A Guide to IMF Stress Testing
Chapter

Chapter 20. External Linkages and Contagion Risk in Irish Banks

Author(s):
Li Ong
Published Date:
December 2014
Share
  • ShareShare
Show Summary Details
Author(s)
Elena Duggar and Srobona Mitra This chapter was previously published in IMF Staf Papers (2009), Vol. 56, No. 4, pp. 758–86 (Duggar and Mitra, 2009). It was prepared in the context of the 2006 Financial Sector Assessment Program Update on Ireland. The authors would like to thank Arabinda Basistha, Martin Čihák, Salim Darbar, James Morsink, Mark O’Brien, and Mark Swinburne for very useful comments and Marianne El-Khoury and Kiran Sastry for excellent research assistance. The authors would especially like to thank staf of the Irish Central Bank and the Financial Regulator for very useful discussions and data.

Increasing financial integration makes the assessment of cross-country linkages crucial for effective financial surveillance. This study estimates contagion risk between large Irish banks and European and U.S. banks during 1994–2005, using distance-to-default measures and the methodology of extreme value theory. Employing an ordered logit model and controlling for Ireland-specific and global shocks, we find evidence of significant contagion risk coming from the United Kingdom, the United States, and the Netherlands toward Ireland. We also find that patterns of contagion to Irish banks have shifted over time, coming from the United Kingdom in the pre-euro period and from the United States in the post-2001 period.

Method Summary

Method Summary
OverviewThe logit model estimates conditional correlations between banking systems 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, and it assumes that 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 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.

In the past decade, Ireland has become increasingly financially open. Global trends in financial liberalization, innovation, and banking consolidation activity, as well as greater bank reliance on wholesale funding, increasing competition from foreign-owned banks, and the development of the Irish Financial Services Center (FSC) have led to increasing international exposure for the banking system. At the same time, the Irish domestic banking system has remained relatively concentrated, with Bank of Ireland (BoI), Allied Irish Banks PLC (AIB), and Anglo Irish Bank Corp. PLC (Anglo IB) representing about 45 percent of total banking assets but nearly 80 percent of the domestic retail market.

Box 20.1.Sources of Contagion—A Literature Survey

Sources of cross-country contagion risk can be grouped into the following:

  • Bank deposit runs. Traditionally, the literature has focused on the implications of bank deposit runs for the payment system, the money supply, and financial intermediation. Liquidity shocks hitting one bank could cause depositors to also run on other solvent banks, in fear of lacking reserves of liquid assets in the banking system (Freixas, Parigi, and Rochet, 2000). The runs could be triggered by rumors on banking system fragility, reputation or operational risk in countries where banks may have subsidiaries, branches, and even representative offices.

  • Wholesale funding channels. With the developments in technological change, deregulation, globalization, and the increased use of financial markets, the focus has shifted to systemic risks arising at the wholesale level (intermediation, investment banking, securities trading, asset management, business banking), and concentrating on the largest and most complex financial institutions.

  • Liquidity and credit risk in interbank markets. One possible channel of contagion is through the interbank market (Allen and Gale, 2000; Freixas, Parigi, and Rochet, 2000) in the form of liquidity shocks when banks withdraw their deposits at other banks, or in the form of credit risk when deposits at other banks are not being repaid.

  • Unidentified channels. There could also be contagion in the financial markets in the absence of explicit links, when in the presence of asymmetric information difficulties in one market are perceived as a signal of possible difficulties in others (Morgan, 2002).

In this study, following De Nicoló and Tieman (2005) and Gropp, Lo Duca, and Vesala (2006),1 we explore whether international openness is accompanied by increasing financial risk for the three major Irish banks.2 We explore trends in international interdependencies between Irish banks and banks in the major European countries and in the United States over the period 1994–2005 and analyze whether international linkages have led to possible contagion risk. We assess whether and to what extent large negative shocks in banks of these countries affected banks in Ireland. In particular, we measure the probability that a higher number of banks in Ireland will experience a large negative shock at the same time as banks in other countries (“co-exceedances”). The existence and the magnitude of these effects have implications for the monitoring of financial stability, as contagion risk is widely perceived to be an important element of banking crises and systemic risk. Understanding the direction of such risk is important for policymakers to better focus limited financial supervisory resources.

We use the term “contagion” to mean the transmission of an idiosyncratic shock affecting one bank or a group of banks to other banks or other banking sectors, using distance to default (DD) as a measure of bank risk. There could be a number of possible contagion channels (Box 20.1) and several ways to measure contagion risk: estimating autocorrelation and survival time tests using historical data on bank failures (Calomiris and Mason, 2000), using the interbank market lending exposure matrix and measuring domino effects of insolvency in one bank spreading to others (Furhine, 2003), or analyzing the correlations of stock returns of banks to measure interdependencies (De Nicoló and Kwast, 2002).

The study follows the approach employed in the recent literature, which estimates contagion risk by using the DD as a comprehensive measure of default risk (Gropp and Moerman, 2004; De Nicoló and Tieman, 2005; Gropp, Lo Duca, and Vesala, 2006). The DD represents the number of standard deviations away from the default point—the point at which the book value of liabilities of the bank is just equal to the market value of assets. Unlike unadjusted stock returns, the DD combines information about stock returns with leverage and asset volatility information, thus encompassing the most important determinants of default risk. The higher the DD, the greater the distance of the bank from default point and the lesser the risk or probability of default of the bank. The benefit of this approach is that co-movements in DD can be analyzed without specifying a particular channel of contagion. Rather, these comovements reflect interdependencies between domestic and cross-border banks encompassing all potential channels of contagion, including that occurring in the absence of explicit links between banks.

We first use rolling correlations of changes in the DD of Irish banks with major European and U.S. banks to analyze trends in cross-country interdependencies. Then, we use an ordered logit model to estimate contagion risk: the probability of Irish banks experiencing a large shock on the same day (co-exceedances) as banks in other countries after controlling for Ireland-specific and global factors—common shocks that could affect all banks simultaneously. Large shocks are defined by the bottom 15th percentile of the weekly difference in the daily DD of all banks. The 15th percentile threshold was used, instead of (say) the bottom 5th percentile, because we wanted to include incidents that have a higher probability of occurrence—those that could be associated with large shocks and not necessarily only with crisis scenarios.

Analyzing trends in rolling correlations in the percentage changes in the DD, we find evidence of increasing simultaneous occurrence of shocks across countries, suggesting increasing global interdependencies over the last decade. Further, following the approach of Gropp, Lo Duca, and Vesala (2006), we find evidence in favor of significant cross-border contagion from the United Kingdom, the United States, and the Netherlands to Ireland. Moreover, it seems that contagion risk has shifted from coming from Europe in the pre-euro period to coming from the United States in the post-euro period. These findings have policy implications for the monitoring of financial stability and could direct scarce supervisory resources in managing foreign exposures.

However, all results come with significant caveats: we are using equity prices of banking groups and yet mainly discussing international links typically associated with banking without explicitly stating links through securities and insurance (the top two Irish banks have links to insurance companies). We are using data available for only a small group of listed banks in each country—so it might not be representative of the system, and the banks in Ireland comprise less than 50 percent of the system (although a much larger share of the retail market). The number of observations experiencing large shocks is low; the results could, therefore, be driven by the large shocks associated with the tech-bubble burst in 2000.

The following section describes various external linkages of Irish banks. Trends in interdependencies using rolling correlations of DDs between the top three Irish banks and banks from other European countries and the United States are analyzed in Section 1. An ordered logit model, to estimate the probability of co-exceedances in Irish banks, is presented in Section 2, while Section 3 summarizes and concludes the discussion.

1. Some Evidence of External Linkages of the Irish Banking Sector

This section aims to find the channels and the direction of potential contagion risk from other countries to Ireland, looking at external linkages of the Irish banks. The Irish financial system has been more closely integrated with the United Kingdom in the past, but links with continental Europe and the United States have strengthened. There are many external linkages of the Irish banking sector, stemming from direct equity exposures in cross-border banks; direct exposure through loan books in other countries; deposit and funding sources from other countries or from numerous foreign banks operating in Ireland; stock market participation in other countries, through securities and asset management firms; and holding of credit risk transfer instruments written on assets located in another country, indirectly exposing Irish banks to international shocks. Each of these is explored in turn.3

First, Irish banks have foreign equity exposure through their expansion overseas. Both AIB and BoI have sizable operations in the United Kingdom and have been growing their niche wholesale international businesses. Both banks own universal banks in Northern Ireland. AIB has a large stake in a U.S. regional bank and majority-owns a Polish bank—the U.S. and Polish investments contributed a combined 16 percent to AIB’s pretax profit in the first half of 2005 (Standard and Poor’s, 2006), although the Polish operations also result in relatively high nonperforming loans.

Second, Irish banks have large loan-book exposures abroad. The two largest banks, AIB and BoI, are geographically diversified—each with almost equal share of domestic and foreign assets. Nearly 28 percent of AIB, 44 percent of BoI, and 41 percent of Anglo IB loan-book exposures were in the United Kingdom. Although AIB held over 20 percent stake in a U.S. bank, U.S. operations were only 2 percent of its loan book. Anglo IB, in contrast, has about 5 percent of its loan book exposed to the United States; without equity exposures in the United States, it operates through a representative office.

Third, the increased reliance on wholesale funding, including interbank borrowing and capital market issues, is another potential source of international interdependencies. The average loan-to-deposit ratio for Ireland exceeds 150 percent, one of the highest for industrial countries. All major Irish banks are dependent for funds on the interbank and securities markets—AIB and BoI fund about 40 percent of lending in the wholesale market, while the market funding requirement for Anglo IB is about 35 percent. The overwhelming bulk of both nonresident interbank borrowing (83 percent) and debt securities issued and held by nonresidents (83 percent) in 2004 were vis-à-vis non-euro area residents.

Fourth, Irish banks buy risk protection mainly from banks of other countries. The underlying asset in structured credit risk transfer (CRT) products include mainly loans and bonds issued by financial and nonfinancial firms; mortgages (for asset-backed securities); and financial and nonfinancial firm debt as underlying asset for the more traditional CRT (e.g., mortgage indemnity guarantee). The United States and the United Kingdom are the main counterparty locations selling risk protection to Irish banks. Other countries include France, Germany, Canada, Switzerland, the Netherlands, Italy, and Poland. The major currencies of denomination are the euro and the dollar, along with the pound sterling. Irish banks also issue covered bonds—BoI has transferred the bulk of its domestic residential mortgage assets to a designated mortgage credit institution, which has a banking license to issue mortgage-covered securities. These covered bonds are used both for hedging interest rate risk and for generating additional funding. Almost 60 percent of these securities are held by other euro area members, while 25 percent are held in dollars by other countries.

Fifth, Irish banks are directly and indirectly exposed to property markets abroad. All the top three banks have loan-book exposures to the United Kingdom property market. At least AIB and Anglo IB sell mortgages in the United States—AIB through its U.S. subsidiary and Anglo IB through its representative office. The latter is more focused on commercial property lending in the United States. BoI has launched a new venture with a leading Spanish bank, La Caixa, to provide extra mortgage options for Irish people buying property in Spain, which includes equity release from existing BoI mortgages. Part of the real estate price risk is mitigated by the Irish banks buying risk protection against these exposures. Irish legislation on covered bonds broadens the scope of risk protection by making loans made from countries such as the United States, Canada, Switzerland, and Japan eligible for the collateral pool.4 However, Irish banks could be indirectly exposed to property markets by selling risk protection (buying of covered bonds, credit default swaps, and mortgage-backed securities) to other banks that are exposed to foreign property markets. From anecdotal evidence, some small FSC banks, exposed to international property markets, are selling credit default swaps to other domestic-oriented banks, indirectly exposing the latter to these property markets even with no loan-book exposure.

Sixth, Bank for International Settlements (BIS) data on banks resident in Ireland show their net asset positions vis-à-vis banks and nonbanks in various countries (Figure 20.1). During 2001–05, the Irish resident banks had a negative net asset position vis-à-vis banks and nonbanks in the United Kingdom; an overall positive net asset position vis-à-vis banks and nonbanks in the United States; and a large positive net asset position with Italy, Spain, and France. However, much of these positions could belong to foreign-owned banks operating through branches in the FSC5 in Dublin—some of these banks operate almost exclusively with nonresidents and have some links, though limited, with the domestic economy.

Figure 20.1Ireland-Resident Banks Net Asset Position vis-à-vis Banks and Nonbanks in Various Countries

(in millions of U.S. dollars)

Source: Authors, based on Bank for International Settlements banking statistics, http://www.bis.org/statistics/bankstats.htm.

Note: AUT = Austria; BEL = Belgium; DEU = Germany; ESP = Spain; FRA = France; GBR = the United Kingdom; ITA = Italy; NLD = the Netherlands.

Seventh, BIS-consolidated statistics report Irish banks’ on-balance-sheet financial claims vis-à-vis the rest of the world. This type of data would give us an idea of the extent of exposure of the Irish banks during a credit event in these countries. Irish banks are exposed mostly to the United Kingdom and Germany among EU countries—and to the United States (although small in comparison) among non-EU countries.

Given the above evidence, we could expect the following possible channels of contagion risk: between Ireland and the United Kingdom and United States, on the basis of equity exposure, loan exposure, and exposure via the interbank and the securities markets; between Ireland and the Netherlands, on the basis of the presence of large Netherlands-owned banks in Ireland; and between Ireland and Italy, Spain, France, and Germany on the basis of interbank market exposure. The direction of contagion is more challenging to establish. However, evidence from BIS data (Figure 20.1) and the discussion in this section suggests the following:

  • Countries with which Ireland has a net nonbank asset position would likely expose Ireland to credit risk in loans and other asset markets: Austria, France, Germany, Italy, the Netherlands, Spain, and the United States.

  • Ireland’s positive net bank exposure in Italy and Spain, and previously the United States, Austria, and France, possibly exposes Ireland to credit risk that could lead to liquidity risk.6

  • Ireland has a negative net bank-asset position, which exposes it to liquidity risk if some problem were to arise in the country from which Ireland is borrowing: Germany, the United Kingdom, Belgium, the Netherlands, the United States, Austria, and France.

There could also be contagion in the absence of explicit links, when due to asymmetric information, difficulties in one country/market are perceived as a signal of possible difficulties in others. The interdependencies mentioned above may be overstated to the extent that they include banks resident in Ireland but operating in the FSC mainly with nonresidents; however, data were not sufficient to exclude these from the BIS data sample. We next look at only the three largest domestic banks—which have nearly 80 percent of the retail market—and their interlinkages and possible contagion risk channels with banks in other European countries and in the United States.

Box 20.2.The Distance to Default Measure and Data Issues

The derivation of distance to default (DD) is described in detail in Gropp, Lo Duca, and Vesala (2006) and in Gropp and Moerman (2004), and in the case of a portfolio of bank assets in De Nicoló and Tieman (2005). The distance-to-default measure is based on the structural valuation model of Black and Scholes (1973) and Merton (1974), and is defined as follows:

where VA,t is the firm’s assets value with mean r and volatility σA, and Xt is the book value of the debt at time t, that has maturity equal to T. The market value of equity of the firm is viewed as a call option on the firm’s assets, VA, with time to expiration equal to T. The strike price of the call option is the book value of the firm’s liabilities, Xt. Default occurs when the value of the firm’s assets is less than the strike price—that is, when the ratio of the value of assets to debt is less than one. The distance to default tells us by how many standard deviations the log of this ratio needs to deviate from its mean in order for default to occur.

An estimation of distance to default requires knowing both the asset value and asset volatility of the firm. The required values, however, correspond to the forward-looking economic values rather than the accounting figures, and it is not appropriate to use balance-sheet data for estimating these two parameters. Instead, the asset value and volatility are estimated using equity data. The distance-to-default measures we use are taken from IMF’s Monetary and Capital Markets Department Distance-to-Default Database, with the methodology described in Vassalou and Xing (2004), except that the value of assets is taken to be equal to the value of equity plus the book value of liabilities. At each date, the value of assets, the return on assets and its volatility are derived using the Black-Scholes option-pricing formula, using one year of daily equity return data preceding the estimation date, and the accounting value of liabilities for the relevant year.

Declines in the (VA,t/Xt) ratio are equivalent to declines in capitalization. Thus, the distance-to-default measure combines information about equity returns with leverage and asset volatility information, hence encompassing the most important determinants of default risk. Empirical studies have shown that the distance to default is a good predictor of corporate defaults (Moody’s KMV), and predicts banks’ downgrades in developed and emerging market countries (Chan-Lau, Jobert, and Kong 2004; Gropp, Vesala, and Vulpes 2006).

We use a data set of daily distance-to-default data for 40 banks in eight countries: France (two banks), Germany (four banks), Ireland (three banks), Italy (six banks), the Netherlands (two banks), Spain (four banks), the United Kingdom (five banks), and the United States (14 banks), for the period January 1994 to November 2005 (Table 20.1). The data set includes all banks in these countries that are listed at a stock exchange and whose distances to default are available from the Distance-to-Default Database. We dropped four banks for which the distances to default were not available for the entire period (one bank for France, one bank for the United Kingdom, and two banks for Italy). In general, the banks in the sample are quite large relative to the population of banks in the European Union, and represent a high fraction of total assets of commercial banks in each country. For each bank, the sample contains 3,105 daily observations (except one U.K. bank with 2,522 observations and one bank from the Netherlands with 2,803).

2. Trends in Interdependences Using Distance-To-Default Indicators

This section describes correlations of banking risks between Ireland and other countries after a discussion of the data. We use the distance-to-default indicator to measure bank financial risk; the DD combines information about stock returns with leverage and asset volatility information, thus encompassing the most important determinants of default risk. The cross-country correlation of changes in the DD would indicate interdependencies arising from a broad set of channels, including contagion occurring in the absence of explicit links between banks.

The DD is based on the Black-Scholes option-pricing model (Black and Scholes, 1973) and is estimated using stock price data (Box 20.2). A bank’s equity is viewed as a call option on the bank’s assets, with strike price equal to the current book value of total liabilities. When the value of the bank’s assets is less than the strike price, its equity value is zero. The DD represents the number of asset value standard deviations that the bank is away from the default point, where the default point is defined as the point at which the liabilities of the bank are just equal to the market value of assets (alternatively, the point where the stock price is zero). The market value of assets is not observable but is estimated using equity values and accounting measures of liabilities.

The banks in the sample are just about five standard deviations from the default point—the (pooled) average DD is 5.2; the median, 5.0. There is some variation among banks—the mean DD by bank ranges from 3.3 (Italy) to 7.7 (Spain). The three Irish banks have a mean DD of about 6 (median about 5.7). Figure 20.2 shows the trends in the systemwide distances to default by country, including an Eastern Europe average comprising Poland and Hungary.7 There seems to be a general increase in the DD starting in 2003 for all countries, indicating a global improvement in bank health in the last two years of the sample period. There also seems to be an upward trend over the past decade for France, Italy, and Eastern Europe. Irish banks suffered a trend decline in the DD around 1997 that continued until end-1999—this partly coincides with the experiences of the United States, the United Kingdom, and (partially) the Netherlands, Germany, and Spain, as does the continued recovery after 2003. They had suffered another negative shock during 2002—likely the aftereffects of September 11, 2001, and possibly connected with the AIB’s U.S. subsidiary scandal in 2002. Irish banks have recovered starting in 2003 and have DD levels comparable to the levels of the late 1990s.

Figure 20.2Hodrick-Prescott (HP) Trend of Banks’ Aggregate Distance to Default (DD) by Country

Source: Authors.

Note: Each panel shows the aggregated DD of sample banks—weighted by assets—in a country. See Table 1 for the sample of banks from each country. The aggregates are HP-filtered. An increase shows a decline in systemic risks.

Table 20.2 shows the correlation of the distances to default between Ireland and the other countries. We find that on average the correlations are positive and quite high for the Netherlands, the United Kingdom, the United States, and Spain. However, if we analyze the past decade in three separate periods—pre-euro period of January 3, 1994, to December 31, 1998; post-euro period of January 1, 1999, to September 11, 2001; and post–September 11 period of September 12, 2001, to November 25, 2005—we find that the correlations become generally much smaller or negative in the post-euro period compared with the pre-euro period and then increase to very high levels in the post–September 11 period.

Table 20.1Sample Banks
BankCountry
1Allied Irish BanksIreland
2ANG.IR.BKIreland
3Bank of IrelandIreland
4CitigroupUSA
5Bank of AmericaUSA
6JP Morgan Chase & Co.USA
7Wells Fargo & Co.USA
8WachoviaUSA
9US BancorpUSA
10Suntrust BanksUSA
11Nat. CityUSA
12Bank of New YorkUSA
13BB&TUSA
14Fifth Third BancorpUSA
15State Street Corp.USA
16KeycorpUSA
17PNC Financial Services GroupUSA
18BarclaysU.K.
19Hongkong and Shanghai BankingU.K.
Corporation Holdings (ORD $0.50)U.K.
20Lloyds TSB GP.
21Royal Bank of ScotlandU.K.
22Standard CharteredU.K.
23Bankgesellschaft BerlinGermany
24Bayerische Hypo-und Vereins BankGermany
25CommerzbankGermany
26Deutsche BankGermany
27BNP ParibasFrance
28Societe GeneraleFrance
29Banco Espanol de CreditoSpain
30Banco Popular EspanolSpain
31Banco Santander Central HispanoSpain
32BBV ArgentariaSpain
33ABN Amro HoldingNetherlands
34Fortis (AMS)Netherlands
35Unicredito ItalianoItaly
36San Paolo IMIItaly
37CapitaliaItaly
38Banca IntesaItaly
39Banca Intesa RNCItaly
40Unicredito Italiano RNCItaly
Source: Authors.
Source: Authors.
Table 20.2Correlations in Distance to Defaults
Pre-EuroPost-EuroPost-September 11
Ireland and1994-20051994-981999-20012001-05
France0.310.34−0.530.89
Germany0.640.61−0.690.81
Italy0.500.370.180.88
Netherlands0.710.61−0.390.94
United Kingdom0.810.680.630.92
United States0.730.450.830.76
Spain0.760.74−0.370.91
Eastern Europe0.03−0.12−0.710.81
Source: Authors.Note: Cross-country correlations in banking systems’ distance to default. The sample covers January 3, 1994–November 25, 2005.
Source: Authors.Note: Cross-country correlations in banking systems’ distance to default. The sample covers January 3, 1994–November 25, 2005.

Figure 20.3 shows the trends in one-year rolling correlations in the DD for all countries, including an average across all country pairs. The latter, which should be more indicative of global trends, seems to be increasing over time, being stronger in the 2003–04 period. It seems that the average pairwise correlation decreased in 1999 with the introduction of the euro and increased after 2001. It also seems that over the last couple of years of the sample period where bank health has been improving, the correlations seem to have diminished. The rolling correlations of Ireland and the United Kingdom seem to have been high during the whole period, with the exception of 1999. The correlations of Ireland with France, Italy, and Spain seem to have increased over time. In general, the correlations of Ireland with continental Europe seem to have been negative during 1999–2001.

Figure 20.3Hodrick-Prescott Trend of One-Year Rolling Correlations of Distance to Default (DD)—Average of All Country Pairs, and between Ireland and Other Countries

Source: Authors.

Note: The top-left corner chart shows the HP-trend of 249-day rolling average correlations of DDs across all country pairs. The rest of the charts show the same for correlations among Ireland and other countries. The HP-filter uses a lambda of 6812100.

Occurrence of shocks, especially large shocks, in banks is captured by the weekly percentage change in the DD for each bank, (ΔDDt / |DDit|). The mean of the percentage change in the distances to default is zero, as expected, and the largest negative change is 237 percent, which represents a sizable shock. Figure 20.4 shows the trends in the percentage change in distances to default for all countries. The most volatile period for Ireland was 1999–2000, which was also volatile for the United Kingdom and France. Conversely, Italy, the Netherlands, Spain, and Germany experienced large shocks in 2001–03. Overall, for most countries, there were fewer shocks in 2003–05 than in the decade before. This could reflect a general drop in credit events across the world mainly due to benign macroeconomic conditions and credit cycle for the last couple of years of the sample period. For instance, FitchRatings (2005) reported a sharp drop in credit events to 37 in 2005 from 94 in 2003.

Figure 20.4Weekly Change in the Distance to Default (DD) by Country

Source: Authors.

Note: The charts show the five-day weekly percent change in the banks’aggregated (weighted by assets) DDs for each country. See Table 20.1 for the sample of banks from each country. The percent changes are expressed as fractions.

In order to describe how interdependencies across countries have evolved over the decade, we try to capture the simultaneous occurrence of shocks across countries through simple correlations (Table 20.3) and the one-year rolling correlations (Figure 20.5) in the percentage change in the DD, between Ireland and the other countries. On average, the correlations between the (ΔDD / |DD|) between Ireland and the other countries are positive. They seem to have a stronger magnitude for the United Kingdom and the United States than for the other European countries. Unlike the correlations in the DD, the correlations in the (ΔDD / |DD|) seem to be strongest in the pre-euro period and weakest in the post-euro period (but still positive), suggesting that the joint improvement in the distances to default in the last couple of years of the sample period were accompanied by fewer simultaneous shocks. Over the decade, the rolling correlations seem to be decreasing over time for Germany, the United Kingdom, and the United States, with no clear trend for the other countries. However, the average of the rolling correlations across all country pairs has a positive trend over the decade, suggesting increasing global interdependencies.

Table 20.3Correlations in Changes in Distance to Defaults
Pre-EuroPost-EuroPost-September 11
Ireland and1994-20051994-981999-20012001-05
France0.320.430.170.42
Germany0.080.450.170.12
Italy0.310.360.260.37
Netherlands0.310.470.230.32
United Kingdom0.450.600.370.44
United States0.360.500.280.34
Spain0.270.410.190.33
Eastern Europe0.020.0300.26
Source: Authors.Note: Cross-country correlations in banking systems’ weekly changes in the distance to default. The sample covers January 3, 1994–November 25, 2005.
Source: Authors.Note: Cross-country correlations in banking systems’ weekly changes in the distance to default. The sample covers January 3, 1994–November 25, 2005.

Figure 20.5Rolling Correlations of Weekly Percent Changes in Distance to Default—Between Ireland and Other Countries

Source: Authors.

Note: Each chart shows the one-year rolling correlations between Ireland and another country’s weekly change in the banking systems’ distance to default; the last chart shows the average of these correlations. The percent changes are expressed as fractions.

3. Contagion Determinants: Multivariate Analysis Using Co-Exceedances

The number of banks in Ireland that experience a large shock on the same day as banks in other countries, after controlling for Ireland-specific and global shocks, is labeled as “co-exceedances.” We use an ordered logit model to estimate such co-exceedances. Such methodology of extreme value theory to assess contagion risk was first proposed by Bae, Karolyi, and Stulz (2003) in the context of stock market returns in emerging markets. Bae, Karolyi, and Stulz (2003) and Gropp and Moerman (2004) show that it is useful to examine only the tails of the distributions of returns and of the distances to default, as the distributions exhibit fat tails, and the correlation among the observations is substantially higher for larger shocks. We use weekly changes in the DD to examine whether shocks in one bank or banking system appear to influence the DD of other banks, controlling for common shocks affecting all banks simultaneously.

A. Methodology and data

The dependent variable is the number of Irish banks simultaneously experiencing large shocks or tail events. We follow Gropp, Lo Duca, and Vesala (2006) in arguing that contagion is associated with extreme negative movements in banks’ default risk. These events can be identified from the negative tail of the distribution of the changes in the DD. We define large shocks as the negative 15th percentile of the common distribution of the percentage change in the DD across all banks. We compute the “co-exceedances” of banks in a given country as the number of banks in a given country that were simultaneously in the tail on the same day (Table 20.4).

Table 20.4Data Description and Summary Statistics, January 3, 1994–November 25, 2005
VariableDefinitionNumber of ObservationsMeanMedianSDMinMax
Bank specific
dditDistance to default (DD) of bank i at time t1,23,3045.195.002.33−2.1627.17
Δddit/|ddit−1|Percentage change in the distance to default1,23,1040.020.003.15−236.53895.32
Country specific
#x0394;ddct/|ddct−1|Percentage change in the distance to default of banking system c at time t
France2,8450.000.010.08−0.260.38
Germany2,8450.010.000.60−22.2511.80
Ireland2,8450.000.000.07−0.340.49
Italy2,8450.010.010.09−0.490.92
Netherlands2,8450.010.010.08−0.570.81
Spain2,8450.000.000.09−0.521.84
United Kingdom2,8450.000.010.07−0.470.35
United States2,8450.000.010.06−0.230.23
Eastern Europe average2,8450.010.010.10−0.680.95
Co-exceedances IRLNumber of banks in the 15th percentile negative tail of Δddit / |ddit−1| in IRL2,8450.360.000.710.003.00
Co-exceedances USNumber of banks in the 15th percentile negative tail of Δddit / |ddit−1| in US2,8451.760.002.880.0014.00
Co-exceedances UKNumber of banks in the 15th percentile negative tail of Δddit / |ddit−1| in UK2,8450.780.001.240.005.00
Co-exceedances GERNumber of banks in the 15th percentile negative tail of Δddit / |ddit−1| in GER2,8450.790.001.080.004.00
Co-exceedances FRNumber of banks in the 15th percentile negative tail of Δddit / |ddit−1| in FR2,8450.330.000.640.002.00
Co-exceedances SPANumber of banks in the 15th percentile negative tail of Δddit / |ddit−1| in SPA2,8450.460.000.860.004.00
Co-exceedances NLNumber of banks in the 15th percentile negative tail of Δddit / |ddit−1| in NL2,8450.320.000.590.002.00
Co-exceedances ITNumber of banks in the 15th percentile negative tail of Δddit / |ddit−1| in IT2,8451.130.001.560.006.00
Memo items
Negative 15th percentile of Δddit / |ddit−1| −0.070
Source: Authors.Note: IRL refers to Ireland, US to United States, UK to United Kingdom, GER to Germany, FR to France, SPA to Spain, NL to Netherlands, and IT to Italy. The banking system changes in DD for each country is calculated as follows: the DD for all the banks is aggregated (weighted by total assets) to get the system-wide DD; the summary statistics of the change in the systemwide DD are reported above.
Source: Authors.Note: IRL refers to Ireland, US to United States, UK to United Kingdom, GER to Germany, FR to France, SPA to Spain, NL to Netherlands, and IT to Italy. The banking system changes in DD for each country is calculated as follows: the DD for all the banks is aggregated (weighted by total assets) to get the system-wide DD; the summary statistics of the change in the systemwide DD are reported above.

Given the discrete ordinal nature of the dependent variable, we use an ordered logit model to estimate the probability of a number of Irish banks simultaneously in the tail as a function of the number of banks in the tail in the other countries, controlling for Ireland-specific and global shocks:

where j = 0, 1, 2 represents the number of banks in the tail simultaneously (“co-exceedances”) in country c—0 if no banks are in the tail, 1 if one bank is in the tail, and 2 if two or more banks are in the tail. The vector Fc comprises Ireland-specific and global shocks affecting Ireland. Cc,t−1 is the lagged number of co-exceedances in country c, and Cd,t represents the co-exceedances in period t in country d. Table 20.5 summarizes such co-exceedances by country.

Table 20.5Description of the Sample by Country
Co-Exceedances = 0Co-Exceedances = 1Co-Exceedances ≥2
CountryNumber of BanksMaximum of Co-ExceedancesCountPercentCountPercentCountPercent
Ireland332,32975.0149716.012798.99
United States14141,61652.0549816.0499131.92
United Kingdom551,90961.4855617.9164020.61
Germany441,71855.3375524.3263220.35
France222,35575.8546514.982859.18
Spain442,23872.0846014.8140713.11
Netherlands222,33075.0456318.132126.83
Italy661,59451.3461119.6890028.99
Total4040
Source: Authors.Note: A “co-exceedance” refers to the simultaneous occurrence of an extreme shock in a number of banks. An extreme shock is defined by one when bank risk (weekly change in the distance to default) increases beyond a 15 percent threshold. This threshold is given by the 15th percentile left-tail cutoff of weekly changes in distance to default of the pooled sample of banks—when a bank’s weekly change in the distance to default ≤ −0.07. Also refer to Table 20.4.
Source: Authors.Note: A “co-exceedance” refers to the simultaneous occurrence of an extreme shock in a number of banks. An extreme shock is defined by one when bank risk (weekly change in the distance to default) increases beyond a 15 percent threshold. This threshold is given by the 15th percentile left-tail cutoff of weekly changes in distance to default of the pooled sample of banks—when a bank’s weekly change in the distance to default ≤ −0.07. Also refer to Table 20.4.

We perform three sets of estimations—the base model, the extended model, and the extended model for each of the three large Irish banks. The base model consists of a set of Ireland-specific and common global shocks, Fc, and persistence in Irish co-exceedances, Cc,t-1. We use four variables to represent Fc:

  • 1. Systemic risk—we use a systemic risk indicator measuring the number of stock markets that are experiencing a large shock at time t. We construct this control similarly to the modeling of large shocks in banks: we use an indicator variable that we set equal to one if a stock market of a given country experienced a shock large enough to be in the bottom 15th percentile of the distribution of the weekly change in returns, and zero otherwise. We calculate such an indicator variable for the euro area stock market (Euronext Top 100 Index, or N100), the U.S. stock market (National Association of Securities Dealers Automated Quotations, or NASDAQ),8 and the Irish stock market (Irish Stock Exchange Overall Index, or ISEQ), from Bloomberg. The systemic risk control is the sum of the indicator variables measuring whether the European stock market, the Irish stock market, and the U.S. stock market were in the tail on a given day, and it ranges from 0 to 3. This control should be positively related to the number of co-exceedances in Ireland.

  • 2. U.S. stock market volatility—we use the weekly change in the volatility of the U.S. stock market to control for volatility spillovers from the United States. There is a large presence of U.S. software firms located in Ireland—nearly all of the largest U.S. software companies use Ireland as their primary base of operations for servicing markets in Europe, Africa, and the Middle East. Also, Irish banks are directly and indirectly exposed to the U.S. stock market through their securities and asset management businesses and through their equity ownership, trade of credit risk transfer instruments, and wholesale funding activities. We estimated the volatility in NASDAQ using a generalized autoregressive conditional heteroskedasticity or GARCH(1,1) model of the form: σt2=α+βσt-12+γεt-12, using maximum likelihood. The regression results are reported in Table 20.6.

  • 3. Irish stock market volatility—similar to U.S. stock market volatility, the weekly change in the volatility of the ISEQ was calculated to account for Ireland-specific shocks.

  • 4. Interest rate shock—the fourth control includes the weekly change in the yield of the Irish 10-year government bond to reflect interest rate shocks.9 We would expect the interest rate control to be positively related to the number of co-exceedances.

Table 20.6Results from GARCH (1,1) Model: σt2=α+βσt12+γϵt12
ConstantCoefficientStandard ErrorZp
U.S. stock market volatility—NASDAQ
7.52E-056.54E-0611.50
ε2t−10.650.0316.970
σ2t−10.380.0124.020
Ireland stock market volatility—ISEQ
7.79E-054.46E-0617.460
ε2t−10.670.0415.340
σ2t−10.220.0212.520
Source: Authors.Note: “U.S. stock market volatility—NASDAQ” (“Ireland stock market volatility—ISEQ”) implies the weekly change in the volatility of the U.S. stock market or NASDAQ (Irish stock market, ISEQ), estimated using a GARCH(1,1) model using maximum likelihood, where σt2 refers to volatility of weekly stock returns.
Source: Authors.Note: “U.S. stock market volatility—NASDAQ” (“Ireland stock market volatility—ISEQ”) implies the weekly change in the volatility of the U.S. stock market or NASDAQ (Irish stock market, ISEQ), estimated using a GARCH(1,1) model using maximum likelihood, where σt2 refers to volatility of weekly stock returns.

The extended model has, in addition, the number of co-exceedances in other countries in the sample—the United States, United Kingdom, Germany, France, Spain, Netherlands, and Italy. The model for each of the three main banks includes, in addition to the extended model regressors, co-exceedances from each of the other two banks.

The summary statistics and the descriptive statistics for the number of co-exceedances per country, that is, the number of banks simultaneously in the tail on a given day, are presented in Tables 20.1, 20.4, and 20.5. Even though the number of banks per country differs somewhat, there is at least one day on which all banks experienced a large adverse shock simultaneously. We limit the number of outcomes to 0, 1, and 2 or more co-exceedances.10 The time pattern of co-exceedances in Ireland is presented in Figure 20.6. The most turbulent period for Irish banks was the period 1999–2000 and the calmest being the mid-1990s and the last couple of years of the sample period.

Figure 20.6Time Pattern of Co-Exceedances in Ireland, January 3, 1994–November 25, 2005

(in number of days each year)

Source: Authors.

Note: The figure shows the number of days in each year, n (n = 0, 1 … 3), banks simultaneously experienced an extreme shock (“co-exceedance”). The extreme event is defined by one when bank risk (weekly change in the distance to default) increased beyond a 15 percent threshold. This threshold is given by the 15 percent left-tail cutoff of weekly changes in distance to default of the pooled sample of banks—when a bank’s weekly change in the distance to default ≤ −0.07.

B. Econometric model results

The results for the basic and the extended models are given in Table 20.7. The dependent variable is the number of banks whose weekly percentage change in the DD was in the 15th percentile negative tail in a given day. In all countries with more than two banks, we limit the model to estimating three outcomes: 0, 1, and 2 or more banks simultaneously in the tail.

Table 20.7Ordered Logit Model: Contagion in Daily Co-Exceedances of the Weekly Change in Distances to Default, January 3,1994-November 25, 2005
IrelandAIBAnglo IBBoI
Basic ModelExtended ModelBasic ModelExtended ModelBasic ModelExtended ModelBasic ModelExtended Model
RegressorsCoeff.SECoeff.SECoeff.SECoeff.SECoeff.SECoeff.SECoeff.SECoeff.SE
Co-Exceedances own lagged2.16**0.072.08**0.083.36**0.143.11**0.163.51**0.143.36**0.153.07**0.142.77**0.15
Systemic risk0.81**0.060.73**0.060.87**0.080.79**0.090.62**0.080.52**0.090.81**0.080.67**0.08
Volatility ISEQ−33.9634.04−34.4133.9737.5653.8911.2355.895.947.39−20.0847.99−19.5641.21−44.740.75
Volatility US−18.6611.74−19.42*11.52−1.9217.4−0.9616.96−1016.1−10.2916.03−28.82*16.67−35.07**15.91
Interest rate IRL2.18**0.411.92**0.413.24**0.583.03**0.591.53**0.591.15**0.592.45**0.542.03**0.55
Contagion FR0.110.080.120.11−0.020.110.130.11
Contagion GER0.010.07−0.040.1−0.020.10.0060.09
Contagion IT0.0020.06−0.130.090.05*0.09−0.050.09
Contagion NL−0.21**0.09−0.36**0.13−0.170.130.0190.12
Contagion SP−0.040.070.21**0.11−0.070.11−0.140.11
Contagion UK0.13**0.0690.17*0.090.150.10.060.09
Contagion US0.27**0.060.20**0.080.23**0.080.38**0.08
AIB0.030.20.51**0.17
Anglo IB0.190.190.34**0.17
Bol0.41**0.180.39**0.19
Pseudo-R20.310.320.3930.4080.3510.3620.3270.347
Source: Authors.Note: Dependent variable: number of Irish banks simultaneously in the tail (“co-exceedances”). “Co-exceedances own lagged” refers to one lag of the dependent variable. “Systemic risk” measures the number of stock markets that are experiencing a large shock at timer. It is constructed from indicators of three stock markets: the euro area stock market index (N100 or Euronext Top 100 Index), the U.S. (NASDAQ), and the Irish stock market (ISEQ or Irish Overall Index) indices. Each indicator is set egual to one if a stock market of a given country experiences a shock large enough to be in the bottom 15th percentile of the distribution of the weekly change in returns, and zero otherwise. Systemic risk is the sum of the indicator variables and it ranges from 0 to 3. “Volatility US” (“Volatility ISEQ”) implies the weekly change in the volatility of the U.S. stock market or NASDAQ (Irish stock market, ISEQ), estimated using a GARCH(1,1) model of the form: σt2 = α + βσt−12 + γεt−12, using maximum likelihood, where σt2 refers to volatility of weekly stock returns (also see Table 20.8). “Interest rate IRL” is the weekly change in the yield of the Irish 10-year government bond. “Contagion&” refers to one lag of co-exceedances in France (FR), Germany (GER), Italy (IT), the Netherlands (ND), and the United States (US). The extreme movements in risk of the three Irish banks are referred to as AIB (Allied Irish Bank), Anglo IB (Anglo Irish Bank), and Bol (Bank of Ireland). The goodness of fit in LOGIT (and other binary) models is given by the “Pseudo-/R2.” This statistic is the likelihood ratio index, computed as R2=1l(β˜)/l(β¯) where l(β¯) is the restricted log-likelihood or the maximized log-likelihood value when all slope coefficients are restricted to zero, which is equivalent to estimating the unconditional mean probability of an observation being in the tail.*p<.05;**p<.01.
Source: Authors.Note: Dependent variable: number of Irish banks simultaneously in the tail (“co-exceedances”). “Co-exceedances own lagged” refers to one lag of the dependent variable. “Systemic risk” measures the number of stock markets that are experiencing a large shock at timer. It is constructed from indicators of three stock markets: the euro area stock market index (N100 or Euronext Top 100 Index), the U.S. (NASDAQ), and the Irish stock market (ISEQ or Irish Overall Index) indices. Each indicator is set egual to one if a stock market of a given country experiences a shock large enough to be in the bottom 15th percentile of the distribution of the weekly change in returns, and zero otherwise. Systemic risk is the sum of the indicator variables and it ranges from 0 to 3. “Volatility US” (“Volatility ISEQ”) implies the weekly change in the volatility of the U.S. stock market or NASDAQ (Irish stock market, ISEQ), estimated using a GARCH(1,1) model of the form: σt2 = α + βσt−12 + γεt−12, using maximum likelihood, where σt2 refers to volatility of weekly stock returns (also see Table 20.8). “Interest rate IRL” is the weekly change in the yield of the Irish 10-year government bond. “Contagion&” refers to one lag of co-exceedances in France (FR), Germany (GER), Italy (IT), the Netherlands (ND), and the United States (US). The extreme movements in risk of the three Irish banks are referred to as AIB (Allied Irish Bank), Anglo IB (Anglo Irish Bank), and Bol (Bank of Ireland). The goodness of fit in LOGIT (and other binary) models is given by the “Pseudo-/R2.” This statistic is the likelihood ratio index, computed as R2=1l(β˜)/l(β¯) where l(β¯) is the restricted log-likelihood or the maximized log-likelihood value when all slope coefficients are restricted to zero, which is equivalent to estimating the unconditional mean probability of an observation being in the tail.*p<.05;**p<.01.

The basic regression results suggest that the probability of Irish banks being simultaneously in the bottom tail varies positively with systemic risk—real shocks experienced through large stock market movements in Ireland and elsewhere increase the probability of a higher number of Irish banks being simultaneously in the tail. However, controlling for global shocks and including that of the ISEQ, Irish co-exceedances still respond positively and significantly to changes in a long-term Irish interest rate. Long-term lending rates (especially mortgage rates) would likely follow movements in long-term government bond yields. Given the large exposure of Irish banks to the real estate market and the prevalence of variable loan rates, sudden increases in interest rates could be associated with credit events that might have a negative impact on banks. In addition, the notion that the number of co-exceedances could be sticky is supported: the lagged (by one day) number of co-exceedances is positive and significant. The stock market volatility controls are not significant, suggesting that the systemic risk variable might be sufficiently capturing stock market spillover effects.

Next, we extend the model to include contagion or co-exceedances from other countries (Table 20.7). We measure contagion by including the first lag of the co-exceedances in the other seven countries. If, after controlling for common shocks, any of these variables turn out to be significant, we interpret this as contagion from that country. We find evidence of contagion from the United Kingdom, the United States, and the Netherlands (with a negative sign) toward Ireland. Adding foreign co-exceedances adds information to the specification, which is reflected in the fact that the significance of the controls remains largely unchanged.

How does the size of the contagion from the United States to the Irish banks compare with that from the United States to other countries? Although the coefficient estimates should not be interpreted as marginal effects, the relative effect of one variable compared with another can be gauged by the relative sizes of the coefficient estimates. To get the relative U.S. effect, we benchmark the U.S. influence with that of the United Kingdom. That is, the relative size of the U.S. influence is the coefficient for the United States divided by the coefficient for the United Kingdom. For the Irish banks, the relative U.S. influence on the three Irish banks is (0.27/|0.13|) 2.1 (from Table 20.7). Next we run similar regressions for each of the other countries (not reported here) and calculate the relative U.S. influence where both the United States and the United Kingdom influences are significant. Figure 20.7 shows a bubble chart for the four euro area countries with significant coefficients for both the United States and the United Kingdom. The relative contagion from the United States to the Irish banks is second only to Germany.

Figure 20.7Relative Size of Contagion from the United States to Selected European Countries

Source: Authors.

Note: The black patches show the relative size of contagion from the United States (U.S.), benchmarked against contagion from the United Kingdom (UK). For example, for Ireland (IRL) it is 0.27/|0.13|=2.1, derived from Table 20.7 (“extended model for Ireland” column). The patches for the other countries—France (FRA), Germany (GER), Italy (ITA), the Netherlands (NLD), Spain (ESP)—are derived from similar regressions that are not reported in the paper. For Italy, the Netherlands, and the United Kingdom, only the U.S. coefficient is shown, since the UK coefficient was not significant for Italy and the Netherlands, and not applicable for the UK.

The final set of estimations involves estimating the extended model for each of the three individual Irish banks (Table 20.7). There is evidence of contagion from the United Kingdom (at the 10 percent significance level), the United States, Spain, and the Netherlands (negative coefficient) toward AIB; contagion from Italy (at the 10 percent significance level) and the United States toward Anglo IB; and from the United States toward BoI. We find evidence of a two-way contagion between AIB and BoI and between Anglo IB and BoI—these could reflect both interbank linkages and off-balance-sheet or derivative positions where one bank is the buyer and the other the seller of various risk protections.

The results for the extended model qualitatively survive several robustness checks. First, when the number of U.S. banks was reduced from 14 to 5 (the top 5), the U.S. influence on Irish banks remained the same. Second, taking monthly changes in DD, rather then weekly changes, does not change the U.S. and U.K. influence; the contagion from the Netherlands disappears. Third, changing the threshold that defines the negative tail of the distribution of large shocks does not qualitatively change the results for the United States and the United Kingdom. Finally, looking at the pattern of contagion risk over time, we find different linkages in the pre-euro, post-euro, and the post–September 11 periods (Table 20.8). We find evidence of contagion from the United Kingdom and Germany (at the 10 percent significance level) in the pre-euro period, from the United States and the United Kingdom (at the 10 percent significance level) in the post-euro period, and only from the United States (at the 10 percent significance level) to Ireland in the post–September 11 period. This suggests a changing pattern of contagion risk, from stronger linkages with Europe in the earlier periods to stronger linkages with the United States later on, consistent with the evidence presented earlier.

Table 20.8Ordered Logit Model: Contagion in Daily Co-Exceedances of the Weekly Change in Distances to Default by Subsample
Pre-Euro PeriodPost-Euro PeriodPost-September 11 Period
(Jan. 1994–Dec. 1998)(Jan. 1999–11 Sept. 2001)(12 Sept. 2001–Nov. 2005)
IrelandCoeff.SECoeff.SECoeff.SE
Co-Exceedances1.73**0.142.20**0.142.05**0.14
IRL lagged
Systemic risk1.17**0.170.46**0.120.76**0.1
Volatility ISEQ−120.12*65.5691.0599.59−27.8660.73
Volatility US33.8648.71−20.65*12.81−12.9839.03
Interest rate IRL2.52**0.611.50*0.91−0.150.94
Contagion FR0.120.14−0.010.150.180.18
Contagion GER0.22*0.120.080.12−0.140.13
Contagion IT0.060.110.0050.11−0.0020.12
Contagion NL−0.240.16−0.140.17−0.030.17
Contagion SP−0.010.13−0.060.140.010.15
Contagion UK0.29**0.120.20*0.12−0.190.13
Contagion US0.140.10.28**0.110.18*0.11
Pseudo-R20.3390.3220.286
Source: Authors.Note: Dependent variable: number of Irish banks simultaneously in the tail (“co-exceedances”). “Co-exceedances IRL lagged” refers to one lag of the dependent variable. “Systemic risk” measures the number of stock markets that are experiencing a large shock at time t. It is constructed from indicators of three stock markets: the euro area stock market index (N100 or Euronext Top 100 Index), the U.S. (NASDAQ), and the Irish stock market (ISEQ or Irish Overall Index) indices. Each indicator is set equal to one if a stock market of a given country experiences a shock large enough to be in the bottom 15th percentile of the distribution of the weekly change in returns, and zero otherwise. Systemic risk is the sum of the indicator variables, and it ranges from 0 to 3. “Volatility US” (“Volatility ISEQ”) implies the weekly change in the volatility of the U.S. stock market or NASDAQ (Irish stock market, ISEQ), estimated using a GARCH(1,1) model of the form σt2 = α + βσt−12 + γεt−12, using maximum likelihood, where σt2 refers to volatility of weekly stock returns. “Interest rate IRL” is the weekly change in the yield of the Irish 10-year government bond. “Contagion &” refers to one lag of co-exceedances in France (FR), Germany (GER), Italy (IT), the Netherlands (NL), Spain (SP), United Kingdom (UK), and the United States (US). The extreme movements in risk of the three Irish banks are referred to as AIB (Allied Irish Bank), Anglo IB (Anglo Irish Bank), and BoI (Bank of Ireland). The goodness of fit in LOGIT (and other binary) models is given by the “Pseudo R2.” This statistic is the likelihood ratio index, computed as R2=1l(β˜)/l(β¯), where l(β¯), is the restricted log-likelihood or the maximized log-likelihood value when all slope coefficients are restricted to zero, which is equivalent to estimating the unconditional mean probability of an observation being in the tail.* p< .05; ** p< .01.
Source: Authors.Note: Dependent variable: number of Irish banks simultaneously in the tail (“co-exceedances”). “Co-exceedances IRL lagged” refers to one lag of the dependent variable. “Systemic risk” measures the number of stock markets that are experiencing a large shock at time t. It is constructed from indicators of three stock markets: the euro area stock market index (N100 or Euronext Top 100 Index), the U.S. (NASDAQ), and the Irish stock market (ISEQ or Irish Overall Index) indices. Each indicator is set equal to one if a stock market of a given country experiences a shock large enough to be in the bottom 15th percentile of the distribution of the weekly change in returns, and zero otherwise. Systemic risk is the sum of the indicator variables, and it ranges from 0 to 3. “Volatility US” (“Volatility ISEQ”) implies the weekly change in the volatility of the U.S. stock market or NASDAQ (Irish stock market, ISEQ), estimated using a GARCH(1,1) model of the form σt2 = α + βσt−12 + γεt−12, using maximum likelihood, where σt2 refers to volatility of weekly stock returns. “Interest rate IRL” is the weekly change in the yield of the Irish 10-year government bond. “Contagion &” refers to one lag of co-exceedances in France (FR), Germany (GER), Italy (IT), the Netherlands (NL), Spain (SP), United Kingdom (UK), and the United States (US). The extreme movements in risk of the three Irish banks are referred to as AIB (Allied Irish Bank), Anglo IB (Anglo Irish Bank), and BoI (Bank of Ireland). The goodness of fit in LOGIT (and other binary) models is given by the “Pseudo R2.” This statistic is the likelihood ratio index, computed as R2=1l(β˜)/l(β¯), where l(β¯), is the restricted log-likelihood or the maximized log-likelihood value when all slope coefficients are restricted to zero, which is equivalent to estimating the unconditional mean probability of an observation being in the tail.* p< .05; ** p< .01.

4. conclusion

This chapter examines the external linkages of the Irish banking sector and estimates an indicator of potential contagion risk—arising from idiosyncratic shocks in other countries—that the three major Irish banks may be exposed to. Aggregate balance sheet data of Irish-resident banks suggest several channels of external interdependencies, including foreign equity exposures, loan-book exposures abroad, and wholesale funding through interbank and capital market issues. However, apart from these links, there could also be foreign exposures through credit risk transfers (e.g., Irish banks selling risk protection to cross-border banks that could be subject to credit events in other countries) and through operational risks that are difficult to measure but can quickly lead to large fluctuations in bank stock prices.

Given that Irish-resident banks include a large number of foreign banks operating in the Financial Services Center that have limited Irish linkages in the retail market, we focus on the three major listed banks—BoI, AIB, and Anglo IB—which have nearly 80 percent of the domestic retail market. We proxy banking risk by distance-to-default measures constructed from bank equity prices and look at correlations of changes in the DD of Irish banks with banks in other countries. Following Gropp, Lo Duca, and Vesala (2006), we then define large changes in DD of each bank—co-exceedances—if the changes fall below the 15th percentile of the negative tail of the joint distribution of the distance-to-default changes across all banks of the sample. We use an ordered logit model to estimate the probability of the number of Irish banks being in the tail at the same time as banks in other countries, controlling for Irish-specific and global factors.

There is evidence suggesting contagion risk from the United States and the United Kingdom to Ireland, although the size of the U.S. influence dominates that of the United Kingdom in almost all regressions and across many robustness checks.11 There are obvious balance sheet linkages of Irish banks with the United Kingdom through subsidiaries. The United States and the United Kingdom also remain the major countries selling risk protection to Irish banks. Aggregate data show that Irish-resident banks (but not necessarily the three banks in the empirical study) have positive net-asset exposures in the United States, which could give rise to credit risk from exposure to nonbank assets and liquidity risk from Irish banks’ (net) interbank borrowing from the United States, from mid-2004. Conversely, Irish-resident banks have very large and negative net-asset exposure to the United Kingdom, suggesting that risks in Irish banks’ on-balance-sheet exposures to the United Kingdom might have been mitigated by off-balance-sheet risk protection bought from United Kingdom banks. Still, both the United States and the United Kingdom have had booming property markets—Irish banks are exposed to them and could be affected in the event of a substantial downturn in any of these markets.

Some tentative policy lessons could be drawn from the results of this exercise. The Central Bank and Financial Services Authority of Ireland may want to stress test specific categories of exposures of Irish banks to both the United States and the United Kingdom. Even though linkages with the United States do not come out strongly from aggregate consolidated balance sheet exposures, there might be derivatives or other off-balance-sheet exposures that the bank supervisors may need to be vigilant of. The Irish authorities may need to collect more information about types and counterparties of derivative positions and risk transfers through structured products of Irish banks, as the use of these is likely to grow rapidly in the future.12 This would especially be necessary if Irish banks are buying CRT products from foreign banks (that is, selling risk protection) that are in turn exposed to property markets or other loan products in the United States or the United Kingdom—thus exposing the Irish banks to these markets even though there is no direct loan exposure.

Finally, some caveats apply to the econometric results. We are using equity prices of banking groups and yet mainly discussing international links typically associated with banking without explicitly stating links through securities and insurance (the top two banks are also involved in insurance and securities business). We are using data available for only a small group of listed banks in each country—so it may not be representative of the system, and the banks in Ireland comprise less than 50 percent of the system in terms of total assets (although nearly 80 percent of the retail market). The number of observations experiencing large shocks is low; the results could therefore be driven by the large shocks associated with the tech-bubble burst of 2000.

References

Gropp, Lo Duca, and Vesala (2006) find evidence of significant cross-border contagion in Europe (Ireland not included in the sample) during January 1994 to January 2003.

The choice of the banks was guided by the availability of data on their equity prices.

Some empirical evidence of foreign contagion risk in Ireland is provided by Gropp and Moerman (2004), who study the joint occurrence of both positive and negative extreme shocks in banks’ DD among large European Union banks in the period January 1991 to January 2003 and identify AIB and BoI as among the systemically important banks in the European Union. Moreover, they find a contagion effect from the United Kingdom to Ireland.

With the exception of Luxembourg, most European countries limit the asset pool to European Economic Area assets.

The FSC was established in Dublin in 1987 (known at that time as the International Financial Services Center, or the IFSC) to facilitate financial operations with nonresidents and was endowed with corporate tax benefits. These tax benefits have now been lifted. There are about 450 international institutions that have established offices in the FSC (accounting for about 40 percent of banking assets), including 50 percent of the 50 largest financial institutions in the world—including major banks from the United States, United Kingdom, the Netherlands, Italy, and Germany. In turn, almost all domestic credit institutions also conduct business from the FSC.

The Irish banking supervisors have a specific requirement for banks on liquidity maintenance: 25 percent of deposits, and short-term liabilities have to be covered by liquid assets.

Weighted average DD of each listed bank, weighted by assets.

The Dow Jones Industrial Average gives similar results.

This assumes that foreign interest rate shocks are reflected in the movement of Irish interest rates and in stock market movements.

The number of co-exceedances depends on the number of banks included in the sample and may not necessarily reflect the strength of the banking system per se. Still, comparing countries with equal number of banks in the sample suggests that Spanish banks tended to experience fewer shocks compared with German banks and that Dutch banks tended to be somewhat less frequently subject to shocks compared with French banks.

Econometric estimates (not reported) suggest that although there is no evidence of contagion from Europe to the United States, the United States is a source of contagion not only for Ireland but also for all the other European countries. Germany seems to be mainly a receiver of contagion risk rather than a source, which is consistent with the evidence from BIS that Germany is mainly “a buyer” of risk.

See Chan-Lau and Ong (2006) for the regulatory and supervisory initiatives taken by the U.S. Office of the Comptroller of the Currency and the U.K. Financial Services Authority in this regard.

    Other Resources Citing This Publication