The Euro and cross-border banking: evidence from bilateral data.

Blank, Sven ; Buch, Claudia M.

MOTIVATION

Has the Euro promoted financial integration in Europe? (1) This question has been addressed in a series of research papers in recent years. This literature shows that money markets are highly integrated and that the introduction of the Euro has been a catalyst for the integration of bond and equity markets. As opposed to this, the integration of banking markets lags behind. Cappiello et al. (2006) point at an increased co-movement of returns in Euro Area bond and equity markets. They also find changes in the term premia following the introduction of the Euro. Hartmann et al. (2003) find strong integration of money markets as well as a convergence of bond and equity markets. However, the integration of banking markets remains incomplete. Lane and Walti (2006) conclude that membership in the Euro Area has strengthened bilateral financial linkages. At the same time, however, they find that global factors are becoming increasingly important.

Only a few studies use bilateral data to assess the impacts of intra-European capital flows and asset holdings. The available evidence points to a positive effect of the Euro. Using data on bilateral portfolio investments and estimating gravity models, De Santis and Gerard (2006) and Lane (2006) show that the Euro has stimulated regional financial integration. Lane finds that common membership in the Euro Area roughly doubles bilateral bond holdings. According to Lane and Walti (2006), similar studies show that bilateral equity holdings increased by about two-thirds.

Why should the Euro have promoted financial integration in Europe? In the literature on international trade in goods, four factors are singled out as potentially contributing to a trade-enhancing effect of currency unions. Bilateral trade costs, markups, or marginal costs of firms may decline, and the number of traded varieties may increase (see eg Baldwin, 2006). As regards international financial integration, similar cost reductions may come into play. These could stimulate inter-regional financial linkages. In contrast, however, portfolio effects could lower inter-regional financial integration. The elimination of exchange rate risks implies that the hedging properties of Euro Area assets and liabilities may deteriorate. Hence, investors may look for diversification potential outside rather than inside the Euro Area.

In this paper, we apply a gravity model to international banking data to estimate the effect of the Euro on cross-border assets and liabilities of banks. Drawing on recent work by Baldwin (2006) and Baldwin and Taglioni (2006), we particularly study the sensitivity of standard gravity models to control for unobserved heterogeneity across countries. We provide evidence on the determinants of banks' bilateral cross-border assets and liabilities over the past decade, focusing on OECD countries and on developments inside and outside the Euro Area. We concentrate on the OECD because gross foreign assets have increased in particular for these countries and because we want to study a relatively homogenous group of countries. We focus on international bank assets since international debt instruments, assets as well as liabilities, still account for about 200% of GDP for industrialised countries and about 100% of GDP for emerging markets and developing countries (Lane and Milesi-Ferretti, 2006a, b). Equity investments are a little more than half as important.

Our research is related to three sets of studies on the determinants of cross-border financial integration. A first set of studies looks at the determinants of aggregated asset holdings of countries. The recently updated data set by Lane and Milesi-Ferretti (2006a), for instance, provides information on aggregated foreign assets and liabilities. However, by focusing on aggregated positions, the effects of international integration on intra-regional asset holdings cannot be analysed.

A second set of papers explains the patterns of cross-border banking assets. These studies show that explaining cross-border capital flows is much more difficult than explaining stocks of foreign assets and liabilities. Buch (2003), for instance, studies a panel of bilateral cross-border asset holdings of banks which report to the Bank for International Settlements (BIS). She finds that, apart from market size, regulations and information costs affect the patterns of cross-border asset holdings. Bilateral distance is used as one proxy for information costs in this study. Aviat and Coeurdacier (2007) use a similar cross-sectional data set of banks' foreign assets and show that the negative impact of distance on asset holdings is largely due to the positive link between trade and international bank assets. The response of bilateral bank lending to cyclical factors has been studied less frequently. Buch et al. (2005), for example, use a data set similar to ours, but focus on a shorter time period (1999-2003). Goldberg (2005) uses bank-level data of US banks. These papers suggest that the impact of macroeconomic variables on changes in cross-border positions is rather weak and not very stable over time.

A third set of papers studies regional financial integration using bilateral data. Kim et al. (2006) use portfolio and banking data to study international financial integration in Asia. Using a gravity-type approach, they find East Asia to be more closely integrated with global markets than among each other. In terms of financial integration in Europe, the papers closest in spirit to ours are by De Santis and Gerard (2006) and Lane (2006). De Santis and Gerard analyse the degree of portfolio reallocation that has taken place in international portfolio investments between 1997 and 2001. Besides, they link the degree of portfolio reallocation to the introduction of the Euro. Their data set comprises bilateral bond and equity holdings for 23 developed countries and seven emerging markets, using the change in portfolio weights between the end of 1997 and the end of 2001 as the dependent variable. They find that financial integration is not a global phenomenon. Instead, they show that the degree of equity and bond home bias has declined significantly only among the European countries, Australia, New Zealand, and Singapore. Also, their results show that the degree of regional integration inside the Euro Area has increased. Lane (2006) uses levels of bilateral bond holdings and changes in these bond holdings between 1997 and 2004 and likewise finds a positive Euro effect.

In contrast to these papers, we focus on banking assets and liabilities, and we have annual information on cross-border financial linkages at hand, rather than data for two points in time. We have data from reporting countries inside the Euro Area, Belgium, Germany, France, Italy, the Netherlands, and outside the Euro Area, Hong Kong, Japan, Switzerland, UK, USA. As to recipient countries, we use information on all OECD members. In terms of the time series dimension, our data set covers the pre-Euro period 1994-1998 as well as the post-Euro period 1999-2004. We also study cross-border lending and borrowing instead of only focusing on cross-border asset holdings.

In the next section, we discuss how the Euro effects can be measured using a gravity framework. In the subsequent section, we describe our data, and provide descriptive statistics on changes in cross-border financial linkages over time, focusing in particular on the re-allocation of assets and liabilities towards the Euro Area. In the penultimate section, we supply empirical evidence on the determinants of banks' foreign activities, using gravity-type equations. The last section concludes and summarises the main results. Overall, our paper provides evidence of a positive Euro effect for assets holdings inside the Euro Area. Evidence for a positive effect on cross-border liabilities is less strong.

MEASURING THE EURO EFFECT

In the international trade literature, Rose (2000) proposes to use gravity models for estimating the effects of currency unions on trade. He uses a gravity regression including a 0/1-dummy that switches on if two countries form a monetary union. The baseline gravity model is given by:

In [X.sub.ij] = [[alpha].sub.0] + [[beta].sub.1]ln[Y.sub.i] + [[beta].sub.2]ln[Y.sub.j] + [[beta].sub.3]ln[D.sub.ij] + [[beta].sub.4]CU + [[epsilon].sub.ij] (1)

where X is the bilateral trade between reporting country i and recipient country j, Y the GDP, D the distance, and CU the dummy for currency union.

Rose (2000) finds a very large impact of currency unions on trade. According to the original estimates, forming a common currency would boost bilateral trade by about 200%. Subsequent work has modified his model by using different empirical techniques and different types of data sets while trying to account for problems due to omitted variables, reverse causality, and model misspecification. Rose and Stanley (2005) place the currency union effect on bilateral trade in the order of magnitude ranging from 30% to 90%. Baldwin (2006, p. 48) provides a critical review of literature on the, as he calls it, 'Rose effect' of currency unions and concludes that 'I believe that we can be fairly sure that some form of Rose effect is occurring in the Eurozone.' However, he suggests the size of this effect only in the order of magnitude of 5%-10%.

Baldwin and Taglioni (2006) point to three main methodological problems of estimating gravity-type regressions in order to find effects of policy measures such as the Euro effect.

The 'gold medal mistake': Omitted variables in gravity regressions which are correlated with trade costs, or--in our case--with costs of cross-border financial transactions, might be correlated with the error term. Omitted factors could be related to characteristics of the country pair that are unobservable for the researcher. One remedy is to include country-pair fixed effects, which pick up this unobserved heterogeneity in the data. Unobserved heterogeneity could also vary over time, which would require using time-varying country-pair fixed effects. However, for all practical purposes, including time-varying country-pair fixed effects has not been feasible in our application. Note that inference about the effects of currency unions in specifications with pair fixed effects amounts to so-called difference-in-difference estimations. Essentially, the estimators compare the before-and-after difference for the Euro Area countries to the before-and-after differences for the non-Euro Area countries. Hence, the identification of the Euro effect derives from the time-series variation in the data, since all cross-country variation is absorbed by fixed effects.

The 'silver medal mistake': Whereas the theoretical gravity equation for trade is typically based on exports, empirical gravity models often use the sum of exports and imports as the dependent variable. This may affect the results because the log of averages does not correspond to the average of the logs. In our empirical application below, we will avoid this problem as we will consider foreign assets and liabilities of banks separately.

The 'bronze medal mistake': Deflating all nominal variables, as is often done in gravity regressions, might induce spurious correlations if there are global trends in inflation. One way to address this problem is to include time dummies in all regressions which capture common business cycle developments. We will follow this approach in our empirical application. Note also that Melitz (2006) defends using data in real constant US dollars to avoid distortions in relative prices to affect results.

Baldwin (2006) also discusses the possible impact of non-linearities in the data, which might be picked by the CU-dummy. In order to moderate the effects of non-linearities, he suggests estimating the CU-effect on a homogenous set of countries, and including non-linear terms for GDP. For this reason, we focus on data for OECD countries only and test the robustness of our results using squared GDP. Besides, we estimate the Euro effect on a reduced sample, namely the more homogenous sub-group of EU countries.

DATA AND DESCRIPTIVE STATISTICS

In this section, we describe the data used in this paper. Details on the data specification are given in Table 1 and the Appendix.

Dependent variables

Our aim within this paper is to analyse the effects of the Euro on cross-border assets and liabilities of commercial banks. Our data on banking sector linkages come from the BIS. We have quarterly data for the years 1995-2004 on bilateral foreign assets and liabilities for 10 BIS reporting countries, Belgium, Germany, France, Hong Kong, Italy, Japan, Netherlands, Switzerland, UK, US. (2) Concerning the recipient countries, we have information on virtually all countries worldwide. Yet, in order to avoid problems due to non-linearities, we restrict our sample of the recipient countries to the OECD area.

We employ data at annual frequency to eliminate seasonal effects and to match the time dimension of some of our explanatory variables. We use the domestic consumer price index to convert the data into real terms. The data are aggregated across banks in each reporting country, but they are specified by the country of destination. For each reporting country, we have data on loans (assets) and deposits (liabilities). These comprise two types of cross-border financial positions: first, loans and deposits vis-a-vis non-residents in all currencies and, second, loans and deposits vis-a-vis non-residents in foreign currency. In terms of data classification, we follow the BIS's convention (BIS, 2006). With regard to loans, the reporting country is a creditor vis-a-vis a foreign country, whereas reported deposits reflect claims of a foreign country vis-a-vis a reporting country.

The BIS collects information from national central banks on the cross-border assets and liabilities of commercial banks. Whereas the reporting area has formerly been restricted mainly to OECD countries, the set of countries has been enlarged to further include large emerging markets and financial centres. Until recently, however, data on bilateral activities among the BIS reporting countries have not been published by the BIS. Hence, we resort to unpublished data, which have kindly been made available by the BIS' Statistics Department. These data do also allow for an analysis of the assets and liabilities among the reporting countries for an extended range of time.

The BIS publishes two sets of banking statistics. The locational statistics are based on the balance of payments principle, that is they include all assets and liabilities of residents vis-a-vis non-residents. In principle, these data are available since the early 1970s on a bilateral basis. In addition to aggregated positions by country, the BIS also collects breakdowns into different types of borrowers, banks and non-banks, and on the currency composition of foreign assets and liabilities. This information is important when calculating changes in cross-border assets and liabilities which are the result of exchange rate valuation changes. In addition to information on banks' total assets and liabilities, we have information on the amounts denominated in Euro, in Yen, in Pounds Sterling, in Swiss Francs, in US dollars, and in other currencies. Each position is given in US dollars.

In contrast to the locational statistics, the second set of statistics, the BIS' consolidated statistics consolidate inter-office positions among banks and their foreign affiliates. Hence, the consolidated statistics provide a more detailed picture of the banks' exposures from specific reporting countries to foreign countries. The consolidated statistics are also more detailed with regard to the sector coverage than locational statistics. However, no breakdown into different currencies is available.

For all countries, except Belgium and the UK, we have information on the currency composition of assets and liabilities, which is displayed in Figure 1. Overall, the share of assets and liabilities denominated in Euro has not changed significantly for the considered reporting countries. If anything, the share of assets denominated in Euro has risen slightly except for Switzerland and the United States. Despite these relative steady patterns in the currency of denomination, valuation changes have, of course, a potential impact on the changes in cross-border assets and liabilities of banks. Therefore, we estimate our main model using data in constant US dollars and additionally provide an estimation based on data that are unadjusted for exchange rate effects. To obtain data in constant US dollars, we convert the original series into constant US dollars (BIS, 2006). Since our data are initially denominated in US dollars, we first transform these series of assets and liabilities back into its original currency for each period t, [x.sub.t.sup.NC], and then adjust for valuation changes by applying the following formula: [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] where [e.sub.0.sup.(NC/USD)] is the exchange rate of the national currency to the US dollar (NC/USD) at the beginning of the sample period, and k is the number of currencies.

[FIGURE OMITTED]

Explanatory variables

Our set of explanatory variables follows the international finance literature; see Aviat and Coeurdacier (2007) for a recent study using banking data similar to ours. As our main explanatory variables, we include domestic and foreign real GDP, bilateral distance, and a common language dummy. The expected signs for GDP and for the common language dummy are positive: international financial linkages should be stronger between large countries and between countries which share a common cultural background. The expected impact of distance is negative, as distance proxies for information costs and for the impact of trade.

We successively add the log of bilateral trade and risk and return proxies. Bilateral trade is included to accommodate the finding by Aviat and Coeurdacier (2007) that bilateral bank assets holdings and bilateral trade are strongly linked and that a negative impact of distance on asset holdings captures to a large extent the impact of distance on trade. The expected impact is positive.

Our main return proxies are the domestic and the foreign real interest rate. The expected impact of the interest rate variables differs between foreign assets and foreign liabilities. Generally, we expect to find a positive (negative) impact of the foreign interest rate on cross-border financial assets (liabilities). We expect the reverse signs for the domestic interest rate. However, we also note that earlier literature using gravity models for capital flows has had difficulties in establishing the expected signs on return proxies.

The main explanatory variables of interest are dummy variables which pick up the fact whether two countries are members of the Euro Area. These dummy variables are specified as follows. A dummy for Euro Area membership of the recipient countries is equal to 1 after 1999 for Austria, Belgium, Finland, France, Germany, Greece after 2001, Ireland, Italy, Luxembourg, the Netherlands, Portugal, and Spain. By analogy, we do not have data for all these countries as reporting countries. Hence, the corresponding dummy for the reporting countries is equal to 1 for Belgium, Germany, France, Italy, and the Netherlands and zero for Hong Kong, Japan, Switzerland, the UK, and the USA. To test whether the introduction of the Euro has strengthened intra-Euro Area financial linkages, we create a combined dummy which indicates whether both, the recipient and the reporting country count among Euro Area members; see Lane (2006) for a similar specification. Note that our identification of the Euro Area effect is based, within the specifications using country-pair fixed effects, on the time series variation in this variable. Besides, we split the sample into EU and non-EU countries. We do so by using a dummy for EU membership which is equal to 1 for Austria after 1995, Belgium, Czech Republic after 2004, Denmark, Finland after 1995, France, Germany, Greece, Hungary after 2004, Ireland, Italy, Luxembourg, the Netherlands, Poland after 2004, Portugal, Slovakia after 2004, Spain, Sweden after 1995, and the United Kingdom.

Our explanatory variables come from the OECD and national sources, and the data have been retrieved via Datastream. We deflate nominal values with the respective national consumer price index. The real interest rate is given by the difference between the nominal interest rate and the percentage change in the consumer price index. The exchange rate series are obtained from Datastream as well. To avoid structural breaks, the exchange rate series for Euro Area member countries are denominated in local currency units versus the US dollar. Even after the introduction of the Euro as a single currency, we adhere to this procedure. Thus, we multiplied the exchange rate denoted in Euro per US dollar with the official conversion rate of the respective member country.

Descriptive statistics

Common consensus suggests that the financial markets of the Euro Area countries have become more integrated over the past years. At the same time, the speed of integration has differed across financial market segments. Despite the on-going integration of financial markets and the deregulation of cross-border banking activities, national banking systems in the Euro Area remain distinctly shaped by their individual characteristics (see eg Sorensen and Puigvert Gutierrez, 2006).

In this section, we check whether the introduction of the Euro has led to a re-structuring of banks' portfolios in favour of the Euro Area or against it. We follow the literature that has studied the impact of the Euro on international portfolio choices. One of the more recent studies uses the IMF's International Portfolio Investment Survey (De Santis and Gerard, 2006) to analyse whether the introduction of the Euro has reduced the home bias in investment portfolios. This is done by comparing the structure of international investment portfolios to a benchmark portfolio, using cross-sectional data for the years 1997 and 2001. One main finding of this study points out that the adoption of the Euro has reduced the home bias on national equity and bond markets inside the Euro Area. This result is based on the observed increase in total asset shares held within the Euro Area member countries (see also Lane, 2006).

The set up of their study differs from our approach in several regards. We use time series banking data as opposed to only cross-sectional evidence. Further we restrict the sample to 10 instead of 30 reporting countries. Yet, it is instructive to look at our data in a similar way. Figure 2 thus plots the shares of assets and liabilities that the reporting countries hold vis-a-vis the Euro Area as a percentage of total cross-border assets and liabilities. For some reporting countries such as the United States, the United Kingdom, Japan, and the Netherlands, these shares have remained relatively stable over the reporting period. For Switzerland, the Euro Area has become less attractive as a destination of banks' foreign activities. Concerning banks in Belgium and France, the Euro Area has gained importance only with respect to assets.

Figure 2 hints at significant differences in the importance of Euro Area assets and liabilities, which are correlated with the distance between markets. For instance, the United States has relatively small financial linkages with the Euro Area. Overall, evidence provided in Figure 2 does not lend strong support to the hypothesis that the introduction of the Euro induced banks to significantly restructure their international portfolios in favour of the Euro Area.

[FIGURE 2 OMITTED]

IS THERE A EURO EFFECT IN CROSS-BORDER BANKING?

Model specification

We now turn to a more systematic analysis of the determinants of banks' foreign assets and liabilities. We use a work-horse of the international trade and, to an increasing degree, of the international finance literature, the gravity model. Theoretical foundations for empirical applications of the gravity model in international finance are less well-developed than in the trade literature. One of the seminal papers in this field is by Martin and Rey (2004) who develop a gravity-type relationship for international equity holdings. Our baseline testing equation is thus based on:

log ([X.sub.ijt]) = [[alpha].sub.0] + [[beta].sub.0] log(GD[P.sub.it]) + [[beta].sub.1] log (GD[P.sub.jt]) + [[beta].sub.2] log([dist.sub.ij]) + [[beta].sub.3][r.sub.it] + [[beta].sub.4][r.sub.jt] + [[beta].sub.5][D.sub.EMU] _ [[epsilon].sub.ij] (2)

where log([X.sub.ijt]) is the log of bilateral bank assets (liabilities), GD[P.sub.it] (GD[P.sub.jt]) is domestic (foreign) GDP, [dist.sub.ij] is the bilateral distance between the two countries as a proxy of information and trade costs, [r.sub.it] ([r.sub.jt]) is a proxy of the return on banks' assets and liabilities on the domestic (foreign) market, and [D.sub.EMU] is a dummy for EMU membership. In the panel estimates of equation 2, we follow Arellano (1987) and compute robust standard errors which allow for both heteroscedasticity and autocorrelation of arbitrary form. Further, in compliance with the approach of Aviat and Coeurdacier (2007), we take the home bias of banks for granted. Our analysis differs from their method, as we use panel rather than cross-sectional data.

We specify our empirical gravity model in three different ways (Baldwin, 2006):

(i) country dummies for reporting and recipient countries plus time fixed effects,

(ii) time-varying country fixed effects for reporting and recipient countries,

(iii) country-pair dummies, time fixed effects,

(iv) country-pair dummies, time fixed effects, robustness tests.

Distance and other time-invariant country-pair characteristics can be included in specifications (i) and (ii) but not in models (iii) and (iv). (3) Estimating the model using time-varying country-pair fixed effects is not possible. We use specification (iii) to vary the set of control variables. Starting from a baseline gravity regression, we include domestic and foreign GDP, distance, and a dummy for members of the Euro Area. We then add the following variables successively: log bilateral trade as the sum of exports and imports to account for the fact that trade and financial integration are related, real interest rates to include a measure of relative rates of return across countries, and squared domestic and foreign GDP to account for possible non-linearities within the data.

Regression results

Results for foreign assets are reported in Table 2; results for foreign liabilities are reported in Table 3. Results show the success of gravity models in explaining bilateral cross-border financial linkages. Domestic and foreign GDP as well as geographic distance have a significant impact of cross-border assets and liabilities, and the signs are as expected. In regressions including time and country fixed effects, gravity models explain about 90% of the variation of assets and liabilities across countries and across time. The impact of domestic GDP is positive with an elasticity of about 0.6; the impact of foreign GDP lies within a similar range. The distance coefficient is comparable among foreign assets and liabilities (-0.6 to -0.7). Speaking a common language also increases cross-border financial linkages, but the variable is only significant at the 10% level.

Running a fixed effects model (columns 3) yields an overall [R.sup.2] of only about 35% for foreign assets and 14% for foreign liabilities. This result shows that the fixed effects pick up a significant part of the variation in the dependent variable. Hence, unobserved heterogeneity is substantial in the baseline model.

Turning next to the main variable of interest in this paper, the coefficient of the Euro Area dummy, we find a positive and significant impact on both, foreign assets and foreign liabilities. The coefficient estimates drop as we move from the OLS regressions including dummy variables to the panel fixed effects estimator. In the baseline panel specifications, we obtain a coefficient estimate of 0.6 for the foreign assets and 0.3 for the foreign liabilities. Since this is a 0/1-dummy variable, the obtained elasticities of foreign assets and liabilities with regard to the Euro area dummy amount to [e.sup.0.6] = 1.8 and [e.sup.0.3] = 1.3. Hence, bilateral assets (liabilities) in the Euro Area are about 80% (30%) higher than those among OECD countries that are not members of the Euro Area.

For foreign assets, this Euro Area effects remain positive and significant, even if we include other explanatory variables or if we change the set up of our empirical model. Moving from an OLS to a panel specification lowers the estimated EMU-effect, but does not change its significance (Columns 1 and 2 versus Column 3 of Table 2). The point estimate for the EMU dummy is also smaller for the EU-only than for the full sample, but it remains significant (Column 4). Including bilateral trade, domestic and foreign real interest rates, and an EU dummy, as is done in Columns (5) through (7) again has no impact on the estimated EMU effect. Finally, estimating the model separately for foreign assets vis-a-vis non-banks (Column 8) shows a significant EMU effect that is even higher than for the full sample. Note that there are some changes in the significance of the GDP terms. For the EU sample, domestic GDP is insignificant. For non-bank assets, foreign GDP is insignificant.

For foreign liabilities (Table 3), the evidence of a robust Euro Area effect is less clear-cut. While we obtain a significant coefficient in the baseline specification, the coefficient becomes insignificant in the specification using EU countries only. Also, we find an insignificant impact of the Euro Area dummy for foreign liabilities of non-banks. Taken together with the results for foreign assets, this would suggest that cross-border lending to non-banks has become easier in the Euro Area while non-banks find it more difficult to borrow from abroad. Generally, finding weaker evidence of a Euro effect for non-banks is consistent with earlier results suggesting that banking integration is relatively high at the wholesale but not at the retail level {Lane and Walti, 2006, p. 6).

For domestic and foreign GDP, we find similar changes in sign and significance as in the equations for foreign assets. Interestingly, most of the additional control variables, which were insignificant in the asset equations, are now significant. More specifically, a greater volume of bilateral trade increases bilateral liabilities, but this effect becomes insignificant if we account for the endogeneity of trade using IV estimates (results not reported).

Robustness tests

We check the robustness of our results in a number of ways. We begin by re-running the baseline model dropping each of the recipient and the reporting countries successively. In unreported regressions, we find that the effects of the Euro Area dummy do not change. Also, estimation of the panel for each of the reporting countries separately hints at a strong positive Euro Area effect with respect to the Euro Area reporting countries. The latter regressions rely on foreign assets as the dependent variable. In the regressions using foreign liabilities, we find a negative Euro effect for the non-Euro Area reporting countries. Subsequently, we add additional explanatory variables, which have been used in the literature on international bank assets. The following results are not reported but available upon request.

Exchange rate effects: We account for exchange rate effects in different ways. First, we include the bilateral exchange rates to the US dollar to pick up valuation effects in the data. The expected signs of these variables are not clear since we lack detailed information about the full currency composition of banks' assets and liabilities. The main results are unaffected, except for domestic GDP which becomes insignificant. In addition, we use foreign assets and liabilities expressed in constant US dollar to reduce the impact of valuation effects on our results. The main results remain unaffected. Most importantly, the Euro Area effect is positive and significant throughout for banks' foreign assets whereas it becomes insignificant for foreign liabilities in the same specifications as above. To further account for exchange rate effects, we include a proxy for exchange rate volatility, but the previous results are not touched.

UK effect: We follow De Santis and Gerard (2006) and specify dummies for the United Kingdom as the reporting and the recipient country as well as for the Euro Area being the reporting and the United Kingdom being the recipient region. Our main results are unchanged.

Return proxies: Since our dependent variable is a composite measure of banks' holdings of loans and deposits, bonds, and equities across countries, it is difficult to compute the effective rates of return, and the corresponding volatilities and correlations, of these assets. In addition to measuring returns through real interest rates, we follow Aviat and Coeurdacier (2007) and construct a measure for risk-adjusted returns by dividing the means of the stock returns by their variance. Since these data contain large outliers, we truncate the risk-adjusted return series at the 10th and 90th percentile. We also enter the two components of this measure separately, that is we use the variance of stock market returns as a measure for risk. These proxies for risk and return may better account for investors' opportunity costs and diversification motives than merely looking at the real interest rate differential. We find that foreign risk-adjusted returns have a weakly significant impact on assets. This impact, however, differs from the one found by Aviat and Coeurdacier (2007): risk-adjusted foreign stock market returns turn out to have a negative, albeit very small, effect on assets. The same holds true for effect of domestic risk-adjusted returns on liabilities. In addition, the variance of domestic stock returns seems to exert an attracting influence on cross-border liabilities. In a sense, our findings thus mirror the results of Portes and Rey (2005) who failed to find the expected effects of risk and return measures on cross-border equity flows.

Corruption: We include an index reflecting the severity of corruption attributed to a country by Transparency International, again similar to Aviat and Coeurdacier (2007). Note that this index should be better called 'index of reliability' since a lower score measures a higher degree of corruption. Since countries with more reliable institutions should positively affect the investment climate, we expect more cross-border assets and liabilities in countries with less corruption. Interestingly, the perceived level of corruption does not significantly affect cross-border asset holdings, but less corruption at home and abroad lowers bilateral liabilities.

Quarterly frequency: Finally, instead of aggregating our data at an annual frequency, we use the data at the original quarterly frequency. Results are unchanged.

SUMMARY AND OUTLOOK

This paper has studied the effect of the Euro on intra-EU banking sector linkages. We use a new panel data set on banks' foreign assets and liabilities covering the pre- and the post-Euro area. In terms of country coverage, we have information on 10 reporting countries at hand of which one half lies inside and the other half lies outside the Euro Area. We use all OECD countries as recipient countries. Our empirical methodology is a gravity regression using a full set of country-pair and time fixed effects to account for unobserved heterogeneity across countries. Our study has the following findings:

For banks' foreign assets, we find a positive and significant increase in intra-Euro Area financial linkages following the introduction of the Euro in 1999. This result is robust against the inclusion of different sets of explanatory variables, and it is significant--albeit smaller--when restricting the sample to the EU countries only.

For banks' foreign liabilities, we find a significant Euro effect as well, but this effect turns out to be insignificant for the EU sub-sample, for regressions including exchange rate effects, and for the foreign liabilities of non-banks.

In terms of other control variables, we find the most robust results for market size--which increases cross-border financial linkages--and distance--which lowers cross-border financial linkages. The effect of other explanatory variables, in particular referring to those measuring returns on financial markets, is less clear-cut.

Overall, our results complement earlier research on financial integration in Europe by showing a positive Euro effect. Since we have focused on banks' cross-border assets and liabilities only, we have nothing to say about the degree of integration of banking markets relative to other financial market segments. In this sense, our findings are not inconsistent with literature showing a relative small degree of integration of national banking markets. In terms of the potentially offsetting effects of lower transaction costs and reduced diversification potential inside the Euro Area, our results support the dominance of the transaction costs effect, in particular for banks' foreign assets.

DATA APPENDIX

Common language: Taken from the Centre d'Etudes Prospectives et d'Informations Internationales (CEPII, www.cepii.fr). A dummy variable that indicates whether two countries share a common language (which is spoken by at least 20% of the population).

Corruption index: Yearly corruption perceptions index taken from Transparency International for each country (http://www.transparency.org). A low score indicates a higher perceived level of corruption.

Cross-border assets and liabilities: Locational statistics of the BIS: Worldwide international on-balance sheet assets and liabilities of BIS reporting banks, covering international positions of banks' head offices in the source countries and all offices at home and abroad, in million US dollar. The data are defined as in Tables 2A, 2B, 3A, 3B and 5A of the BIS Quarterly Review (BIS, various issues). Unpublished bilateral data have kindly been provided by the Statistics Department of the BIS. Details on the data definition are given in the main text.

Distance: Taken from CEPII. Bilateral distances between countries are calculated by using a formula that takes the most populated cities of each country into account.

Exchange rates: National currency against the US dollar, provided by Datastream. Exchange rates of members of the European Monetary Union are expressed in the former national currency versus the US dollar by multiplying the exchange rate of the Euro versus the US dollar by the official conversion rate of the respective EMU member country.

Gross domestic product (GDP): Seasonally adjusted data as provided by the OECD, in million US dollars. Owing to lack of availability or short length of the time series, seasonally unadjusted data were used for Iceland, Luxembourg, Mexico, Poland, Sweden, Turkey, and Hong Kong, with this last GDP taken from national sources as reported by Datastream. These series are adjusted using quarterly dummies. Data for the Netherlands were taken from the International Financial Statistics (IMF 2006).

Interest rates: For most countries, we use a monthly average of the 3-month interbank offered rate as reported by Datastream. We take 90-day certificates of deposits for Japan, Korea, and the US and treasury bills with the same maturity for Australia, Canada, Hungary, Iceland, New Zealand, and, Sweden. The interest rate series for Luxembourg was taken from Belgium.

Prices: Represented by each country's consumer price index taken from Datastream, not seasonally adjusted.

Stock market indices: Provided by Datastream. We take the following indices: All Ordinaries (Australia), ATX (Austria), BEL (Belgium), TSX (Canada), PX-50 (Czech Republic), KFX (Denmark), HEX (Finland), SBF (France), DAX (Germany), ATG (Greece), BUX (Hungary), ICEX (Iceland), Price Index of Ordinary Stocks & Shares (Ireland), MIB (Italy), TOPIX (Japan), KOSPI (Korea), IPC (Mexico), Amsterdam SE All Share (Netherlands), OSEBX (Norway), WIG (Poland), PSI (Portugal), SAX (Slovak Republic), Madrid General Index (Spain), Affarsvarlden Index (Sweden), ISE National (Turkey), FTSE All Share (UK), Dow Jones Industrial Average (US), and Hang Seng (Hong Kong); owing to limited data availability, we assigned Belgium and Australian stock market data to Luxembourg and New Zealand, respectively.

Trade volume: Calculated as the sum of bilateral exports and imports. Bilateral trade data are taken from 'Direction of Trade Statistics on CD-Rom' (DOTS) of the International Monetary Fund. Data are denominated in US dollar. Because data for Belgium is only available since 1997 and only the total value of exports and imports for Belgium and Luxembourg together are available before that date, we assign 90% of these values to Belgium's exports and imports for the missing observations.

Note: Quarterly data have been converted into annual data using the mean values for each year except for GDP where we use the sum of the quarterly observations.

REFERENCES

Arellano, M. 1987: Computing robust standard errors for within-groups estimators. Oxford Bulleting of Economics and Statistics 49(4): 431-434.

Aviat, A and Coeurdacier, N. 2007: The geography of trade in goods and asset holdings. Journal of International Economics 71(1): 22-51.

Baldwin, R. 2006: The Euro's trade effects. Working Paper 594, European Central Bank: Frankfurt a.M., March.

Baldwin, R and Taglioni, D. 2006: Gravity for dummies and dummies for gravity equations. NBER Working Paper 12516, National Bureau of Economic Research: Cambridge, MA, September.

Bank for International Settlements (BIS). 2006: Guidelines to the International Locational Banking Statistics. Monetary and Economic Department: Basle.

Bank for International Settlements (BIS). (various issues): BIS Quarterly Review. Bank for International Settlements (BIS): Basle.

Buch, CM. 2003: Information or regulation: What drives the international activities of commercial banks. Journal of Money, Credit, and Banking 35(6): 851-870.

Buch, CM, Carstensen, K and Schertler, A. 2005: Macroeconomic shocks and foreign bank assets. Working Paper 1254, Kiel Institute for World Economics, July.

Cappiello, L, Hordahl, P, Kadareja, A and Manganelli, S. 2006: The impact of the Euro on financial markets. Working Paper 598, European Central Bank: Frankfurt a.M., March.

De Santis, R and Gerard, B. 2006: Financial integration, international portfolio choice and the European monetary anion. Working Paper 626, European Central Bank: Frankfurt a.M., May.

Goldberg, L. 2005: The international exposure of U.S. banks. NBER Working Paper 11365, National Bureau of Economic Research: Cambridge, MA, May.

Hartmann, P, Maddaloni, A and Manganelli, S. 2003: The Euro area financial system: Structure, integration, and policy initiatives. Oxford Review of Economic Policy 19(1): 180-213.

Kim, S, Lee, J-W and Shin, K. 2006: Regional and global financial integration in East Asia. MPRA Paper 695, Munich Personal RePEc Archive: Munich, May.

Lane, P and Milesi-Ferretti, GM. 2006a: The external wealth of nations mark II: Revised and extended estimates of foreign assets and liabilities, 1970 2004. IMF Working Paper 06/69, International Monetary Fund: Washington, DC, March.

Lane, P and Milesi-Ferretti, GM. 2006b: Examining global imbalances. Finance and Development 43(1): 38 41.

Lane, P and Walti, S. 2006: The Euro and financial integration. IIIS Discussion Paper No. 139, Institute for International Integration Studies: Dublin, May.

Lane, P. 2006: Global bond portfolios and EMU. Institute for international integration studies. IIIS Discussion Paper No. 168, Institute for International Integration Studies: Dublin, June.

Martin, P and Rey, H. 2004: Financial super-markets: Size matters for asset trade. Journal of International Economics 64(2): 335-361.

Melitz, J. 2006: Comment on "The Euro's Trade Effects'. Working Paper 594, European Central Bank: Frankfurt a.M., March.

Portes, R and Rey, H. 2005: The determinants of cross-border equity flows. Journal of International Economics 65(2): 269-296.

Rose, AK. 2000: One money, one market: Estimating the effect of common currencies on trade. Economic Policy 15(30): 7-45.

Rose, AK and Stanley, T. 2005: A meta-analysis of the effect of common currencies on international trade. Journal of Economic Surveys 19 (3): 347-365.

Sorensen, CK and Puigvert Gutierrez, JM. 2006: Euro area banking sector integration--using hierarchical cluster analysis techniques. Working Paper 627, European Central Bank: Frankfurt a.M., May.

(1) An earlier version of this paper was presented at the 12th Dubrovnik Economic Conference organized by the Croatian National Bank held in Dubrovnik from 28 June to 1 July, 2006. The authors would like to thank the Statistics Department of the Bank for International Settlements (BIS) in Basel, in particular Philippe Mesny and Karsten von Kleist, for the provision of data. We thank Evzen Kocenda, Paul Wachtel, participants of the 12th Dubrovnik Economics Conference, and two anonymous referees for most helpful comments on an earlier draft. Catherine Tahmee Koch has provided excellent research assistance. All errors and inaccuracies are solely in our own responsibility.

(2) For reasons of data confidentiality, we do not report descriptive statistics for Hong Kong and Italy but we use data for these countries in our regressions.

(3) A dummy variable for the legal origins of countries, for instance, has a positive and significant impact on foreign assets in the cross-country model estimated by Aviat and Coeurdacier (2007). We do not include variables of this type here since they would drop out ill our--preferred--fixed effects regression.

SVEN BLANK (1) & CLAUDIA M BUCH (2,3)

(1) Department of Economics, University of Tuebingen, Mohlstrasse 36, Tuebingen 72074, Germany. E-mail: sven.blank@uni-tuebingen.de

(2) Department of Economics, University of Tuebingen, Germany

(3) Institute for Applied Economic Research TAW, Germany

Table 1: Descriptive statistics

Variable Obs. Mean Std. dev.

Bilateral foreign assets
 (million USD) 2,222 29,220 66,051
Bilateral foreign liabilities
 (million USD) 2,222 22,230 51,583
Bilateral foreign assets
 (constant USD) 1,738 24,807 56,738
Bilateral foreign liabilities
 (constant USD) 1,738 18,044 40,793
Domestic GDP (million USD) 2,222 5,355,771 11,400,000
Foreign GDP (million USD) 2,222 3,023,725 8,383,925
Distance (kilometers) 2,222 5,284 4,921
Common language 2,222 0.168 0.374
Bilateral exports (million USD) 2,222 2,869 5,073
Bilateral imports (million USD) 2,222 2,777 5,685
Trade volume (million USD) 2,222 5,647 10,476
Domestic real interest
 rate (percent) 2,222 1.73 1.77
Foreign real interest
 rate (percent) 2,222 2.26 2.53
Domestic exchange rate relative
 to the USD 2,222 181.42 519.50
Foreign exchange rate relative
 to the USD 2,222 162.50 400.43
Domestic stock market index 2,222 3938.34 4380.04
Foreign stock market index 2,222 6113.01 6018.47
Domestic risk-adjusted
 stock returns 1,634 9.17 14.63
Foreign risk-adjusted
 stock returns 1,623 7.15 10.03
Domestic variance of
 stock returns 2,020 0.011 0.014
Foreign variance of
 stock returns 2,020 0.014 0.023
Domestic corruption index 2,222 7.47 1.30
Foreign corruption index 2,197 7.04 1.94
Domestic CPI 2,222 115.09 29.35
Foreign CPI 2,222 143.63 163.78

Variable Minimum Maximum

Bilateral foreign assets
 (million USD) 10 899,607
Bilateral foreign liabilities
 (million USD) 3 657,002
Bilateral foreign assets
 (constant USD) 10 553,012
Bilateral foreign liabilities
 (constant USD) 3 519,806
Domestic GDP (million USD) 38,500 49,900,000
Foreign GDP (million USD) 5,040 49,900,000
Distance (kilometers) 173.00 19,263
Common language 0 1
Bilateral exports (million USD) 16 52,855
Bilateral imports (million USD) 7 72,985
Trade volume (million USD) 34 125,840
Domestic real interest
 rate (percent) -1.68 9.09
Foreign real interest
 rate (percent) -4.3 15.91
Domestic exchange rate relative
 to the USD 0.54 2164.10
Foreign exchange rate relative
 to the USD 0.54 2164.10
Domestic stock market index 30.43 24970.73
Foreign stock market index 30.43 24970.73
Domestic risk-adjusted
 stock returns -4.91 58.10
Foreign risk-adjusted
 stock returns -4.62 41.23
Domestic variance of
 stock returns 0.00 0.08
Foreign variance of
 stock returns 0.00 0.25
Domestic corruption index 2.99 9.69
Foreign corruption index 2.66 10
Domestic CPI 85.52 195.30
Foreign CPI 42.22 1,162

Table 2: Results of crass-sectional regressions for
cross-border assets

 (1) (2) (3)
 OLS, country OLS, time Panel fixed
 and time varying country effects
 fixed effects fixed effects

Domestic GDP 0.698 *** 0.248 *** 0.666 ***
 [3.32] [5.29] [3.24]
Foreign GDP 0.679 *** 0.660 *** 0.650 ***
 [3.42] [9.61] [3.26]
Both EMU (0/1 dummy) 0.891 *** 0.999 *** 0.637 ***
 [7.50] [4.51] [8.01]
Distance -0.603 *** -0.596 ***
 [7.62] [6.96]
Common language 0.335 ** 0.330 *
 [2.15] [1.95]
Log bilateral trade
Domestic real
 interest rate
Foreign real
 interest rate
Both EU (0/1 dummy)
Observations (N x T) 2,222 2,222 2,222
Cross-sections (N) 202
[R.sup.2] 0.87 0.88 0.39

 (4) (5) (6)
 Panel fixed Panel fixed Panel fixed
 effects effects effects
 (EU only)

Domestic GDP -0.066 0.652 *** 0.668 ***
 [0.15] [2.97] [3.28]
Foreign GDP 1.312 *** 0.630 *** 0.676 ***
 [3.45] [2.93] [3.39]
Both EMU (0/1 dummy) 0.361 *** 0.634 *** 0.635 ***
 [3.09] [7.71] [8.27]
Distance
Common language
Log bilateral trade 0.031
 [0.19]
Domestic real 0.009
 interest rate
 [0.66]
Foreign real -0.011
 interest rate
 [1.18]
Both EU (0/1 dummy)
Observations (N x T) 639 2,222 2,222
Cross-sections (N) 72 202 202
[R.sup.2] 0.78 0.39 0.39

 (7) (8)
 Panel fixed Panel fixed
 effects effects
 (non-banks)

Domestic GDP 0.613 *** 0.905 ***
 [2.98] [3.41]
Foreign GDP 0.522 *** 0.317
 [2.71] [1.47]
Both EMU (0/1 dummy) 0.645 *** 0.971 ***
 [8.06] [8.92]
Distance
Common language
Log bilateral trade
Domestic real
 interest rate
Foreign real
 interest rate
Both EU (0/1 dummy) 0.443 **
 [2.18]
Observations (N x T) 2,222 2,222
Cross-sections (N) 202 202
[R.sup.2] 0.39 0.41

This table reports results of cross-section regressions for
the log of cross-border assets of banks. The data are pooled
across reporting countries. Robust t-values in brackets.
***, **, * Significant at the 1%, 5%, 10% level, respectively.

Table 3: Results of cross-sectional regressions for
cross-border liabilities

 (1) (2) (3)
 OLS, country OLS, time- Panel fixed
 and time varying effects
 fixed effects country
 fixed effects

Domestic GDP 0.713 *** 0.526 *** 0.665 ***
 [3.08] [8.16] [2.88]
Foreign GDP 0.815 *** 0.746 *** 0.771 ***
 [4.78] [10.28] [4.53]
Both EMU (0/1 dummy) 0.678 *** 1.026 *** 0.294 ***
 [5.21] [4.87] [3.03]
Distance -0.703 *** -0.680 ***
 [6.52] [5.87]
Common language 0.349 ** 0.333 **
 [2.32] [2.06]
Log bilateral trade
Domestic real
 interest rate
Foreign real
 interest rate
Both EU (0/1 dummy)
Observations (N x T) 2,222 2,222 2,222
Cross-sections (N) 202
[R.sup.2] 0.87 0.88 0.39

 (4) (5) (6)
 Panel fixed Panel fixed Panel fixed
 effects effects effects
 (EU only)

Domestic GDP -0.974 * 0.448 ** 0.644 ***
 [1.89] [2.01] [2.78]
Foreign GDP 0.668 0.458 ** 0.735 ***
 [1.29] [2.24] [4.14]
Both EMU (0/1 dummy) 0.002 0.254 ** 0.258 ***
 [0.01] [2.58] [2.66]
Distance
Common language
Log bilateral trade 0.474 ***
 [2.63]
Domestic real -0.06 ***
 interest rate
 [3.51]
Foreign real 0.012
 interest rate
 [1.16]
Both EU (0/1 dummy)
Observations (N x T) 639 2,222 2,222
Cross-sections (N) 72 202 202
[R.sup.2] 0.38 0.39 0.16

 (7) (8)
 Panel fixed Panel fixed
 effects effected
 (non-banks)

Domestic GDP 0.696 *** 0.563 **
 [3.01] [2.36]
Foreign GDP 0.847 *** 0.631 ***
 [4.79] [3.34]
Both EMU (0/1 dummy) 0.289 *** 0.109
 [2.98] [1.14]
Distance
Common language
Log bilateral trade
Domestic real
 interest rate
Foreign real
 interest rate
Both EU (0/1 dummy) -0.262
 [1.61]
Observations (N x T) 2,222 2,222
Cross-sections (N) 202 202
[R.sup.2] 0.15 0.29

This table reports results of cross-section regressions
for the log of cross-border liabilities of banks. See Table
2 for further notes.