A Guide to IMF Stress Testing
Chapter

Chapter 21. Identifying Spillover Risk in the International Banking System: An Extreme Value Theory Approach

Author(s):
Li Ong
Published Date:
December 2014
Share
  • ShareShare
Show Summary Details
Author(s)
Jorge A. Chan-Lau, Martin Čihák, Srobona Mitra and Li Lian Ong This chapter combines material from the International Journal of Finance and Economics (2012), Vol. 17, No. 4, pp. 390–406 (Chan-Lau and others, 2012) and IMF Working Paper No. 07/267 (Čihák Ong, 2007). The authors would like to thank Arabinda Basistha, Jörg Decressin, Dale Gray, Francois Haas, Daniel Hardy, Paul Mills, Jason Mitchell, James Morsink, Klaus Schaeck, Peter Wilding, Bank of England and HM Treasury participants at a seminar held at HM Treasury and participants at the conference on Information in Bank Asset Prices: Theory and Empirics, Ghent University, for useful comments, and Chanpheng Dara for excellent research assistance.

We use the extreme value theory framework to analyze spillover risk across the international banking system. We test for the likelihood that an extreme shock affecting a major, systemic global bank would affect another large local or foreign counterpart, and vice versa. Our results reveal several key trends among major global banks: spillover risk among banks exhibits “home bias”; individual banks are affected differently by idiosyncratic shocks to their major counterparts; and banks are affected differently by common shocks to the real economy or financial markets. In general, bank soundness appears more susceptible to common (macro and market) shocks when the global environment is turbulent; this may have important implications for global financial stability especially during stressful periods. Not surprisingly, our findings also suggest that bank spillover risk has risen over time, which emphasizes the need for continuing collaboration on cross-border supervision and crisis management.

Method Summary

Method Summary
OverviewThe logit model estimates conditional correlations between banks during extreme movements in distance-to-default indicators.
ApplicationThe method is appropriate for modeling spillover risk.
Nature of approachParametric approach—conditional correlations are based on an econometric model.
Data requirementsAccounting information on bank liabilities; market-price-based data on bank equity prices, interest rate (term) spreads, stock market indices.
StrengthsThe results provide insights into the direction of spillovers across banking systems during crises; the methodology is transparent and can be easily replicated; data on actual exposures between institutions are not required in order to deduce spillovers.
WeaknessesThe model assumes that:
  • there are enough extreme movements in the data to provide insights about actual spillovers during a crisis;

  • the movements in the market-based data are sufficient to summarize information on both common exposures and actual intra-institution exposures, when in reality the market does not have adequate information.

ToolThe EViews example program codes are available in the toolkit, which is on the companion CD and at www.elibrary.imf.org/stress-test-toolkit.

Contact author: S. Mitra.

The international banking system has expanded rapidly since the late 1990s. In addition to the overall size of banking assets, the international positions of banks from many of the major banking systems have grown significantly, by several multiples, during this period (Figure 21.1).1 Thus, the banking sector, which continues to be of key importance for global financial stability, represents a potentially important channel for spillovers through the international financial system.

Figure 21.1International Positions by Nationality of Ownership of BIS Reporting Banks

Sources: Bank for International Settlements; and Table 8A of International Banking Statistics.

Note: BIS = Bank for International Settlements.

Shocks to a major global or regional bank, originating from their operations in the home country or in a host country, potentially could affect the international banking system. Spillovers could occur through its linkages with other major banks or through its operations in global financial centers.2 There are numerous specific ways in which this could occur. Notably, external linkages could stem from direct and indirect equity exposures of local banks to overseas banks or, conversely, from shareholdings of local banks by foreign banks; direct exposures through loan books; deposit and funding sources from overseas and/or from foreign banks operating in a particular country; payments and settlement systems; and holdings of credit risk transfer instruments written on assets held by local and/or overseas institutions. Spillovers could also occur without any explicit links between banks when a negative shock in one bank is misinterpreted by investors as a signal of diminished soundness in other banks, either in the same country or in a different country.

Several studies have looked at the various transmission channels for spillovers across banks. These include liquidity shocks affecting one bank, causing deposit runs at other solvent banks (Freixas, Parigi, and Rochet, 2000); shocks via the interbank market when banks withdraw their deposits at other banks (Allen and Gale, 2000; Freixas, Parigi, and Rochet, 2000); or, in the absence of explicit links, difficulties in one market perceived as a signal of possible difficulties in others due to asymmetric information (Morgan, 2002).

Key to the financial stability debate is that banks also have become increasingly intertwined with other segments of the financial sector, both locally and across countries. For instance, banks provide insurers with liquidity facilities to pay current claims and letters of credit as evidence of their ability to pay future claims. The formation of bancassurance groups through the merger of banks with insurers is another example of cross-sector linkages. Banks are also increasingly providing services to investors such as hedge funds, mutual funds, and pension funds, in the form of devising, intermediating, and making markets for financial instruments. Cross-border lending activities also have increased strongly in recent years, especially with the sharp growth in private equity deals.

This chapter focuses on determining spillover risk among some of the largest, systemic banks in the world and within the European Union, using the extreme value theory (EVT) framework. Gropp and Moerman (2004) and Gropp, Lo Duca, and Vesala (2006) apply EVT in testing for spillovers across EU banking systems. It should be noted that the exact nature of the links between the financial institutions is not explored in this study. Rather, the results are intended to represent “maps” that could guide the allocation of limited surveillance and supervisory resources, so that more detailed links may then be identified as necessary. In addition to highlighting the relationships among the major global banks, it could also focus cross-border collaboration among supervisors in these countries.3

Our aim is to identify potential risk concentrations among systemically important banks. Arguably, even if a crisis cannot ultimately be averted, early detection of vulnerabilities in the financial system could provide country authorities with additional time to prepare contingency plans and to focus their attention on likely stress points. To this end, understanding the interdependencies between individual banks with potential systemic impact and their exposure to economic and financial risks is crucial (Bank of England, 2006).

We attempt to answer the following questions on the global and EU banking systems: First, given the increasing internationalization of financial services, are all banks similarly affected by common shocks to the global economy or financial system? Alternatively, is “home bias” the most dominant factor, notwithstanding the effects of globalization? In other words, are banks predominantly influenced by domestic shocks, either because of their domestic focus or the local regulatory environment, despite being largely integrated into the global financial system? Or are different banks—irrespective of domicile—affected differently by shocks, because of their increasingly different business and geographic mixes?

Our results reveal several key trends among major global banks. Notably, “home bias” is an important factor in terms of spillovers. Banks also are affected generally by common shocks to the real economy or financial markets, although the global banking system as a whole tends to be more exposed to these shocks during more turbulent periods, compared with the more benign periods. Further, the risk of spillovers across the major global banks has risen in recent years. Several important relationships are also highlighted at a more specific, bank-by-bank level. Shocks to U.S. banks—especially to Morgan Stanley and Citigroup—appear to have become increasingly important for foreign banks over time, while some of the major U.S. banks appear largely insulated from foreign banks’ idiosyncratic shocks. In contrast, shocks to Japan’s major banks have limited impact on their counterparts and vice versa. Within the EU, ABN Amro and Fortis represent important cross-border spillover risks.

The remainder of this chapter is organized as follows: Section 1 describes the method and data. The empirical analysis of spillover risk across major global banks is presented in Section 2. The analysis is extended in Section 3 to focus specifically on the interlinkages among the largest banks from each EU country. Section 4 concludes the chapter.

1. Empirical Method

The EVT approach to identifying spillovers better captures the information that large, extreme shocks are transmitted across financial systems differently than small shocks.4 Multivariate EVT techniques are used to quantify the joint behavior of external realizations (or “co-exceedances”) of financial prices or returns across different markets. The body of literature on EVT has grown in recent years. Recent empirical work using EVT to model spillovers in financial markets, including bond, equity, and currency markets, are Starica (1999); Forbes and Rigobon (2001); Longin and Solnik (2001); Quintos (2001); Bae, Karolyi, and Stulz (2003); Chan-Lau, Mathieson, and Yao (2004); Hartmann, Straetmans, and de Vries (2004); and Poon, Rockinger, and Tawn (2004). Gropp and Moerman (2004) subsequently apply this approach to changes in the distances to default of 67 individual EU banks.5 They use nonparametric tests of banks’ changes in distances to default to test for spillover risk between two banks, after adjusting for bank size. Separately, Gropp, Lo Duca, and Vesala (2006) use a multinomial logit model to the changes in the distances to default of European banks to determine spillovers among banking systems within the region.

Within this framework, we test for co-exceedances, that is, the likelihood that an extreme shock affecting a major, systemic global bank also would affect another large local or foreign counterpart. We assume that spillover risk is associated with extreme negative comovements in bank soundness measures. In other words, we try to determine whether extreme but plausible negative shocks to a particular bank’s stability could be associated with stresses experienced by other major banks in the international banking system. In our tests, we are able to identify the co-exceedances, or evidence of spillover risk, attributable to idiosyncratic shocks in the banking sector, because we factor out the impact of domestic and global shocks.

Our analysis differs from existing studies in several important ways:

  • We test for spillover risk among individual, systemically important banks, rather than aggregating the effects for a particular country. Gropp and Moerman (2004) and Gropp, Lo Duca, and Vesala (2006) incorporate most listed banks in the EU, in their respective papers, including smaller banks that are likely nonsystemic. This could have the effect of overestimating the impact of certain banking systems on others. Indeed, Gropp and Moerman (2004) observe that “an unreasonable number of very small banks” appear to have systemic importance in their results.

  • We select the 24 biggest banking groups in the world, by total assets, on the basis that these banks could individually pose systemic risk to the domestic or foreign banking systems. In particular, we include institutions from two of the biggest banking systems in the world—that is, Japan and the United States—which contribute significantly to international banking activity. The influence of big banks from Japan and the United States, which represent 9 out of the 24 biggest banks, is likely to be very important, especially given the sharp increase in international banking activity in recent years. The focus on major banks is very pertinent given that the financial authorities in our sample countries are looking to improve cross-border collaboration on supervision issues and are thus highly concerned about the impact of systemically important banks.

  • As an extension, we map spillover risks among major EU banks. Much of the existing literature on the financial soundness of EU banks does not cover spillover effects and focuses instead on the impact from country-level and EU-level shocks.

  • We incorporate local, regional (where relevant), and global market and real economy factors into our models. Given the global nature of our data set, we utilize regional and world stock market indices to capture shocks that are broader in nature, in addition to using individual local stock market indices and domestic interest rate yield spreads to reflect domestic developments.

A. Model

Distance to Default

First, we calculate distance to default (DD) as a comprehensive measure of a bank’s default (solvency) risk. The DD measure is based on the structural valuation model of Black and Scholes (1973) and Merton (1974). The metric represents the number of standard deviations away from the point where the book value of a bank’s liabilities is equal to the market value of its assets, that is, the default threshold. The DD is an attractive measure in that it measures the solvency risk of a bank by combining information from stock returns with information from leverage and volatility in asset values—key determinants of default risk. It does not require specification of a particular channel through which the transmission of shocks occurs. An increase in the DD implies greater stability/soundness, or a lower risk of default.6

Black and Scholes (1973) and Merton (1974) first drew attention to the concept that corporate securities are contingent claims on the asset value of the issuing firm. This insight is clearly illustrated in the simple case of a firm issuing one unit of equity and one unit of a zero-coupon bond with face value D and maturity T. At expiration, the value of debt, BT, and equity, ET, are given by

where VT is the asset value of the firm at expiration. The interpretation of equations (21.1) and (21.2) is straightforward. Bondholders get paid fully only if the firm’s assets exceed the face value of debt; otherwise, the firm is liquidated and assets are used to partially compensate bondholders. Equity holders, thus, are residual claimants in the firm because they get paid only after bondholders.

Note that equations (21.1) and (21.2) correspond to the payoff of standard European options. The first equation states that the bond value is equivalent to a long position on a risk-free bond and a short position on a put option with strike price equal to the face value of debt. The second equation states that equity value is equivalent to a long position on a call option with strike price equal to the face value of debt. Given the standard assumptions underlying the derivation of the Black-Scholes option pricing formula, the default probability in period t for a horizon of T years is given by the following formula:

where N is the cumulative normal distribution, Vt is the value of assets in period t, r is the risk-free rate, and σA is the asset volatility.

The numerator in equation (21.3) is the DD. An examination of equation (21.3) indicates that estimating default probabilities requires knowing both the asset value and the asset volatility of the firm. The required values, however, correspond to the economic values rather than the accounting figures. It is thus not appropriate to use balance-sheet data for estimating these two parameters. Instead, the asset value and volatility can be estimated. It is possible to solve the following equations (21.4) and (21.5) for the asset value and volatility:

if Et the value of equity; σE, the equity price return volatility; and D, the face value of liabilities, are known; and d1 and d2 are given by equations (21.6) and (21.7):

We derive the parameters from market data in the following manner:

  • The time horizon T is fixed at one year.

  • The value of equity, Et corresponds to the market value of the firm. The data are obtained from Bloomberg by multiplying the number of shares outstanding for a firm by the closing share price on a particular day.

  • The equity volatility, σE, corresponds either to historical equity volatility or implied volatility from equity options. This is calculated from the standard deviation of daily share price returns over a one-year period (which we define as 260 days).

  • The face value of liabilities, D, usually is assumed equal to the face value of short-term liabilities plus half of the face value of long-term liabilities.7 This number represents the “default barrier.” The liability data are obtained from Bankscope. The item “Deposits and Short-Term Funding” is used to represent short-term liabilities, while the long-term liabilities are derived by deducting the short-term liabilities from the “Total Liabilities” item. To obtain daily liability data from annual balance sheets, the data are intrapolated between two year-end balances.

  • The risk-free rate, r, is the one-year government bond yield, in the same currency as those of the market and balance sheet data.

Once the asset value and volatility are estimated, the default probability of the firm, and thus the DD, could be derived from equation (21.3).

The Binomial Logit Model

We employ a binomial logit model to determine the likelihood that a large shock to one major bank would cause stress to a large counterpart. Specifically, we apply a model similar to that used by Gropp, Lo Duca, and Vesala (2006) to estimate the probability that the (percentage) change in the DD of one bank falls in a prespecified percentile in the negative tail, following large negative shocks to the DDs in the rest of the banks in the sample and after controlling for country-specific and global factors.

Defining the “Extreme Values.” First, we calculate the percentage change in the DD (which we denote ΔDD) from the generated series of DDs. The ΔDD is calculated over five trading days for the following reasons: (1) extreme events are more significant if they are prolonged; events that last for only a day are of little concern; and (2) the use of weekly changes reduces “noise” in the data.8 Corresponding ΔDDs between banks reflect interdependencies that incorporate all potential channels for spillovers, thus precluding the need to define explicit links between banks or to specify a particular channel for potential spillovers. We define large shocks (or “extreme values”) as the 10th percentile left tail of the common distribution of the ΔDDs across all banks (Figure 21.2).9 We calculate the ΔDD, on a daily basis, as follows:

Figure 21.2Distribution of Changes in Distance to Default, 18-Bank Sample

Source: Authors.

Note: The 10th percentile left tail threshold for the stacked DD data of 18 banks is –0.018.

We then stack all ΔDDit observations from equation (21.8) and calculate the threshold, T10, for the bottom 10 percent tail. For estimation purposes, we initially omit six banks—the three Japanese banks, Credit Agricole (France), Credit Suisse (Switzerland), and Halifax Bank of Scotland (HBOS) (United Kingdom)—because of the shorter periods for which their respective data are available. From the remaining 18 banks over the sample period May 30, 2000, through August 2, 2006, the threshold for the 10th percentile left tail is calculated at –0.018. Observations that fall below this threshold, that is, in the bottom 10 percent tail, are the “extreme values.”

Applying the Econometric Model. A co-exceedance is defined as the probability that a particular bank will experience a large negative shock as a result of shock to another bank in the sample, after controlling for common shocks. That is, the observation of one extreme value as a result of another. The co-exceedances for each bank i at time t are defined as binary variables, yit, such that

where T10 is the 10th percentile threshold in the left tail of the distribution.

We estimate the conditional probability that bank i will be in distress at time t conditional on bank j (j≠i) being in distress, after controlling for other country-specific and global factors, as

which is based on the cumulative distribution function for the logistic distribution, where x represents the explanatory variables F and C, and β represents the slope coefficients α, ρ, γ. The parameter α represents the sensitivity of bank i to real and financial developments in its own country and in the global market, Fit; ρ represents the sensitivity of bank i to extreme shocks it has experienced itself in the previous periods of up to s lags, Cit–s10 and γ represents the sensitivity of bank i to extreme shocks experienced by the rest of the banks in the sample during the previous period, Cji–1 (where j≠i), or, in other words, the co-exceedance of bank i with other banks. All the C variables are lagged by one period to capture the impact on bank i from developments at the other banks, taking into account the differences in trading hours across the different time zones. The other explanatory variables are defined in the next subsection.11

The goodness of fit in logit (and other binary) models is given by the McFadden R2. This statistic is the likelihood ratio index, computed as

where l(β˜) is the restricted log likelihood—this is the maximized log likelihood value when all slope coefficients are restricted to zero and is equivalent to estimating the unconditional mean probability of an observation being in the tail.

Incorporating Nonbank Explanatory Variables

1. Country-specific market shocks

We use the local stock market return volatility to control for country-specific market shocks. We calculate the weekly (five trading-day) returns on each country-specific benchmark stock index by taking the weekly log-difference of the stock index in the local currency. The volatility of returns is proxied by the conditional variance estimated from a generalized autoregressive conditional heteroskedasticity or GARCH(1,1) model of the weekly returns,12 such that

where Xt is the weekly local currency return in the country’s stock price index and σt2 is the GARCH volatility, at time t. The autoregressive conditional heteroskedasticity or ARCH effect is captured by the lagged square residual, εt12 We predict this period’s variance by forming the weighted average of a long-term mean (the constant, w), the forecast variance from the previous period (σt12) and information about volatility observed in the last period (εt12). This model is consistent with the volatility clustering associated with financial returns data, where large changes in returns are likely to be followed by further large changes. Lagrange multiplier tests show significant ARCH(1) effects for all the stock market returns used in this study.

2. Developments in the real economy

We use (five trading-day) changes in term structure spreads to represent expectations of changes in the business cycle in a bank’s home country. Put another way, the changes in the spreads reflect the broader real economy developments in that country. The term structure spread is calculated as the difference between a long-term interest rate (the 10-year government bond yield) and a short-term rate (the one-year government bond yield) in any one country. Thus, the change in yield curve slope—our explanatory variable—is defined as follows:

where yct is the term structure spread at time t.

3. Global market shocks

We apply a global stock market return volatility variable to control for common shocks affecting global markets. This index is published in U.S. dollars but is converted to the currency of the country in which the bank associated with the dependent variable is located. We use the same method as that for the local stock markets and estimate the GARCH(1,1) volatility for the global stock market index.

B. Data

Our global data set includes the world’s top 24 largest exchange-listed banking groups by total assets, as at end-2005, according to Bankscope.13 These comprise institutions from other major banking systems such as France, Germany, Japan, the Netherlands, Switzerland, the United Kingdom, and the United States; Spanish, Belgian, and Italian banking groups also make up the top 24, although banking activity in these three countries is much smaller by comparison. All these banks have a presence in the major financial centers of the world.

Our EU data sample comprises the 33 largest listed EU banks, accounting for about a half of total EU banking system assets. We originally selected the top 50 largest banks in the EU and added the biggest bank in each EU country that would otherwise not have a representative in this category. We then refined the sample to include only banks for which good quality and sufficient data are available, which reduced the sample to 33 banks.14

We use separate control variables to account for common factors affecting local financial markets, the local real economy, and regional and global market developments. Specifically, we incorporate the following:

  • The price return volatility of the local stock market index returns to capture local market influences.

  • Changes in the slope of the local term structure (between one- and 10-year government bonds) to represent developments in the domestic real economy.15

  • The price return volatility in the Morgan Stanley Capital International (MSCI) All-Country World Index returns to account for global market factors.

  • For the EU-specific analysis, the price return volatility in the MSCI All-Country Europe Index returns to account for regional shocks as well.

Balance sheet data for the individual banks, used in all the DD calculations, are obtained from Bankscope, while the requisite market data are available from Bloomberg.16

Several caveats apply to our data. First, some of the banking groups in our sample represent important constituents in their respective country’s stock market indices and may also be represented in the global and regional indices. This means that some of the stock market volatility effects captured in the results could be driven partly by the volatility in the individual bank stocks. This suggests that the impact of idiosyncratic shocks on interbank spillovers represents “conservative” estimates.17 Second, the balance sheet data on banks’ long-and short-term liabilities are available only on an annual basis from Bankscope. Thus, our calculation of daily DDs requires extrapolation between two data points. In this case, we assume that the liabilities change proportionally each day between the two reporting dates. Finally, the DD risk measure does not factor in default risk arising from off-balance-sheet exposures, which could be substantial especially for major international banks engaged in proprietary trading activities.18

2. Default and Spillover Risks for the Global Banking System

A. Analysis

The sample period covered in the analysis is May 30, 2000, to August 2, 2006. However, data for six banks—Credit Agricole (France), Credit Suisse (Switzerland), HBOS (United Kingdom), Mitsubishi UFJ (Japan), Mizuho (Japan), and Sumitomo Mitsui (Japan)—are available only from later dates. Thus, only 18 banks (the “main sample”) are tested for the full sample period; the other banks are subsequently added to the main sample as their data become available, and we rerun the tests for each expanded sample.

An examination of banks’ DD suggests that bank soundness broadly deteriorated across countries during the mid-2000 to mid-2003 period.19 The collective decline in DDs has coincided with the period following the bursting of the global information technology bubble; the slowdown in global economic growth; and the economic and financial difficulties experienced in some Latin American countries, such as Argentina and Brazil, where some of the major banks have direct business interests. The U.K. and U.S. banks appear to have been less affected by the general turbulence, relative to banks from other countries, as their DDs remained relatively stable during this time.

The stresses on the global banking system during the first half of the sample period are also evident from the number of negative extreme values, or left-tail events, across banks.20 The health of the global banking system then improved vastly over the mid-2003 to end-2005 period. The DDs of all banks in our sample rose strongly during this period; correspondingly, the overall number of left-tail events fell substantially. However, the global banking system came under some pressure in 2006. The number of left-tail events increased across many banks in our sample during this period. Although the exact causes of the observed stress are unclear, it has coincided with the oil and commodity price shocks experienced in early 2006, as well as with the sharp corrections in global asset prices observed in the second quarter of 2006.

In the world’s biggest banking market, we find that U.S. banks are vulnerable to spillover risk from banks from their own country as well as from overseas. Table 21.1 suggests that some of these banks also are susceptible to shocks affecting European banks; however, no European bank represents a consistently common spillover risk. Among U.S. banks, Goldman Sachs appears to be largely insulated from external shocks, while shocks to U.S. banks have had some impact on their European counterparts. There appears to be little interaction between several U.S. banks with domestic stock market and interest rate shocks. This suggests that stresses to U.S. banks during this period have not necessarily been tied to developments in the local market or economy. Rather, bank soundness appears to be more closely related to volatilities in global markets, potentially reflecting the global nature of U.S. banking businesses.

Table 21.1Contagion Risk among the World’s Biggest Banking Groups (18 Banks), May 30, 2000, to August 2, 2006(in percent level of significance)
Source: Authors.Note: Factors in the columns are dependent on factors in the rows.Blank cells represent nonsignificance at any level below 10 percent. represents coefficients on own lags, of up to 5 lags. represents significance at the 1 percent level. represents significance at the 5 percent level. represents significance (negative sign) at the 5 percent level or lower. represents the grouping of major banks in any one country, where 3 or more banks are represented in the sample.
Source: Authors.Note: Factors in the columns are dependent on factors in the rows.Blank cells represent nonsignificance at any level below 10 percent. represents coefficients on own lags, of up to 5 lags. represents significance at the 1 percent level. represents significance at the 5 percent level. represents significance (negative sign) at the 5 percent level or lower. represents the grouping of major banks in any one country, where 3 or more banks are represented in the sample.

Spillover risk appears quite significant among European banks. Shocks to ABN Amro (Netherlands) appear to affect banks across several countries, including those in the United Kingdom and the United States, while Societe Generale (France), Deutsche Bank (Germany), and ING (Netherlands) are among those most vulnerable to spillover risk from other major international banks. With the exception of Barclays, U.K. banks appear mostly insulated from spillover risk from foreign banks. Spillovers among same-country banks are difficult to determine, given the limited number of major global banks from each country. Among U.K. banks, for which we have a larger sample, spillover risk is significant. Both Hongkong and Shanghai Banking Corporation (HSBC) and the Royal Bank of Scotland (RBS) are exposed to spillover risk from Barclays; in turn, Barclays appears to be exposed to shocks from both of these banks as well.

Interestingly, our results thus far show some common threads with those of Gropp and Moerman (2004), notwithstanding the differences in time periods, model, and bank samples. Like us, the authors identify ABN Amro (Netherlands) and HSBC (United Kingdom) to be among the more systemically important banks for those outside their own country. They also find close links among banks within countries.

B. Robustness tests with subsamples

Next, we split the sample period into two subsamples, to determine the robustness of our initial findings. A natural structural break would be around mid-2003, which separates the more turbulent period in global economic and market conditions from the benign period that followed. Thus, we define the first subsample as May 30, 2000, to May 30, 2003, and the second as June 1, 2003 to August 2, 2006.

Our results reveal several broad trends among the major international banks. We find evidence of a “home bias” for spillover risk (Table 21.2). For example, spillovers from U.K. banks to local counterparts are more prevalent than to foreign banks. Similarly, shocks to U.S. banks also have a proportionately greater impact on their local counterparts. That said, individual banks are affected differently by idiosyncratic shocks to their major counterparts, possibly because of their different business and geographic mixes. Banks are also not similarly affected by common shocks to the domestic real economy or to financial markets, although the global banking system as a whole tends to be more exposed to these shocks during more turbulent periods, compared with the more benign times.

Table 21.2Significant Co-Exceedances, 2000–2006(in percent of total possible bank transmission channels)
Contagion to Banks in
Initial Shock to

Banks in
Continental

Europe
United

Kingdom
United

States
Continental Europe1749
United Kingdom6676
United States6623
Source: Authors.
Source: Authors.

Importantly, spillover risk across the major global banks has risen in recent years (Table 21.3). The exposure of U.S. banks to each other appears to have intensified; the impact of shocks in U.S. banks on foreign banks has also increased, as have spillovers from U.K. to continental European banks. Individually, shocks to Societe Generale (France), HSBC (United Kingdom), and Morgan Stanley (United States) have had increasingly greater impact on foreign banks. Within Europe, spillover risk from the French banks appears to have increased over time.21

Table 21.3Change in the Number of Significant Co-Exceedances, 2000–2003 to 2003–2006(in percent)
Contagion to Banks in
Initial Shocks to

Banks in
Continental

Europe
United

Kingdom
United

States
Continental Europe200−25
United Kingdom40000
United States300*40
Source: Authors.* The number increased from zero to 1.
Source: Authors.* The number increased from zero to 1.

We subsequently add the remaining six banks to the sample of 18 banks as their data become available, and we rerun the tests for each expanded sample.22 The banks are added in the following order: Mitsubishi UFJ (Japan), HBOS (United Kingdom), Credit Agricole (France), Credit Suisse (Switzerland), Sumitomo Mitsui (Japan), and Mizuho (Japan).23 Our analysis of each of the six sets of results reveals several notable trends:

  • Among U.S. banks, Morgan Stanley and Goldman Sachs largely are insulated from spillovers. Morgan Stanley consistently represents the biggest spillover risk for foreign banks; Citigroup also has become increasingly important in recent years.

  • In continental Europe, shocks to Societe Generale has had the widest impact over time, while banks such as Fortis (Belgium) and Santander (Spain) have become more exposed to shocks from elsewhere. In the United Kingdom, Barclays is the consistent risk factor for its local counterparts, while HSBC remains the most important U.K. spillover risk factor for foreign banks.

  • Spillover risk for major Japanese banks has been limited. Japanese banks appear to pose little risk to the other major international banks, despite the size of the banking system, which is the fourth largest in the world. Similarly, these banks are largely insulated from shocks to foreign banks.

3. Extensions: Spillover Risk Within the EU

We subsequently examine the spillover risks among major banks from EU countries. The sample consists of 33 banks.24 The period covered is May 30, 2000, to April 30, 2007. However, the data for six banks—Bank Austria Creditanstalt (Austria), Credit Agricole (France), Deutsche Postbank (Germany), HBOS (United Kingdom), National Bank of Greece (Greece), and PKO (Poland)—are available only from later dates. Thus, only 27 banks are tested for the full sample period (the “main sample”); the other banks are added subsequently to the main sample as their data become available, and we rerun the tests for each expanded sample. The spillover risks among EU banks are summarized in Table 21.4.25 We derive the following main observations from the results: 26

  • Consistent with the results for global banks, “home bias” is also evident within the EU. Spillovers among banks in the same country appear to be relatively more frequent than among banks from different countries. For the whole sample period, significant spillovers were found in about 40 percent of all possible domestic links, compared with about 9 percent of all possible cross-border links. This result is significant (at the 5 percent level), and it also seems robust over time: for all the subperiods, the relative frequency of co-exceedances among domestic banks was higher than the relative frequency of co-exceedances among banks from different countries.

  • However, cross-border linkages remain an important consideration. The absolute number of significant cross-border spillovers in the EU sample was higher than the number of significant domestic spillovers. This is driven by the number oipotential cross-border linkages among the large banks, which is much higher than the number of potential domestic linkages. So, even with the lower relative frequency, cross-border co-exceedances are more numerous than domestic co-exceedances (57 compared with 19 for the full sample). This finding may seem trivial, but it serves as a reminder that significant cross-border linkages, even if relatively less frequent than domestic linkages, may still be quite important and may require more attention (e.g., in terms of supervisory time) than suggested by the relative frequencies.

  • Moreover, the relative frequency of spillovers across borders has been increasing. When we divide the sample period into two, we find the proportion of banks for which international spillovers is significant rising, from 7.6 percent in May 2000-November 2003 to 8.3 percent in December 2003-April 2007 and 8.7 in November 2005-April 2007; the proportion decreases for domestic linkages, from 28.6 percent in May 2000-November 2003 to 18.8 percent in December 2003-April 2007 and 18.6 in November 2005-April 2007.27

  • The spillover risks posed by the large EU banks differ across institutions. Some banks (e.g., OTP or Bank of Ireland) have no significant spillover impact on other banks, while others have significant impact on a number of domestic and foreign banks at the same time. Interestingly, the bank with the biggest potential for cross-border spillovers within the EU is Fortis (which ranks 19th in the EU in terms of total assets), which has significant impact on eight other banks (six cross-border and two domestic). HSBC is second, with six spillover links (five cross-border and one domestic). The largest number of banks (19) affect between one and three other cross-border banks.

Table 21.4Significant Co-Exceedances among EU Banks, May 2000–April 2007
PeriodNumber of

Significant Links 1
As Percent of All

Possible Links 2
May 2000–April 2007
Domestic1939.6
Cross-border578.7
May 2000–November 2003
Domestic1428.6
Cross-border507.6
December 2003–April 2007
Domestic918.8
Cross-border548.3
November 2005–April 2007
Domestic1318.6
Cross-border868.7
Source: Authors.

Number of bank pairs for which co-exceedances were significant at the 5 percent level in the given period.

Number of significant links in percent of all possible contagion channels (i.e., as a percentage of all possible domestic and cross-border pairings of banks, respectively).

Source: Authors.

Number of bank pairs for which co-exceedances were significant at the 5 percent level in the given period.

Number of significant links in percent of all possible contagion channels (i.e., as a percentage of all possible domestic and cross-border pairings of banks, respectively).

4. Conclusion

This chapter uses market-based indicators to highlight potential interrelationships among the world’s biggest banks from three regions, as well as among the EU’s biggest banking groups, and their exposure to spillover risk from their counterparts. Specifically, the main objective is to identify the direction of spillovers among those banks through the international banking system. In doing so, our results also provide some information on areas where risks may be concentrated, thus highlighting relationships that may require closer supervision and surveillance and a more detailed understanding of linkages by the local authorities. Our findings could also help country authorities focus their collaborative supervisory efforts on specific areas, given their limited resources.

Using an EVT framework, our results yield several clear trends in the interrelationships among the major banks. Overall, the risk of spillovers among local banks is important (“home bias”). Banks also are affected by common shocks to the real economy or financial markets, although they tend to be more vulnerable to these shocks during more turbulent periods, compared with more benign times. Meanwhile, spillover risk among these banks appears to have increased over time.

In light of these findings, ensuring sound risk management practices continues to be a key challenge, both for supervisors and for banks. Given the continuing growth and increased complexity of banks’ businesses, risk management techniques need to be continually enhanced and improved. In many countries, supervisors also are promoting greater use of stress testing as a key risk management tool.

Appropriately, greater emphasis is being placed on improving cooperation in cross-border financial crisis prevention and management. European regulators and supervisors and the European Commission support more efficient, risk-based, cross-border collaboration among supervisors. Internationally, the existing tripartite of Switzerland, the United Kingdom, and the United States is considered one of the most fully developed examples of home—host collaboration in supervision. In other collaborative efforts, the U.K. Financial Supervisory Authority and the New York Federal Reserve have worked closely and continuously with major participants in the credit risk transfer market to resolve the issue of backlogs in trade confirmations and assignments and continue to emphasize the need for “borderless” solutions in the oversight of the credit derivatives market (Geithner, McCarthy, and Nazareth, 2006). EU authorities have also signed the Memorandum of Understanding for crisis management, which includes performing crisis simulation exercises at the EU level. Nonetheless, country authorities largely acknowledge that there is a need for further work on cross-border coordination and information sharing between national authorities in promoting financial stability (Gieve, 2006).

Appendix I. Data Sets
Table 21.5World’s Largest Banking Groups: Exchange-Listed Institutions
Rank by Total Assets as at End-2005Banking GroupNationalityBloomberg TickerDate from Which DD Data Are Available
1Barclays PLC*United KingdomBARC LNMay 17,2000
2UBS AG*SwitzerlandUBSN VXMay 30, 2000
3Mitsubishi UFJ Financial Group, Inc.Japan8306 JPApril 3, 2002
4HSBC Holdings PLC*United KingdomHSBA LNMay 17,2000
5Citigroup, Inc.*United StatesC USMay 17,2000
6BNP Paribas*FranceBNP FPMay 30, 2000
7ING Groep NVNetherlandsINGA NAMay 30, 2000
8Royal Bank of Scotland Group PLC*United KingdomRBS LNMay 17,2000
9Bank of America Corporation*United StatesBAC USMay 17,2000
10Crédit Agricole SAFranceACA FPDecember 16,2002
11Mizuho Financial GroupJapan8411 JPMarch 12, 2004
12JP Morgan Chase & Co.*United StatesJPM USMay 17,2000
13Deutsche Bank AG*GermanyDBK GRMay 30, 2000
14ABN Amro Holding NV*NetherlandsAABA NAMay 30, 2000
15Credit Suisse Group*SwitzerlandCSGN VXJanuary 1,2003
16Société Générale*FranceGLE FPMay 30, 2000
17Banco Santander Central Hispano SASpainSAN SMMay 30, 2000
18HBOS PLC*United KingdomHBOS LNSeptember 11,2002
19UniCredito Italiano SpAItalyUC IMMay 30, 2000
20Morgan Stanley*United StatesMS USMay 17,2000
21Sumitomo Mitsui Financial Group, Inc.Japan8316 JPDecember 3,2003
22FortisBelgiumFORB BBMay 30, 2000
23Goldman Sachs Group, Inc.*United StatesGS USMay 17,2000
24Merrill Lynch & Co., Inc.*United StatesMER USMay 17,2000
Source: Bloomberg.Note: DD = distance to default.

Identified by the Bank of England (BoE, 2006) as large, complex financial institutions (LCFIs), which carry out a diverse and complex range of activities in major financial centers.

Source: Bloomberg.Note: DD = distance to default.

Identified by the Bank of England (BoE, 2006) as large, complex financial institutions (LCFIs), which carry out a diverse and complex range of activities in major financial centers.

Table 21.6Stock Market Indices and Government Bond Yields
Stock MarketBond
CountryIndexBloomberg TickerMaturityBloomben Ticker
BelgiumBEL 20BEL20EUR Belgium sovereign zero coupon yield, 1 year

EUR Belgium sovereign zero coupon yield, 10 year
I90001Y

190010Y
FranceCAC 40CACEUR France sovereign zero coupon yield, 1 year

EUR France sovereign zero coupon yield, 10 year
I01401Y

I01410Y
GermanyDAX 30DAXEUR Germany sovereign zero coupon yield, 1 year

EUR Germany sovereign zero coupon yield, 10 year
F91001Y

F91010Y
ItalyS&P MIBSPMIBEUR Italy sovereign zero coupon yield, 1 year

EUR Italy sovereign zero coupon yield, 10 year
F90501Y

F90510Y
JapanNikkei 225NKYJPY Japan sovereign 10- to 30-year zero coupon yield, 1 year

JPY Japan sovereign 10- to 30-year zero coupon yield, 10 year
F10501Y

F10510Y
NetherlandsAEXAEXEUR Netherlands sovereign zero coupon yield, 1 year

EUR Netherlands sovereign zero coupon yield, 10 year
F92001Y

F92010Y
SpainIBEX 35IBEXEUR Spain sovereign zero coupon yield, 1 year

EUR Spain sovereign zero coupon yield, 10 year
F90201Y

F90210Y
SwitzerlandSMISMICHF Switzerland sovereign zero coupon yield, 1 year

CHF Switzerland sovereign zero coupon yield, 10 year
F25601Y

F25610Y
United KingdomFTSE 100UKXGBP United Kingdom zero coupon yield, 1 year

GBP United Kingdom zero coupon yield, 1 year
I02201Y

I02210Y
United StatesS&P 500SPXUSD Treasury actives zero coupon yield, 1 year

USD Treasury actives zero coupon yield, 10 year
I02501Y

I02510Y
WorldMSCI All-Country WorldMXWD
Source: Bloomberg.
Source: Bloomberg.
Appendix II. Results

Figure 21.3World’s Largest Banking Groups: Distances to Default

Sources: Authors; Bankscope; and Bloomberg.

Figure 21.4World’s Largest Banking Groups: Binomial Logit Representations of the 10th Percentile Left Tail

(“extreme values” or “exceedances”)

Sources: Authors; Bankscope; and Bloomberg.

Appendix III. Binomial Logit Results
Table 21.7Contagion Risk among the World’s Biggest Banking Groups (24 Banks), March 12, 2004, to August 2, 2006(in percent level of significance)
Source: Authors.Note: Factors in the columns are dependent on factors in the rows.Blank cells represent nonsignificance at any level below 10 percent. represents coefficients on own lags, of up to 5 lags. represents significance at the 1 percent level. represents significance at the 5 percent level. represents significance (negative sign) at the 5 percent level or lower. represents the grouping of major banks in any one country, where 3 or more banks are represented in the sample.
Source: Authors.Note: Factors in the columns are dependent on factors in the rows.Blank cells represent nonsignificance at any level below 10 percent. represents coefficients on own lags, of up to 5 lags. represents significance at the 1 percent level. represents significance at the 5 percent level. represents significance (negative sign) at the 5 percent level or lower. represents the grouping of major banks in any one country, where 3 or more banks are represented in the sample.
Appendix IV. EU-Only Sample: Data Sets and Results
Table 21.8EU Banks: Exchange-Listed Institutions
Date of Availability
Stock Market
Major Banking GroupsNationalityStock TickerCurrencyStandard Deviation

of Returns
CapitalizationRisk-Free RateFinancial

Statement
Distance to Default
Erste Bank der Oesterreichischen Sparkassen AGAustriaEBS AVEURρFrom May 30, 2000ρςFrom May 30, 2000
Bank Austria Creditanstalt AGAustriaBACA AVEURFrom July 8, 2004From July 9, 2003ρςFrom 8 July, 2004
Fortis GroupBelgiumFORB BBEURρFrom May 30, 2000ρςFrom May 30, 2000
KBC Group—KBC Groep NV/ KBC Groupe SABelgiumKBC BBEURρFrom May 30, 2000ρςFrom May 30, 2000
Dexia SALuxembourgDEXB BBEURρFrom May 30, 2000ρ1999–2005From May 30, 2000
Danske Bank A/SDenmarkDANSKE DCDKKρFrom May 30, 2000ρςFrom May 30, 2000
BNP ParibasFranceBNP FPEURρFrom May 30, 2000ρςFrom May 30, 2000
Crédit Agricole SAFranceACA FPEURFrom December 16, 2002From December 14, 2001ρ1999–2005From December 16, 2002
Société GénéraleFranceGLE FPEURρFrom May 30, 2000ρςFrom May 30, 2000
NatixisFranceKN FPEURρFrom May 30, 2000ρ1999–2005From May 30, 2000
Deutsche Bank AGGermanyDBK GREURρFrom May 30, 2000ρςFrom May 30, 2000
Commerzbank AGGermanyCBK GREURρFrom May 30, 2000ρςFrom May 30, 2000
Deutsche Postbank AGGermanyDPB GREURFrom June 23, 2005From June 23, 2004ρςFrom June 23, 2005
National Bank of GreeceGreeceETE GAEURρFrom May 30, 2000From August 1, 2000ςFrom August 1, 2000
National Savings and Commercial Bank of Hungary (OTP Bank)HungaryOTP HBHUFρFrom May 30, 20001999–2005From May 30, 2000
Bank of IrelandIrelandBKIR IDEURρFrom May 30, 2000ρςFrom May 30, 2000
Allied Irish Bank PLCIrelandALBK IDEURρFrom May 30, 2000ρςFrom May 30, 2000
UniCredito Italiano SpAItalyUC IMEURρFrom May 30, 2000ρςFrom May 30, 2000
Intesa SanpaoloItalyISP IMEURρFrom May 30, 2000ρςFrom May 30, 2000
ING Group—ING Groep NVNetherlandsINGA NAEURρFrom May 30, 2000ρςFrom May 30, 2000
ABNAmro Holding NVNetherlandsAABA NAEURρFrom May 30, 2000ρςFrom May 30, 2000
PKO BPPolandPKO PWPLNFrom November 9, 2005From November 10, 2004ρ2003–6From November 9,2005
Santander Central Hispano Group—Banco Santander Central HispanoSpainSAN SMEURρFrom May 30, 2000ρςFrom May 30, 2000
Banco Bilbao Vizcaya Argentaría SASpainBBVA SMEURρFrom May 4, 2000ρςFrom May 4, 2000
Nordea Bank AB/Nordea GroupSwedenNDA SSSEKρFrom May 30, 2000ρςFrom May 30, 2000
Skandinaviska Enskilda Banken ABSwedenSEBA SSSEKρFrom May 30, 2000ρςFrom May 30, 2000
Svenska HandelsbankenSwedenSHBA SSSEKρFrom May 30, 2000ρςFrom May 30, 2000
Barclays PLCUnited KingdomBARC LNGBPρFrom May 17, 2000ρςFrom May 17,2000
HSBC Holdings PLCUnited KingdomHSBA LNGBPρFrom May 17, 2000ρςFrom May 17,2000
Royal Bank of Scotland Group PLCUnited KingdomRBS LNGBPρFrom May 17, 2000ρςFrom May 17,2000
HBOS PLCUnited KingdomHBOS LNGBPFrom September 11, 2002From September 10, 2001ρςFrom September 11,2002
Lloyds TSB Group PLCUnited KingdomLLOY LNGBPρFrom May 17, 2000ρςFrom May 17,2000
Standard Chartered PLCUnited KingdomSTAN LNGBPρFrom May 17, 2000ρςFrom May 17,2000
Sources: Bankscope and Bloomberg.ρ At least from May 2000.ς At least from 1999.
Sources: Bankscope and Bloomberg.ρ At least from May 2000.ς At least from 1999.
Table 21.9EU Countries: Stock Market Indices and Government Bond Yields
Stock MarketSovereign Bond
CountryIndexCurrencyBloomberg

Ticker
MaturityBloomberg Ticker
AustriaAustria Traded ATX IndexEURATXEUR Austria sovereign zero coupon yield, 1 yearF90801Y
EUR Austria sovereign zero coupon yield, 10 yearF90810Y
BelgiumBEL 20 IndexEURBEL20EUR Belgium sovereign zero coupon yield, 1 yearF90001Y
EUR Belgium sovereign zero coupon yield, 10 yearF90010Y
DenmarkOMX Copenhagen 20 IndexDKKKFXDKK Denmark sovereign zero coupon yield, 1 yearF26701Y
DKK Denmark sovereign zero coupon yield, 10 yearF26710Y
FranceCAC 40 IndexEURCACEUR France sovereign zero coupon yield, 1 yearI01401Y
EUR France sovereign zero coupon yield, 10 yearI01410Y
GermanyDAX IndexEURDAXEUR Germany sovereign zero coupon yield, 1 yearF91001Y
EUR Germany sovereign zero coupon yield, 10 yearF91010Y
GreeceAthens Stock Exchange General IndexEURASEEUR Greece sovereign zero coupon yield, 1 yearF90401Y
EUR Greece sovereign zero coupon yield, 10 yearF90410Y
HungaryBudapest Stock Exchange IndexHUFBUXHUF Hungary sovereign zero coupon yield, 1 yearF11401Y
HUF Hungary sovereign zero coupon yield, 10 yearF11410Y
IrelandIrish Overall IndexEURISEQEUR Ireland sovereign zero coupon yield, 1 yearF91801Y
EUR Ireland sovereign zero coupon yield, 10 yearF91810Y
ItalyS&P MIB IndexEURSPMIBEUR Italy sovereign zero coupon yield, 1 yearF90501Y
EUR Italy sovereign zero coupon yield, 10 yearF90510Y
NetherlandsAmsterdam Exchanges IndexEURAEXEUR Netherlands sovereign zero coupon yield, 1 yearF92001Y
EUR Netherlands sovereign zero coupon yield, 10 yearF92010Y
PolandWSE WIG 20 IndexPLNWIG20PLN Poland sovereign zero coupon yield, 1 yearF11901Y
PLN Poland sovereign zero coupon yield, 10 yearF11910Y
SpainIBEX 35 IndexEURIBEXEUR Spain sovereign zero coupon yield, 1 yearF90201Y
EUR Spain sovereign zero coupon yield, 10 yearF90210Y
SwedenOMX Stockholm 30 IndexSEKOMXSEK Sweden sovereign zero coupon yield, 1 yearF25901Y
SEK Sweden sovereign zero coupon yield, 10 yearF25910Y
United KingdomFTSE 100 IndexGBPUKXGBP United Kingdom zero coupon yield, 1 yearI02201Y
GBP United Kingdom zero coupon yield, 1 yearI02210Y
RegionMSCI All-Country Europe IndexEURMXER
WorldMSCI All-Country WorldEURMXWD
Source: Bloomberg.
Source: Bloomberg.
Table 21.10Contagion Risk among the Biggest Banking Groups in the EU (27 Banks), May 30,2000, to April 30, 2007(in percent level of significance)
Source: Authors.Note: Factors in the columns are dependent on factors in the rows.Blank cells represent nonsignificance at any level below 10 percent. represents coefficients on own lags, of up to 5 lags. represents significance at the 1 percent level. represents significance at the 5 percent level. represents significance (negative sign) at the 5 percent level or lower. represents the grouping of major banks in any one country, where 3 or more banks are represented in the sample.
Source: Authors.Note: Factors in the columns are dependent on factors in the rows.Blank cells represent nonsignificance at any level below 10 percent. represents coefficients on own lags, of up to 5 lags. represents significance at the 1 percent level. represents significance at the 5 percent level. represents significance (negative sign) at the 5 percent level or lower. represents the grouping of major banks in any one country, where 3 or more banks are represented in the sample.
Table 21.11Contagion Risk among the Biggest Banking Groups in the EU (33 Banks), November 9,2005, to April 30, 2007(in percent level of significance)
Source: Authors.Note: Factors in the columns are dependent on factors in the rows.Blank cells represent nonsignificance at any level below 10 percent. represents coefficients on own lags, of up to 5 lags. represents significance at the 1 percent level. represents significance at the 5 percent level. represents significance (negative sign) at the 5 percent level or lower. represents the grouping of major banks in any one country, where 3 or more banks are represented in the sample.
Source: Authors.Note: Factors in the columns are dependent on factors in the rows.Blank cells represent nonsignificance at any level below 10 percent. represents coefficients on own lags, of up to 5 lags. represents significance at the 1 percent level. represents significance at the 5 percent level. represents significance (negative sign) at the 5 percent level or lower. represents the grouping of major banks in any one country, where 3 or more banks are represented in the sample.
References

Schoenmaker and van Laecke (2006) examine the internationalization of cross-border banking, the public policy issues, and the appropriate supervisory response.

We define contagion risk as the transmission of an idiosyncratic shock from one banking group to another. In other words, by adopting this definition, we are allowing for normal linkages among banks to be a possible channel of contagion and not restricting our analysis to cover only changing cross-market linkages following a shock (referred to as “shift contagion” by Forbes and Rigobon, 2001).

For example, Duggar and Mitra (2009) demonstrate that the major Irish banks are also vulnerable to shocks emanating from the Netherlands and the United States, contrary to the focus of the Irish supervisory authorities largely on the United Kingdom.

For example, Granger causality tests may not accurately depict the common occurrence of extreme events between banks, because relationships between banks are likely to be very different across tranquil and turbulent periods. High correlations in bank soundness during normal times provide little information on the likelihood of contagion. Conversely, contagion—as we define it—during extreme events could well arise from interlinkages that also exist during tranquil times.

The distance to default (DD) is an indicator of default risk based on Merton (1974). Generally, empirical studies have shown that the DD is a good predictor of corporate defaults (Moody’s KMV, or MKMV) and is able to predict banks’ downgrades in developed and emerging market countries (Chan-Lau, Jobert, and Kong, 2004; Gropp, Vesala, and Vulpes, 2004).

Note that DDs are risk neutral, that is, they do not take into account that risk preferences may be different between volatile and benign periods.

This is based on work done by MKMV (see Crosbie and Bohn, 2003).

For instance, stock price returns exhibit day-of the-week effects (French, 1980; Keim and Stambaugh, 1984; Jaffe and Westerfield, 1985; Lakonishok and Smidt, 1988; Chang, Pinegar, and Ravichandran, 1993), while nonsynchronous trading effects related to the overnight or weekend non-trading periods affect the calculation of daily close-to-close returns (Rogalski, 1984), effects of which could be “smoothed” using weekly data.

Ideally, a 1st or even 5th percentile left tail would capture the very extreme events; however, either cutoff would have resulted in much too few observations for this period of data.

This operation adjusts for any serial correlation in the residuals, which may be induced by our use of overlapping weekly ΔDDs.

The results are not significantly different when we apply the Gompertz distribution instead of the logistic distribution.

This method was introduced by Ding and Engle (1991) and subsequently applied by De Santis and Gerard (1997, 1998); Bae, Karolyi, and Stulz (2003); and Ledoit, Santa-Clara, and Wolf (2003).

We originally selected the top 35 largest banks in the world but subsequently refined the sample to the 24 largest exchange-listed banks for which good quality and sufficient data are available. The list of banks in our data set and their corresponding data availability are presented in Appendix I, Table 21.5.

The EU adopted a regulation requiring public companies to convert to International Financial Reporting Standards (IFRS) beginning in 2005. All publicly traded EU companies were required to prepare their consolidated accounts using IFRS from 2005. Thus, the Bankscope balance sheet data from 2005 onward incorporate IFRS requirements.

Details and sources of the market data are presented in Appendix I, Table 21.6.

We test for robustness by omitting the local stock market variable and rerunning the binomial logit model. Our results show that the contagion effects remain largely the same; some of the local market effects are captured by the global market variable. However, the McFadden R2 is slightly stronger for the existing model.

Notwithstanding this limitation, empirical studies have shown that the DD is still a good indicator of default risk in the banking sector (Chan-Lau, Jobert, and Kong, 2004; Gropp, Vesala, and Vulpes, 2004).

The detailed results are available on request.

The 10th percentile left tail for each sample, expanded by one bank at a time, remains at–0.018.

See Appendix III, Table 21.7, for the 24-bank results. Detailed results for the 19—23 bank samples are available on request.

See Appendix IV, Tables 21.8 and 21.9, for lists of exchange-traded banks and associated market data.

See Appendix IV, Tables 21.10 and 21.11, for 27- and 33-bank results, respectively.

As a side result, the tables also show the significance levels for the control variables. In many cases, the “common factors” turn out to be insignificant, but they are significant for some banks. Also, the number of significant cases is higher for the more recent subperiods.

Detailed results are available upon request.

    Other Resources Citing This Publication