The Decline of EMU Core Countries’ Portfolio Equity Investments in the Euro Area: The Role of Stock Return Correlations

This paper investigates the effect of time-varying stock market correlations on the abrupt and persistent decrease of the reciprocal foreign investments by Euro area member countries after 2007. A strong stock market correlation reduces the diversification opportunities and it may therefore have affected the reciprocal investment by EMU countries. The 2007 represents both the outbreak of the global financial crisis and the beginning of the enlargement process. The two events have had a very different impact on the stock returns’ correlation among EMU members. While the enlargement to new countries has reduced the average returns’ correlation within the Euro area as a whole, the financial crisis and the sovereign debt crisis have led to an increase in stock returns’ correlation among old member states. We find that, among old EMU countries, core countries have been particularly affected by stock returns’ correlation. They reduced their equity investments both in foreign EMU core and in foreign EMU periphery economies after the crisis, with a particularly pronounced contraction witnessed by those country-pairs that displayed highly correlated asset returns. These results highlight the importance of the diversification motive for international portfolio investments and complement the institution-based explanation of the decline of the Euro area investments.


Introduction
After the inception of the European Economic and Monetary Union (EMU), Eurozone countries displayed disproportionately large investment in other members' assets, both in bonds (Lane, 2006;Giofré, 2013), and in equities (Lane and Milesi-Ferretti, 2007;Balta and Delgado, 2009;Slavov, 2009). However, after the financial crisis, bilateral cross-border portfolio equity holdings within the EMU area experienced an even more abrupt and persistent fall than the general downfall of international financial flows (Lane, 2013;Milesi-Ferretti and Tille, 2011). This peculiar evidence occurred in conjunction with the combination of two outstanding events, such as the enlargement process, on the one hand, and the financial crisis, later turned into the sovereign debt crisis, on the other hand. Giofré and Sokolenko (2022) highlighted that the crisis has drastically weakened the linkages among original members. A peculiar decline in economic development and, more importantly, a deterioration of the control of corruption standards of Euro periphery countries, those more severely injured in the sovereign debt crisis, induced a sharp decrease of their inward investments by the Euro area as a whole. Vermeulen (2013) found a significant negative relationship between foreign equity holdings and stock market correlations during the financial crisis, while no significant relationship was found before the crisis.
We conjecture that the stronger stock market correlation induced by the crisis might have reduced the diversification opportunities, and then the reciprocal investment by Euro area member countries. We find, indeed, that the increase in stock returns' correlation induced by the global crisis has played a significant role in explaining the change in the investment pattern of a subset of old EMU economies, that is, core countries. We highlight that the latter reduced equity investment in foreign core and periphery EMU economies after the crisis, and that the decrease has been particularly pronounced for EMU country pairs that displayed highly correlated asset returns. This paper contributes to the literature about the time-varying common currency effect on bilateral portfolio investments, by adding the diversification motive to the drivers of the decline of the within EMU investment after 2007, and complementing the institution-based explanation already highlighted in the literature.
The rest of the paper is structured as follows. In Section 2, we frame our analysis within the extant literature. In Section 3, we sketch the estimable equation. In Section 4, we describe the data and provide some descriptive statistics. In Section 5, we report the results of the empirical analysis. Section 6 concludes.

A Brief Literature Review
Since international investments aim at improving portfolio performance by reducing risk, large stock returns' correlations may jeopardize the benefits of international diversification.
More recently, however, Bergin and Pyun (2016) developed a testable model that incorporates the overall covariance structure of returns and found that the latter could help resolve the correlation puzzle: as predicted by the theory, international investors do seek destination countries with low comovement of returns with the home country.
The literature has furthermore emphasized that the correlation structure among stock returns is not stable over time: it is stronger during periods of high volatility than it is during periods of low volatility (Forbes & Rigobon, 2002;Ang & Bekaert, 2015). Christoffersen et al. (2012;2014), adopting various models to capture the dynamic dependence across equity markets, found that correlations have increased markedly, especially in developed markets. The benefits from international diversification therefore have reduced over time for advanced economies, while emerging markets still offer significant diversification benefits, especially during large market downturns.
A prolific literature investigated more specifically the change in returns' correlation after the global financial crisis and the European sovereign debt crisis. Wang et al. (2017) investigated stock market contagion during the global financial crisis from the US to the other six G7 and BRIC countries and found that cross-market correlations during the crisis period increased significantly. Gkillas et al. (2019) found that international equity portfolios experienced contagion effects through increased co-movements during the financial crisis and the UE debt crisis. Vermeulen (2013) disclosed a significant negative relationship between foreign equity holdings and stock market correlations during the financial crisis, while no relationship was detected before the crisis. Mollah et al. (2016) indicated that the benefits of portfolio diversification had significantly decayed during both crises for developed countries.
Importantly, the increase in correlation due to a crisis can persist. Correlation coefficients rise during economic crisis and decline again when the crash is over, but do not drop back to the original level as before the crisis (Teti et al., 2018). Zhang et al. (2013) found that the recent financial crisis has changed permanently the correlations between BRICS, U.S. and European stock markets, suggesting that the reducing diversification benefits are a longrun and world-wide phenomenon, especially after crises.
The importance of returns' correlation may be especially relevant for EMU countries because comovement of stock returns is stronger for countries that are close both in geographical and in cultural terms. Recent studies suggest indeed that the correlation of stock returns increases with decreasing geographical distance (Eckel et al., 2011) and that the number of overlapping opening hours and sharing a common border tend to increase crosscountry stock market correlation (Flavin et al., 2002).
Focusing on the European market integration, Polanco-Martínez et al. (2018) presented an analysis of EU peripheral stock market indices and the S&P Europe 350 index, as a European benchmark market, over the precrisis (2004)(2005)(2006)(2007) and crisis (2008)(2009)(2010)(2011) periods. They found that the correlation is stronger for the crisis than for the pre-crisis period.
Conversely, Goldstein et al. (2019) disclosed that equity markets of the advanced European Union countries tend to comove more closely with each other than with the peripheral economies, but that the relative comovement structure before, during, and after the global financial crisis has been very stable.
When restricting the analysis to the Euro area, the empirical evidence depends on both the time and the crosscountry dimension.
Pizzutilo (2012) considered the pre-crisis period (2001)(2002)(2003)(2004)(2005)(2006)(2007) and highlighted that, notwithstanding the high degree of correlation among the Eurozone stock markets, opportunities for diversification were still present. Sehgal et al. (2017) examined the stock market integration process amongst EMU countries from 2002 to 2013, thus encompassing a normal period as well as the global financial crisis and Eurozone debt crisis periods. The results suggested that large-sized EMU economies exhibited stronger stock market integration when moving from normal to crisis periods.
The evidence of heterogeneity in the correlation patterns within the Euro area over time and across-countries, suggests the need to investigate further the evolution of the effects of returns' correlation on international investment after the crisis and to identify the peculiar patterns of different EMU members.

Estimable Equation
Our baseline estimation builds on the following specification: (Note 1) (1) The dependent variable, log(FPE sh ), is the logarithm of the foreign portfolio equities (FPE) of source country s in host country h.
Our regression specification accounts for pair-specific regressors (Z sh or W sh ), such as the returns' correlation, country-specific variables (X h , Y s , Q h , T s ), such as size and institutional variables, and time factors (D).
Continuous regressors (Q h , T s and W sh ) are expressed in logarithmic terms, so that their coefficients can be easily interpreted in elasticity terms (e.g., if a significant coefficient is equal to 0.3, then a 10% increase in the regressor induces a 3% increase in the dependent variable). Conversely, the effect of a dichotomous variable (X h , Y s and Z sh ) on a dependent variable expressed in logs is captured by the following transformation of its coefficient β: e β -1. (Note 2) EMU sh is a bilateral-specific dummy variable taking value 1 when both the source country s and the host country h are EMU members, and 0 otherwise; EMU s (or EMU h ) is instead a country-specific dummy variable equal to 1 when the source country (or host country) is a EMU member, and 0 otherwise.
Finally, D is a dummy capturing the time dimension, such as pre-or post-crisis period, which allows us to detect any global shift in foreign investment due to macroeconomic shocks.
We adopt a Difference-in-Difference approach: the econometric specification (1) is enriched to include the interactions of the EMU dummies with other factors (A), to estimate to which extent a factor A affects the FPE among EMU countries, on top of the global effect played by A on FPE. (Note 3) (2) To address the problem of inflation of zero investment observations and the need to get estimates robust to different patterns of heteroskedasticity, typical of standard trade log gravity models, we follow the econometric strategy developed by Santos Silva and Tenreyro (2006). Accordingly, we model the dependent variable FPE sh as following a Poisson distribution: we apply the Poisson Pseudo-Maximum Likelihood estimator, with year dummy, individual fixed effect -that in our case corresponds to country-pair fixed effects-and with standard errors adjusted for twoway clustering, at the investing-destination country pair and year levels.

Data
We draw the bilateral equity portfolio investments for the sample of 68 countries, in the period 2001-2017, from the Coordinated Portfolio Investment Survey (CPIS), released by the IMF. (Note 4) Netherlands, Luxembourg, Hong Kong SAR, the British Virgin Islands, Bermuda, the Cayman Islands, Ireland, and Singapore-[hosting] more than 85 percent of the world's investment in special purpose entities, which are often set up for tax reasons" . (Note 6) Table 1 reports the variables included in our analysis and their main descriptive statistics. Note. This table reports the descriptive statistics of the dependent variable and the regressors used in the analysis. The subscript sh refers to the country-pair sh,* indicates that the corresponding variable is included in the analysis for both the destination and the investing country.

Descriptive Statistics
The first panel reports the dependent variable, i.e., the bilateral portfolio equities holdings expressed in US$. They range from zero to 1295 billions of US$, with a median of 8.10 millions and a standard deviation of 29 billions.
The second panel refers to the main regressor, that is, the bilateral stock returns' correlation variable, with a mean equal to 0.34, a median of 0.37 and a standard deviation equal to 0.62.
The third panel comprises all other regressors, and is further split into sub-groups. The size variables included are GDP per capita and GDP in US$. The GDP per capita of source and host countries shows a large dispersion among countries: the GDP per capita mean is 24327 US$, while 50% of the sample has a GDP per capita lower than 16681 US$. The minimum value is equal to 447 US$, while the maximum is 119225 US$, with a standard deviation of 21977 US$. A notable degree of dispersion is also present in the GDP in US$ variable.
With the exception of the distance, the bilateral gravity covariates are all binary variables and capture the presence of a common border, language, legal origin or colonial linkages.
The capital mobility variable ranges from zero to 10, to indicate increasing levels of capital mobility: its mean is equal to 4.48, the first quartile is equal to 1.54, while the third quartile is equal to 6.92.
Finally, the institutional variables refer to the control of corruption and are drawn from the Worldwide Governance Indicators (WGI, World Bank) and from Transparency International. These variables' index goes from 0 to 100, reflecting increasing country governance standards. (Note 7)

The Dynamics of FPE and EMU Linkages
In Figure 1, we show the dynamics of bilateral foreign portfolio equities (FPE), reported in Giofré and Sokolenko (2022). Panel a) reports the trend of bilateral foreign investment over years for all countries in the sample. The ijbm.ccsenet.org International Journal of Business and Management Vol. 17, No. 5;2022 figure displays an increasing pattern of world FPE until 2007, a drop in 2008, reflecting the abrupt fall in financial flows due to the crisis (Lane, 2013;Milesi-Ferretti and Tille, 2011), and then a recovery up to a level more than 3 times larger than its initial level. Panel b) reports the trend of FPE among EMU member countries over time. This figure shows, similarly to panel a), a fall in the FPE among EMU countries since 2007; differently from panel a), there has been no recovery, but rather a slow decline down to 40 percent of its initial level. This suggests that Euro area specific dynamics can be responsible of the persistence in the decline of FPE among member countries. To properly address this issue, we analyze below the EMU effect in a multivariate regression, properly accounting for the heterogeneity within the EMU group. Table 2 replicates, as a benchmark, the analysis in Giofré and Sokolenko (2022), but considers the time period split in Period 1 (benchmark, excluded category) and Period 2 (2007-2017), rather than into crisis and post-crisis dummy. The dependent variable is the log of bilateral foreign equity investment (FPE). As specified above, the coefficients of all regressors expressed in logs can be interpreted in elasticity terms, while the effect of dummy variables on the dependent variable is captured by the coefficient β as follows: e β -1.
In column 1, we include standard gravity variables, used in literature to define the cultural and geographic proximity between two countries for equity flows (Portes and Rey, 2005;Portes et al., 2001), and equity holdings (Chan et al., 2005). We expect geographical and cultural proximity to have a positive impact on foreign portfolio equities, as a decrease in physical and cultural distance reduces information costs, and then enhances investment by foreign investors. Consistently with the gravity model approach, we also include the size variables, that express the economic weight of the investing and host countries, such as market capitalization and GDP per capita, and finally we control for capital mobility.
The results, as predicted, show that the gravity variables have a strong impact on the FPE allocation. The distance variable displays a negative coefficient (-0.069), meaning that an increase in distance is associated with a decrease in foreign portfolio equities. Sharing a common border leads to an increase of FPE by 49% (e 0.401 -1 = 0.49), having an official language in common increases FPE by 84%, having a common colonial past increases equities share by 4.1 times, while the common legal origin seems to have no impact. The contribution of the size variables also appears important.    is normalized to 1, so that the other coefficients are defined in relative terms.
Stock market capitalization has a significant and positive impact on the FPE, which appears to be stronger for the host (0.783) than for the source country (0.567). The opposite happens to the GDP per capita variable: an increase in the source country's GDP strongly fosters foreign investment, with a non-significant effect for the host countries. Capital mobility variables play a positive significant role on the investing size, and a modest negative impact on the host side.
The coefficient of the bilateral EMU dummy variable over the whole period is equal to 0.562, that is, EMU countries invest one another 75% more than other country-pairs. In column (2) of Table 2, we include a dummy for the period 2007-2017 (Period 2), that is, the declining period observed in panel (b) of Figure 1. Since we also include the interaction between the Period 2 dummy and the EMU sh dummy, the coefficient of the Period 2 dummy (-0.299) captures the general fall (-26%) of bilateral FPE for non-EMU country pairs, while the coefficient of the EMU sh dummy (0.733), is referred to the pre-financial crisis period, and is large, positive and statistically significant (+108%). The effect of the EMU dummy in the subsequent period is computed by adding up the coefficient of the corresponding interaction term (EMU sh · Period2) to the non-interacted one (EMU sh ).
The negative coefficient of the interaction term EMU sh · Period2 can be interpreted as the change of the EMU effect (on FPE) induced by the crisis (or, symmetrically, as the change of the crisis effect on FPE for EMU country pairs). It is negative and significant, thus suggesting a significant drop from 108% (e 0.733 -1) to 67% (e 0.733-0.220 -1) in the EMU effect on FPE, relatively to the pre-crisis period (or, symmetrically, a more negative effect of the crisis for EMU country pairs).  dummies, the Period 2 dummy, and their interactions. The regression specification also includes year dummy, individual country pair fixedeffects and standard errors adjusted for two-way clustering at the investing-destination country pair and year levels. ***, **, and * indicate significance at the 1, 5, and 10% levels, respectively.
The results in Table 2 confirm, in a multivariate setting, that after 2007, the common currency effect on bilateral FPE has significantly fallen, and therefore the linkages among EMU countries have significantly loosened. The results are quite similar, when restricting to the sample of OLD EMU members, either on the investing side ( (2a) and (2b)), or on the destination side ((3a) and (3b)), or on both ((4a) and (4b)). As emphasized by Giofré and Sokolenko (2022), this points to a slackening of the linkages among the original members because of the crisis, more than to the inclusion of new countries eventually less connected with the euro area.

OLD EMU Group Decomposition
In Table 3, following Lane and Milesi-Ferretti (2017), among others, we distinguish within the OLD EMU group, the Euro core countries (Austria, Belgium, Finland, France, Germany, Luxembourg and the Netherlands) and the Euro periphery or Euro crisis countries (Greece, Ireland, Italy, Portugal and Spain), and investigate their investment patterns.
In column (1a) and (1b) of Table 3, we observe that bilateral FPE among Euro core countries fell from 93% to 64%, after the crisis. When considering the cross investment of Euro core countries in Euro periphery countries (columns (2a) and (2b)), the drop is much more relevant, starting from 94% to only 11% larger investment, after the crisis. ijbm.ccsenet.org International Journal of Business and Management Vol. 17, No. 5;2022 Table 2 are included, but not reported. ***, **, and * indicate significance at the 1, 5, and 10% levels, respectively.
When considering the investments by Euro periphery countries, we observe that their investment in Euro core countries ((3a) and (3b)) are on average 51% larger than other country pairs, while their investment in periphery economies are 102% larger than other country pairs, with no substantial difference between the pre-crisis and the subsequent period.
These results solicit a specific deeper investigation of the investment by core countries in other OLD countries' assets.
In particular, we conjecture that the investment diversion out of the Euro area is attributable to the pursuit of diversification benefits, that might have shrunk within the area, as a consequence of the stronger correlation induced by the global and sovereign debt crisis.

The role of Returns' Correlation
In Figure 2, we report the dynamics of the bilateral one-year lagged correlation of monthly returns, after normalizing to 1 its 2001 value. Panel a) shows the dynamics of the worldwide stock correlation, with a not clearcut behavior. The returns' correlation in the EMU group (panel b)), shows instead a slightly decreasing pattern. This trend appears however mostly related to the enlargement of the Euro area: when restricting in fact to the OLD EMU sample, we observe an increasing trend, which appears particularly pronounced for the core countries, as evident from the y-axis scale. ijbm.ccsenet.org International Journal of Business and Management Vol. 17, No. 5;2022 In Table 4, we report the results of a regression with the stock return correlation as dependent variable. The first column shows an increase in worldwide stock correlation in the second period. While it declines for the whole EMU group, when focusing on OLD EMU countries, this correlation significantly increases after the crisis, thus confirming what shown in Figure 2: the decreasing trend of the whole EMU group is attributable to the entrance of new members featuring less correlated stock markets. Note. This table is the same as Table 5, but the binary High stock returns' correlation variable (H correl s,h ) is defined relative to the median, rather than to the mean.
In Table 5, we add to the econometric specification the stock returns' correlation variable and its interaction with the EMU dummies. We aim to assess how foreign portfolio investments in the Euro area have responded to diversification benefits and to detect to which extent their contraction contributed to explain the fall of core countries' investment in OLD EMU assets shown in Table 3.
To enhance the interpretability of the results' size, we include, rather than the continuous version of the correlation variable, its dichotomic counterpart, H correl sh , equal to 1 if the correlation between the stock returns of country s and h is larger than the mean, and 0 otherwise. (Note 8) Columns (#a) add only the correlation binary variable, while columns (#b) also include the interaction between this variable and the corresponding EMU dummy. The coefficient of the binary variable H correl sh in columns (#a) is non-significant. This finding is consistent with the literature, which highlights a non-systematic role of stock returns' correlation in explaining foreign portfolio equities for Euro area investors, in the first decade of the EMU (Lane & Milesi-Ferretti, 2007). However, the results in columns (#b) show that a systematic diversification motive of Euro area investors emerges after 2007, when the returns' correlation among member countries gets stronger.
We observe that, in the second period (column (1b)), the linkages among EMU members are even tightened, from 106% (e 0.724 -1) to 213% (e 0.724+0.417 -1), but only for those country pairs with weakly correlated stock returns (H correl sh = 0). The investments are instead hardly undercut, down to 62% (e 0.724+0.417-0.661 -1) for those country pairs ijbm.ccsenet.org International Journal of Business and Management Vol. 17, No. 5;2022 34 whose stock returns are highly correlated (H correl sh = 1). This effect in the EMU group is led by the OLD EMU dynamics and, more specifically, by the core countries' dynamics. Indeed, the negative impact of the correlation variable in the second period is larger for the OLD EMU sub-group than for the whole EMU group, and even larger for the CORE sub-sub-group. In particular, the linkages among core countries fall from 91% to 57% in the second period, but only for highly correlated country pairs, while the weakly correlated country pairs witness an even stronger reciprocal attractiveness. The underweighting of highly correlated assets is particularly dramatic when considering the investment of core countries in periphery countries: the strong attractiveness before the crisis (92%) turned into a deterrence (-28%), compared to other country pairs' FPE. This latter finding is consistent with the results in Giofré and Sokolenko (2022), who identify the core-periphery linkages as the ones that have experienced the harshest deterioration after the crisis.

Robustness: alternative regressors and sample specifications
To provide consistency to our findings, in Table 6 and 7, we subject our main results to robustness checks.
In Table 6, we test the sensitivity of our analysis to a redefinition of our main regressor, that is, the dichotomic correlation variable, H correl sh : we re-compute it relative to the median benchmark, rather to the mean, and the results are left substantially unchanged.   Table 2 are included, but not reported. ***, **, and * indicate significance at the 1, 5, and 10% levels, respectively.
In Table 7, we consider alternative definitions of the offshore centers. In columns (#a), we follow a classification, which, among EMU countries, excludes the Netherlands from the offshore list, but adds Cyprus, Latvia and Malta (Zoromé, 2007). In columns (#b), we extend the list of offshore centers to other EMU countries, such as Cyprus, Malta and Belgium, following the classification in Lane and Milesi-Ferretti (2017). (Note 9) We find that, under alternative definitions of offshore centers, the results, though marginally different in size, confirm the baseline specification's ones.

Returns' correlation, size and institutions
Finally, we try to reconcile our results with the previous findings in the recent literature. Giofré and Sokolenko (2022) find that the worsening of the economic condition and the weakened control of corruption mechanisms in peripheral countries have triggered the decline in bilateral FPE among EMU countries.
In Table 8, we check if the role of the returns' correlation is robust to the inclusion of these factors.
The first index considered (columns (1a) and (1b)) is the "Control of Corruption" drawn from Worldwide Governance Indicators (WGI, Source: World Bank), which captures the perceptions of the extent to which public power is exercised for private gain. As an alternative, in columns (2a) and (2b), we consider the "Perceived Control of Corruption index" (Source: Transparency International), which captures the perceived level of corruption in the public sector, relying on different country sources. As far as the size indicators are concerned, we consider the GDP per capital, as a proxy of the standard of living, and the GDP in US$, as a standard size measure.  Note. This table is the same as Table 8, but focuses on OLD countries.
Since in Giofré and Sokolenko (2022) the size and control of corruption indexes are dichotomic and defined relatively to their time-varying median, for consistency and the ease of comparability, the high returns' correlation binary variable is also defined relative to the median (as in Table 6).
Let us focus on columns (#b) of Table 8, reporting all interactions. Column (1b) shows that EMU investments in destination countries which display high values of control of corruption (less corrupted economies) are significantly larger (from 58% to 200%), but this larger effect is significantly dampened by a high returns' correlation with domestic assets (174%). The same evidence emerges when considering the role of the alternatively defined perceived corruption (column (2b)), of GDP per capita, or nominal GDP (columns (3b) and (4b)).
We replicate the analysis of Table 8 restricting to OLD EMU in Table 9 and to core countries in Table 10. In Table  9, the results found in Giofré and Sokolenko (2022) are confirmed, while the evidence of the diversification motive is weaker.
When focusing on core countries in Table 10, the results resume significance: hence, the non-significant interaction coefficients for OLD EMU in Table 9 are due to periphery investing countries. Table 10 shows that core countries are significantly driven by diversification motives, on top of other concurrent factors: a higher returns' correlation, inevitably produced by the persistent crisis, has reduced the diversification opportunities of core countries in member countries' assets, thus contributing to the general shrunk in bilateral portfolio investment within the Euro area. ijbm.ccsenet.org International Journal of Business and Management Vol. 17, No. 5;2022  Note. This table is the same as Table 8, but focuses on CORE countries.

Conclusions
This paper focuses on the contraction of core EMU countries' investments in the Euro area. We find that the increase in stock returns' correlation induced by the global crisis has played a significant role in explaining the change in the investment pattern of core countries towards other EMU members after 2007. An increase in returns' correlations implies lower diversification opportunities and may turn into lower investments. We find indeed that core countries reduced equity investment in core and periphery economies after the crisis, and that the decrease has been significantly larger in those economies offering fewer diversification opportunities, due to the higher return correlation with the domestic assets.
This paper adds the diversification motive to the drivers of the fall of the within EMU investment after 2007, and complements the size-and institution-based explanation already highlighted in the recent literature. Core countries investors are confirmed to have underweighted assets issued by countries with poor control of corruption, but, as a novel finding, this effect has proven to be particularly harsh for destination EMU countries with returns highly correlated with the domestic ones.