European Financial Market Integration and German Cross Border Portfolio Flows

Transcription

1 European Financial Market Integration and German Cross Border Portfolio Flows Barbara Berkel MEA, Mannheim University First Version: January 31, 2006 Abstract The paper analyzes the effect of European financial integration, especially of the EMU, on gross portfolio flows between Germany and 47 countries for 1987 to A gravity model of asset trade `a la Martin and Rey (2004) is estimated. The following results are found: (1) After accounting for country pair fixed effects, German gross portfolio flows are about 43 to 50 percent higher for countries participating in the EMU. (2) Developments intertwined with the formation of the EMU - measured by exchange rate volatility, financial development and real economic integration - do not seem to account for this effect. (3) The size of the EMU effect on gross portfolio flows is higher for bank-based countries, more developed financial markets, countries with increased real integration as well as for countries that are geographically further away from Germany and thus associated with higher information costs. Keywords: European financial market integration, EMU, gravity model of bilateral asset trade, gross portfolio flows JEL classification: F21, F36, G15 Barbara Berkel, MEA - Mannheim Research Institute for the Economics of Aging, Department of Economics, Mannheim University, D Mannheim, Germany, phone: , fax: , berkel@mea.uni-mannheim.de

2 1 Introduction The formation of the European Economic and Monetary Union (EMU) has been the most important development affecting European financial markets in the last two decades. Starting in 1990, all twelve participating countries abolished exchange controls. 1 Finally, eleven countries gave up their national currencies on January 1, 1999, Greece following in 2001, thereby eliminating any exchange rate risk among themselves. In order to analyze this process, three main questions are addressed in this paper: (1) How large is the effect of European financial integration, especially of the EMU, on German cross border portfolio flows? 2 (2) What kind of reforms or underlying country characteristics can explain (part of) this effect? (3) Are there heterogeneous responses within this set of countries to increased European financial market integration with respect to portfolio investment? It is worth investigating European financial integration in more detail for two main reasons: First, learning more about the underlying factors and driving forces that determine market integration helps to understand changes in international asset trade. In the last two decades, the volume of capital flows has increased dramatically. Second, in light of economic globalization, financial market integration is very likely to further increase over time and across countries. The results shed light on the implications of future enlargements of the EMU and formations of new currency unions. The level of financial integration can be assessed by using price based or quantity based indicators. Most studies dealing with the European integration of financial markets use price based measures (Baele (2005), Bartram, Taylor and Wang (2005), Fratzscher (2002)). By contrast, this paper investigates the role of European financial integration and the EMU on German cross border portfolio flows. The estimates are based on a gravity model of asset trade à la Martin and Rey (2004) using annual data on gross portfolio flows between Germany and 47 countries for the period of 1987 to This approach is led by the following considerations: First, gross flows reveal frictions in asset trading and the degree of segmentation between markets more directly than asset holdings, net flows or stock returns. Second, Germany is economically the largest country within the EMU. In 2002, it accounted for 19 percent of total portfolio investment within 1 Austria, Belgium, Finland, France, Germany, Italy, Luxembourg, Netherlands in 1990; Ireland, Portugal and Spain in 1993, and Greece in Portfolio flows mainly include equity, mutual funds, bonds and money market papers. 1

3 the EMU. 3 While panel data on bilateral portfolio flows within a broader set of countries is not yet publicly available, Germany is a promising starting point to analyze European financial market integration. Third, as opposed to studies using high frequency data this study is based on annual data, thereby allowing to employ explaining variables that are only available at annual frequencies. These variables are used in order to account for enhanced financial market integration in Europe and to characterize heterogeneous country responses. Results reveal that gross portfolio flows are about 43 to 50 percent higher among countries participating in the EMU after controlling for country pair fixed effects. Reforms intertwined with the formation of the EMU - measured by exchange rate volatility, financial market development and real integration - are not able to account for this effect. The EMU effect on gross portfolio flows is larger for countries with bank-based systems, more developed financial markets, countries with increased real economic integration as well as for countries that are further away geographically from Germany and thus associated with higher information costs. The paper is organized as follows: Section 2 reviews important results of the literature on financial market integration and presents a gravity model of bilateral asset trade used for the estimations. It is followed by a description of the data in Section 3. Section 4 summarizes and discusses the empirical results which are concluded in Section 5. 2 Financial market integration and bilateral asset flows In order to develop hypotheses on the effect of the EMU on asset flows, the existing literature on European financial market integration is being reviewed. The subsequent Section outlines a standard empirical gravity model for bilateral asset flows that is used for the empirical analysis. 3 This number is based on the Coordinated Portfolio Investment Survey (CPIS) issued by the IMF. 2

4 2.1 Financial market integration in Europe - What do we know? A number of recent studies analyzes the degree of European financial market integration from various angles. Baele (2005), Bartram, Taylor and Wang (2005) and Fratzscher (2002) investigate the degree of financial integration within Europe by analyzing stock market returns across countries using high frequency data and time-series methods. Baele (2005) investigates the effect of further globalization and regional integration on the intensity by which global and regional market shocks are transmitted to local equity markets. He finds that the interdependence of 13 European equity markets with the US, and especially within European countries increased over the 1980s and 1990s. According to his findings, equity market development, trade integration and price stability enhance the extent of interdependence within European equity markets. Bartram, Taylor and Wang (2005) conjecture that the degree of dependence between equity markets of countries within the Euro area increased in late 1997 or early 1998 after the Euro membership had been determined and announced. Similarly, Fratzscher (2002) suggests that European equity markets have become more integrated since He also shows that reduced exchange rate uncertainty as well as monetary policy convergence of interest rates and inflation rates have been the central driving force behind the financial integration process in Europe. Baele et al. (2004) present a comprehensive assessment of the impact of the EMU with respect to different price and quantity indicators and find that all measures of integration indicate a rising degree of equity market integration. From a microeconomic perspective, Guiso, Haliassos and Japelli (2003) find that households equity market participation has increased. They analyze the current state of equity ownership in several European countries. However, considerable country-specific differences remain, which they explain by different levels of participation costs in the Euro area. This suggests that there are still a number of barriers that need to be overcome before full integration of European equity markets is reached. A comparison of the European equity market with other market segments reveals that while the money market has almost fully converged after the introduction of the single currency, important barriers to international investment still remain in the equity market. The markets for government and corporate bonds as well as the credit market lie in between these two extremes. Despite being characterized by different levels of integration, all sectors have shown a 3

5 marked increase in integration, underlining the hypothesis that monetary unions facilitate cross-border asset flows (Baele et al. (2004), Adam et al. (2002)). Pagano and von Thadden (2004) focus on the impact that the monetary union has on the markets for Euro area sovereign and private bonds. They find that the adoption of a single currency and the elimination of currency risk is not sufficient to integrate markets if institutional, legal and fiscal barriers persist. In this respect, the sequence of policy actions in the wake of the formation of the EMU aimed at removing most remaining obstacles therefore facilitating a huge improvement of capital markets integration. These findings help to understand the different levels of financial integration in different segments of the capital market. A more indirect measure of financial market integration looks at investment savings correlations. In a world of perfect capital markets the two macroeconomic variables should be independent of each other. Empirically, this is not the case - a phenomenon that is well know in the literature as the so called Feldstein- Horioka puzzle. Blanchard and Giavazzi (2002) show that the correlation between domestic saving and investment has declined over time, especially in the Euro area, suggesting higher integration in financial markets. As shown in the cited papers, financial market integration altogether increased substantially in Europe over the last two decades. Different levels of integration among financial market segments exist, though, in which still some institutional, legal and fiscal barriers remain. Integration of European equity markets increased especially in the late 1990s, but is still lower compared to other segments. This paper adds three new aspects to the existing literature: First, a comprehensive econometric analysis of European financial integration on gross portfolio flows is undertaken. The analysis is based on a gravity model approach of asset trade that has performed well in explaining volumes of bilateral cross border asset trade in earlier studies. 4 Second, the question of whether countries respond differently to European financial integration depending on financial market structure, information costs and real economic integration is addressed explicitly. This aspect might bear important implications for future EMU enlargements. Third, the effect of the EMU formation on German financial market integration is tried to be disentangled from the effect of increased integration with the EU-15 countries. 4 E.g. Buch (2005) applies a gravity model to bank lending data, Portes and Rey (2005) to cross-border portfolio investment, Di Giovanni (2005) to M&A activity and Portes, Rey and Oh (2001) to corporate, government bonds and equities. 4

6 Increased trading activity of portfolio assets induced by the formation of the EMU is interpreted as increased financial integration in this paper. This measure entails only a specific quantity-based angle of financial integration. While increased trading activity does not imply any conclusions for the actual outcome of arbitrage between countries or the extent of asset diversification across countries, it gives an informative picture of German financial openness towards portfolio asset trading with EMU and EU countries as opposed to non-european countries. 2.2 A gravity model of bilateral asset trade - Econometric framework Martin and Rey (2004) propose a theory of asset trade based on a general equilibrium model from which a gravity equation emerges. 5 Their model is characterized by fully optimizing agents, endogenous market capitalization and frictions in the asset market. 6 Its main implication is that gross flows of asset trade between two countries should depend inversely on transaction costs between the countries such as information costs and efficiency of transaction technology and depend proportionally on market size which is proxied by stock market capitalization. 7 The basic estimation equation takes the following form: log(t ij,t ) = β 0 + β 1 log(mcap i,t ) + β 2 log(mcap j,t ) + β 3 log(distance ij,t ) + β 4 log(credit i,t ) + β 5 log(credit j,t ) + β 6 foreigner ij,t N Z n + β 7 return corr ij,t + ij,t + ij,t, n=1 where i denotes the source or transacting country, j the country invested in and t time. The dependent variable, T ij,t, is defined as country i s transactions of country j s portfolio investment. The volume of transactions is the sum of portfolio asset purchases and sales. The explaining variables are market capitalization 5 An empirical gravity model equation also emerges from a model by Obstfeld and Rogoff (2001) that introduces transaction costs solely in the goods market thereby generating substantial amounts of home bias. See Lane and Milesi-Ferretti (2004) for an N-country extension of the Obstfeld and Rogoff (2001) model. 6 As the model is static asset transactions and holdings of foreign assets coincide. So far, dynamic models that explain transactions volumes do not yet exist. 7 Generally, either stock market capitalization or GDP are used to measure market size. The use of GDP leads to similar results. 5

7 of country i and j, mcap i,t and mcap j,t, distance between country i and country j, distance ij, as well as private credit provided by the banking sector, credit i,t and credit j,t. These variables enter in logarithms. Moreover, the percentage of foreigners of nationality i or j living in Germany, 8 and the monthly correlation of returns in year t, return corr ij,t are included. β 0 is a constant and ij,t refers to an error term, which is assumed to be distributed N(0, σ 2 ). Additional variables, Zij,t, n are included that account for time and country-fixed effects: a full set of time dummies, dummies for financial centers as well as a dummy variable describing whether a banking crisis is present in country i or country j in the relevant or precedent year, crisis i,t and crisis j,t. The theory by Martin and Rey (2004) suggests that β 1 = β 2 and that β 3 < 0. Private credit provided by the banking sector is a common proxy for financial development. More developed markets provide more efficient investment opportunities thus leading to reduced transaction costs which promote international asset trade. As Di Giovanni (2005) shows, financial development within a country helps to aid its firms investing abroad. This paper investigates whether credit provided by the banking sector in country i and j has a positive effect on portfolio transactions, i.e. β 4 > 0, β 5 > 0. foreigner ij,t depicts a proxy of familiarity between countries and is expected to have a positive influence on transactions, i.e. β 6 > 0. This is in line with French and Poterba (1991), who stress the importance of cultural familiarity as an explanation for international investment. If investors follow diversification considerations, β 7 should be negative. The basic specification of the gravity model is subsequently augmented in order to analyze different hypotheses concerning the course of European integration and its underlying driving forces. The estimation strategy is as follows: First, the basic gravity model described above is estimated in order to compare the results with similar empirical studies. Second, effects of European financial integration are identified by adding dummy variables that stand for stage one and three of the EMU. 9 Moreover, dummy variables that are equal to one over the whole time period if both countries are part of the EMU or EU-15. Further checks are undertaken to show the robustness of the results. Third, variables that proxy financial and macroeconomic reforms are added in order to account for (part of) the EMU 8 This variable always refers to foreigners living in Germany but not to Germans living abroad. 9 The EMU started in 1990 with stage one, the abolition of all existing exchange controls. Stage three took place in 1999 with the fixation of exchange rates. Exceptions from this time pattern are Greece, Ireland, Portugal and Spain. These countries entered stage one not before Greece entered the third stage in

8 effect. This proceeding tries to identify and distinguish the importance of factors related to European financial integration. Fourth, heterogeneous responses to increased European financial market integration are investigated by adding interaction effects with variables that measure financial structure, information costs and real economic integration. Many papers using the gravity model approach to analyze asset trade only exploit the cross-sectional variation of the data. In order to measure the impact of financial market integration - which is a continuous process over time - the time-series dimension is of major interest. Therefore, not only standard pooled ordinary least squares (POLS) estimates but also fixed effects estimations are undertaken Data and descriptive statistics This study investigates German cross-border portfolio investment. 11 The dependent variable, T ij,t, includes the amount of foreign purchases and sales of German portfolio assets - with i and j referring to the foreign transacting country and Germany respectively - as well as the amount of German sales and purchases of foreign portfolio investment - in this case i refers to Germany and j to the foreign country. The data in use was provided by the Deutsche Bundesbank. Portfolio investments are part of the balance of payments and include equity, mutual funds, bonds and notes as well as money-market papers. 12 The data is available for 47 countries from 1987 to 2002 (see Table 1). The period covers the three stages of the formation of the EMU from 1990 to Further financial and macroeconomic variables are necessary for the empirical analysis. Table 2 summarizes their definitions and sources. Table 3 reports summary statistics for all variables. As can be seen in the last two columns, mean values of all variables except insider i,t and insider j,t are significantly different for EMU countries compared to a sample excluding EMU countries. Descriptive statistics of portfolio investments by direction of investment are presented in Table 4 for single years. Starting from the early 1990s, there is a very strong increase in overall portfolio investment (purchases and sales) for both directions, i.e. German assets purchased and sold by foreign countries as well 10 See Cheng and Wall (2005) for a comparison of different panel estimation methods for the estimation of bilateral goods trade. 11 So far, a full cross-country panel data set of bilateral portfolio flows is not publicly available. 12 See Deutsche Bundesbank, Monthly Reports, 7

9 as foreign assets purchased and sold by Germany. Moreover, percentage shares of portfolio investment within OECD, EU-15 and EMU countries are reported. About 98 percent of investments in either direction are undertaken with OECD countries. This share stays constant throughout the whole sample period. This is contrasted by an increase in the shares of investments with EU-15 and EMU countries. Level and timing of this increase depend on the direction of investment: The share of German investment (again purchases and sales) in EU-15 and EMU countries increases strongly in the late 1980s and early 1990s. The EU-15 share rises from 45 percent in 1987 to 72 percent in The EMU share grows even more strongly, namely from 27 percent to 66 percent. A slightly different pattern arises for the other direction: Investment in German portfolio assets by EU countries is also rising mainly in the early 1990s but less significantly: from 74 percent in 1987 to 90 percent in Investments undertaken by EMU countries increase later, in the mid-1990s, and both less dramatically, from 21 percent in 1987 to 30 percent in Overall, the main difference between the two directions constitutes by a larger share of foreign investment in German portfolio assets by EU-15 countries and a lower share of German sales and purchases of EMU and EU portfolio assets. This is partly driven by a large share of German portfolio assets purchased and sold by the UK due to its importance as a leading financial center. The empirical analysis accounts for this fact by including financial center dummies 13 and by undertaking additional robustness checks. 4 Empirical results First, the impact of European financial integration on German portfolio investment is investigated in general. Second, potential underlying forces driving European integration are explicitly taken into account. It is very likely that countries responses to European financial integration differ. This issue is addressed in the last part of this Section. 13 Financial centers are Hong Kong, Ireland, Luxembourg, UK, Singapore and Switzerland. For each of these countries separate dummies enter that refer to country i and country j. 8

10 4.1 German portfolio investment and European financial integration The standard gravity regression equation described in Section 2.2 is used to identify the effect of European financial integration on portfolio investment. The effect is modelled by different dummy variables that mirror the formation of the EMU or, more generally, EMU or EU membership. The effect of EU versus EMU financial integration is disentangled and further robustness checks are undertaken The formation of the EMU In a first step, results of the standard gravity model are compared with results in the existing literature. In a second step, the effect of financial market integration in Europe is investigated. The explanatory variables include a constant, absolute stock market capitalization for both the transacting country and the country invested in, mcap i,t and mcap j,t, geographical distance between the two countries, distance ij, private credit provided by the banking sector as a proxy for financial development for both countries, credit i,t and credit j,t. As indicated by the gravity model equation, all these variables enter in logarithms. Furthermore, the percentage of foreigners of nationality i or j living in Germany, foreigner ij,t, and the correlation between stock market indices, return corr ij,t, are calculated and included. Additional year dummies, dummies for financial centers and dummies for financial crises in the transacting country i or the country invested in, country j, are added in all specifications but not explicitly reported. 14 The coefficients of the first three columns in Table 5 are based on pooled OLS estimates with White-heteroscedasticity robust standard errors. Specification (1) is the most parsimonious one including only the scaling variables, the percentage of foreigners and distance. Additional variables are added subsequently in specifications (2) and (3). 15 The results are consistent with earlier estimates of gravity models in the literature. The distance coefficient is - as expected - negative and ranges in absolute size between 0.67 and one. Portes and Rey (2005) 14 In order to take German reunification into account an additional dummy variable was considered. As the effect turns out to be insignificant, specifications without this additional dummy are presented in the following. 15 Note that even in specification (1) the R 2 amounts to 47 percent, which shows that the model performs very well in explaining the variation in the data. 9

11 report a coefficient around minus 0.6 estimating bilateral portfolio flows between 14 countries from 1989 to Buch (2005) considers assets and liabilities of commercial banks from five countries (France, Germany, Italy, UK and the US) for 1983 and 1999 and estimates a distance coefficient between minus 0.3 and 1.25 depending on the respective estimation sample. By and large the estimated coefficients of the remaining variables in specifications (1) - (3) are in line with the theoretical considerations discussed in Section 2.2: (1) The coefficients of the scaling variables, mcap i,t and mcap j,t, are close to one which is a common finding in the literature (Portes and Rey 2005). However, they are not exactly equal in size, as suggested by theory. 16 (2) The percentage of people of the respective foreign country living in Germany, foreigner ij,t, is associated with a positive effect on transactions. This result suggests that familiarity between two countries plays a role which is in line with French and Poterba (1991). (3) Private credit provided by the banking sector is significant for the transacting country i as well as for the country invested in, country j, (specification (2) and (3)). The effect is almost equal in size for both countries. The more developed financial markets are at home and abroad the more foreign portfolio assets are transacted. (4) The correlation of equity returns has a positive effect (specification (3)). 17 This finding is not consistent with theoretical considerations which suggest a negative coefficient if investors follow a diversification motive. However, this result is in line with earlier empirical findings in gravity model estimations. 18 It might be driven by the common impact of financial integration on the dependent variable and on return correlations. 19 In specification (4) and (5) panel estimation methods are employed. Specification (4) is a country pair fixed effects estimation that controls for country heterogeneity. 20 The estimated coefficients on the scaling variables decrease substantially and the coefficient on mcap i,t is not significant. As geographical distance is collinear to the estimated country pair fixed effects, it adds no explanatory power. In contrast to specification (2) and (3), the variable credit i,t 16 F-Tests of both coefficients being equal are undertaken and reported in Table The same results are obtained if lagged correlations are included in order to proxy expected correlations. 18 See e.g. Portes and Rey (2005), Aviat and Coeurdacier (2004). 19 Coeurdacier and Guibaud (2005) show that when instrumenting current stock market correlation with stock return correlation over the period of a diversification motive can be identified. However, as such data is not present for the country sample used in this paper, this approach can not be undertaken. 20 Separate country pair fixed effects for each direction are considered, i.e. country pair ij is distinguished from country pair ji. 10

12 as a proxy for financial development of the transacting country i is not significant. Obviously, financial development of the country invested in, country j, has considerable power to explain changes in portfolio investment over time. The differences between the pooled OLS and the fixed effects estimates reveal that variables such as the percentage of foreigners in Germany, foreigner ij,t, mainly explain cross sectional variation and heterogeneity across countries. The same applies to credit provided by the banking sector in the transacting country i as a proxy for financial development, credit i,t, as well as market capitalization of country i, market i,t. As an alternative to fixed effects, a random effects model is estimated in specification (5). The coefficients on the scaling variables, mcap i,t and mcap j,t, lie in between the values for the pooled OLS and the fixed effects model. The coefficients on foreigner ij,t, as well as credit i,t and credit j,t are significant but smaller in size compared to the pooled OLS estimations. The Hausman χ 2 -test clearly rejects the assumptions of the random effects model, thus suggesting that the fixed effects model is preferred. 21 The fixed effects estimation is also preferred from an intuitive point of view. Fixed effects are due to omitted variables that are specific to cross-sectional units. Some of the main forces behind the fixed effects should be information or transactions costs, e.g. due to effective capital controls. Most of these effects are not random but deterministically associated with certain historical, political or geographical facts (Egger 2000). Table 6 describes results where in addition to the variables entering in specification (1) dummy variables capturing omitted variable effects of European financial market integration are included. First, a dummy variable, d1990, is added which equals one in the respective year if both countries are taking place in the first stage of the EMU that came into effect in 1990 with the abolishment of exchange controls. 22 Since then capital movements are completely liberalized among the participating countries. Second, a dummy variable, d1999, is added that is equal to one if both countries are members of the EMU and conversion rates to the Euro are fixed. In 1999, the third stage of the EMU begins and one real single currency is established among the eleven participating countries. Greece joins this stage only in Since then, one single monetary policy is in use under the authority of the European Central Bank (ECB). A three-year transition period follows before the introduction of actual Euro notes and coins, 21 The Hausman statistic tests for the orthogonality of the random effects and the regressors. A significant test statistic reveals a high importance of group-specific effects and their correlation with the independent variables (Wooldridge (2002), chap ). 22 Ireland, Spain, Portugal joined in 1993, Greece in

13 but legally the national currencies have already ceased to exist in Third, a dummy enters that is equal to one for all years if both countries belong to the EMU. For each specification, two estimates are reported: pooled OLS and fixed effects. The dummies are highly significant in almost all specifications and the coefficients on the other variables do not change. After 1990 Germany experiences higher portfolio investment volumes with countries that are also part of the first stage of the EMU. Specifications (6) and (7) indicate that ceteris paribus gross flows with EMU countries are on average 190 percent larger in the pooled OLS estimation and 40 percent higher in the fixed effects estimation. The third stage of the EMU ceteris paribus leads to portfolio investments that are on average 130 percent higher in the pooled OLS and 49 percent higher in the fixed effects estimation (specifications (8) and (9)). When both dummies enter simultaneously in the estimation equation (specifications (10) and (11)), the dummy for stage one is roughly as large as before, which underlines the importance of the abolition of all official impediments to free capital mobility. The dummy for the fixation of exchange rates only remains significant in the fixed effects estimates. As specification (12) shows, there exists not only a change in trading volumes due to these two events but also a positive level effect of enhanced portfolio flows over the whole estimation period from : German cross border portfolio investment with EMU countries is on average three times as large as with non-emu countries. Both, the OLS and the fixed effects results document that the formation of the EMU leads to a change in portfolio investment over time. The pooled OLS estimator captures both the effect over time and the cross-sectional effect of higher trade within EMU countries. As both effects are positive, overall coefficients on the dummy variables of interest are higher in pooled OLS estimations The role of EMU versus EU The question arises whether the results are really driven by increased integration in the wake of the EMU formation or rather by increased economic and financial integration within all EU countries. Therefore, additional dummies for EU country pairs are considered in the regression analysis. 23 Corresponding results are summarized in Table This dummy refers to the EU-15 countries Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain, Sweden and UK. 12

14 Across the whole sample period, trade with EU countries is on average 260 percent higher (specification (13)). This effect is even stronger than the comparable effect for EMU countries mentioned above. Once both the EMU and the EU dummy enter simultaneously the regression equation, both dummies are significant and positive (specification (14)). Thus for EMU countries, a positive effect on top of the EU effect is revealed. Specification (15) points into the same direction. Instead of the EU dummy, a dummy for countries being part of the EU but not of the EMU enters, EUnonEMU, this being namely UK, Denmark and Sweden. The F-Test yields that both coefficients are not significantly equal at the 5 percent level. However, the results show that the size of the estimated coefficients of both dummies differs only marginally, i.e. 240 versus 260 percent. The two effects are disentangled further across time: The next four specifications ((16)-(19)) analyze to what extent the effects of the first and third step of the EMU are driven by the fact that they belong to the EMU or EU. Artificial dummy variables for the first and third stage of the EMU are also constructed for EU countries that are not part of the EMU. As changes over time are of interest, each specification is again not only estimated by pooled OLS but also by fixed effects. As expected, the abolition of exchange controls as well as the fixation of exchange rates yield significantly higher effects on gross portfolio flows for EMU countries compared to EU countries not participating in the Euro. This effect is especially pronounced in the fixed effects estimations. Next, these effects are estimated separately for each year. Interaction terms between the EMU dummy variable and year dummies as well as the dummy variable for EU countries that are not part of the EMU and year dummies enter the regression. The estimated coefficients of the interaction terms are plotted in Figure 1. In the late 1980s, the estimated coefficients of the EMU-countries are slightly smaller compared to the coefficients of the countries not being part of the EMU. After 1992, an increasing gap arises. As the standard error bands show, the coefficients are always significantly different from zero. Non-reported F-Tests reveal that the estimated yearly coefficients are significantly different from one another for all years since Further robustness checks The regressions discussed so far are reproduced for two sub-samples in order to check whether results are driven by one or the other: German investment in foreign portfolio assets is separated from investment in German assets undertaken 13

15 by foreigners (see Table 8). Employing pooled OLS one finds that the effect of European financial integration measured either by d1990, d1999 or EMU is always larger for German purchases and sales of foreign assets compared to the full sample. In fixed effects estimations, there is no significant effect of stage one of the EMU for German transactions of foreign assets. Changes over time with respect to stage two of the EMU are larger for foreign transactions of German portfolio assets. Significant differences in size depending on whether pooled OLS or fixed effects estimations are used remain as in the full sample. Qualitatively, the results are very similar for both sub-samples. 24 Further regressions have been undertaken to check the robustness of the results. 25 First, single year regressions are run to reproduce Figure 1. In this case, the estimated coefficients on the non-interacted variables are not restricted to be constant over time. Similar effects for the dummies are estimated. Second, the regressions described in Table 3 are redone for a sample of EU-15 countries. The results are robust with respect to the EMU dummy effect. Third, the UK is excluded from the sample because there is a large volume of German portfolio investment purchased or sold from the UK that might drive some of the results. The estimated coefficients do not change. Note that a dummy for the UK as a financial center is included in all regressions discussed so far. Fourth, lagged values for the scaling variables as well as private credit provided by the banking sector are used instead of contemporaneous values in order to address a potential endogeneity problem. The estimated coefficients on these variables do not change. Fifth, instead of yearly time dummies a linear and quadratic time trend enter the regressions. The estimated effects, especially on the various dummy variables measuring the financial integration within EMU and EU countries, do not change. 24 One exception is the coefficient of foreigner ij,t. When foreign transactions of German portfolio assets are considered, the coefficient on the percentage of foreigners of country i living in Germany has a significantly negative impact whereas in all other estimations and also in the complementary sub-sample the coefficient is significantly positive. The negative effect in the fixed effects estimations is mainly driven by the years 1987 to 1993 where foreign transactions of German portfolio assets increased substantially and by more than German transactions of foreign portfolio assets (see Table 3). At the same time the percentage of foreigners decreased on average, especially for foreigners of EMU countries. 25 They are not reported in the paper but can be received from the author upon request. 14

16 4.2 Accounting for European financial integration The dummy variables reflecting stage one and three of the EMU capture the effect of increased financial integration. What exactly are the factors that are driving this financial integration? In this Section, financial and macroeconomic factors are investigated that might account for at least part of the omitted variable effect Financial reforms The introduction of the Euro has resulted in the elimination of exchange rate risk within the Euro area. It has also reduced the differences of investment opportunities across the member countries. The absence of exchange rate risk allows corporations to raise funds across countries with fewer constraints and costs. This can in general have a large effect on financial integration because exchange rate risk is an important source of risk priced on capital markets (e.g. Dumas and Solnik (1995) and Hardouvelis et al. (2006)). In addition, lower costs of cross-country transactions, improved liquidity and better developed European capital markets are associated with the launch of the common European currency. These factors have been noted as important drivers of integration in the Euro area by Fratzscher (2002) as well as by Danthine, Giavazzi and von Thadden (2001). Not only in developed countries but also in developing countries, there has been ample evidence that the removal of legal and non-legal barriers leading to financial market liberalizations has promoted financial integration substantially (Bekaert, Harvey and Lundblad (2005) and Bekaert and Harvey (1995)). Table 9 summarizes regression results where in addition to the dummy variables for stage one and three of the EMU another set of variables enters that proxy exchange rate risk, financial development and liquidity and quality of institutions. Exchange rate risk is measured as the standard deviation of the mean monthly bilateral exchange rate over its mean in year t, exrate vol ij,t. In order to account for enhanced financial development, private credit provided by the banking sector is included in the regression, credit i,t and credit j,t. Quality of financial institutions is proxied by the degree of insider trading, insider i,t and insider j,t, which is a survey-based measure. These additional variables are not always available for the full sample and time period. 26 Therefore, a benchmark regression is reported that is based on the same observations but excludes the variable of interest. Then, the variable of interest is added and the coefficients of the dummy 26 For the availability of variables across years refer to Table 3. 15

17 variables of stage one or three can be compared across both regressions. The following effects are expected: (1) More volatile exchange rates lead to increased uncertainty and make it more risky for both companies and individuals to invest outside their own currency. As a result, investors need to be compensated for higher risk premiums or costly hedging of exchange rate risk by higher returns. Gross flows will ceteris paribus be lower if exchange rate volatility is higher. (2) Enhanced financial development leads to more competition for and availability of financial products which reduces transaction costs further. Consequently, financial development leads to enhanced cross-border portfolio investment. (3) The enforcement of insider trading laws is associated with a reduction in the cost of equity in a country (Bhattacharya and Daouk 2002). Beny (2005) shows that countries with more prohibitive insider trading laws have greater stock ownership dispersion, more informative stock market prices and greater stock market liquidity. The variables insider i,t and insider j,t range from 1 to 10 with higher values denoting less insider trading. Less insider trading makes stock markets more attractive to outside investors. A positive coefficient is thus expected. Table 9, Panel A shows results of pooled OLS, whereas Panel B refers to fixed effects estimations. The estimated coefficients of credit i,t and credit j,t as well as insider j,t and insider j,t are statistically significant and yield the expected signs in the OLS estimation. The reported F-Test statistics show that the impact of d1990 decreases significantly once credit provided by the banking sector enters the regression (specification (6)). This shows that part of the positive effect of stage one of the EMU is due to increased financial development. None of the other additional variables leads to a significantly lower coefficient on the dummies of stage one or stage three. Exchange rate volatility is only significant when using fixed effects estimates but not in a pooled OLS estimation. Consequently, not the level of exchange rate volatility adds explanatory power but the change over time. Again this effect does not alter the dummy variables of interest. Alternative measures of exchange rate volatility have been tried without yielding different results. 27 In addition, credit j,t explains some time-series variation, but not enough to significantly alter the coefficient of the dummy variable for stage one or stage two of the EMU. insider i,t and insider j,t are not significant in fixed 27 Reinhart and Rogoff (2004) constructed an exchange rate classification index ranking exchange rate regimes from freely floating to a peg. The employment of their measure does not yield any conceivable results as freely floating regimes are favored in general, whereas at the same time the common currency area of the Euro, being classified as a peg, has a positive effect on transactions. Also lagged values of exchange rate volatility have been used instead of contemporaneous values. Results do not change. 16

18 effects estimations. Overall, only financial development can explain part of the estimated effect of stage one of the EMU in a pooled OLS estimation. Most strikingly, reduced exchange rate volatility does not even account for part of the effect. Obviously, the effect of the formation of the EMU captures something more than is measured by the above mentioned explicit factors in this framework Real economic integration The empirical literature on real and financial integration has established the so called quantity puzzle : A positive association between financial integration and GDP correlations is revealed in the data, whereas theory predicts a negative relation if anything. 28 It is very likely that this empirical puzzle also applies to the EMU. The member states of the EMU have committed themselves to real convergence criteria such as price stability, limited government deficit and debt levels as well as to one single monetary policy. At the same time financial integration was pushed forward by the introduction of the Euro. There is growing evidence that real integration among members has been strengthened through real convergence in Europe. 29 As both developments intertwine, real integration might account for part of the financial integration measured by the dummy variables of stage one and three. Using the same methodology as in Section 4.2.1, GDP growth correlation as a proxy for real economic convergence is included in the regressions (Table 10). Panel A refers to pooled OLS regressions whereas Panel B reports fixed effects estimations. The results show that GDP growth correlations have a significant positive effect on transactions, except for specification (14) in Panel A. This finding is in line with the quantity puzzle revealed in other empirical studies. The inclusion of GDP growth correlations leads to smaller coefficients of the stage one and stage three dummies. However, the F-Test indicates that coefficients do not significantly change in size. Real integration in this empirical setting is not able to account for financial integration. Regression results in Table 10 might be biased due to endogeneity of GDP growth correlations, reflecting real integration, and transaction volumes, mir 28 See Imbs (2004) and Imbs (2006) for a detailed discussion of the puzzle. 29 See e.g. Frankel and Rose (1997), Berger and Nitsch (2005). 17

19 roring financial integration. Accounting for endogeneity would lead to a lower impact of business cycle correlations on transactions, thereby lowering also its ability to reduce the coefficients of the stage one and three dummies. As there is no significant reduction in the size of coefficients anyway, endogeneity does not affect the interpretation of the coefficients of interest on the dummy variables, d1990 and d Do countries respond differently to European financial integration? European financial market integration and the formation of the EMU are likely to have a different impact in every country. The differences should depend on how much countries financial markets grow and become more efficient and how important costs associated with cross border portfolio investment (e.g. information costs) are. Countries with low transaction costs and relatively developed, efficient and large financial markets might see a large increase in gross flows induced by European financial market integration. At the same time, it is also possible that an increase in cross border trading activity might be larger for countries starting with relatively high transactions costs as well as less developed, efficient and smaller financial markets. Countries can also be differentiated with respect to the extent of real economic integration. As shown in Section financial integration is associated with real economic integration. Against the background of this quantity puzzle, one expects countries with more synchronized business cycles, i.e. countries that are more integrated in real terms, to be more financially integrated in the sense that they experience larger transaction volumes. In order to explore differences in the effect of financial integration on portfolio investment across countries, the basic regression is expanded by an interaction term between the dummy variable for the first stage of the EMU and the variable of interest: 30 log(t ij,t ) = β 0 + β 1 log(mcap i,t ) + β 2 log(mcap j,t ) + β 3 log(distance ij,t ) + β 4 X ij,t + β 5 d β 6 X ij,t d ij,t + ij,t, n=1 30 The interaction effect with the first stage of the EMU, d1990, is reported. Similar results can be obtained using interactions with EMU or d1999. N Z n 18

20 where X ij,t refers to the variable of interest. The estimated coefficient of the dummy variable for the first stage of the EMU, β 5, the effect of the interaction term, β 4, as well as the own effect of the variable of interest, β 6, are reported in Table 11. First, financial market structure and development are considered as factors of interest that differentiate the EMU effect on transactions across countries. Second, the role of information costs proxied by distance is explored. Finally, the link between the volume of transactions and real integration is tested in order to investigate the quantity puzzle Financial market structure and development An indicator variable that equals one if a country is more market-based and zero if it is more bank-based, called market i,t and market j,t, is included. It is used to investigate whether economies with higher stock market or with higher banking activity respond differently. In bank-based systems, banks play a leading role in mobilizing savings and allocating capital whereas in market-based systems securities markets take the role of getting society s savings to firms, exerting corporate control and easing risk management. Demirgüc-Kunt and Levine (2001) find that financial systems tend to be more market-based in higher income countries where stock markets become more active and efficient than banks. In order to proxy the development and capacity of the banking sector to allocate capital, the amount of credit provided by the banking sector relative to GDP is included, credit i,t and credit j,t. In more developed markets, i.e. larger and more liquid markets, market prices are supposed to be more informative (Beck, Demirgük-Kunt and Levine (2001) and Wurgler (2000)). Efficient investment depends on the investor s ability to distinguish promising investment opportunities from mediocre ones and on the existence of sufficient financial intermediaries. Results in Table 11 show that, first, the effect of higher transactions due to stage one of the EMU is smaller for more market-based transacting countries i or, put differently, larger for bank-based transacting countries i. Second, transacting countries with more developed financial markets, in terms of more private credit provided by the banking sector relative to GDP, experience higher transaction volumes during stage one of the EMU compared to transacting countries with less developed financial markets. Taking both results together, bank-based financial systems ceteris paribus experience larger transaction volumes with stage one of the EMU. At the same time, the increase in transaction volumes depends positively on the development of the banking sector. Both effects only refer to 19