The Role of Country Concentration in the International Portfolio Investment Positions for the European Union Members
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

Contributor Notes

Authors’ E-Mail Addresses: ibrushko@cerge-ei.cz, YHashimoto@imf.org

This paper examines the international portfolio flows of European Union. Our analysis includes three dimensions: (1) the level of countries portfolio investment concentration (those who invest evenly among counterparties versus those who invest more heavily in some counterparties); (2) the share of total portfolio investment assets invested at the destination; and (3) pre- and during the crisis periods. We find that portfolio investment positions respond differently to macroeconomic variables depending on the level of investment concentration and the share of invested assets. In particular, variables of health of the financial system become important determinants for portfolio investment during the crisis.

Abstract

This paper examines the international portfolio flows of European Union. Our analysis includes three dimensions: (1) the level of countries portfolio investment concentration (those who invest evenly among counterparties versus those who invest more heavily in some counterparties); (2) the share of total portfolio investment assets invested at the destination; and (3) pre- and during the crisis periods. We find that portfolio investment positions respond differently to macroeconomic variables depending on the level of investment concentration and the share of invested assets. In particular, variables of health of the financial system become important determinants for portfolio investment during the crisis.

I. Introduction

Although it is known that investing internationally provides opportunities for diversification, investors do not use these opportunities fully. Among the reasons they have cited for not doing so are information asymmetry, preferences for home assets, transaction costs, and high risk-aversion. All these factors determine international investors’ motivation to invest abroad and, consequently, their investment strategies. Thus, for example, countries with higher risk aversion could be expected to diversify more, while countries with lower risk aversion might exploit more risky strategies by specializing more and forgoing the benefits of diversification. Home investors’ preferences for the assets of a particular country also determine their investment behavior. It was also shown that changes in risk aversion and risk appetite during an economic slowdown have an impact on portfolio rebalancing (Fu, 1993; Kumar and Persaud, 2001; Coudert and Gex, 2006; and Caceres et al., 2010).

Concentration may therefore play a crucial role in portfolio management. On the one hand, by forming more concentrated portfolios, investors may forgo the benefits of international diversification,2 on the other hand, they may enjoy an information advantage by investing in a limited set of assets (Kacperczyk et al., 2005; Ivkovic, 2008; and Huij and Dereall, 2009).

These findings suggest to us that different investment strategies are driven by different incentives for investing abroad. As the consequence of differences in the motivation to invest, one may expect correspondingly different responses to the same changes in the investment environment. Although there is an extensive literature on the relative benefits of diversification and concentration, studies are scarce on the specific question of how the choice of one or the other can explain changes in international portfolio investment positions. To fill this gap in the literature, the present study tries to determine whether an investment strategy matters for international investment reallocation. Investment strategy is characterized by two dimensions: investment type—low or high concentration—of the source countries; and investment share at the destination. Investment concentration measures the extent to which a source country has concentrated its investments; that is, a country that invests more evenly among many counterparties is considered a low-concentration country, whereas a country that invests heavily in only a few counterparties is considered a high-concentration country.

The investment share at the destination measures the proportion of a source country’s total portfolio investment that is invested in the destination country. This measure highlights the importance and exposure of a source country to the destination countries. The hypothesis is that the investment type is driven by particular investment motives and considerations, while the second dimension (the share of the portfolio a country invests in a counterparty) might be determined more by investor’s preferences for particular sectors, asset types, or a particular country. We expect that both of these factors (investment concentration and share) determine changes in portfolio positions. We also compare portfolio positions before and during the crisis. The portfolio positions could be affected considerably, especially during the crisis, because the risk-aversion rises (Guiso and Paiella, 2008) and the correlation of the asset returns across countries increases sharply during economic slowdowns (Forbes and Rigobon, 2002).

We contribute to the literature in several ways. First, we take into account macroeconomic conditions when investigating the country’s portfolio positions and the level of uncertainty. Second, using the Coordinated Portfolio Investment Survey (CPIS) data, we look at the investment type (high- or low-concentration) and the investment share at the destination to identify the countries’ portfolio investment decisions. Finally, we split the sample period into two—before and after the global financial crisis—to examine the impact of the crisis on portfolio investment positions.

The empirical results of this study suggest that investment type (concentration) plays a role in explaining changes in international portfolio investment positions. The results also show that low-concentration countries respond in the opposite direction relative to the high-concentration type: when the portfolio flows exhibit a negative reaction to the changes in the macro variables for the high-concentration type, the low-concentration type reaction to such variables is less negative or is positive; conversely, the positive reaction of the high-concentration type is accompanied by a less positive reaction of the low-concentration type. Moreover, the analysis reveals that the crisis changes the set of the variables that elicit these differing responses from the two types: before the crisis, the types mainly differ in their responses to the general macroeconomic conditions (GDP growth, CPI, unemployment, government debt, etc.), while during the crisis the types put different weights on the variables that can signal the health of the financial system (short- and long-term interest rates, stock index and stock index growth). We believe that our findings can help policymakers to predict and manage severe capital outflows that can occur when a country faces unexpected (exogenous) market liquidity shocks and contagion.

The paper is organized as follows. Section II introduces the methodology. The main variables and data sources are discussed in Section III. The main empirical results are discussed in Section IV. Section V concludes.

II. Model and Methodology

The main aim of this paper is to investigate the extent to which the international portfolio investments change depending on macroeconomic conditions. In our model we include both macroeconomic variables and the uncertainty of these macroeconomic variables in a given year, because uncertainty will have a crucial role for the international investment decision.

Our basic dynamic panel model of international portfolio investment positions is represented by equation (1).

Yi,j,t=α+ρYi,j,t1+βXj,t+μi,j+ηt+ei,j,t(1)

where subscripts i, j, and t denote the country of origin, the country of destination, and time, respectively. Yi,j,t is the vector of dependent variables of either the logarithm of the ratio of total portfolio assets, equity securities, or debt securities, invested by country i into country j in year t evaluated at the market value to GDP. Yi, j, t – 1 is the vector of the logarithm of the portfolio investment to GDP ratio in the previous year, and it is included into the estimation equation in order to take into account the persistence of the series. Xj,t is the matrix of control variables for country j in year t and consists of the macroeconomic variables and their standard deviations listed in footnote 3.3 μi,j represents the vector of unobservable fixed effect between countries i and j such as culture, history, geography, economical or social interconnection. ηt denotes the vector of common time-specific unobservable effects. ei, j, t is the vector of the error term with zero mean and constant variance.

In estimating equation (1), endogeneity problems will arise in our specification, as changes in the portfolio flows affect short- and long-term interest rates: the outflows move rates up because less funds are available for borrowing (i.e., the excess demand for money) and inflows move them down (Warnock and Warnock, 2009; Kuori and Porter, 1974), but changes in the short- and long-term interest rates also determine the decision to invest Through interest rates channels, other right-hand side macroeconomic variables are also influenced by changes in the portfolio flows. For example, changes in interest rates affect the government debt through the borrowing costs, making the government debt as an endogenous variable.

To overcome the endogeneity issues, we use the system dynamic panel GMM estimator developed by Arellano and Bond (1995) and extended by Blundell and Bond (1998). To solve the endogeneity problem, Arellano and Bond (1991) suggest using the lags, starting from the second one as instruments for the endogenous variables in the first difference equation. The system dynamic panel provides efficiency by estimating both the equations in levels and in the first differences. In the equation in the levels, the estimator uses the lags of first differences of the endogenous variables, while in the equation in differences, the second lag of the endogenous variables is used as an instrumental variable. For the lags and the lags of first differences of the variable to be the appropriate instruments for the endogenous variables, the following conditions should hold:

For the equation in levels:

E[ΔYi,j,t1(μi,j+ei,j,t)]=0andE[ΔXi,j,t1(μi,j+ei,j,t)]=0(2)

For the equation in first differences:

E[Yi,j,t2Δei,j,t]=0andE[Xj,t2Δei,j,t]=0(3)

for t≥3.4

For distinguishing high- and low-concentration types, we use the approach of Kacperczyk et al. (2007) who developed the industry concentration index. We calculate the concentration index according to the following formula:

CIi,t=ΣjN(ωi,j,tω¯j,t)2(4)

where ωi,j,t is the share of total portfolio investment assets of country i invested in country j in year t and ω¯j,t5 is the region average share of total portfolio investment assets invested in country j.6

The concentration index should reflect how the portfolio of a particular country deviates from the benchmark portfolio. The benchmark portfolio measures the average weights of invested assets in a destination country, and the time trend of concentration index is shown in Figure 1. There are two main reasons for using this index. First, the index is adjusted for a country’s attractiveness for investments (i.e. if the country is perceived to offer better investment opportunities, the average weight of investments in this country will be higher). Second, it takes into account the time-varying optimal investment share. The second feature is currently especially topical, because we can observe the flight to quality and higher capital inflows into such countries as Germany and Unites States. Once the concentration index is calculated for all sample countries, the countries are classified either as a high- or low-concentration group. Countries with concentration index equal to or higher than the 65th percentile are classified as the high-concentration type. The low-concentration investment types are those countries with the concentration index equal to or below the 35th percentile.7

Figure 1.
Figure 1.

Time trend of concentration index by country

Citation: IMF Working Papers 2014, 074; 10.5089/9781475543759.001.A001

In order to test whether high- and low-concentration types differ in their responses to the changes in the macroeconomic variables, we modify equation (1) and estimate equation (5):

Yi,j,t=α+ρYi,j,t1+βXj,t+γdlow*Xj,t+ηt+ei,j,t(5)

where dlow is a dummy variable and equals 0 if the data come from the sub-sample of the high-concentration type and 1 if the data are for the sub-sample of the low-concentration type. If we find the estimates of γ to be significant, this will provide evidence that these investment types differ in the responses to the macro variables.

III. Data and Main Variables

The set of control macroeconomic variables includes short-term interest rate, long-term interest rate, stock index, growth of stock index, unemployment rate, the growth of seasonally adjusted consumer price index (inflation), real effective exchange rate, GDP growth, the ratio of current account to GDP, and government debt to GDP ratio.8 To take into account the uncertainty associated with the macro variables, we also include the standard deviations during a given year of such variables as short-term interest rate, long-term interest rate, growth of stock index, stock index, unemployment rate, and seasonally adjusted consumer price index growth.

We include the short-term interest rate because it can signal market liquidity (Bomfim, 2003; Chordia, 2000) and reveal the spread and the cost of offsetting the position by the security issuer or trader (Aiyagari and Gertler, 1999). The change in the short-term interest rate results in the profitability of the investment and subsequently in the motivation to invest in or withdraw from a country. Besides, a change in the short-term interest rate may be perceived as a change in monetary policy and an increase (decrease) in interest rates pushes stock prices and yields to maturities down (up) (Rigobon and Sachs, 2002), which results in portfolio outflows (inflows). The standard deviation of the short-term interest rate will proxy for the uncertainty of market liquidity.

Long-term interest rates are usually used by investors to discount the future cash flows that determine asset prices. That is why a change in the long-term interest rate results in asset price movements and capital gains or losses from the investments, while the volatility of long-term interest rates contributes to the risk of asset returns. Gagnon et. al. (2011) show that a decline in the long-term interest rate may signal a reduction in the risk premium which will also have an impact on portfolio withdrawals.

Because our dependent variable is the international portfolio investment position (investments in total portfolio investment assets, equity securities, or debt securities) at market value, we also include the change in the stock index which should control for the changes in the asset valuation as well as for the change in the position itself. The standard deviation of the stock index is used to control for the volatility of the asset prices.9

Investors’ decisions to invest or disinvest are also influenced by the returns they earn from investing in a particular country (Brennan and Cao, 1997). Because we do not have the data on the returns, we use instead the growth of the stock index. The inclusion of the growth of the stock index is motivated by the fact that passive strategies are optimal and one cannot outperform the market (Burton, 2003 and Monnier and Rulik, 2012). The standard deviation of the growth of the stock index will also be a proxy for the risk of returns.

Valuation Changes10

In estimating the impact of international portfolio investment positions, it is important to separate valuation changes from actual changes in investment positions. Simply taking a difference of two time series end-period investment positions does not fully reveal information on country’s investment strategy because these data include changes in asset prices (valuation changes). Asset price changes could mask the actual transaction values especially when the market is volatile and could make misleading interpretation of shift in investment positions. In order to measure valuation changes, detailed data such as asset type, maturity, and prices are necessary. However, it is difficult to make an accurate estimation of valuation changes mainly due to the lack of such data. The IIP and CPIS data provide information on the broad composition of assets held in the form of equity and debt securities, but they do not provide details such as maturity and currency. Also, data on returns and bond indices are limited. One way to handle valuation changes is to include macroeconomic variables that proxy volatility in estimation equations. Interpretations of these variables are as follows. First, considering valuation changes as noise in the market and inclusion of macroeconomic variables as control variables can separate out the noise. Second is to view asset prices as being part of the outcome of the portfolio allocation decision. When demand for a certain type of asset increases, its price goes up as a consequence of that portfolio choice, so that portfolio reallocation is achieved in part by valuation changes, rather than by actual flows. In this regard, it is assumed that the portfolio reallocation through valuation changes (exchange rate) is not the case for our sample countries because they share the same currency.

The bilateral data on the portfolio returns are not available, but the balance of payments provides the total income on the portfolio investment for a country. We construct a new variable, which is calculated as the yearly portfolio income divided by the assets at the end of the previous year, and call it the returns on assets. This variable should measure the profitability of countries’ portfolio investments, while the standard deviation is supposed to signal the risk of the investments. Figure 2 depicts the relation between the standard deviation of returns and the concentration index. For both debt and equity securities, one might observe that the dispersion of standard deviation of returns on portfolio investments decreases with the increase in the concentration index. This may imply that low-concentration countries may tolerate higher risk associated with investments.

Figure 2.
Figure 2.

Standard deviation (SD) of returns on portfolio investments

Citation: IMF Working Papers 2014, 074; 10.5089/9781475543759.001.A001

The unemployment rate is included as a leading indicator of stock performance, because the higher unemployment today implies lower GDP tomorrow and, as a result, lower stock returns (Flannery and Protopapadakis, 2002; Boyd et al., 2005); the standard deviation of the unemployment rate will also reflect the uncertainty of the stock performance. Inflation is included as a lagging indicator for the security analysis; the standard deviation of this variable should also take into account the risk of the asset returns.

We also include the government debt to GDP ratio as one of the explanatory variables. An increase in government debt drives up the demand for financing, which results in higher interest rates and consequently in higher borrowing costs, lower profit margins, and lower returns on assets. Moreover, an increase in the government debt motivates the government to increase the government bond supply which affects, as Maier (2006) shows, market liquidity and the returns that investors require.

In our analysis, we also use the time dimension because investing countries face different global macroeconomic conditions. We provide the evidence in Figure 3, which represents the comparison of distributions of the main variables used in the analysis before the crisis and after the crisis. As one would expect, the mass of distribution of short-term interest rate shifts towards the right, which implies that more of the destination countries were facing liquidity constraints that pushed the short-term interest rate upwards. What is interesting is that there was also a shift of the mass of the distribution to the left. We believe that could happen due to capital reallocation whereby some countries enjoyed capital inflows as the result of “flight-to-quality”. This hypothesis can be also supported by the leftward shifts of the mass in the long-term interest rate, which can proxy for the yields on long-term government debt instruments. The increased demand for these instruments could push prices up and yields (long-term interest rate) downwards. These shifts also tell us that the countries became more heterogeneous in the level of short-term interest rates during the crisis.

Figure 3.
Figure 3.
Figure 3.

The comparison of distributions of main variables before and during the crisis

Citation: IMF Working Papers 2014, 074; 10.5089/9781475543759.001.A001

We do not observe severe changes in the distributions of the unemployment rate, the stock index, and stock index growth; however, there is evidence of a slight increase in the dispersion of the stock index and a shift of mass in its growth, as expected, to the left during the crisis. On the contrary, the distributions of such variables as GDP growth, inflation, relative effective exchange rate, current account to GDP ratio, and government debt to GDP become more dispersed and flat, which also points to an increase in the heterogeneity in global macroeconomic conditions.

Figure 3 also gives evidence of an increase in all our measures of risk (except for the standard deviation of the stock index): there was a shift to the right in the mass of the distributions of such variables as the standard deviation of short-term and long-term interest rates, the standard deviation of stock index growth, the unemployment rate, and inflation.

We use the data of the international investment positions from the CPIS, which are available on an annual basis over 2001 to 2010. The dataset provides bilateral cross-country data for individual countries, i.e. the international portfolio investment of country i in country j, in addition to the global (aggregated) annual data. Data are available for the total portfolio investment positions and are also broken down by asset type: equity securities and debt securities. For debt securities, data are further classified into long- and short-term debt. The CPIS data provide two dimensions—the asset side and the liability side—for total portfolio assets, equity and debt securities. For both of the dimensions and for every concept, the CPIS contains two data entries. The asset side data represent residents’ holding of securities issued by nonresidents (outward investment), i.e., portfolio investment by country i in country j is recorded as portfolio investment assets of country i. The liability side data include securities issued by residents and owned by nonresidents (inward investment), i.e., portfolio investment by country j in country i is recorded as portfolio investment liabilities of country i.

The dataset is unbalanced, and if there is a missing value for any of the variables, the missing values are substituted with the derived value if available.11 For our analysis, we focus on the portfolio investments of European Union members as the country-investors. The set of countries includes Austria, Belgium, Bulgaria, Cyprus, the Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Poland, Portugal, Romania, the Slovak Republic, Slovenia, Spain, Sweden, and United Kingdom. We take the annual data on GDP, government debt, and current account balance from the World Economic Outlook database. From the Global Data Source dataset we take quarterly data of the short- and long-term interest rates, stock index, and unemployment rate. The quarterly data on seasonally adjusted CPI and relative effective exchange rates are taken from the IMF dataset.

IV. Empirical Results

The estimation results for the full sample period (2001-10) are presented in Table 5. The first three columns (columns 1, 2, and 3) of the table provide the estimates of equation (5), where the dependent variable is total portfolio investment assets, assets invested in equity securities, and assets invested in debt securities, respectively. The last three columns (columns 4, 5, and 6) are the estimates of γ from equation (5).

Table 1.

Variables and data sources

article image
CPIS—Coordinated Portfolio Investment SurveyWEO—World Economic OutlookGDS—Global Data Source
Table 2.

Investment type across countries

article image
Table 3.

Correlation matrix

article image
article image
*** p<0.01, ** p<0.05 * p<0.1
Table 4.

Descriptive statistics of the variables

article image
Table 5.

Portfolio investments, 2001–10

article image
Robust standard errors are in parentheses*** p<0.01, ** p<0.05, * p<0.1

Tables 6, 7, and 8 represent the estimation results of equation (5) for the full sample period but for the subset with the share of invested assets equal to or below 1 percent, in the range between 1 and 7 percent, and equal to or above 7 percent, respectively. Tables 912 provide the estimation results for years 2001-06 using the full sample, and the sub-samples with the share of invested assets below or equal to 1 percent, between 1 and 7 percent, and equal or above 7 percent, respectively. The estimation results for the sample years 2007–10 for the full sample, and with share of invested assets less or equal to 1 percent, in the range of 1 and 7 percent, respectively, and equal to or above 7 percent, are presented in Tables 1316. We also provide the map for the significant γ and their signs for the full sample period, for years 2001-06, and for years 2007-10, respectively, in Tables 17, 18, and 19.

Table 6.

Portfolio investments, 2001-10, share below or equal to 1 percent

article image
Robust standard errors are in parentheses*** p<0.01, ** p<0.05, * p<0.1
Table 7.

Portfolio investments, 2001-10, share in the range of 1 and 7 percent

article image
Robust standard errors are in parentheses*** p<0.01, ** p<0.05, * p<0.1
Table 8.

Portfolio investments, 2001-10, share equal to or above 7 percent

article image
Robust standard errors are in parentheses*** p<0.01, ** p<0.05, * p<0.1
Table 9.

Portfolio investments, 2001–06

article image
Robust standard errors are in parentheses*** p<0.01, ** p<0.05, * p<0.1
Table 10.

Portfolio investments, 2001–06, share below or equal to 1 percent

article image
Robust standard errors are in parentheses*** p<0.01, ** p<0.05, * p<0.1
Table 11.

Portfolio investments, 2001–06, share in the range of 1 and 7 percent

article image
Robust standard errors are in parentheses*** p<0.01, ** p<0.05, * p<0.1