has the global banking system become more fragile over time?

J

H

Da

b

a

ARRAA

JFGGG

KBSSBDM

1

gtotpbhaoivo

a

a

h1

ARTICLE IN PRESSG ModelFS-278; No. of Pages 12

Journal of Financial Stability xxx (2014) xxx–xxx

Contents lists available at ScienceDirect

Journal of Financial Stability

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

as the global banking system become more fragile over time?

eniz Anginera,∗, Asli Demirguc-Kuntb,1

Pamplin College of Business, Virginia Tech, 7054 Haycock Road, Suite 352, Falls Church, VA 22043, USAThe World Bank, 1818 H Street, NW, Washington, DC 20433, USA

r t i c l e i n f o

rticle history:eceived 12 June 2012eceived in revised form 18 January 2013ccepted 18 February 2014vailable online xxx

EL classification:36111215

a b s t r a c t

We examine the evolution of credit risk co-dependence in the banking sectors of over 65 countries. Wefind that there has been a significant increase in default risk co-dependence over the 3-year period leadingup to the financial crisis. We also find that countries that are more integrated with liberalized financialsystems have seen a greater rise in co-dependence in their banking sectors. This negative effect of financialopenness is mitigated by a strong institutional environment that allows for efficient public and privatemonitoring of financial institutions.

© 2014 Elsevier B.V. All rights reserved.

eywords:anking crisesystemic riskystemic fragilityank default risk

oodscficbacttwB

istance-to-defaulterton model

. Introduction

The last decade has seen a tremendous transformation in thelobal financial sector. Globalization, innovations in communica-ions technology and de-regulation have lead to significant growthf financial institutions around the world (Rajan, 2006). Theserends had positive economic benefits and have led to increasedroductivity, increased capital flows, lower borrowing costs, andetter price discovery and risk diversification. But the same trendsave also led to greater linkages across financial institutions as wells an increase in exposure of these institutions to common sourcesf risk. The recent financial crisis has demonstrated that financialnstitutions around the world are highly inter-connected and thatulnerabilities in one market can easily spread to other markets

Please cite this article in press as: Anginer, D., Demirguc-Kunt, A., HaFinancial Stability (2014), http://dx.doi.org/10.1016/j.jfs.2014.02.003

utside of national boundaries.In this paper, we examine whether the global trends described

bove have led to an increase in co-dependence in default risk

∗ Corresponding author. Tel.: +1 917 400 2951.E-mail addresses: [email protected] (D. Anginer),

[email protected] (A. Demirguc-Kunt).1 Tel.: +1 202 473 7479.

iaAlsterc

ttp://dx.doi.org/10.1016/j.jfs.2014.02.003572-3089/© 2014 Elsevier B.V. All rights reserved.

f commercial banks around the world. The growing expansionf financial institutions beyond national boundaries over the pastecade has resulted in these institutions competing in increasinglyimilar markets, exposing them to common sources of market andredit risk. During the same period, rapid development of newnancial instruments such as such as credit default swaps andollateralized debt obligations, has increased the complexity ofank balance sheets and created new channels of inter-dependencycross these institutions. Both increased interconnections andommon exposure to risk make the banking sector more vulnerableo economic, liquidity and information shocks. There is a substan-ial theoretical literature that models the various channels throughhich such shocks can culminate in a systemic banking crisis.attiston et al. (2012) and Gai and Kapadia (2010) show that an

ncreasing network of credit relationships can amplify the effect ofn initial economic shock leading to a full-fledged systemic crisis.llen and Gale (2000) analyze financial contagion resulting from

inkages across financial intermediaries. In their model, the pos-ibility of contagion from a liquidity shock depends strongly on

s the global banking system become more fragile over time? J.

he completeness of the structure of inter-regional claims. Freixast al. (2000) show that financial structures in which all banks haveelations with a center bank, but not with each other, are also sus-eptible to contagion. Diamond and Rajan (2005), Cifuentes et al.

dx.doi.org/10.1016/j.jfs.2014.02.003

dx.doi.org/10.1016/j.jfs.2014.02.003

http://www.sciencedirect.com/science/journal/15723089

http://www.elsevier.com/locate/jfstabil

mailto:[email protected]

mailto:[email protected]

dx.doi.org/10.1016/j.jfs.2014.02.003

ARTICLE IN PRESSG ModelJFS-278; No. of Pages 12

2 al of Fi

(taicsoos

id(dett1

uWgsitata(atrl2dA

diccseefTfidmi

hmmtrt2bi(eato

fifibbfiHasiiBo2e

owsdedp

t(r

2

pwaWildopowtfihiedpaied

cf

d1 = log(VA/X) + (r − d + (S2A/2))T

SA

√T

; d2 = d1 − SA

√T

(1)

D. Anginer, A. Demirguc-Kunt / Journ

2004) and Brunnermeier and Pedersen (2009) analyze contagionhrough fire sales of illiquid assets. In these models, insolvencynd illiquidity become mutually reinforcing leading to contagiouslliquidity spirals. Information asymmetry also provides a potentialhannel in which shocks can be propagated through the bankingystem. In the presence of asymmetric information, difficulties inne bank may be perceived as a signal of possible difficulties inther banks, especially if banks’ assets are opaque and balanceheet information is uninformative (Morgan, 2002; Yuan, 2005).

To examine whether the global banking sector has become morenterdependent and more susceptible to shocks, we construct aefault risk measure for all publicly traded banks using the Merton1974) contingent claim model. We compute weekly time series ofefault probabilities for over 1900 banks in over 65 countries andxamine the importance of common global factors in explaininghe variation in changes in banks’ default risk. We also examinehe evolution of the correlation structure of default risk over the998–2010 time period.

We decompose default risk into global and regional effectssing the empirical methods of Heston and Rouwenhorst (1994).e show that changes in default risk depend more strongly on

lobal effects than on regional effects. The variance decompositionuggests that around 38% of the systematic variance in changesn default risk from the Merton (1974) model can be attributedo global effects. Changes in default risk for larger banks (withssets greater than $10 billion) and for banks operating in rela-ively open economies, depend less strongly on regional factorsnd more strongly on global factors. Following Bekaert and Wang2009), we use the variance ratio as our co-dependence measurend examine its evolution over the sample period. We show thathere has been a substantial increase in co-dependence in defaultisk of publicly traded banks starting around the beginning of 2004eading up to the global financial crisis starting in the summer of007. Although we observe an overall trend toward convergence inefault risk globally, this trend has been much stronger for Northmerican and European banks.

We also examine variation in co-dependence across countriesuring and before the financial crisis. In particular, we test to see

f the trends of globalization and de-regulation mentioned abovean explain cross-country differences in systemic risk. We find thatountries that are more integrated and have liberalized financialystems have seen higher co-dependence in their banking sectors,specially during the crisis. The increase in co-dependence, how-ver, is mitigated by strong institutional environments that allowor efficient public and private monitoring of financial institutions.hese results are consistent with the view that liberalization andnancial openness that are not accompanied by sufficient pru-ential regulation and supporting institutions to ensure effectiveonitoring, are likely to result in excessive risk-taking and higher

ncidence of crises (Demirguc-Kunt and Detragiache, 2005).Increased co-dependence in credit risk in the banking sector

as important implications for capital regulations. In the after-ath of the financial crisis, there has been renewed interest inacro-prudential regulation and supervision of the financial sys-

em. There has also been a growing consensus to adjust capitalequirements to better reflect an individual bank’s contribution tohe risk of the financial system as a whole (Brunnermeier et al.,009; Financial Stability Forum, 2009a, 2009b). Recently a num-er of papers have tried to measure and quantify systemic risk

nherent in the global banking sector. Adrian and Brunnermeier2011), Huang et al. (2009), Chan-Lau and Gravelle (2005), Avesani


t al. (2006), and Elsinger et al. (2006) use a portfolio credit riskpproach to compute the contribution of an individual bank tohe risk of a portfolio of banks. Our paper is related to this strandf literature, but rather than quantifying systemic risk of large B

nancial Stability xxx (2014) xxx–xxx

nancial institutions, we focus on examining time series trendsor a large cross-section of banks. A number of papers have exam-ned the correlation structure of equity returns of a subsample ofanks. De Nicolo and Kwast (2002) find rising correlations betweenank stock returns in the U.S. from 1988 to 1999. Schuler (2002)nds similar results for Europe using a sample from 1980 to 2001.awkesby et al. (2007) analyze co-movements in equity returns for

set of US and European large complex financial institutions usingeveral statistical techniques and find a high degree of commonal-ty. This paper is also related to the literature that studies contagionn financial markets (see among others Forbes and Rigobon, 2002;oyson et al., 2010) and also the literature that examines the impactf globalization on convergence of asset prices (Bekaert and Wang,009; Longin and Solnik, 1995; Bekaert and Harvey, 2000; Bekaertt al., 2012).

We contribute to the existing literature in three respects. First,ur empirical analyses cast a wider net than the existing literaturehich focuses only on a particular region or a country and covers a

horter time period.2 Second, we examine time series trends in co-ependence and test for structural changes over time. Finally, wexamine cross-country differences in co-dependence and link theifferences to measures of financial and economic openness andublic and private monitoring structures in different countries.

The rest of the paper is organized as follows. Section 2 describeshe data sources and describes the construction of the Merton1974) default risk measure. Section 3 presents the empiricalesults, and Section 4 concludes.

. Data sources, credit risk and co-dependence measures

The key variables for our analysis come from BANKSCOPE whichrovides bank-level balance sheet information, and DATASTREAMhich provides information on stock prices, market capitalization

nd stock volume. We examine the time period from 1998 to 2010.e use weekly market data and annual accounting information

n creating the credit risk measure. All data items are in US dol-ars to make comparisons across countries possible. We computeefault probabilities implied from the structural credit risk modelf Merton (1974). This approach treats the equity value of a com-any as a call option on the company’s assets. The probabilityf default is computed using the “distance-to-default” measure,hich is the difference between the asset value of the firm and

he face value of its debt, scaled by the standard deviation of therm’s asset value. The Merton (1974) distance-to-default measureas been shown to be good predictor of defaults outperform-

ng accounting-based models (Campbell et al., 2008; Hillegeistt al., 2004; Bharath and Shumway, 2008). Although the Mertonistance-to-default measure is more commonly used in bankruptcyrediction in the corporate sector, Merton (1977) points out thepplicability of the contingent claims approach to pricing depositnsurance in the banking context. Bongini et al. (2002), Bartramt al. (2007) and others have used the Merton model to computeefault probabilities of commercial banks.

We follow Campbell et al. (2008) and Hillegeist et al. (2004) toalculate Merton’s distance-to-default. The market value of equityor a bank is modeled as a call option on the bank’s assets:

VE = VAe−dT N(d1) − Xe−rT N(d2) + (1 − e−dT )VA


2 See for instance, De Nicolo and Kwast (2002), Schroder and Schuler (2003) andrasili and Vulpes (2005).

dx.doi.org/10.1016/j.jfs.2014.02.003


al of Financial Stability xxx (2014) xxx–xxx 3

arvt

S

v

oWddbdhceptr1TputVVmiM

M

tf

pcbItaattmWfiJ

dccsfid

a

0

1

2

3

4

5

6

7

8

9

10

2010200920082007200620052004200320022001200019991998

All

Exc. USA

Dist

ance

-to-

Defa

ult

Fig. 1. Distance-to-default over time. This figure shows the weekly weighted-average distance-to-default of all banks (All), and for all banks excluding thosel2

IccaedDm

eatoctjamfBrrt

vmsfottBBi


Above VE is the market value of a bank. VA is the value of bank’sssets. X is the face value of debt maturing at time T. r is the risk-freeate and d is the dividend rate expressed in terms of VA. sA is theolatility of the value of assets, which is related to equity volatilityhrough the following equation:

E = VAe−dT N(d1)SA

VE(2)

We simultaneously solve the above two equations to find thealues of VA and sA.

We use the market value of equity for VE and short-term plusne half long-term liabilities to proxy for the face value of debt X.e have found similar results using short term debt plus currently

ue portion of long term liabilities plus demand deposits as theefault barrier. Since the accounting information is on an annualasis, we linearly interpolate the values between two accountingates following Bartram et al. (2007). The interpolation methodas the advantage of producing a smooth implied asset value pro-ess and avoids jumps in the implied default probabilities at yearnd.3 sE is the standard deviation of weekly equity returns over theast 12 months. In calculating the standard deviation, we requirehe company to have at least 36 non-zero and non-missing weeklyeturns over the previous 12 months. T equals 1 year, and r is the-year treasury bill rate, which we take to be the risk free rate.he dividend rate, d, is the sum of the prior year’s common andreferred dividends divided by the market value of assets. Wese the Newton method to simultaneously solve the two equa-ions above. For starting values for the unknown variables we use,A = VE + X and sA = sEVE(VE + X). Once we determine asset values,A, we then compute asset returns as in Hillegeist et al. (2004):t = max(VA,t/VA,t−1 − 1, r). As expected returns cannot be negative,

f asset returns are below zero they are set to the risk-free rate.4

erton’s distance-to-default is finally computed as:

ertonDD = log(VA/X) + (m − d − (S2A/2))T

SA

√T

(3)

The default probability is the normal transform of the distance-o-default measure, defined as: PD = MertonDD. In the analyses thatollow we focus on log changes in default probability: �log(PD).

The summary statistics for the distance-to-default measure arerovided in Table 1. In the table we also report the number of banksovered by both BANKSCOPE and DATASTREAM, and the number ofanks that remain after we impose the data filters described above.

n all, we have 1942 banks in 68 countries for which we are ableo calculate Merton DD measure. Fig. 1 plots the value weightedverage distance-to-default measure over time for all banks andll banks excluding those located in the USA. We see a similar pat-ern in both series with a substantial increase in default risk duringhe crisis. Table 2 reports annual average bank distance-to-default

easure for different regions. The region definitions are based onorld Bank country classifications. We group countries not classi-

ed by the World Bank, under Western Europe, North America andapan, Australia and New Zealand.

We use a number of country level variables to explain co-ependence in the banking sector. In particular, we link theross-country differences in co-dependence to measures of finan-ial and economic openness, and public and private monitoring


tructures in each country. We collect three sets of variables. Therst set of variables is country level controls. We obtain economicevelopment measures from the World Bank’s World Development

3 We obtain similar results without interpolation and using year-end values.4 We obtain similar results if we use a 6% equity premium instead of asset returns

s in Campbell et al. (2008).

tfvv

D

ocated in the USA (Exc. USA), satisfying the data requirements outlined in Section of the paper. Asset values of banks are used as weights.

ndictor (WDI) database. We use the natural logarithm of GDP perapita (log GDP/cap) to measure the economic development of aountry. We use the variance of GDP growth rate (GDP/cap vari-nce) over the past 5 years to measure economic stability (Karolyit al., 2012; Morck et al., 2000). In addition, we obtain bank depositsivided by GDP (Bank deposits/GDP) from the Financial Structuresataset of Beck et al. (2009) to control for banking sector develop-ent.The second set of variables relate to measures of financial lib-

ralization and openness. We use imports plus exports of goodsnd services divided GDP (Trade/GDP) to measure global integra-ion (Bekaert and Wang, 2009). The Chin-Ito measure financialpenness measure (Chin-Ito financial openness) quantifies capitalontrol policies and other regulations and restrictions on capi-al flows (Chinn and Ito, 2008). Higher values indicate greater deure financial openness. We also use foreign ownership of bankss a measure of financial integration. Foreign ownership variableeasures the fraction of banks that are 50% or more owned by

oreign institutions, and is obtained from surveys conducted byarth et al. (2008). Our final measure of liberalization quantifiesestrictions and regulations on international financial transactionseforms (International capital liberalization) and is obtained fromhe database created by Abiad et al. (2010).

The final set of variables measure the strength of public and pri-ate monitoring in each country. Banking supervision is an indexeasuring supervisory authorities’ power and authority to take

pecific preventive and corrective actions. The measure rangesrom zero to fourteen, with fourteen indicating the highest powerf the supervisory authorities. Capital stringency index measureshe amount of capital a bank must maintain. It ranges from zeroo eight, with a higher value indicating higher capital stringency.oth measures are obtained from the Barth et al. (2008) survey.ased on the notion that efficient private monitoring depends on

nformation availability and sharing (Djankov et al., 2007), we usehe depth of credit information sharing (Credit information depth)


rom the World Bank Doing Business Survey as a measure of pri-ate monitoring.5 This variable ranges from zero to six, with higheralues indicating deeper credit information. Finally, we use deposit

5 Details on how these variables are constructed are available on World Bank’soing Business Survey website at http://www.doingbusiness.org/methodology.

dx.doi.org/10.1016/j.jfs.2014.02.003

http://www.doingbusiness.org/methodology

Please cite this article in press as: Anginer, D., Demirguc-Kunt, A., Has the global banking system become more fragile over time? J.Financial Stability (2014), http://dx.doi.org/10.1016/j.jfs.2014.02.003


4 D. Anginer, A. Demirguc-Kunt / Journal of Financial Stability xxx (2014) xxx–xxx

Table 1Data coverage and summary statistics of distance-default.This table shows the summary statistics for the distance-to-default measure for countries for which we have at least one observation. Our original data from both BankScopeand Datastream covers 2294 publicly traded banks across 86 countries. Of these banks, 1942 of them satisfy our requirement that (i) they have at least 36 weekly non-zerovolume and non-missing returns over the previous 12 months, (ii) they have non-missing liabilities and market capitalization. P5, P50 and P95 denote the 5th, 50th and 95thpercentile values respectively.

Country Number of banks withBANKSCOPE andDATASTREAM coverage

Number of banks withdistance-to-defaultmeasure

Mean Standarddeviation

P5 P50 P95

Argentina 8 8 4.09 1.81 1.87 3.80 7.81Austria 14 13 9.33 3.22 2.77 10.83 12.00Australia 20 19 6.47 1.95 3.05 6.56 9.73Belgium 6 6 6.99 2.98 1.84 6.78 12.00Bahrain 16 11 5.30 2.26 2.30 4.87 9.98Bermuda 17 13 4.02 2.24 1.46 3.53 8.84Brazil 30 25 3.88 2.07 1.35 3.39 8.14Canada 20 18 6.60 2.63 2.71 6.28 12.00Switzerland 24 24 7.63 3.44 2.56 7.24 12.00Chile 9 8 6.00 2.41 2.40 5.76 11.70People’s Republic of China 12 12 4.68 1.86 2.21 4.48 8.03Colombia 9 6 4.83 1.63 2.59 4.80 7.87Czech Republic 3 2 5.33 2.62 1.94 4.67 11.68Germany 31 30 6.90 3.48 1.99 6.54 12.00Denmark 20 18 7.56 2.59 3.33 7.51 12.00Egypt 3 3 3.83 1.40 2.20 3.40 6.03Spain 12 12 6.85 2.57 3.06 6.62 11.65Finland 7 7 6.66 2.71 2.34 6.41 11.65France 58 55 8.42 3.41 2.73 8.71 12.00United Kingdom 53 50 5.78 2.75 1.87 5.32 11.30Greece 18 17 3.90 1.75 1.69 3.68 6.93Hong Kong 17 16 4.72 2.45 1.55 4.24 9.70Hungary 3 3 4.25 1.60 1.80 3.93 7.64Indonesia 20 15 2.47 1.04 0.91 2.39 4.11Ireland 6 5 6.02 2.53 1.29 5.79 10.32Israel 12 8 6.92 2.23 3.56 6.72 11.16India 29 26 4.14 1.86 1.73 3.84 7.45Iceland 7 6 4.22 1.35 1.87 4.29 6.58Italy 43 38 6.58 2.76 2.60 6.17 12.00Jordan 12 11 4.86 1.43 2.80 4.61 7.35Japan 164 160 7.06 3.05 2.46 6.98 12.00Kenya 9 7 4.57 3.03 1.97 3.34 12.00Republic of Korea 16 13 4.35 2.25 1.47 4.00 9.76Kuwait 22 22 4.59 1.87 1.74 4.47 7.93Kazakhstan 9 1 2.73 0.35 2.28 2.61 3.40Lebanon 6 1 6.95 2.07 4.25 7.25 10.21Liechtenstein 2 2 6.76 2.16 3.50 6.76 12.00Sri Lanka 13 8 5.04 1.94 2.30 4.91 8.39Lithuania 5 3 4.01 1.38 1.67 4.18 6.07Luxembourg 7 6 7.37 3.04 3.19 6.70 12.00Morocco 6 6 6.56 2.09 3.50 6.40 10.31Mauritius 2 2 8.67 1.95 6.52 7.67 12.00Mexico 14 10 5.70 3.04 1.87 4.89 12.00Malaysia 29 29 4.71 2.37 1.14 4.42 9.18Netherlands 12 12 6.21 3.07 2.44 5.37 12.00Norway 22 22 9.18 3.02 3.35 10.18 12.00New Zealand 1 1 8.12 2.22 5.11 7.79 11.68Oman 3 3 4.33 1.41 2.12 4.45 6.44Peru 4 4 4.87 1.87 2.42 4.45 8.68Philippines 17 16 4.18 1.97 1.67 3.85 8.19Pakistan 17 14 4.05 1.73 1.68 3.77 7.61Poland 15 15 3.69 1.42 1.65 3.55 6.21Portugal 9 8 6.96 2.87 3.02 6.39 12.00Qatar 6 6 3.49 0.94 2.09 3.38 5.41Romania 3 2 3.41 1.32 1.25 3.49 5.41Russian Federation 15 7 3.61 1.90 1.16 3.19 7.31Saudi Arabia 10 10 4.06 1.64 2.19 3.59 7.39Sweden 11 11 5.05 2.41 1.83 4.64 9.82Singapore 17 17 5.47 2.82 1.64 4.96 11.11Slovenia 7 2 5.78 1.01 4.24 5.77 7.64Slovakia 4 3 7.78 3.24 2.53 8.00 12.00Thailand 29 28 3.55 1.66 1.04 3.37 6.50Turkey 24 22 2.56 0.84 1.34 2.55 3.90Taiwan 36 34 4.59 2.00 1.97 4.19 8.57Ukraine 6 1 2.69 0.41 2.23 2.61 3.54USA 1064 916 5.80 2.42 2.20 5.54 10.34Venezuela 8 6 4.02 2.24 1.86 3.34 9.60South Africa 32 27 3.77 1.62 1.53 3.63 6.48

dx.doi.org/10.1016/j.jfs.2014.02.003


D. Anginer, A. Demirguc-Kunt / Journal of Financial Stability xxx (2014) xxx–xxx 5

Table 2Regional time series of distance-to-default.This table shows the weighted-average distance-to-default measure for nine different regions computed each year. We calculate weighted-average distance-to-default bymarket capitalization each week for each region. We then compute arithmetic averages across 52 weeks in a given year.

Region 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Africa 3.49 1.92 3.42 4.17 3.83 4.81 4.87 5.02 4.72 4.27 4.04 2.62 4.39Central Asia and Eastern Europe 2.29 2.07 2.8 3.12 3.77 4.5 4.2 4.34 3.74 4.08 3.64 1.65 2.73East Asia and Pacific 3.33 2.63 3.62 4.43 4.28 5.46 5.54 7.26 6.61 5.50 4.35 3.30 5.02Japan, Australia and New Zealand 5.49 4.62 4.89 6.15 5.96 6.27 6.25 8.01 6.48 7.19 5.01 4.26 6.34Latin America and Caribbean 3.23 2.93 3.77 4.29 4.78 4.42 5.39 5.18 4.19 4.32 3.32 2.11 4.11Middle East and North Africa 6.97 5.89 6.78 8.19 8.97 6.77 7.12 5.17 4.2 4.67 4.62 2.88 4.98North America 4.87 3.89 3.46 3.89 5.19 5.65 8.05 8.79 8.92 8.47 4.47 1.68 3.88South Asia 4.75 3.65 3.56 4.85 5.64 6.39 4.32 4.2 4.33 3.9 3.07 2.35 3.36Western Europe 4.86 4.21 5.45 5.18 5.39 5.88 8.41 10.18 8.87 8.33 7.07 2.97 5.26

Table 3Regional distance-to-default correlations.This table reports the time-series correlations between different regions over the 1998–2010 time period. First, we calculate weighted-average distance-to-default by marketcapitalization in each region per week. Then the pair-wise Pearson correlation coefficients are calculated from the weekly data.

Africa CAEE EAP JANZ LAC MENA NA SA WE

Africa 1Central Asia and Eastern Europe 0.7709 1East Asia and Pacific 0.7705 0.7834 1Japan, Australia and New Zealand 0.6800 0.7297 0.7386 1Latin America and Caribbean 0.7357 0.8035 0.6210 0.6539 1Middle East and North Africa 0.1221 0.2428 −0.0960 0.0207 0.4400 1

ifihiaf(

3

dmrip

P

rlbWnaisWc

dfitl

cTTa9i

tcdbsnections and integration as well as similarities in risk exposures,which can contribute to systemic fragility. We do the principalcomponent decomposition using all banks over the 1998–2010

-5.0

-4.5

-4.0

-3.5

-3.0

-2.5

-2.0

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

Varia

nce

Ra�o

All

Exc. USA

Fig. 2. Variance ratio. This figure shows the variance ratio calculated as the ratio of

North America 0.6356 0.7283 0.8137

South Asia 0.3013 0.4732 0.3157

Western Europe 0.7297 0.7864 0.8478

nsurance coverage as a barrier to effective private monitoring ofnancial institutions. If investors are insulated from losses they willave less of an incentive to monitor financial institutions. Deposit

nsurance coverage ratio is the amount of deposit insurance cover-ge divided by deposits per capita. It is set to 1 if a country offersull coverage. We obtain this variable from Demirguc-Kunt et al.2008).

. Co-dependence in the banking sector

There are a number of different approaches to measuring co-ependence. In this paper, we use the variance ratio as the maineasure of co-dependence. The variance ratio is calculated as the

atio of the average variance of changes in log default probabil-ty divided by the variance of the average changes in log defaultrobability:

Rt = log

(Var(

1N

∑i� log(PDi,s)

)(1N

∑iVar(� log(PDi,s))

))

for s ∈ [t − 52, t] (4)

The variance ratio increases as correlations in changes in defaultisk between banks increase. If the correlations are one, then theog variance ratio takes on a value of zero. The variance ratio haseen previously used by Ferreira and Gama (2005) and Bekaert andang (2009) in examining convergence of asset prices in inter-

ational markets. Fig. 2 plots the variance ratio calculated on annnual basis for all banks, and for all banks excluding those locatedn the USA. Both series show a similar pattern. There has been aubstantial increase in co-dependence during the financial crisis.

e also observe a gradual increase in the period leading up to therisis.

Before examining time-series and cross-country variation in co-ependence, we first explore commonality in changes in default


or the full sample period. We begin by examining correlationn changes in default risk between different regions. To computehese correlations, we first calculate value-weighted changes inog default probability, �log(PD), for each region and then compute

toaia

0.7447 0.7010 −0.0086 10.4045 0.4986 0.6411 0.2996 10.7474 0.7166 −0.0844 0.9044 0.1371 1

orrelations over the sample period. We use total assets as weights.able 3 presents the matrix of pairwise correlations across regions.he correlations are fairly high across regions except for the Middlend North Africa region. The correlations range from −8% to over0% with an average of 54%. High correlations indicate commonal-

ty in credit risk changes.Next, we conduct a standard principal components analysis on

he covariance of weekly changes in default probabilities. Increasedommonality among default risk of banks can be empiricallyetected by using principal components analysis. In particular,y examining the importance of a single component to explainubstantial variation in credit risk, we can detect increased con-


he average variance of changes in log probability of default divided by the variancef the average changes in log default probability (Eq. (4)) for all banks(All) and forll banks excluding those located in the USA (Exc. USA) in the dataset. Variance ratios calculated each year using weekly data. We use log changes in default probabilitynd include banks with at least 26 observations to compute the variance ratio.

dx.doi.org/10.1016/j.jfs.2014.02.003



Table 4Principal component decomposition of changes in credit risk.This table reports the first five components from a principle components analysisof default risk. First, we compute changes in log default probability for banks overthe 1998–2010 time period. We then perform a principal component decompositionusing the estimated covariance matrix. We report only the cumulative proportion ofthe variance explained by the first five components. Same decomposition excludingUS banks and excluding the 2007–2010 time period is reported in the second andthird columns of the table respectively.

Proportion of variance explained (cumulative)

All Exc. USA Exc. 2007–2010

Component 1 0.7181 0.5878 0.5283Component 2 0.8582 0.8346 0.7771Component 3 0.9277 0.9113 0.8647Component 4 0.9593 0.9404 0.9279Component 5 0.9857 0.9575 0.9504

tctpcewamfiocr

cboditte

bR

Fig. 3. Clustering in default risk. This figure shows the percentage of banks in agiven week that have simultaneous worst change in default risk over a 12-monthperiod. We compute log changes in default probability each week for each bank inoct

ga

�

dtnFiewnrCogvariance explained by regional effects is computed in a similar fash-ion. Table 5 shows the results from this decomposition. We report

TVIv

a

aop

c

ime period. The results are reported in Table 4. As a robustnessheck, we also report results excluding US banks and excludinghe 2007–2010 crisis period. For the full sample, the first princi-al component explains about 70 percent of the total variation inhanges in credit risk, while the first three principal componentsxplain more than 90 percent. The results are slightly weaker whene exclude US banks and the crisis years. The principal component

nalyses results suggest that there is a significant amount of com-onality in the variation of default risk changes. Furthermore, the

rst principal component consists of a roughly uniform weightingf default risk changes for banks in our sample. The first principalomponent, thus, resembles a global factor affecting the defaultisk changes of all banks.

Consistent with this result we observe significant clustering ofhanges in bank default probabilities. Fig. 3 shows the percentage ofanks that had their worst change in default risk in the same weekver a 12-month period. If the changes in credit risk are indepen-ent we would expect to see an even distribution of worst changes

n default risk for banks over time. In other words we would expecto see a flat line with no spikes. Instead, we observe significant clus-ering. The extent of clustering during the recent financial crisis isspecially dramatic.

To explore further the systematic variation in changes in


ank default risk, we follow the methodology of Heston andouwenhorst (1994) and decompose changes in default risk into

rt

able 5ariance decomposition of default risk.

n this table, we report the systematic variance attributable to global and regional effectsariance of changes in log default probability (�log(PD)). Each week of the sample period

nd nine region dummy variables: � log (PD)ikt = at +∑K

k=1I(C)ikC + εikt . In the regressi

nd zero otherwise. To avoid multi-collinearity, we impose the region effects weighted bf banks in each region k. For each period t, we run a cross-sectional regression to estimroportion of systematic variance explained by global effects is approximately given by:

omputed in a similar fashion. We repeat the analyses for all banks, banks with assets gre

Region Number of banks All banks

Global effect Region

Africa 36 0.28 0.72

Central Asia and Eastern Europe 76 0.29 0.71

East Asia and Pacific 145 0.43 0.57

Japan, Australia and N.Z. 200 0.24 0.76

Latin America and Caribbean 73 0.38 0.62

Middle East and North Africa 81 0.13 0.87

North America 920 0.72 0.28

South Asia 46 0.11 0.89

Western Europe 289 0.55 0.45

ur dataset. Each year, we then count the number of banks that had their worsthange in log default probability in the same week. We then divide this count byhe total number of banks in the dataset in that particular year.

lobal and regional effects. We model log changes in default prob-bility as follows:

log (PD)ijkt = at +K∑

k=1

I(C)ikCkt + εijkt (5)

Above, at is the global component common to all banks. I(C)ik is aummy variable that takes on a value if one, if bank i is headquar-ered in region k, and zero otherwise. We partition the data intoine regions (K = 9) using the regional definitions described above.ollowing Heston and Rouwenhorst (1994), we impose restrictionsn order to avoid multi-collinearity when estimating the param-ters of the model. In particular, we impose the region effectseighted by the number of banks to be zero:

∑Kk=1n(C)kCkt = 0.

(C)k is the number of banks in each region k. For each week t, weun a cross-sectional regression to estimate the coefficients at andkt. For each individual bank belonging to region k, the proportionf systematic variance explained by global effects is approximatelyiven by: var(at)/(var(at) + var(ckt)) The proportion of systematic


esults for the full sample, as well for large banks with assets greaterhan $10 billion and smaller banks with assets less than $10 billion.

. We use Heston and Rouwenhorst (1994)’s method to decompose the systematicfrom 1998 to 2010, we run a cross-sectional regression of �log(PD) onto a constant

on, I(C)Ik is a dummy variable equal to one, if bank i is headquartered in region k,

y the number of banks to be zero:∑K

k=1n(C)kCkt = 0. n(C)k is equal to the number

ate the coefficients, at and Ckt . For each individual bank belonging to region k, thevar(akt )

var(at )+var(ckt ) . The proportion of systematic variance explained by region effects is

ater than $10B, and banks with assets less than $10B.

Banks with assets >$10B Banks with assets <$10B

effect Global effect Region effect Global effect Region effect

0.29 0.71 0.22 0.780.47 0.53 0.15 0.850.49 0.51 0.29 0.710.33 0.67 0.22 0.780.39 0.61 0.23 0.770.24 0.76 0.05 0.950.78 0.23 0.69 0.310.23 0.77 0.10 0.900.62 0.38 0.41 0.59

dx.doi.org/10.1016/j.jfs.2014.02.003


al of Financial Stability xxx (2014) xxx–xxx 7

Tiotb

arlhetw4iii

3

dioWdoor

P

oItptescsatmfibbi2at

Tpn

tfe

ion

esti

mat

es

from

Eq. (

6)

in

the

pap

er. F

ollo

win

g

Bek

aert

and

Wan

g

(200

9)

we

use

tren

d

test

s

to

det

ect

pot

enti

al

chan

ges

in

co-d

epen

den

ce. F

irst

, th

e

vari

ance

rati

o

(PR

t)

is

com

pu

ted

for

52

wee

k

roll

ing

inte

rval

s.

We

use

log

chan

ges

in

def

ault

pro

babi

lity

and

incl

ud

e

ban

ks

wit

h

at

leas

t 26

obse

rvat

ion

s

to

com

pu

te

the

vari

ance

rati

o.

We

then

run

the

foll

owin

g

regr

essi

on:

t∈19

98.0

1−20

03.1

2}· t

+

˛2I {

t∈20

04.0

1−20

07.0

6}+

ˇ2I {

t∈20

04.0

1−20

07.0

6}· t

+

˛3I {

t∈20

07.0

6−20

09.1

2}+

ˇ3I {

t∈20

07.0

6−20

09.1

2}· t

+

ε t. I

n

this

equ

atio

n, I

{t∈p

erio

d}is

a

du

mm

y

vari

able

that

take

s

on

a

valu

e

of

one

over

the

the

lin

ear

tim

e

tren

d. t

-Sta

tist

ic

are

rep

orte

d

belo

w

coef

fici

ent

esti

mat

es

and

are

corr

ecte

d

for

auto

corr

elat

ion

usi

ng

New

ey–W

est

wit

h

thre

e

lags

.

Wor

ld

Dev

elop

ed

Dev

elop

ing

Euro

pea

nU

nio

nU

SA

Jap

an,

Au

stra

lia

and

N.Z

.

Afr

ica

Eeas

tern

Euro

pe

and

Cen

tral

Asi

a

East

Asi

aan

d

Paci

fic

Lati

nA

mer

ica

and

Car

ribe

an

Mid

dle

East

and

N

Afr

ica

Sou

th

Asi

a

Nor

thA

mer

ica

Wes

tern

Euro

pe

−3.4

24**

*−3

.434

***

−2.2

73**

*−3

.044

***

−3.3

78**

*−2

.846

***

−1.4

14**

*−1

.800

***

−2.9

01**

*−2

.237

***

−0.7

24**

*−1

.446

***

−3.3

93**

*−3

.057

***

(−26

.291

)

(−26

.015

)

(−29

.887

)

(−25

.238

)

(−29

.785

)

(−36

.25)

(−13

.769

)

(−44

.343

)

(−13

.37)

(−41

.373

)

(−7.

450)

(−17

.253

)

(−30

.109

)

(−24

.015

)−0

.002

***

−0.0

01**

*−0

.003

***

0.00

0

−0.0

01**

*0.

001*

−0.0

01**

*−0

.001

***

0.00

4

−0.0

02**

*−0

.002

***

−0.0

01**

*−0

.001

***

0.00

0(−

3.64

5)(−

2.83

9)(−

12.3

27)

(0.3

07)

(−3.

670)

(1.6

9)

(−3.

259)

(−10

.576

)(1

.06)

(−7.

024)

(−7.

254)

(−4.

581)

(−3.

373)

(0.6

99)

−6.5

88**

*−6

.413

***

−2.4

21**

*−7

.111

***

−5.4

51**

*−3

.786

***

0.43

2*−3

.169

***

−3.1

46**

*−2

.482

***

−0.7

06

−2.6

40**

*−5

.336

***

−7.9

41**

*

(−21

.959

)(−

18.9

18)

(−4.

665)

(−16

.365

)(−

22.8

76)

(−32

.71)

(1.6

53)

(−22

.073

)(−

31.2

1)

(−12

.651

)

(−1.

093)

(−7.

338)

(−20

.613

)

(−17

.706

)0.

004**

*0.

004**

*−0

.004

***

0.00

8***

0.00

2***

0.00

2***

−0.0

05**

*0.

001**

*−0

.001

**−0

.001

***

−0.0

05**

*0.

001

0.00

2***

0.00

9***

(6.6

08)

(5.8

97)

(−3.

188)

(8.9

21)

(4.1

61)

(7.3

)

(−8.

789)

(4.6

08)

(−3.

59)

(−3.

482)

(−3.

637)

(1.6

21)

(3.2

97)

(10.

630)

−11.

953**

*−1

1.61

6***

−12.

650**

*−1

1.42

2***

−10.

315**

*−4

.489

***

−8.0

00**

*−1

1.17

5***

−10.

045**

*−8

.404

***

−10.

293**

*−9

.735

***

−10.

443**

*−1

0.69

4***

(−25

.455

)

(−24

.897

)

(−19

.020

)

(−24

.689

) (−

17.3

76)

(−14

.97)

(−14

.525

)

(−10

.538

)

(−41

.32)

(−32

.621

)

(−17

.243

)

(−26

.888

)

(−17

.570

)

(−26

.209

)0.

015**

*0.

014**

*0.

015**

*0.

015**

*0.

012**

*0.

001**

0.01

1***

0.01

4***

0.01

2**0.

010**

*0.

012**

*0.

013**

*0.

012**

*0.

014**

*

(19.

754)

(19.

380)

(14.

475)

(19.

703)

(12.

893)

(8.8

3)

(12.

084)

(8.6

60)

(29.

83)

(23.

109)

(12.

933)

(21.

483)

(13.

128)

(20.

395)

662

662

662

662

662

662

662

662

662

662

662

662

662

662

3

3

3

3

3

3

3

3

3

3

3

3

3

3

0%

two

tail

ed

leve

l.%

two

tail

ed

leve

l.%

two

tail

ed

leve

l.


he larger banks tend operate beyond national borders, and engagen risk-transfer activities and share linkages with a greater numberf institutions, exposing them common sources of risk. We would,herefore, expect greater portion of systematic variance of largeranks to be explained by global effects.

The variance decomposition suggests that the global effectccounts for 36% of the systematic variation in changes in defaultisk. We see substantial differences across different regions. Banksocated in South Asia, and Middle East and North Africa regionsave a lower percentage of their systematic variation in credit riskxplained by global effects. Most of the systematic variation forhese banks comes from sources common to the region. Consistentith our conjecture, banks with assets greater than $10 billion have

3% of the systematic variation coming from global effects, while its only 26% for banks with assets less than $10 billion. These resultsndicate that that there is a significant global component to changesn default risk in the banking sectors across different regions.

.1. Time series analyses

In this section, we examine time series variation of co-ependence in the banking sector. In particular, we are interested

n whether there have been significant changes in co-dependencever the sample period from 1998 to 2010. Following Bekaert andang (2009), we use trend tests to detect structural changes in co-

ependence. We compute the variance ratio (PRt) for each regionver 52 week rolling intervals. We include banks with at least 26bservations in a given 52-week window to compute the varianceatio. For trend tests, we use the following empirical model:

Rt = ˛1I{t∈1998.01−2003.12} + ˇ1I{t∈1998.01−2003.12} · t

+ ˛2I{t∈2004.01−2007.06} + ˇ2I{t∈2004.01−2007.06} · t

+ ˛3I{t∈2007.06−2009.12} + ˇ3I{t∈2007.06−2009.12} · t + εt (6)

Above, I{t∈period} is a dummy variable that takes on a value ofne over the specified time period, and t is the linear time trend.n estimating the coefficients, we correct for auto correlation upo three lags. We split the sample into three time intervals. Timeeriod from June 2007 to December 2009 corresponds roughly tohe global financial crises. Although it is difficult pin down thexact date, it was toward the end of June 2007 when the firstignificant signs of the crisis began to appear. There was a signifi-ant increase in market uncertainty as subprime mortgage backedecurities were discovered in portfolios of banks and hedge fundsround the world. Few weeks later, BNP ceased redemptions inhree of its funds due to “complete evaporation of liquidity” in the

arkets.6 January 2004–June 2007 is the period leading up thenancial crisis. It was around the beginning of 2004 when thereegan a substantial increase in subprime lending, growth shadowanking (Gorton, 2009), increase in leverage of major financial

nstitutions and reliance in short-term borrowing (Adrian and Shin,008; Morris and Shin, 2009) as well as increase in global imbal-nces (Jagannathan et al., 2009) that culminated in a crisis startinghe summer of 2007.

The results from the empirical model in Eq. (6) are reported in


able 6. There is an increase in co-dependence during the criseseriod (July 2007–December 2009) for all regions. In fact, there isot a single country which did not see an increase in co-dependence

6 http://invest.bnpparibas.com/cid3162415/bnp-paribas-investment-partners-emporaly-suspends-the-calculation-of-the-net-asset-value-of-the-following-unds-parvest-dynamic-abs-bnp-paribas-abs-euribor-and-bnp-paribas-abs-onia.html?pid=769. Ta

ble

6Ti

me-

seri

es

anal

ysis

.Th

is

tabl

e

show

s

the

regr

ess

each

regi

on

each

wee

k

over

PRt=

˛1I {

t∈19

98.0

1−20

03.1

2}+

ˇ1I {

spec

ified

tim

e

per

iod

, an

d

t

is

Int

1998

.01–

2003

.12

Slop

e

1998

.01–

2003

.12

Int

2004

.01–

2007

.06

Slop

e

2004

.01–

2007

.06

Int

2007

.07–

2009

.12

Slop

e

2007

.06–

2009

.12

N

Lags

*In

dic

ates

sign

ifica

nce

at

1**

Ind

icat

es

sign

ifica

nce

at

5**

*In

dic

ates

sign

ifica

nce

at

1

dx.doi.org/10.1016/j.jfs.2014.02.003

http://invest.bnpparibas.com/cid3162415/bnp-paribas-investment-partners-temporaly-suspends-the-calculation-of-the-net-asset-value-of-the-following-funds-parvest-dynamic-abs-bnp-paribas-abs-euribor-and-bnp-paribas-abs-eonia.html?pid=769





8 al of Fi

dnpstJhddd

3

dcwipfihaac(sosNfirifit

tltsfcbtatfietc2psacm

c

cofbir

Gaf

PPt(tiiivr

tisaaroot(ruTwba

oAptmcvapsiAiscaiafinancial institutions. If investors are insulated from losses they willhave less of an incentive to monitor (Demirguc-Kunt and Huizinga,


uring this time period. However, we do see variation in the mag-itude of the increase across countries and regions. For the timeeriod leading up to the crises (January 2004–June 2007), weee greater variation. There was an increase in co-dependencehroughout most of the developed world.7 Banks in United States,apan, Australia and New Zealand, and especially European Unionave seen a significant rise in co-dependence. Banks located ineveloping countries, on the other hand, have seen a decline in co-ependence over the same time period.8 It is these cross-sectionalifferences that we explore next.

.2. Cross-country analyses

In this section, we examine the cross-country differences inefault risk co-dependence. We are interested in testing if finan-ial and economic openness and greater integration are associatedith higher co-dependence in the banking sector. We are also

nterested in whether institutional structures that allow for betterrivate and public monitoring can mitigate the negative effects ofnancial integration and openness. Although, a number of papersave linked commonality in asset returns to economic opennessnd financial liberalization (see for instance Karolyi et al., 2012nd Bekaert and Wang, 2009), the evidence of the effects of finan-ial openness on a country’s vulnerability to crises remains mixedBekaert et al., 2006; Kose et al., 2006). Focusing on the bankingector, a number of papers have found liberalization and financialpenness to be associated with higher incidence of banking cri-is (Mehrez and Kaufmann, 1999; Arteta and Eichengreen, 2002;oy, 2004). However, others suggest that the relationship betweennancial openness and fragility varies with the institutional envi-onment (Demirgüc -Kunt and Detragiache, 1999). Countries withnstitutions that foster efficient public and private monitoring ofnancial institutions respond better to shocks to their banking sys-ems.

We use an empirical model based on Djankov et al. (2007) toest the relationship between realized co-dependence with countryevel measures of financial openness and monitoring. We examinewo time periods. We compute realized co-dependence for the cri-is period from 2007 to 2010, and for the pre-crisis boom periodrom 2004 to 2006. We compute the variance ratio (PR) for eachountry each year. As before, we use log changes in default proba-ility and include banks with at least 26 observations to computehe variance ratio. For each country, we then compute an aver-ge variance ratio over the 2007–2010 time period by taking theime-series average over those 3 years. We then use measures ofnancial openness and monitoring calculated at end of 2006 toxplain cross-country differences in average variance ratios overhe 2007–2010 time period. We repeat the analyses for the pre-risis time period by computing an average variance ratio over the004–2006 time period and using country level measures com-uted at the end of 2003. We exclude countries with less thaneven banks from the analyses to make sure that computed vari-


nce ratios are not noisy. In the regressions, we use the followingontrols: the natural logarithm of GDP per capita (log GDP/cap) toeasure the economic development of a country, the variance of

7 We use World Bank country classifications to define developed and developingountries.

8 It is possible that the variance ratio for the countries located in developingountries may be increasing mechanically as a result of an increase in the numberf banks over the sample period. As a robustness check, we replicated the analysesor a balanced panel that includes only the banks that were in the sample at theeginning of the period in 1998 and were also in the sample at the end of the period

n 2010. Although this analysis introduces a survivorship bias, we obtain similaresults for the developing countries using this methodology.

2

et

na

pii2


DP growth rate (GDP/cap variance) to measure economic stability,nd bank deposits divided by GDP (Bank deposits/GDP) to controlor banking sector development.

Cross-sectional regression results are reported in Table 7. Inanel A, we report the results for the crisis time period, and inanel B we report the results for the pre-crisis time period. Consis-ent with the prior literature that has examined asset correlationsKarolyi et al., 2012; Morck et al., 2000), we find GDP per capitao be associated with reduced co-dependence. Macroeconomicnstability as measured by variance of GDP growth has a positivempact on co-dependence. That is, greater macroeconomic instabil-ty increases credit risk co-dependence. Bank sector developmentariable, bank deposits over GDP, is insignificant in most of theegression specifications.

To examine linkages between financial integration, liberaliza-ion and co-dependence, we use four different measures. We usemports plus exports of goods and services divided by GDP to mea-ure overall economic integration. The chin-Ito financial opennessnd the Abiad et al. (2010) international capital liberalization vari-bles are used to proxy for de jure liberalization of restrictions andegulations on international financial transactions. We use foreignwnership of banks as a measure of financial integration. A numberf studies show that shocks to parent banks can be transmitted toheir foreign subsidiaries with negative effects for credit creationAcharya and Schnabl, 2010; Chava and Purnanandam, 2011).9 Theesults of the cross-country regressions using each of these meas-res separately are reported in columns 2–5 in Panels A and B ofable 7. These different measures give consistent results. Overall,e find greater openness, integration and financial liberalization to

e associated with increased co-dependence during the pre-crisisnd the crisis time periods.

Next, we examine impact of private and public monitoringn systemic fragility and on the openness-fragility relationship.s bank failures can be very costly due to the crucial role bankslay within the economic system, supervisors are naturally incen-ivized to monitor fragility in the banking system. But effective

onitoring and supervision requires authorities to take timelyorrective action. We use bank supervision, which measures super-isory authorities’ power and authority to take specific preventivend corrective actions, to proxy for public monitoring. To proxy forrivate monitoring, we use three different measures. The first mea-ure is capital stringency. Stringent capital requirements increasencentives to monitor and control risk taking.10 Repullo (2004) andllen et al. (2011) show that investors tend to prefer well capital-

zed banks, as these banks have greater incentives to monitor. Theecond measure we use is credit information depth, which measuresredit information availability and sharing. Greater informationvailability and sharing allows for better monitoring of financialnstitutions (Djankov et al., 2007).11 Finally, we use deposit insur-nce coverage, as a measure of barrier to private monitoring of


004; Ioannidou and Penas, 2010). Generous safety nets also tend

9 Popov and Udell (2012) and Claessens and van Horen (2012) also find that for-ign subsidiaries reduced their lending more compared to domestic banks duringhe financial crisis.10 Capital stringency may also reduce co-dependence through other channels. Aumber of papers, for instance, emphasize the role of capital as a potential buffer inbsorbing liquidity, information and economic shocks (Repullo, 2004).11 Information asymmetry also provides a potential channel in which shocks can beropagated through the banking system. A number of papers have used constrained

nformation asymmetry framework to explain risk contagion and crises (see fornstance Gennotte and Leland, 1990; Kodres and Pritsker, 2002; Hong and Stein,003; Barlevy and Veronesi, 2003; Yuan, 2005).

dx.doi.org/10.1016/j.jfs.2014.02.003



D. Anginer, A. Demirguc-Kunt / Journal of Financial Stability xxx (2014) xxx–xxx 9

Table 7Cross-country regressions.This table reports results from cross-country regressions of variance ratio on country level controls and measures of financial openness and monitoring. The dependentvariable is the variance ratio, calculated for each country and each year using log changes in default probabilities and include banks with at least 26 observations. Thevariance ratios are then averaged over the crisis period from 2007 to 2010 and the pre-crisis period from 2004 to 2006. Independent variables are computed at the end of2006 for the crisis period, and at the end of 2003 for the pre-crisis period. All the variables are described in detail in Section 2 of the text. In the regressions, we excludecountries with less than 7 banks. The regression results for the crisis period are reported in Panel A, and for the pre-crisis period in Panel B. Standard errors are reported inparentheses below their coefficient estimates and adjusted for heteroskedasticity.

Panel A (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Variables

log GDP/cap −0.076** −0.079** −0.125*** −0.077** −0.110*** −0.145** −0.092*** −0.064 −0.068* −0.122*

(0.035) (0.036) (0.041) (0.035) (0.039) (0.055) (0.034) (0.039) (0.037) (0.069)GDP/cap variance 0.041* 0.031 0.047** 0.032 0.046** 0.014 0.030 0.040 0.037 −0.008

(0.023) (0.021) (0.021) (0.022) (0.023) (0.019) (0.023) (0.025) (0.023) (0.028)Bank deposits/GDP −0.071 −0.071 −0.099 −0.182** −0.039 0.006 −0.042 −0.036 −0.085 0.191

(0.081) (0.118) (0.082) (0.091) (0.083) (0.132) (0.079) (0.083) (0.081) (0.140)

Global integration and financial opennessForeign ownership 0.352* −0.128

(0.182) (0.228)Chin-Ito financial openness 0.095*** −0.056

(0.033) (0.061)Trade/GDP 0.001** 0.007***

(0.001) (0.002)International capital liberalization 0.146** 0.040

(0.062) (0.114)

Public and private monitoringDeposit Insurance Coverage 0.001** 0.001**

(0.000) (0.000)Capital stringency −0.059*** −0.034

(0.021) (0.034)Banking supervision −0.028* −0.233***

(0.017) (0.074)Credit information depth −0.006* 0.034

(−0.003) (0.043)Constant −0.437 −0.501* −0.121 −0.449 −0.571** 0.105 0.010 −0.524* −0.453 0.677

(0.280) (0.274) (0.309) (0.277) (0.279) (0.448) (0.301) (0.283) (0.286) (0.720)

N 53 51 53 53 45 53 53 45 53 45R-squared 0.057 0.091 0.088 0.080 0.075 0.105 0.089 0.056 0.056 0.294

Panel B (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Variables

log GDP/cap −0.152*** −0.161*** −0.264*** −0.151*** −0.211*** −0.208*** −0.156*** −0.159*** −0.064 −0.266**

(0.045) (0.047) (0.055) (0.045) (0.054) (0.071) (0.045) (0.047) (0.065) (0.101)GDP/cap variance 0.047** 0.059*** 0.089*** 0.036** 0.064*** 0.045** 0.044** 0.046*** 0.027 0.071

(0.019) (0.021) (0.021) (0.017) (0.021) (0.019) (0.020) (0.017) (0.023) (0.047)Bank deposits/GDP −0.033 0.067 −0.049 −0.194* 0.035 0.095 −0.023 0.011 −0.145 0.452*

(0.094) (0.158) (0.089) (0.117) (0.097) (0.168) (0.095) (0.095) (0.124) (0.260)

Global integration and financial opennessForeign ownership 0.615*** 0.403

(0.197) (0.248)Chin-Ito financial openness 0.187*** −0.009

(0.040) (0.072)Trade/GDP 0.002** 0.010***

(0.001) (0.003)International capital liberalization 0.172** 0.162

(0.077) (0.194)

Public and private monitoringDeposit Insurance Coverage 0.001** 0.001*

(0.000) (0.001)Capital stringency −0.032 −0.016

(0.026) (0.041)Banking supervision −0.013 −0.253**

(0.071) (0.121)Credit information depth −0.102** 0.031

(0.045) (0.065)Constant −0.090 −0.279 0.569 −0.116 −0.164 0.305 0.117 −0.084 −0.262 −0.184

(0.346) (0.331) (0.392) (0.342) (0.337) (0.559) (0.385) (0.344) (0.475) (0.845)

N 47 43 47 47 44 43 47 44 47 43R-squared 0.053 0.084 0.084 0.064 0.071 0.068 0.056 0.063 0.071 0.243

* Indicates significance at 10% two tailed level.** Indicates significance at 5% two tailed level.

*** Indicates significance at 1% two tailed level.

dx.doi.org/10.1016/j.jfs.2014.02.003



Table 8Financial openness and monitoring interactions.This table reports results from cross-country regressions of variance ratio on two indices computed using measures of financial openness (Openness Index) and measures ofmonitoring (Monitoring Index). The dependent variable is the variance ratio, calculated for each country and each year using log changes in default probabilities and includebanks with at least 26 observations. The variance ratios are then averaged over the crisis period from 2007 to 2010 and the pre-crisis period from 2004 to 2006. The indicesare computed at the end of 2006 for the crisis period and at the end of 2003 for the pre-crisis time period. In the regressions, we exclude countries with less than sevenbanks. The crisis period results are reported in columns 1–3, and the pre-crisis results are reported in columns 4–6. The Standard errors are reported in parentheses belowtheir coefficient estimates and adjusted for heteroskedasticity.

Variables Crisis period 2007–2010 Pre-crisis period: 2004–2006

(1) (2) (3) (4) (5) (6)

log GDP/cap −0.153*** −0.086** −0.165*** −0.239*** −0.139*** −0.221***

−0.04 −0.035 −0.04 (0.071) (0.049) (0.067)GDP/cap variance 0.027 0.034 0.017 0.081*** 0.043 0.072***

−0.02 −0.024 −0.021 (0.026) (0.027) (0.026)Bank deposits/GDP −0.128 −0.052 −0.121 −0.015 −0.007 0.029

−0.079 −0.078 −0.084 (0.120) (0.120) (0.129)Openness Index 0.166*** 0.195*** 0.197*** 0.564***

−0.034 −0.04 (0.074) (0.140)Monitoring Index −0.183*** −0.190** −0.157*** −0.149

−0.059 −0.074 (0.060) (0.095)Openness Index × Monitoring Index −0.035** −0.156***

−0.013 (0.054)Constant −0.087 −0.22 0.135 0.134 0.133 −0.336

−0.287 −0.288 −0.312 (0.437) (0.442) (0.431)

N 45 45 45 43 43 43R-squared 0.176 0.096 0.23 0.163 0.157 0.256

tamcie

riawiniireaeoictupmIArcep

ca

rphlcvcgo

ttisem

4

icMWttifi

* Indicates significance at 10% two tailed level.** Indicates significance at 5% two tailed level.

*** Indicates significance at 1% two tailed level.

o be correlated with other implicit state guarantees. If there isn implicit guarantee provided by the State to cover losses stem-ing from a systemic crisis, banks will have incentives to take on

orrelated risks (Acharya, 2009). Guaranteed banks will not havencentives to diversify their operations, since the guarantee takesffect only if other banks fail at the same time.

The baseline results for the public and private monitoring areeported in columns 6–9 in Panels A and B. The results are support-ve of our conjecture that greater public and private monitoring isssociated with lower levels of systemic fragility. The results areeaker for the pre-crisis time period. The coefficients on the cap-

tal stringency and bank supervision variables are negative but areot statistically significant in the pre-crisis period. The last column

n Panels A and B reports regression results with all the measuresncluded together. Although the sample used in the regression iseduced, trade/GDP, banking supervision and deposit insurance cov-rage variables remain significant and with the correct sign. Welso test to see if better monitoring can mitigate the negativeffects of financial openness on co-dependence. Since the numberf cross-country observations is low, we cannot run a regressionncluding interactions of all the different measures. Instead, wereate two indices, one for financial openness and one for moni-oring, by aggregating the different measures we have previouslysed. The index starts at a value of zero and is increased by one if aarticular measure for a given country is above the cross-countryedian.12 We create an index for financial openness (Openness

ndex) and for public and private monitoring (Monitoring Index).s before, we run two cross-sectional regressions of the varianceatio averaged over the pre-crisis years from 2004 to 2006 and the


risis period from 2007 to 2010, on the indices computed at thend of 2003 and 2006 respectively for the pre-crisis and crisis timeeriods. The results are reported in Table 8. In columns 1–3 we

12 Since the Deposit Insurance Coverage variable has an expected positive coeffi-ient, we increase the Monitoring Index by one if the Deposit Insurance Coverage for

given country is less than the median.

anpwiUwEc

eport the results for the crisis period, and in columns 4–6 for there-crisis period. Consistent with the earlier results, Openness Indexas a significant positive association with co-dependence. Simi-

arly, Monitoring Index has a significant negative association witho-dependence. In the columns 3 and 6 of the table, we include aariable interacting the two indices as an additional covariate. Theoefficient on the interaction term is negative and significant, sug-esting that monitoring reduces the positive association betweenpenness and co-dependence.

Overall, our results provide interesting preliminary associa-ions between country policies and systemic fragility; they suggesthat countries which are more integrated, and which have liberal-zed financial systems have higher co-dependence in their bankingectors. However, the negative effects of greater openness and lib-ralization can be mitigated through greater public and privateonitoring.

. Conclusion

This paper examines time-series and cross-country variationn default risk co-dependence in the global banking sector. Weompute weekly changes in default probabilities based on theerton (1974) model for over 1900 banks in over 65 countries.e show that default risk has a significant global component in

he banking sector accounting for 36% of the systematic varia-ion. During the global financial crisis, there has been a uniformncrease in co-dependence across all countries. However, we dond cross-sectional differences in the magnitude of the increasecross different countries. We also find that there has been a sig-ificant increase in default risk co-dependence over the 3-yeareriod leading up to the financial crisis. During this time periode find even greater cross-country variation, with banks located

n the developed countries (and especially banks located in the


S and the European Union) seeing an increase co-dependencehile banks located in developing countries seeing a decrease.

xamining cross-country differences in co-dependence during therisis period, we find that countries that are more integrated

dx.doi.org/10.1016/j.jfs.2014.02.003


al of Fi

wdfieo

isflcltTvsai

A

DaMiaE

R

A

A

A

A

A

A

AA

A

B

B

B

B

B

B

B

B

B

B

B

B

B

B

B

C

C

C

C

C

C

D

D

D

D

D

D

D

E

F

F

F

F

F

G

G

G

H

H

H

H

H

I

J

K

K


ith liberalized financial systems have experienced higher co-ependence in their banking sectors. This negative effect ofnancial openness, however, is mitigated by a strong institutionalnvironment that allows for efficient public and private monitoringf financial institutions.

This paper has important policy implications. Co-dependencen bank default risk has important consequences for systemictability. We find increasing co-dependence in banks located in dif-erent national jurisdictions, particularly in more integrated andiberalized financial systems, which suggests the importance ofoordination in bank supervision and regulation. Our results alsoend support to the view that fostering the appropriate incen-ive framework is very important for ensuring systemic stability.hese incentives are shaped by prudential regulation and super-isory power, existence and generosity of deposit insurance andafety net policies, and availability of information. These findingsre especially relevant for larger banks which have grown morenterconnected over the past decade.

cknowledgements

We thank Martin Cihak, Gordi Susic and participants at theubrovnik Economic Conference and the Development Policy in

Riskier World Workshop at the World Bank for comments andian Wang for excellent research assistance. This paper’s findings

nterpretations and conclusions are entirely those of the authorsnd do not necessarily represent the views of the World Bank, theirxecutive Directors, or the countries they represent.

eferences

biad, A., Detragiache, E., Tressel, T., 2010. A new database of financial reforms. IMFStaff Papers 57 (2), 281–302.

charya, V., 2009. A theory of systemic risk and design of prudential bank regulation.Journal of Financial Stability 5 (3), 224–255.

charya, V., Schnabl, P., 2010. Do global banks spread global imbalances? The caseof asset-backed commercial paper during the financial crisis of 2007–2009. IMFEconomic Review 58, 37–73.

drian, T., Brunnermeier, M., 2011. CoVaR. In: Federal Reserve Bank of New YorkStaff Report 348.

drian, T., Shin, H., 2008. Financial intermediary leverage and value at risk. In:Federal Reserve Bank of New York Staff Report No. 338.

llen, F., Carletti, E., Marquez, R., 2011. Credit market competition and capital reg-ulation. Review of Financial Studies 24 (4), 983–1018.

llen, F., Gale, D., 2000. Financial contagion. Journal of Political Economy 108, 1–33.rteta, C., Eichengreen, B., 2002. Banking crises in emerging markets: presump-

tions and evidence. In: Blejer, M., Skreb, M. (Eds.), Financial Policies in EmergingMarkets. MIT Press, Cambridge, MA.

vesani, R., Pascual, A.G., Li, J., 2006. A new risk indicator and stress testing tool: amultifactor Nth-to-default CDS basket. In: IMF Working Paper No. 06/105.

arlevy, G., Veronesi, P., 2003. Rational panics and stock market crashes. Journal ofEconomic Theory 110 (2), 234–263.

arth, J., Caprio, G., Levine, R., 2008. Bank Regulation and Supervision Database. TheWorld Bank, Washington, D.C.

artram, M., Gregory, B., Hund, J., 2007. Estimating systemic risk in the internationalfinancial system. Journal of Financial Economics 86 (3), 835–869.

attiston, S., Gatti, D., Gallegati, M., Greenwald, B., Stiglitz, J., 2012. Liaisons dan-gereuses: increasing connectivity, risk sharing and systemic risk. Journal ofEconomic Dynamics and Control 36 (8), 1121–1141.

eck, T., Demirgüc -Kunt, A., Levine, R., 2009. Financial institutions and marketsacross countries and over time: data and analysis. In: World Bank PolicyResearch Working Paper 4943.

ekaert, G., Harvey, C.R., 2000. Foreign speculators and emerging equity markets.Journal of Finance 55, 565–613.

ekaert, G., Harvey, C.R., Lundblad, C., 2006. Growth volatility and financial liberal-ization. Journal of International Money and Finance 25, 370–403.

ekaert, G., Hodrick, R.J., Zhang, X., 2012. Aggregate idiosyncratic volatility. Journalof Financial and Quantitative Analysis 47 (6), 1155–1185.

ekaert, G., Wang, X., 2009. Globalization and asset prices. In: Working Paper.Columbia University.


harath, S.T., Shumway, T., 2008. Forecasting default with the Merton distance todefault model. Review of Financial Studies 21 (3), 1339–1369.

ongini, P., Laeven, L., Majnoni, G., 2002. How good is the market at assessing bankfragility? A horse race between different indicators. Journal of Banking & Finance26 (5), 1011–1028.

K

L

nancial Stability xxx (2014) xxx–xxx 11

oyson, N., Stahel, C., Stulz, R., 2010. Hedge fund contagion and liquidity shocks.Journal of Finance 65 (5), 1789–1816.

rasili, A., Vulpes, G., 2005. Co-movements in EU banks fragility: a dynamic factormodel approach. In: Working Paper.

runnermeier, M., Crockett, A., Goodhart, C., Persaud, A., Shin, H., 2009. The funda-mental principles of financial regulation. In: 11th Geneva Report on the WorldEconomy.

runnermeier, M., Pedersen, L.H., 2009. Market liquidity and funding liquidity.Review of Financial Studies 22 (6), 2201–2238.

ampbell, J.Y., Hilscher, J., Szilagyi, J., 2008. In search of distress risk. Journal ofFinance 63, 2899–2939.

han-Lau, J.A., Gravelle, T., 2005. The END: a new indicator of financial and nonfi-nancial corporate sector vulnerability. In: IMF Working Paper No. 05/231.

hava, S., Purnanandam, A., 2011. The effect of banking crisis on bank-dependentborrowers. Journal of Financial Economics 99, 116–135.

hinn, M.D., Ito, H., 2008. New measure of financial openness. Journal of ComparativePolicy Analysis 10, 309–322.

ifuentes, R., Ferrucci, G., Shin, H., 2004. Liquidity risk and contagion. Journal of theEuropean Economic Association 3 (2–3), 556–566.

laessens, S., van Horen, N., 2012. Foreign banks: trends, impact and financial sta-bility. In: International Monetary Fund Working Paper.

e Nicolo, G., Kwast, M., 2002. Systemic risk and financial consolidation. Are theyrelated? Journal of Banking and Finance 26, 861–880.

emirgüc -Kunt, A., Detragiache, E., 1999. Financial liberalization and financialfragility. In: Pleskovic, B., Stiglitz, J. (Eds.), Proceedings of the World Bank AnnualConference on Development Economics.

emirguc-Kunt, A., Detragiache, E., 2005. Cross-country empirical studies of sys-temic bank distress: a survey. In: World Bank Policy Research Paper Number7471.

emirguc-Kunt, A., Huizinga, H., 2004. Market discipline and deposit insurance.Journal of Monetary Economics 51 (2), 375–399.

emirguc-Kunt, A., Kane, E.J., Laeven, L., 2008. Determinants of deposit-insuranceadoption and design. Journal of Financial Intermediation 17 (3), 407–438.

iamond, D., Rajan, R., 2005. Liquidity shortages and banking crises. Journal ofFinance 60 (2), 615–647.

jankov, S., McLiesh, C., Shleifer, A., 2007. Private credit in 129 countries. Journal ofFinancial Economics 84, 299–329.

lsinger, H., Lehar, A., Summer, M., 2006. Systemically important banks: an analysisfor the European banking system. International Economics and Economic Policy3 (1), 73–89.

erreira, M.A., Gama, P.M., 2005. Have world, country, and industry risks changedover time? An investigation of the volatility of developed stock markets. Journalof Financial and Quantitative Analysis 40, 195–222.

inancial Stability Forum, 2009a. Reducing procyclicality arising from the bank cap-ital framework. In: Joint FSF-BCBS Working Group on Bank Capital Issues.

inancial Stability Forum, 2009b. Report of the financial stability forum onaddressing procyclicality in the financial system. In: Working Groups on BankCapital Issues, Provisioning, and Leverage and Valuation.

orbes, K., Rigobon, R., 2002. No contagion, only interdependence: measuring stockmarket comovements. Journal of Finance 57 (5), 2223–2261.

reixas, X., Parigi, B., Rochet, J.C., 2000. Systemic risk, interbank relations and liquid-ity provision by the Central Bank. Journal of Money, Credit, and Banking 32 (3/2),611–640.

ai, P., Kapadia, S., 2010. Contagion in financial networks. Proceedings of the RoyalSociety A 466, 2401–2423.

ennotte, G., Leland, H., 1990. Market liquidity, hedging, and crashes. AmericanEconomic Review 80, 999–1021.

orton, G., 2009. The subprime panic. European Financial Management 15 (1),10–46.

awkesby, C., Ian, M., Ibrahim, S., 2007. Comovements in the equity pricesof large complex financial institutions. Journal of Financial Stability 2 (4),391–411.

eston, S., Rouwenhorst, G., 1994. Does industrial structure explain the benefits ofinternational diversification? Journal of Financial Economics 36, 3–27.

illegeist, S., Keating, E., Cram, D., Lundstedt, K., 2004. Assessing the probability ofbankruptcy. Review of Accounting Studies 9, 5–34.

ong, H., Stein, J.C., 2003. Differences of opinion, short-sales constraints, and marketcrashes. Review of Financial Studies 16 (2), 487–525.

uang, X., Zhou, H., Zhu, H., 2009. A framework for assessing the systemic risk ofmajor financial institutions. Journal of Banking and Finance 33, 2036–2049.

oannidou, V.P., Penas, M.F., 2010. Deposit insurance and bank risk-taking: evi-dence from internal loan ratings. Journal of Financial Intermediation 19 (1),95–115.

agannathan, R., Kapoor, M., Schaumberg, E., 2009. Why are we in a recession? Thefinancial crisis is the symptom not the disease!. In: Northwestern UniversityWorking Paper.

arolyi, A., Lee, K.-H., Van Dijk, M., 2012. Understanding commonality in liquidityaround the world. Journal of Financial Economics 15 (1), 82–112.

odres, L.E., Pritsker, M., 2002. A rational expectations model of financial contagion.Journal of Finance 57 (2), 769–799.

ose, A., Prasad, E., Terrones, M.E., 2006. How do trade and financial integration


affect the relationship between growth and volatility? Journal of InternationalEconomics 69, 176–202.

ongin, F., Solnik, B., 1995. Is the correlation in international equity returns constant:1960 to 1990? Journal of International Money and Finance 14 (1), 3–26.

dx.doi.org/10.1016/j.jfs.2014.02.003

http://refhub.elsevier.com/S1572-3089(14)00026-6/sbref0005




































































































































































































































































































































































































































































































































































































































































































































































































































































































































































































1 al of Fi

M

M

M

M

M

M

N

P

R

R

S

Schuler, M., 2002. The threat of systemic risk in European banking. Quarterly Journal

2 D. Anginer, A. Demirguc-Kunt / Journ

ehrez, G., Kaufmann, D., 1999. Transparency, liberalization, and financial crises.In: World Bank Working Paper.

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

erton, R.C., 1977. On the pricing of contingent claims and the Modigliani–Millertheorem. Journal of Financial Economics 15 (2), 241–250.

orck, R., Yeung, B., Yu, W., 2000. The information content of stock markets: whydo emerging markets have synchronous stock price movements? Journal ofFinancial Economics 58, 215–260.

organ, D., 2002. Rating banks: risk and uncertainty in an opaque industry. Amer-


ican Economic Review 92, 874–888.orris, S., Shin, H.S., 2009. Illiquidity component of credit risk. In: Working paper.

Princeton University.oy, I., 2004. Financial liberalization, prudential supervision, and the onset of bank-

ing crises. Emerging Markets Review 5, 341–359.

Y


opov, A., Udell, G.F., 2012. Cross-border banking, credit access, and the financialcrisis. Journal of International Economics 87 (1), 147–161.

ajan, R., 2006. Has finance made the world riskier? European Financial Manage-ment 12, 499–533.

epullo, R., 2004. Capital requirements, market power, and risk-taking in banking.Journal of Financial Intermediation 13, 156–182.

chroder, M., Schuler, M., 2003. Systemic risk in European banking: evidence frombivariate GARCH models. In: ZEW Discussion Papers 03-11. Center for EuropeanEconomic Research.


of Business and Economics 41 (3), 45–165.uan, K., 2005. Asymmetric price movements and borrowing constraints: a rational

expectations equilibrium model of crisis, contagion, and confusion. Journal ofFinance 60, 379–411.

dx.doi.org/10.1016/j.jfs.2014.02.003




































































































































































































































has the global banking system become more fragile over time?

Documents