 −0.01 0.15 1

10 Share of diversified banks −0.02 0.00 −0.20 −0.22 −0.25 −0.06 0.15 −0.28 0.03 111 Tobin’s Q −0.30 −0.23 −0.13 0.00 −0.01 −0.12 0.00 0.18 0.19 0.00 1

0.42

wpelonrtiroa

pvmtuwi

TBTtrybit

12 Dummy SP −0.20 −0.11 0.54

ith higher riskiness. Our proxies for operational complexity alsorovide a mixed picture. Banks with a larger number of employ-es over total assets and higher income diversification tend to beess risky. Although these results are inconsistent with our previ-us hypothesis, they are most likely due to the fact that they doot capture only higher complexity. Indeed, more traditional andetail banking activities, such as lending and deposit taking, areypically more labor intensive and less risky than investment bank-

ng activities. At the same time, income diversification typicallyequires higher complexity, but it also guarantees lower variabilityf total revenue, and therefore lower risk. On the contrary, the neg-tive correlation of risk and Tobin’s Q confirms the results of the

cbfim

able 3aseline specification.he dependent variable is one of the following risk measures: (i) the logarithm of the 5-yeransformed Z′-score, which is the symmetric value of (CAR + EQT)/�(ROA) that gauges the

atio, and �(ROA) is the volatility of returns. International reach is the ratio of nj,t to nmax

ear t, and nmax,t is the maximum number of foreign countries in which the most diversank’s number of employees divided by its assets. Income diversity is computed as 1 − |(n

s the ratio of the sum of the market value of common stocks, the book value of preferenotal assets. All regressions include country and year fixed effects. Risk measures are trim

EDF

OLS reg. Robust reg.

(1) (2)

International reach (n/nmax) 0.967*** 0.607**

(0.326) (0.264)

Log of assets 0.074 −0.065

(0.171) (0.137)

Deposits to liabilities −0.184 −0.419

(0.432) (0.296)

Employees to assets −0.177** −0.112*

(0.073) (0.060)

Income diversity −0.167 −0.047

(0.106) (0.105)

Tobin’s Q −2.522*** −3.301***

(0.599) (0.387)

Country effects Yes Yes

Year effects Yes Yes

Number of observations 671 670

* Significance at the 10% level.** Significance at the 5% level.

*** Significance at the 1% level.

0.42 0.53 −0.26 0.15 0.08 −0.10 −0.06

revious literature, showing that banks with a lower franchisealue are more exposed to agency problems, which in turn deter-ine a higher level of risk. Finally, the sample correlation between

he measures of international diversification and the two meas-res of bank risk is in all cases negative and statistically significant,ith values ranging from −0.08 to −0.16, suggesting that higher

nternational diversification implies lower risk.However, simple correlations do not control for a number of

onfounding factors that might affect the relationship betweenank risk and diversification. For this reason, we move next to aner-grained analysis, based on a set of multivariate econometricodels.

ar expected default frequency (EDF), which proxies the likelihood to default; (ii) aproximity to default, where CAR is the capital asset ratio, EQT is the equity-to-assets,t , where njt is the number of foreign countries in which bank j has a subsidiary inified bank has subsidiaries in year t. Employees to assets are the logarithm of theet interest income − |other operating income)/(total operating income)|. Tobin’s Qtial shares and minority interests, and the book value of debt to the book value ofmed at the 1st and 99th percentiles. Standard errors are in parentheses.

Z′-score

Median reg. OLS reg. Robust reg. Median reg.(3) (4) (5) (6)

0.543 6.973*** 2.913*** 4.358***

(0.427) (1.048) (0.667) (1.001)−0.136 −0.675 0.108 −0.915*

(0.212) (0.554) (0.328) (0.529)−0.329 −2.436*** −3.139*** −3.919***

(0.679) (0.913) (0.720) (1.141)−0.079 −0.135 −0.062 0.219(0.096) (0.228) (0.138) (0.240)−0.025 −0.835** −0.443* −0.643(0.098) (0.394) (0.262) (0.401)−3.453*** −5.771*** −1.383 −2.156(0.787) (1.737) (0.842) (1.890)

Yes Yes Yes YesYes Yes Yes Yes671 793 793 793

36 M.A. Gulamhussen et al. / Journal of Financial Stability 13 (2014) 30–43

Table 4Alternative measure of internationalization: international share.The dependent variable is one of the following risk measures: (i) the logarithm of the 5-year expected default frequency (EDF), which proxies the likelihood to default; (ii) atransformed Z′-score, which is the symmetric value of (CAR + EQT)/�(ROA) that gauges the proximity to default, where CAR is the capital asset ratio, EQT is the equity-to-assetsratio, and �(ROA) is the volatility of returns. International share is the geographic dispersion of subsidiaries (geographic share), which is estimated by foreign subsidiariesassets/total subsidiaries assets. Employees to assets are the logarithm of the bank’s number of employees divided by its assets. Income diversity is computed as 1 − |(netinterest income − |other operating income)/(total operating income)|. Tobin’s Q is the ratio of the sum of the market value of common stocks, the book value of preferentialshares and minority interests, and the book value of debt to the book value of total assets. All regressions include country and year fixed effects. Risk measures are trimmedat the 1st and 99th percentiles. Standard errors are in parentheses.

EDF Z¢-score

OLS reg. Robust reg. Median reg. OLS reg. Robust reg. Median reg.(1) (2) (3) (4) (5) (6)

International share 0.345* 0.348** 0.309 1.906*** 0.819** 1.245**

(0.182) (0.153) (0.227) (0.524) (0.356) (0.570)Log of assets −0.165 −0.333** −0.294 −0.694 0.122 −0.329

(0.178) (0.136) (0.266) (0.591) (0.318) (0.477)Deposits to liabilities −0.599 −1.177*** −0.835 −4.073*** −4.247*** −4.428***

(0.646) (0.379) (1.108) (1.273) (0.895) (1.360)Employees to assets 0.026 0.087 0.086 −0.082 0.096 0.159

(0.081) (0.059) (0.120) (0.260) (0.137) (0.255)Income diversity −0.113 −0.006 −0.042 −1.125*** 0.114 −0.134

(0.114) (0.117) (0.136) (0.394) (0.290) (0.383)Tobin’s Q −2.658*** −3.715*** −3.300*** −9.324*** −4.659*** −6.910**

(0.692) (0.434) (0.847) (2.577) (1.091) (2.726)

Country effects Yes Yes Yes Yes Yes YesYear effects Yes Yes Yes Yes Yes YesNumber of observations 568 568 568 621 621 621

* Significance at the 10% level.** Significance at the 5% level.

*** Significance at the 1% level.

Table 5Alternative measure of internationalization: international concentration (HHI).The dependent variable is one of the following risk measures: (i) the logarithm of the 5-year expected default frequency (EDF), which proxies the likelihood to default; (ii)a transformed Z′-score, which is the symmetric value of (CAR + EQT)/�(ROA) that gauges the proximity to default, where CAR is the capital asset ratio, EQT is the equity-to-assets ratio, and �(ROA) is the volatility of returns. International concentration is the transformed Hirsch–Herfindhal index (HHI): 1 −

∑j(subsidiaryj assets/total subsidiaries

assets)2. Employees to assets are the logarithm of the bank’s number of employees divided by its assets. Income diversity is computed as 1 − |(net interest income − otheroperating income)/(total operating income)|. Tobin’s Q is the ratio of the sum of the market value of common stocks, the book value of preferential shares and minorityinterests, and the book value of debt to the book value of total assets. All regressions include country and year fixed effects. Risk measures are trimmed at the 1st and 99thpercentiles. Standard errors are in parentheses.

EDF Z′-score

OLS reg. Robust reg. Median reg. OLS reg. Robust reg. Median reg.(1) (2) (3) (4) (5) (6)

International concentration (HHI) 0.636** 0.597*** 0.762** 2.501*** 0.513 0.866(0.273) (0.230) (0.356) (0.682) (0.499) (0.876)

Log of assets −0.124 −0.242* −0.155 −0.746 0.385 −0.225(0.184) (0.146) (0.293) (0.572) (0.324) (0.571)

Deposits to liabilities −0.196 −0.701* −0.603 −3.573**** −3.863*** −3.715**

(0.665) (0.411) (1.056) (1.281) (0.908) (1.550)Employees to assets −0.016 0.040 −0.021 −0.074 0.022 0.150

(0.080) (0.063) (0.127) (0.251) (0.138) (0.281)Income diversity −0.151 −0.035 −0.062 −0.991** −0.063 −0.455

(0.137) (0.135) (0.129) (0.437) (0.318) (0.459)Tobin’s Q −2.298*** −3.830*** −2.993*** −9.518*** −4.001*** −6.525**

(0.771) (0.495) (1.117) (3.020) (1.188) (3.039)

Country effects Yes Yes Yes Yes Yes YesYear effects Yes Yes Yes Yes Yes YesNumber of observations 481 481 481 527 527 527

5

5

(s

99% percentiles.14 When EDF is used as a risk measure, our analy-sis is conducted on 671 bank/year observations; in the case of theZ′-score, they are 793.

* Significance at the 10% level.** Significance at the 5% level.

*** Significance at the 1% level.

. Econometric analysis

.1. Baseline specification

Table 3 reports the baseline results of the estimation of Eq.1), where we include country and year fixed effects in allpecifications, and the measures of risk are trimmed at the 1% and t

14 In unreported regressions, available upon request, we verified that includinghe more extreme values, the results are even starker than those of Table 3.

M.A. Gulamhussen et al. / Journal of Financial Stability 13 (2014) 30–43 37

Table 6Components of the Z′-score.The dependent variable is one of the components of a transformed Z′-score, which is the symmetric value of (CAR + EQT)/�(ROA) that gauges the proximity to default, whereCAR is the capital asset ratio, EQT is the equity-to-assets ratio, and �(ROA) is the volatility of returns. International reach is the ratio of nj,t to nmax,t , where njt is the number offoreign countries in which bank j has a subsidiary in year t, and nmax,t is the maximum number of foreign countries in which the most diversified bank has subsidiaries in yeart. International share is the geographic dispersion of subsidiaries (geographic share), which is estimated by foreign subsidiaries assets/total subsidiaries assets. Internationalconcentration is the transformed Hirsch–Herfindhal index (HHI): 1 −

∑j(subsidiaryj assets/total subsidiaries assets)2. Employees to assets are the logarithm of the bank’s

number of employees divided by its assets. Income diversity is computed as 1 − |(net interest income − |other operating income)/(total operating income)|. Tobin’s Q is theratio of the sum of the market value of common stocks, the book value of preferential shares and minority interests, and the book value of debt to the book value of totalassets. All regressions include country and year fixed effects. Standard errors are in parentheses.

OLS regressions

ROA �(ROA) EQT ROA �(ROA) EQT ROA �(ROA) EQT(1) (2) (3) (4) (5) (6) (7) (8) (9)

Intern. reach (n/nmax) −0.106 0.525*** 0.012(0.150) (0.168) (0.008)

International share −0.222* 0.169*** −0.007(0.126) (0.057) (0.004)

Intern. concent. (HHI) −0.111 0.290*** −0.001(0.159) (0.076) (0.006)

Log of assets −0.144 −0.181** −0.012*** −0.145 −0.208*** 0.033 −0.204** −0.212*** −0.009**

(0.089) (0.077) (0.004) (0.093) (0.073) (0.014) (0.090) (0.075) (0.004)Deposits to liabilities 0.396** −0.442*** 0.005 0.902*** −0.805** −0.001** 0.836** −0.704** 0.030**

(0.185) (0.165) (0.010) (0.344) (0.340) (0.002) (0.333) (0.346) (0.015)Employees to assets 0.077** 0.054** −0.001 0.061 0.076** 0.012 0.075** 0.073** −0.002

(0.036) (0.027) (0.002) (0.037) (0.032) (0.003) (0.037) (0.031) (0.002)Income diversity 0.272*** 0.007 0.012*** 0.203** 0.148*** −0.002*** 0.262*** 0.173*** 0.013***

(0.071) (0.047) (0.003) (0.083) (0.052) (0.017) (0.089) (0.058) (0.003)Tobin’s Q 1.253*** 0.162 0.033*** 1.869*** −0.136 −0.136 1.666*** −0.100 0.005

(0.329) (0.134) (0.012) (0.375) (0.175) (0.175) (0.445) (0.197) (0.020

Country effects Yes Yes Yes Yes Yes Yes Yes Yes YesYear effects Yes Yes Yes Yes Yes Yes Yes Yes YesNumber of observations 793 793 793 621 621 621 527 527 527

* Significance at the 10% level.

mdrmto(m1a

rawsbsteobnf

tiTnnfii

isbbrattTtritb

rmtaTriRotstf

** Significance at the 5% level.*** Significance at the 1% level.

Panels 1–3 of Table 3 report the estimates using EDF as aeasure of bank risk and international reach as the measure of

iversification. Panels 4–6 have the same setup, but present theesults obtained measuring risk with the Z′-score. Our first esti-ates are based on OLS (Panels 1 and 4). However, to account for

he presence of influential observations and for the non-normalityf the dependent variable, we also estimate a robust regressionPanels 2 and 5) and a quantile regression model evaluated at the

edian (Panels 3 and 6), with standard errors bootstrapped with00 replications (Efron and Tibshirani, 1993). In both cases, wedopt the same specification used in the OLS estimates.

In Panels 1–3, the coefficient of international reach is positive, itanges from 0.543 to 0.967, and it is significantly different from zerot least at the 5% level in the case of OLS and Robust regressions,hile it is not statistically significant in the case of median regres-

ions. This result is confirmed and reinforced when measuring risky the Z′-score, a backward-looking and balance-sheet-based mea-ure (Panels 4–6). The coefficient of international reach ranges inhis case from 2.913 to 6.973 and it is always significantly differ-nt from zero at the 1% level. Overall, this is convincing evidencef a positive, statistically and economically significant relationshipetween international reach and bank risk, confirming that inter-ational banks are riskier than their counterparts with nationally

ocused activities.The coefficients of the other control variables provide addi-

ional insights into the determinants of bank risk. Bank size has annsignificant impact on bank risk, with only one exception (Panel 6).his result suggests that although larger banks have more opportu-

ities for risk diversification, and that they can benefit from safetyets which influence their ability to weather adverse financial dif-culties as compared to smaller financial institutions, this does not

mpact on their overall riskiness, once international diversification

ton

s controlled for. The incidence of deposits on total liabilities, a mea-ure of the size and stability of funding, has a negative effect onank risk, that is statistically significant only when risk is measuredy the Z′-score. Finally, we find evidence that banks with a largeratio of employees to total assets or higher income diversificationre less risky. The negative coefficient of Tobin’s Q, that is statis-ically significant at the 1% level in Panels 1–4, suggests insteadhat banks with stronger agency problems, which have a lowerobin’s Q are perceived by the market as more risky Interestingly,he positive relationship between international diversification andisk is confirmed also when we control for organizational complex-ty and agency costs, suggesting that these results are indeed dueo the cross-country dimension of the activities of multinationalanks.

Table 4 has the same structure of Table 3, but it reports theesults of the estimation of Eq. (1) using international share as aeasure of international diversification. The number of observa-

ions is in this case smaller, ranging from 568 when EDF is useds a risk measure, to 621 in the case of the Z′-score. Panels 1–3 ofable 4, reporting the estimates using EDF as a measure of bankisk, show that also in this case the coefficient of international shares positive and it is statistically significant in the case of OLS andobust regressions, while it is not statistically significant in the casef median regressions. When measuring bank risk by the Z′-score,he coefficient of international share is also positive and it is alwaystatistically significant at least at the 5% level. The coefficients ofhe other control variables broadly confirm the picture emergingrom Table 3.

Table 5 reports the results obtained using international concen-ration as a measure of international diversification and showing,nce again, that the coefficient on this additional measure of inter-ational diversification is always positive. It is always statistically

38 M.A. Gulamhussen et al. / Journal of Financial Stability 13 (2014) 30–43

Table 7Non-linear effects – baseline specification: international reach.The dependent variable is one of the following risk measures: (i) the logarithm of the 5-year expected default frequency (EDF), which proxies the likelihood to default; (ii) atransformed Z′-score, which is the symmetric value of (CAR + EQT)/�(ROA) that gauges the proximity to default, where CAR is the capital asset ratio, EQT is the equity-to-assetsratio, and �(ROA) is the volatility of returns. International reach is the ratio of nj,t to nmax,t , where njt is the number of foreign countries in which bank j has a subsidiary in yeart, and nmax,t is the maximum number of foreign countries in which the most diversified bank has subsidiaries in year t. Geon2–Geon6 are dummies for values of internationalreach in each quintile of the strictly positive support of the distribution of international reach (the excluded dummy is for values of international reach that are equal to zero).Employees to assets are the logarithm of the bank’s number of employees divided by its assets. Income diversity is computed as 1 − |(net interest income − |other operatingincome)/(total operating income)|. Tobin’s Q is the ratio of the sum of the market value of common stocks, the book value of preferential shares and minority interests, andthe book value of debt to the book value of total assets. All regressions include country and year fixed effects. Risk measures are trimmed at the 1st and 99th percentiles.Standard errors are in parentheses.

EDF Z′-score

International reach quintiles Interquantile reg. International reach quintiles Interquantile reg.(1) (2) (3) (4)

International reach (n/nmax) −0.962* −4.578**

(0.502) (2.093)International reach – Geon2 0.027 1.406***

(0.117) (0.330)International reach – Geon3 0.076 1.157**

(0.184) (0.527)International reach – Geon4 0.403** 1.410**

(0.171) (0.552)International reach – Geon5 0.553*** 1.375**

(0.173) (0.590)International reach – Geon6 0.980*** 4.166***

(0.228) (0.669)Log of assets −0.137 −0.038* −1.243** −0.638

(0.189) (0.261) (0.618) (0.890)Deposits to liabilities −0.349 −1.268* −2.835*** −1.204

(0.457) (0.729) (0.958) (2.047)Employees to assets −0.176** −0.006 0.014 0.415

(0.080) (0.114) (0.220) (0.440)Income diversity −0.186* 0.118 −0.680* 0.598

(0.111) (0.162) (0.396) (0.693)Tobin’s Q −3.338*** 0.434 −6.157*** 0.475

(0.618) (1.034) (1.778) (3.083)

Country effects Yes Yes Yes YesYear effects Yes Yes Yes YesNumber of observations 662 671 793 793

sesmt

aisftdBorwc

sa

nas

rnfijdudstbmn

* Significance at the 10% level.** Significance at the 5% level.

*** Significance at the 1% level.

ignificantly different from zero at the standard confidence lev-ls when risk is measured by the EDF, while it is not statisticallyignificant when bank risk is measured by the Z′-score and the esti-ates are obtained using either the robust or median regression

echniques.In the case of the Z′-score, it is interesting to understand what

re the components that determine a higher overall riskiness ofnternational banks.15 In Table 6 we therefore report the results ofeparate OLS estimates of the effect of bank internationalizationor each one of the three components of Z′-score: the capital-to-otal-assets ratio, the equity-to-total-assets ratio, and the standardeviation of the return on assets. Consistent with the results oferger et al. (2013) for the U.S., we also find that the higher riskiness

f international banks is mainly due to the higher variability of theireturns on assets (Panels 2, 5, and 8). Contrary to Berger et al. (2013),e do not find any evidence that international banks have a higher

apitalization level.16

15 Since EDF is a market measure based on the value of equity determined by thetock markets, it is not expected to be driven by a specific measurable component,nd it would be rather difficult to give an interpretation to each component.16 Estimates obtained using robust and median regression techniques confirm theegative relationship between internationalization and the variability of return onssets, and provide some evidence that banks with a higher internationalizationhare and concentration are less capitalized. Results are available upon request.

trtcbUcrc

ciit

In unreported regressions, available from the authors uponequest, we have verified that the relationship between bank inter-ational diversification and riskiness is less pronounced in therst part of our sample and much more significant in the years

ust before the crisis. Second, running a regression of internationaliversification on size, the share of deposit funding, the three meas-res of operational complexity and including year and countryummies, we have disentangled the part of international diver-ification that can be explained by these specific bank features andhe residual part, that can be interpreted as a neater measure ofank internationalization. We have then estimated our baselineodel including the predicted value from the actual level of inter-

ational diversification, and the difference between the actual andhe predicted value. Consistent with the findings of the partial cor-elations reported in Tables 3–5, the results confirm that it is nothat part of international diversification that can be explained byomplexity that accounts for the higher riskiness of multinationalanks. In additional unreported regressions, we also excluded the.S. and Japan from the baseline specification, since these twoountries might drive the results. However, our previous findingsemain unchanged after excluding these two potentially influentialountries.

The overall picture emerging from our baseline specification

learly shows that internationally diversified banks with broadernternational footprints are riskier than their peers. In the follow-ng, we present the results of a number of additional regressionshat qualify and test the robustness of our findings.

M.A. Gulamhussen et al. / Journal of Financial Stability 13 (2014) 30–43 39

Table 8Non-linear effects – alternative measure of geographic diversification: international share.The dependent variable is one of the following risk measures: (i) the logarithm of the 5-year expected default frequency (EDF), which proxies the likelihood to default; (ii) atransformed Z′-score, which is the symmetric value of (CAR + EQT)/�(ROA) that gauges the proximity to default, where CAR is the capital asset ratio, EQT is the equity-to-assetsratio, and �(ROA) is the volatility of returns. International share is the geographic dispersion of subsidiaries (geographic share), which is estimated by foreign subsidiariesassets/total subsidiaries assets. Employees to assets are the logarithm of the bank’s number of employees divided by its assets. Income diversity is computed as 1 − |(netinterest income − other operating income)/(total operating income)|. Tobin’s Q is the ratio of the sum of the market value of common stocks, the book value of preferentialshares and minority interests, and the book value of debt to the book value of total assets. All regressions include country and year fixed effects. Risk measures are trimmedat the 1st and 99th percentiles. Standard errors are in parentheses.

EDF Z′-score

International share quintiles Interquantile reg. International share quintiles Interquantile reg.(1) (2) (3) (4)

International share −0.257 −1.451*

(0.290) (0.846)International share – Geon2 −0.004 0.872

(0.180) (0.607)International share – Geon3 0.146 0.599

(0.198) (0.706)International share – Geon4 0.053 1.064

(0.196) (0.648)International share – Geon5 0.386* 1.823***

(0.191) (0.635)International share – Geon6 0.403* 2.106***

(0.237) (0.677)Log of assets −0.166 −0.695** −0.947 −0.495

(0.206) (0.317) (0.636) (0.950)Deposits to liabilities −1.046 −3.679*** −3.650*** −1.348

(0.660) (0.946) (1.335) (2.311)Employees to assets −0.001 0.180 −0.080 0.418

(0.096) (0.152) (0.253) (0.498)Income diversity −0.166 0.286 −1.117*** 1.119

(0.115) (0.195) (0.394) (0.807)Tobin’s Q −3.695*** 0.865 −9.249*** 6.989*

(0.672) (1.263) (2.628) (3.705)

Country effects Yes Yes Yes YesYear effects Yes Yes Yes YesNumber of observations 555 568 621 621

* Significance at the 10% level.** Significance at the 5% level.

5

tittiiotclb

twet3rda5esa

oefiatbahiaussrtir

minternational concentration. The results are broadly consistent withthose for international reach with respect to the increasing effect ofinternationalization on bank risk, while we find somewhat weaker

*** Significance at the 1% level.

.2. Non-linear effects

Gulamhussen et al. (2010) show that international diversifica-ion has an inverse U-shaped effect on bank value. In other words,f a bank is operating in a small number of countries, then interna-ional diversification causes the bank’s value to increase, but oncehe bank reaches a given threshold, further expansion abroad hasnstead a negative effect. In that paper, we argued that this find-ng might be due to the high costs and the excessive complexityf managing large multinational banks. One possible reason is thathe management of very large multinational banks might entailonsiderable risks. If this were the case, we should find that veryarge banks have a higher degree of riskiness than what is expectedased on the linear relationship estimated above.

To test this hypothesis, we re-estimate the baseline specifica-ion substituting our continuous measure of international reachith 6 dummies: one for the values of international reach that are

qual to zero and the other 5 for each quintile of the strictly posi-ive support of the distribution of international reach. Panels 1 and

of Table 7 report the results for our two different measures ofisk. The key finding is that the relationship between internationaliversification and bank risk increases with the level of internation-lization, with particularly strong effects for levels at the 4th and

th quintiles. This is consistent with the hypothesis that high lev-ls of international diversification further augment riskiness. In allpecifications, the coefficients of the other bank-specific controlsre in line with the baseline findings. c

An additional nonlinearity could be instead related to the levelf bank risk. For example, less risky banks might be willing to adoptxpansionary strategies precisely to increase their risk-return pro-le, while banks that are already relatively risky might wish toddress the agency problems posed by international diversifica-ion more carefully. This would lead to a negative relationshipetween the effect of international diversification on bank risknd the bank’s initial level of riskiness. To verify this additionalypothesis, we have estimated an interquantile regression, test-

ng whether the effect of international diversification is differentt the 25th and 75th percentiles of the distribution of our meas-res of risk. The results, presented in Panels 2 and 4 of Table 7,how that the coefficient of international reach is negative andtatistically significant using both measures of banks riskiness,espectively at the 10% level for the EDF and at the 5% level forhe Z′-score. This provides some evidence that less risky banks usenternational diversification strategies precisely to increase theiriskiness.17

Tables 8 and 9 report the results obtained using our two othereasures of international diversification: international share and

17 In unreported regressions, available upon request, we have found similar resultsomparing the effect at the 20th and 80th percentiles.

40 M.A. Gulamhussen et al. / Journal of Financial Stability 13 (2014) 30–43

Table 9Non-linear effects – alternative measure of geographic diversification: international concentration.The dependent variable is one of the following risk measures: (i) the logarithm of the 5-year expected default frequency (EDF), which proxies the likelihood to default; (ii)a transformed Z′-score, which is the symmetric value of (CAR + EQT)/�(ROA) that gauges the proximity to default, where CAR is the capital asset ratio, EQT is the equity-to-assets ratio, and �(ROA) is the volatility of returns. International concentration is the transformed Hirsch–Herfindhal index (HHI) (1 −

∑j(subsidiaryj assets/total subsidiaries

assets)2. Employees to assets are the logarithm of the bank’s number of employees divided by its assets. Income diversity is computed as 1 − |(net interest income − otheroperating income)/(total operating income)|. Tobin’s Q is the ratio of the sum of the market value of common stocks, the book value of preferential shares and minorityinterests, and the book value of debt to the book value of total assets. All regressions include country and year fixed effects. Risk measures are trimmed at the 1st and 99thpercentiles. Standard errors are in parentheses.

EDF Z′-score

International concentrationquintiles

Interquantile reg. International concentrationquintiles

Interquantile reg.

(1) (2) (3) (4)

International concentration (HHI) −0.653 −1.657*

(0.563) (1.003)International concentration (HHI) – Geon2 0.117 0.824

(0.179) (0.639)International concentration (HHI) – Geon3 0.340 1.135

(0.202) (0.695)International concentration (HHI) – Geon4 0.029 1.061

(0.201) (0.653)International concentration (HHI) – Geon5 0.403* 1.717**

(0.217) (0.710)International concentration (HHI) – Geon6 0.576*** 2.175***

(0.210) (0.638)Log of assets −0.104 −0.380 −0.983 −0.249

(0.218) (0.315) (0.628) (0.999)Deposits to liabilities −0.598 −3.315*** −3.066** 0.098

(0.676) (1.109) (1.378) (2.280)Employees to assets −0.059 0.095 −0.053 0.188

(0.096) (0.138) (0.246) (0.455)Income diversity −0.246* 0.359* −1.030** 0.339

(0.139) (0.205) (0.440) (0.951)Tobin’s Q −3.245*** 1.544 −9.550*** 7.414

(0.792) (1.341) (3.113) (5.024)

Country effects Yes Yes Yes YesYear effects Yes Yes Yes YesNumber of observations 471 481 526 527

*

eirP

dipTafroc

aeatptinmcoo

tb

5

seofidtIaen

nestimation techniques. First, we use a two stage approach, instru-menting the measure of degree of international diversificationwith: (i) its lagged value at time t − 1 to t − 2;19 (ii) the share of

Significance at the 10% level.** Significance at the 5% level.

*** Significance at the 1% level.

vidence that less risky banks increase their international share andnternational concentration with the purpose of augmenting theirisk levels when they are measured using EDF (Tables 8 and 9,anels 1).

Finally, we have verified the hypothesis that when internationaliversification is associated with a higher degree of bank complex-

ty, this might have an even stronger effect on bank risk. To thisurpose, we have split the sample between banks with values ofobin’s Q and of the number of employees over total assets abovend below the median, and we have estimated the model of Eq. (1)or each sub-sample. The results, available from the authors uponequest, do not provide convincing evidence of a non-linear effectf internationalization on bank risk depending on the degree ofomplexity.

Finally, to test the hypothesis that organizational complexitynd agency problems are the contributing factors of increased riskffect of international diversification, in unreported regressionsvailable upon request, we have introduced in our specifica-ion two terms obtained interacting international reach with ourroxies for organizational complexity and agency costs, respec-ively. If these were the contributing factor of the higher risk ofnternational banks, we should have obtained a positive and sig-ificant coefficient of these terms. Moreover, to unveil possible

ulticollinearity problems, we have orthogonalized organizational

omplexity measures and Tobin’s Q. However, in every instanceur previous results remained broadly unchanged, confirmingur view that our results are indeed due to something specific

st

o the cross-country dimension of the activities of multinationalanks.18

.3. Endogeneity

In the previous sections we have been rather careful in pre-enting our results as partial correlations, avoiding to stress thexistence of a causal relationship of international diversificationn bank risk. Indeed, many determinants of international diversi-cation are the same that underpin bank risk. Finding that moreiversified banks are exposed to higher levels of risk does notherefore constitute sufficient proof per se of a causality effect.nternational diversification itself may be an endogenous choice,nd commercial banks that decide to pursue riskier business mod-ls may decide to do so also by diversifying their activities acrossational borders.

To uncover the existence of a causal relationship from inter-ational diversification to riskiness, we employ two different

18 We thank an anonymous referee for suggesting this check.19 In unreported regressions, available upon request, we have verified that theame results hold also excluding the lagged value at time t − 1 and including insteadhose at time t − 2 and t − 3, to reduce the potential issues of time dependence.

M.A. Gulamhussen et al. / Journal of Financial Stability 13 (2014) 30–43 41

Table 10Controlling for endogeneity – baseline specification: international reach.The dependent variable is one of the following risk measures: (i) the logarithm of the 5-year expected default frequency (EDF), which proxies the likelihood to default; (ii) atransformed Z′-score, which is the symmetric value of (CAR + EQT)/�(ROA) that gauges the proximity to default, where CAR is the capital asset ratio, EQT is the equity-to-assetsratio, and �(ROA) is the volatility of returns. International reach is the ratio of nj,t to nmax,t , where njt is the number of foreign countries in which bank j has a subsidiary inyear t, and nmax,t is the maximum number of foreign countries in which the most diversified bank has subsidiaries in year t. Employees to assets are the logarithm of thebank’s number of employees divided by its assets. Income diversity is computed as 1 − |(net interest income − |other operating income)/(total operating income)|. Tobin’s Qis the ratio of the sum of the market value of common stocks, the book value of preferential shares and minority interests, and the book value of debt to the book value oftotal assets. Excluded instruments are: (i) the lagged value of international reach at time t − 1 and t − 2; (ii) the share of internationally diversified banks in the country, asindirect evidence of an environment that favors internationalization; (iii) a dummy variable for banks that are included in the Standard and Poor’s 500 on the grounds thatthey might be better equipped to finance international diversification. All regressions include country and year fixed effects. Risk measures are trimmed at the 1st and 99thpercentiles. Standard errors are in parentheses. Significance at the 1%, 5%, and 10% levels is denoted by ***, **, and *, respectively. In the case of the Kleibergen-Paap Wald rkF-statistic, ***, **, and *, indicate rejection of the null hypothesis that the endogenous regressor is weakly identified, respectively, at the 10%, 15% and 20% maximal IV relativebias according to the Stock and Yogo (2005) weak ID test critical values for single endogenous regressor.

EDF Z′-score

IV Dynamic GMM IV Dynamic GMM(1) (2) (3) (4)

International reach (n/nmax) 1.271*** −0.941 5.141*** 1.606**(0.440) (0.783) (2.004) (0.671)

Lagged dependent variable 0.928*** 0.920***(0.236) (0.025)

Log of assets 0.081 −0.083 −1.053 −0.144*(0.232) (0.084) (0.800) (0.077)

Deposits to liabilities 0.013 −1.560 −1.241 1.630(0.516) (1.821) (1.269) (1.580)

Employees to assets −0.200* −0.034* 0.031 −0.007(0.103) (0.018) (0.358) (0.017)

Income diversity −0.208 −0.053 −0.742 −0.117(0.145) (0.141) (0.567) (0.185)

Tobin’s Q −2.791*** −1.286 −8.608** −0.158(0.765) (0.871) (3.928) (0.612)

Kleibergen-Paap Wald rk F statistic 631.80*** 626.08***Hansen test for over-identification restrictions 1.25 6.24 2.81 13.45

Country effects Yes Yes Yes YesYear effects Yes Yes Yes YesNumber of observations 381 517 374 608

Table 11Controlling for endogeneity – alternative measure of geographic diversification: international share.The dependent variable is one of the following risk measures: (i) the logarithm of the 5-year expected default frequency (EDF), which proxies the likelihood to default; (ii) atransformed Z′-score, which is the symmetric value of (CAR + EQT)/�(ROA) that gauges the proximity to default, where CAR is the capital asset ratio, EQT is the equity-to-assetsratio, and �(ROA) is the volatility of returns. International share is the geographic dispersion of subsidiaries (geographic share), which is estimated by foreign subsidiariesassets/total subsidiaries assets. Employees to assets are the logarithm of the bank’s number of employees divided by its assets. Income diversity is computed as 1 − |(netinterest income − |other operating income)/(total operating income)|. Tobin’s Q is the ratio of the sum of the market value of common stocks, the book value of preferentialshares and minority interests, and the book value of debt to the book value of total assets. The excluded instruments are the lagged value of international reach at time t − 1.All regressions include country and year fixed effects. Risk measures are trimmed at the 1st and 99th percentiles. Standard errors are in parentheses. Significance at the 1%,5%, and 10% levels is denoted by ***, **, and *, respectively. In the case of the Kleibergen-Paap Wald rk F-statistic, ***, **, and *, indicate rejection of the null hypothesis thatthe endogenous regressor is weakly identified, respectively, at the 10%, 15% and 20% maximal IV relative bias according to the Stock and Yogo (2005) weak ID test criticalvalues for single endogenous regressor.

EDF Z′-score

IV Dynamic GMM IV Dynamic GMM(1) (2) (3) (4)

International share 1.001* −0.877 6.388** 2.629**(0.602) (0.557) (2.607) (1.213)

Lagged dependent variable 0.800*** 0.962***(0.105) (0.082)

Log of assets −0.404* 0.002 −2.488** 0.235(0.244) (0.116) (1.164) (0.278)

Deposits to liabilities 0.270 −1.289 −1.071 2.399(0.806) (1.374) (2.964) (2.258)

Employees to assets −0.094 −0.016 0.377 −0.037(0.085) (0.021) (0.336) (0.037)

Income diversity −0.056 −0.191* −0.904 −0.089(0.137) (0.115) (0.567) (0.253)

Tobin’s Q −2.604*** −1.630*** −7.803* −0.568(0.916) (0.500) (4.200) (1.271)

Kleibergen-Paap Wald rk F statistic 13.58** 13.34**Hansen test for over-identification restrictions 9.80 6.72

Country effects Yes Yes Yes YesYear effects Yes Yes Yes YesNumber of observations 387 381 382 413

42 M.A. Gulamhussen et al. / Journal of Financial Stability 13 (2014) 30–43

Table 12Controlling for endogeneity – alternative measure of geographic diversification: international concentration.The dependent variable is one of the following risk measures: (i) the logarithm of the 5-year expected default frequency (EDF), which proxies the likelihood to default; (ii)a transformed Z′-score, which is the symmetric value of (CAR + EQT)/�(ROA) that gauges the proximity to default, where CAR is the capital asset ratio, EQT is the equity-to-assets ratio, and �(ROA) is the volatility of returns. International concentration is the transformed Hirsch–Herfindhal index (HHI): 1 −

∑j(subsidiaryj assets/total subsidiaries

assets)2. Employees to assets are the logarithm of the bank’s number of employees divided by its assets. Income diversity is computed as 1 − |(net interest income − |otheroperating income)/(total operating income)|. Tobin’s Q is the ratio of the sum of the market value of common stocks, the book value of preferential shares and minorityinterests, and the book value of debt to the book value of total assets. Excluded instruments are: (i) the lagged value of international reach at time t − 2 and t − 3; (ii) theshare of internationally diversified banks in the country, as indirect evidence of an environment that favors internationalization; (iii) a dummy variable for banks that areincluded in the Standard and Poor’s 500 on the grounds that they might be better equipped to finance international diversification. All regressions include country and yearfixed effects. Risk measures are trimmed at the 1st and 99th percentiles. Standard errors are in parentheses. Significance at the 1%, 5%, and 10% levels is denoted by ***, **,and *, respectively. In the case of the Kleibergen-Paap Wald rk F-statistic, ***, **, and *, indicate rejection of the null hypothesis that the endogenous regressor is weaklyidentified, respectively, at the 5%, 10% and 20% maximal IV relative bias according to the Stock and Yogo (2005) weak ID test critical values for single endogenous regressor.

EDF Z′-score

IV Dynamic GMM IV Dynamic GMM(1) (2) (3) (4)

International concentration (HHI) 0.892 −0.626 6.622** 2.590***(0.615) (0.419) (2.711) (0.714)

Lagged dependent variable 1.137 0.932***(0.139) (0.029)

Log of assets −0.094 −0.172 −3.318*** −0.375**(0.258) (0.118) (1.131) (0.146)

Deposits to liabilities (0.962 −2.456* 1.546 1.747*(0.840) (1.325) (2.958) (0.922)

Employees to assets −0.074 −0.005 0.554* −0.076**(0.097) (0.031) (0.337) (0.033)

Income diversity −0.113 0.257 −0.195 0.702***(0.206) (0.195) (0.792) (0.244)

Tobin’s Q −5.197*** −0.706 −3.237 −0.070(1.717) (0.563) (5.814) (0.244)

Kleibergen-Paap Wald rk F statistic 18.65*** 18.03***Hansen test for over-identification restrictions 6.23 12.28 4.93 16.53

iddPfiPiaHtrts

lt(BTaintt

dc

D

e

6

taniip

rfmsdvni

fiwciti

Country effects Yes

Year effects Yes

Number of observations 184

nternationally diversified banks in the country, as indirect evi-ence of an environment that favors internationalization; (iii) aummy variable for banks that are included in the Standard andoor’s 500 on the grounds that they might be better equipped tonance international diversification. In all cases, the Kleibergen-aap Wald rk F-statistic rejects the null hypothesis of weakdentification at the standard confidence levels proposed by Stocknd Yogo (2005), as suggested by Baum et al. (2007), and theansen J test for over-identifying restrictions also fails to reject

he null hypothesis at the standard confidence levels. The resultseported in Panels 1 and 3 of Table 10 show that also in this casehe coefficients of international reach are positive and statisticallyignificant at the 1% level.

Second, we augment the baseline specification including theagged dependent variable, since our measure of bank value is likelyo be time-persistent. We use the generalized method of momentsGMM) developed for dynamic panel data model (Arellano andond, 1991; Arellano and Bover, 1995).20 Panels 2 and 4 ofable 10 show that indeed the lagged dependent variable is positivend highly statistically significant. Reassuringly, the coefficient ofnternational reach is also in this case positive, and statistically sig-ificant although only in the case of the Z′-score. Also in this case,he Hansen J statistic for over-identifying restrictions fails to rejecthe null hypothesis at the standard confidence levels.

Tables 11 and 12 confirm the same results when international

iversification is measured by international share and internationaloncentration.21

20 Estimation is conducted using the XTABOND2 program for Stata written byavid Roodman (2006).

21 In the case of international reach, only its lagged value at time t-1 was used as anxcluded instrument, since with the inclusion of the whole set of instruments used

iio

ft

Yes Yes YesYes Yes Yes231 180 243

. Conclusions

The design of a new regulatory framework capable of addressinghe many fallacies uncovered by the recent financial crisis requires

precise understanding of the characteristics of the different busi-ess models followed by banks and of their riskiness. One of the

ssues that has captured much attention from regulators and pol-cymakers is the international dimension of the financial markets,articularly the role of multinational players.

In this paper we show that multinational banks are indeediskier. The higher value entailed by international diversificationound by Gulamhussen et al. (2010) comes therefore at a cost:

ultinational banks, especially the largest players that have sub-idiaries all over the world, have a higher expected probability ofefault, as measured by EDFs, lower Z-scores, and higher returns’ariability. Several robustness checks indicate that our findings areot driven by few influential observations, or biased by endogene-

ty problems.As there are no grounds to exclude the benefits from the diversi-

cation of the idiosyncratic shocks to the asset and loan portfolios,e can infer that higher riskiness is due to the business model

hosen by multinational banks. In particular, it is most likely thatncentive problems lie at the root of this higher riskiness. Multina-ional banks are not riskier per se, but they can take on more riskf the management decides to do so. A regulatory framework that

ncreases the costs of holding cross-border activities and partic-pations might have a negative adverse-selection effect such thatnly those who are ready to assume high levels of risk will diversify

or the other two measures the Kleibergen-Paap Wald rk F-statistic failed to rejecthe null hypothesis of weak identification.

of Fin

ibtl

R

A

A

A

A

A

A

A

A

B

B

B

B

B

B

B

B

B

B

B

B

B

CC

CD

D

D

D

D

D

D

D

D

D

E

F

F

F

G

G

G

H

H

J

J

J

K

K

K

K

L

L

L

L

L

L

L

M

M

M

MO

P

R

S

S

S

S

S

V

M.A. Gulamhussen et al. / Journal

nternationally. As recently argued by Diamond and Rajan (2009), aetter approach would be to directly adjust the mechanism behindhe incentives that lead multinational banks to take on excessiveevels of risk.

eferences

charya, V.V., Hasan, I., Saunders, A., 2006. Should banks be diversified? Evidencefrom individual bank loan portfolios. J. Bus. 79 (3), 1355–1412.

ckermann, J., 2008. The subprime crisis and its consequences. J. Finan. Stab. 4 (4),329–337.

llen, F., Carletti, E., 2013. New theories to underpin financial reform. J. Finan. Stab.9 (2), 242–249.

mihud, Y., DeLong, G., Saunders, A., 2002. The effects of cross-border bank mergerson bank risk and value. J. Int. Money Finance 21 (6), 857–877.

ltunbas, Y., Gambacorta, L., Marques-Ibanez, D., 2010. Bank risk and monetarypolicy. J. Finan. Stab. 6 (3), 121–129.

rellano, M., Bond, S.R., 1991. Some tests of specification for panel data: Monte Carloevidence and an application to employment equations. Rev. Econ. Stud. 58 (2),277–297.

rellano, M., Bover, O., 1995. Another look at the instrumental variable estimationof error-components models. J. Econom. 68 (1), 29–51.

vgouleas, E., Goodhart, C., Schoenmaker, D., 2011. Bank Resolution Plans as a cat-alyst for global financial reform. J. Finan. Stab. 9 (2), 210–218.

aele, L., De Jonghe, O., Vander Vennet, R., 2007. Does the stock market value bankdiversification? J. Bank Finance 31 (7), 1999–2023.

arba Navaretti, G., Calzolari, G., Pozzolo, A.F., Levi, M., 2010. Multinational bank-ing in Europe: financial stability and regulatory implications lessons from thefinancial crisis. Econ. Pol. 25 (64), 703–753.

attiston, S., Delli Gatti, D., Gallegati, M., Greenwald, B., Stiglitz, J.E., 2012. Defaultcascades: when does risk diversification increase stability? J. Finan. Stab. 8 (3),138–149.

aum, C.F., Schaffer, M.E., Stillman, S., 2007. Enhanced routines for instrumentalvariables/GMM estimation and testing. Stata J. 7, 465–506.

CBS, Basel Committee on Banking Supervision, 2011. Global systemically importantbanks: assessment methodology and the additional loss absorbency require-ment.

ecker-Blease, J.R., Kaen, F.R., Etebari, A., Baumann, H., 2010. Employees, firm sizeand profitability in U.S. manufacturing industries. Investment Manage. Finan.Innovat. 7 (2), 7–23.

erger, A.N., De Young, R., 1997. Problem loans and cost efficiency in commercialbanking. J. Bank Finance 21 (6), 849–870.

erger, A.N., Ghoul, E.S., Guedhami, O., Roman, R.A., 2013. Bank internationalizationand risk-taking., http://dx.doi.org/10.2139/ssrn.2249048, Available on line atSSRN: http://ssrn.com/abstract=2249048

lack, F., Scholes, M., 1973. The pricing of options and corporate liabilities. J. Polit.Econ. 81 (4), 637–659.

oyd, J.H., De Nicolò, G., Jalal, A.M., 2006. Bank risk-taking and competition revisited:New theory and new evidence. In: IMF Working Paper no. 06/297.

uch, C.M., Lipponer, M., 2007. FDI versus exports: evidence from German banks. J.Bank Finance 31 (3), 805–826.

uch, C.M., DeLong, G.L., 2009. Banking globalization: international consolidationand mergers in banking. In: Berger, A.N., Molyneux, P., Wilson, J.O. (Eds.), TheOxford Book of Banking. University Press, Oxford.

uch, C.M., Driscoll, J.C., Ostergaard, C., 2010. Cross-Border diversification in Bankasset portfolios. Int. Finance 13 (1), 79–108.

ameron, A.C., Trivedi, P.K., 2009. Microeconometrics using Stata. Stata Press, USA.hiaramonte, L., Casu, B., 2010. Are CDS spreads a good proxy of bank risk? Evidence

from the financial crisis. In: Cass Business School Working paper no. 05/10.ochrane, J., 2001. Asset Pricing. Princeton University Press, Princeton, NJ.e Haan, J., Poghosyan, T., 2011. Bank size, market concentration, and bank earnings

volatility in the U.S. In: DNB Working paper no. 282.e Nicolò, G.D., Bartholomew, P., Zaman, J., Zephirin, M., 2004. Bank consolidation,

internationalization, and conglomeration: trends and implications for financialrisk. Finan. Markets Inst. Instrum. 13 (4), 173–217.

emirgüc -Kunt, A., Huizinga, H., 2010. Bank activity and funding strategies: theimpact on risk and returns. J. Finan. Econ. 98 (3), 626–650.

emirguüc -Kunt, A., Detragiache, E., 2011. Basel core principles and bank sound-ness: dose compliance matter? J. Finan. Stab. 7 (4), 179–190.

emsetz, R., Saidenberg, M., Strahan, P., 1996. Banks with something to lose: thedisciplinary role of franchise value. Econ. Policy Rev. 2 (2), 1–14.

emsetz, R., Saidenberg, M., Strahan, P., 1997. Agency problems and risk taking atbanks. In: FRB of New York Staff Report 29.

eng, S., Elyasiani, E., Mao, C., 2007. Diversification and the cost of debt of bankholding companies. J. Bank Finance 31 (12), 2453–2473.

eng, S., Elyasiani, E., 2008. Geographic diversification, bank holding company value,

and risk. J. Money Credit. Bank. 40 (6), 1217–1238.

etragiache, E., Tressel, T., Gupta, P., 2008. Foreign banks in poor countries: theoryand evidence. J. Fin. 63 (5), 2123–2160.

iamond, W.D., Rajan, R., 2009. The credit crisis: Conjectures about causes andremedies. In: NBER Working paper no. 14739.

W

Z

ancial Stability 13 (2014) 30–43 43

fron, B., Tibshirani, R.J., 1993. An Introduction to the Bootstrap. Chapman and Hall,New York.

iordelisi, F., Marques-Ibanez, D., Molyneux, P., 2010. Efficiency and risk in Europeanbanking. In: ECB Working paper no. 1211.

ocarelli, D., Pozzolo, A.F., 2005. Where do banks expand abroad? An empiricalanalysis. J. Bus. 78 (6), 2435–2463.

ocarelli, D., Pozzolo, A.F., Salleo, C., 2008. Do M&As in the financial industry modifysystematic risk? In: paper presented at the Conference on Performance Mea-surement in the Financial Services Sector: Frontier Efficiency Methodologiesand Other Innovative Techniques, London, July 4–5.

arlappi, L., Shu, T., Yan, H., 2006. Default risk, shareholder advantage, and stockreturns. Rev. Finan. Stud. 21 (6), 2743–2778.

orton, G., Metrick, A., 2012. Getting up to speed on the financial crisis: a one-weekend-reader’s guide. J. Econ. Lit. 50 (1), 128–150.

ulamhussen, M.A., Pinheiro, C., Pozzolo, A.F., 2010. Do multinational banks createor destroy economic value? In: MoFiR Working paper no. 36.

uang, X., Zhou, H., Zhu, H., 2011. Assessing the systemic risk of a heterogeneousportfolio of banks during the recent financial crisis. J. Finan. Stab. 8 (3), 193–205.

ughes, J.P., Lang, W., Mester, L.J., Moon, C., 1996. Efficient banking under interstatebranching. J. Money Credit. Bank. 28 (4), 1045–1071.

ensen, M.C., 1986. Agency costs of free cash flow, corporate finance, and takeovers.Amer. Econ. Rev. 76 (2), 323–329 (papers and proceedings).

ensen, M.C., Meckling, M.C., 1976. Theory of the firm: Managerial behavior, agencycosts, and ownership structure. J. Finan. Econ. 3 (4), 305–360.

ohn, K., Litov, L., Yeung, B., 2008. Corporate governance and managerial risk taking:theory and evidence. J. Finance 63 (4), 1679–1728.

aen, F.R., Baumann, H.D., 2003. Firm size, employees and profitability in U.S. man-ufacturing industries., http://dx.doi.org/10.2139/ssrn.382402, Available on lineat SSRN: http://ssrn.com/abstract=382402

arolyi, G.A., Stulz, L.M., 2002. Are financial assets priced locally or globally? In:Constantinides, G., Harris, M., Stulz, R.M. (Eds.), Handbook of the Economics ofFinance. Elsevier/North-Holland, Amsterdam, ISBN 978-0-444-59406-8.

ealhofer, S., 2003. Quantifying credit risk II: debt valuation. Finan. Analysts J. 59(3), 30–44.

lein, P.G., Saidenberg, M.R., 2010. Organizational structure and the diversificationdiscount: evidence from commercial banking. J. Ind. Econ. 58 (1), 127–155.

aeven, L., Levine, R., 2007. Is there a diversification discount in financial conglom-erates? J. Finan. Econ. 85 (2), 331–367.

aeven, L., Levine, R., 2008. Complex ownership structures and corporate valuations.Rev. Finan. Stud. 21 (2), 579–604.

aeven, L., Levine, R., 2009. Bank governance, regulation and risk taking. J. Finan.Econ. 93 (2), 259–275.

ewellen, W.G., 1971. A pure financial rationale for the conglomerate merger. J.Finance 26 (2), 521–537.

indenberg, E.B., Ross, S.A., 1981. Tobin’s q ratio and industrial organization. J. Bus.54 (1), 1–32.

i, G., 1985. Robust regression. In: Shapes, D., Hoaglin, C., Mosteller, F., Tukey, J.W.(Eds.), Exploring Data Tables, Trends. Wiley, New York.

o, A.W., 2012. Reading about the financial crisis: a twenty-one-book review. J. Econ.Lit. 50 (1), 151–178.

arkowitz, H.M., 1959. Portfolio selection: efficient diversification of investments.John Wiley & Sons, New York.

ercieca, S., Schaek, J., Wolfe, S., 2007. Small European banks: Benefits from diver-sification? J. Bank Finance 31 (7), 1975–1998.

erton, R.C., 1974. On the pricing of corporate debt: the risk structure of interestrates. J. Finance 29 (2), 449–470.

yers, S.C., Rajan, R., 1998. The paradox of liquidity. Q. J. Econ. 113 (3), 733–771.ngena, S., Popov, A., Udell, G., 2013. When the cat’s away the mice will play:

does regulation at home affect bank risk taking abroad? J. Finan. Econ. 108 (3),727–750.

ozzolo, A.F., 2009. Bank cross-border mergers and acquisitions (causes, conse-quences and recent trends). In: Alessandrini, P., Fratianni, M., Zazzaro, A. (Eds.),The Changing Geography of Banking and Finance. Springer, Norwell.

oodman, D.M., 2006. How to do xtabond2: an introduction to difference and systemGMM in Stata, Center for Global Development. In: Working paper no. 103.

aldias, M., 2013. Systemic risk analysis using forward-looking distance-to-defaultseries. J. Finan. Stab. 9 (4), 498–517.

tiroh, K.J., 2009. Diversification in banking. In: Berger, A.N., Molyneux, P., Wilson,J.O. (Eds.), The Oxford Book of Banking. Oxford University Press, Oxford.

tiroh, K.J., Rumble, A., 2006. The dark side of diversification: the case of U.S. financialholding companies. J. Bank Finance 30 (8), 2131–2161.

tock, J.H., Yogo, M., 2005. Testing for weak instruments in linear IV regression. In:Andrews, D.W.K., Stock, J.H. (Eds.), Identification and inference for econometricmodels: essays in honor of Thomas Rothenberg. Cambridge University Press,Cambridge, pp. 80–108.

weeney, R.J., Warga, A.D., Winters, D., 2001. The market value of debt, market versusbook value of debt, and returns to assets. Finan. Manage. 26 (1), 5–21.

ander Vennet, R., De Jonghe, O., Baele, L., 2004. Bank risks and the business cycle.

In: University of Ghent Working Paper no. 04/264.

illiams, J., 2004. Determining management behaviour in European banking. J. BankFinance 28 (10), 2427–2460.

hang, H., 1995. Wealth effects of U.S. bank takeovers. Appl. Finan. Econ. 5 (5),329–336.