Access to information and international portfolio allocation

Journal of Banking & Finance 37 (2013) 2255–2267

Contents lists available at SciVerse ScienceDirect

Journal of Banking & Finance journal homepage: www.elsevier.com/locate/jbf

Access to information and international portfolio allocation Chandra Thapa a,⇑, Krishna Paudyal a,1, Suman Neupane b,2 a b

Department of Accounting and Finance, University of Strathclyde, Glasgow G4 0LN, Scotland, United Kingdom Department of Accounting, Finance and Economics, Griffith University, Brisbane, Queensland 4111, Australia

a r t i c l e

i n f o

Article history: Received 1 March 2012 Accepted 8 January 2013 Available online 9 February 2013 JEL classification: G11 G14

a b s t r a c t We examine whether foreign equity holdings of portfolio investors depend on the level of information accessibility between the investors’ home and host countries. Using a comprehensive data set, alternative measures of information accessibility and robust analytical techniques, we show that differences in access to cross-country information significantly influence investors’ portfolio allocation decisions. Furthermore, the results suggest that for a given level of access to information, investors prefer to invest more in countries with a higher quality of legal/macro-institutions. Finally, the findings also confirm that the implications of information accessibility are more pronounced when markets are turbulent. Ó 2013 Elsevier B.V. All rights reserved.

Keywords: Access to information International equity portfolio Panel data models Legal/macro-institutions Turbulent market condition

1. Introduction Extant literature shows that, relative to the theoretical suggestions of the International Capital Asset Pricing Model (ICAPM), portfolio investors do not optimally allocate funds across foreign securities (Chan et al., 2005). This drawback in asset allocation is often conjectured due to the presence of the deadweight costs of holding foreign securities arising from different barriers to international investments, including the costs of accessing information about foreign markets and securities (see Cooper and Kaplanis, 1986, 1994; Solnik and McLeavey, 2009; Lau et al., 2010). Such arguments imply that if access to information on the investment environment of a particular foreign market is lower (higher), then the deadweight costs resulting from information accessibility should be higher (lower) for that market. Consequently, investors should invest less in a foreign country where the access to information is lower and vice versa. We examine this prediction by analysing the implications of cross-country information accessibility on the foreign equity holdings of portfolio investors across a large number of countries. Primarily, this paper contributes to the following three major fronts of portfolio allocation and information accessibility issues. ⇑ Corresponding author. Tel.: +44 (0) 141 548 3891; fax: +44 (0) 1415 523547. E-mail addresses: [email protected] (C. Thapa), [email protected] (K. Paudyal), s.neupane@griffith.edu.au (S. Neupane). 1 Tel.: +44 (0) 141 548 2894; fax: +44 (0) 1415 523547. 2 Tel.: +61 (0) 73735 3500; fax: +61 (0) 73735 3719. 0378-4266/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jbankfin.2013.01.011

First, we investigate whether the temporal and cross-country variations in foreign equity allocations can be explained by the level of information accessibility, by focusing on the role of factors that represent access to information between investors’ home and foreign (host) markets. Although Chan et al. (2005)3 also examine this issue using cross section data; we extend the analysis by using a substantially large panel data set of foreign equity holdings of investors based in 33 home countries and investing in 36 host countries, and covering a long time period of 9 years (2001–2009). The use of long time series data in our panel set up allows us to investigate the possible implications of cross-country differences in access to information during periods of tranquil and turbulent economic conditions, especially in the context of the 2007–2009 financial crises. Moreover, Chan et al. (2005) use cross-sectional estimation methods while we employ more efficient and robust panel data techniques that can exploit the cross-sectional as well as the temporal variations in the foreign portfolio allocations of portfolio investors. Second, it is possible that the value of information depends on the quality of legal and macro-institutions in host countries.4 3 Gelos and Wei (2005) examine a similar issue on a sample of emerging markets as host countries and focus on the issue of country-specific transparency measures. 4 For example, earlier studies show that a strong legal/macro-framework leads to higher firm valuation (Claessens et al., 2000), higher dividend pay-outs (La Porta et al., 2000), reduced level of underpricing in IPOs (Engelen and van Essen, 2010) and managerial private benefits (Zingales, 1994; Nenova, 2003). Similarly, Fernandes and Ferreira (2008) demonstrate that the enforcement of insider trading laws enhances stock price informativeness mostly in developed countries and the positive effect of enforcement on price informativeness becomes significantly weaker in countries with poor macro institutions.

2256

C. Thapa et al. / Journal of Banking & Finance 37 (2013) 2255–2267

Therefore, we examine how the quality of macro-institutions in host countries interacts with the cross-country differences in access to information to shape the composition of the foreign equities in the portfolios of international investors. To the best of our knowledge, this is the first study to examine this effect. If the cross-country differences in information accessibility interact with legal/macro-institutions, then the efforts to improve the bilateral information environment alone may not be enough to attract foreign investors, unless matching actions are undertaken to enhance the quality of the country’s legal and macro-institutions. Therefore, the findings of this paper should be important for policy makers as well. Finally, the information environment is also likely to interact with the market condition, affecting investors’ exposure to risk. In an uncertain environment, investors are likely to rebalance their portfolio in favour of markets of which they possess or could seek higher levels of information. For instance, using fund level international country portfolio flow data, Gelos and Wei (2005) show that during the contagion effect of the 1997 East Asian crisis, foreign investors reversed their investment flows more from opaque markets (i.e. markets with lower levels of corporate and macroinformation) compared with transparent markets. This suggests that the implication of cross-country differentials in access to information on investors’ choice is also likely to be dependent on market conditions. Since our sample covers periods of both tranquil and turbulent market conditions, it offers an excellent opportunity to test the implications of the financial crisis that started in 2007 on the portfolio choice of international investors. As such, we extend our analysis to examine whether investors allocate relatively less in countries with lower levels of access to information when the market is turbulent. Furthermore, understanding the relationship between market conditions (tranquil vs. turbulent), information accessibility and the country allocation of international investors, should help policy makers to manage their stock market conditions, particularly the information environment, to make them attractive to foreign investors. Several interesting findings emerge from our analysis which uses an extensive data set, alternative proxies of information accessibility, and a spectrum of regression specifications. First, we find that the differences in access to cross-country information significantly influence investors’ foreign portfolio allocation decisions. Investors tend to allocate more funds to countries that have the same working language as in the investors’ home country, are geographically closer, and have higher bilateral trade. Second, the results further suggest that, for a given level of access to information, investors prefer to invest more in countries that have a higher quality of legal and macro-institutions. Finally, the findings also confirm that the implications of information accessibility are more pronounced when markets are turbulent. The rest of the paper is structured as follows: Section 2 describes the data; Section 3 discusses the methodology and provides a brief description of the features of the sample; empirical results are presented in Section 4; and, Section 5 concludes the paper.

Second, the key explanatory variables of interest are three proxy measures of information accessibility which capture the degree of accessibility and flow of information between the home (source) country of investors and the foreign (host) country. They are: (i) Geographic proximity (i.e. the distance between capital cities of the host and source countries), (ii) Common language, and (iii) Trade based information, i.e. the degree of bilateral trade (as a proportion of total trade of the source country) between the pair countries. The first two measures (Geographic proximity and Common language) are obtained from www.nber.org/wei/data.html (as used by Subramanian and Wei (2005)). The proportionate bilateral trade variable is based on bilateral trade (import plus export) data obtained from the IMF’s periodic publication, International Financial Statistics (IFS). The final set of data includes control variables that are known to affect international investors’ foreign portfolio choices, such as stock market development, market microstructure, capital control, host market volatility, exchange rate risk, historical return, minority shareholder protection and cross-country correlation of equity returns. In the paragraphs below we describe data on foreign equity portfolio holdings, followed by a brief description of the proxy measures of the cross-country information environment and the control variables respectively. 2.1. Measure of foreign equity portfolio holdings Following the work of Cooper and Kaplanis (1986), we use the dependent variable (proportion of risky assets allocated to a foreign country, wi,j,t) as the logarithmic value of the proportion of equity from country j in the portfolio of investors based in country i at time t, whereby wi,j,t is defined as in Eq. (1)5:

wi;j;t ¼

FPHi;j;t

P36

j¼1 FPHi;j;t

! ;

i–j

ð1Þ

where FPHi,j,t is the stock of foreign equity portfolio holdings in US$ million. The sample includes bilateral data on the portfolio investments of investors based in 33 countries (source) in the equities of 36 foreign (host) countries (Appendix A provides the list of the countries). The analysis of the experience of 33 source and 36 host countries in this paper is dictated by the availability of data on the dependent, explanatory and control variables. For instance, estimates of trading cost (a proxy measure of market microstructure condition) are manually hand collected from S&P’s Global Stock Market Factbook and data for the year 2001 are reported for 40 countries of which 36 are included in this study. The remaining four countries are either off-shore financial centres or not enough data points are available for other control variables.6 In the coverage of the CPIS, most participants are the primary end-investors (e.g. banks, security dealers, pension funds, insurance companies, mutual funds, non-financial corporations, households) and primary custodians. However, investments below US$ 500,000 are not reported. Furthermore, some investments are not reported by countries for reasons of confidentiality. 2.2. Measures of access to information (AI)

2. Data This study uses three sets of data. First, the data for the main (dependent) variable of interest (i.e. the foreign equity portfolio holdings of investors) is obtained from the Co-ordinated Portfolio Investment Survey (CPIS) of the International Monetary Fund (IMF). This database provides detailed annual cross-country international equity portfolio holdings in US$ million. It covers 33 source (home) and 36 host (foreign) countries for a period of 9 years from 2001 to 2009. Overall, we have over 7000 usable observations.

As in earlier studies, motivated by the use of gravity models in the field of international trade in goods, services and financial 5 Since the purpose of the analysis is to assess the distribution of only the foreign portfolio, not domestic allocation, the variable (wi,j,t) does not include domestic investments (i.e. wi,i is not included). 6 As in previous studies (e.g. Fidora et al., 2007) on international equity investment, we do not include Luxembourg in our sample. Luxembourg is perceived as an offshore financial centre and is associated with the issue of third-country holdings and round-tripping. For example, in 2003 the total equity holdings reported by German investors alone in Luxembourg was US$ 152 billion, whereas Luxembourg’s total market capitalization for that year was less than US$ 40 billion.

C. Thapa et al. / Journal of Banking & Finance 37 (2013) 2255–2267

transactions (see for instance, Chan et al., 2005; Portes and Rey, 2005), in the main analysis we use three proxy measures (Geographic proximity, Common language and Trade based information) to capture the investors’ level of bilateral crosscountry information.7 Since these factors measure the variations in the availability of cross-country information, they are suitable proxies of access to bilateral information (see Chan et al., 2005). In our analysis we refer to them as access to information (AI). It is highly likely that cross-country investments are influenced by geographic proximity, language information and long-term bilateral trade relationships. In this study the information on language is presented by a dummy variable (Common language) that takes the value of one if the source and host countries share the same language. The original source of the data is the CIA’s World Factbook and includes the nonofficial language(s) spoken by a significant proportion of the population. The Geographic proximity is measured by the log (natural) of distance between the capital cities of the source and host countries. Chan et al. (2005) argue that investors are more confident in holding stocks of foreign companies whose goods and services are well known to them. To represent this factor we incorporate the Trade based information variable in the model. This is measured by the ratio of the pair’s total export and import, scaled by the source country’s total trade (export plus import) for the year. For example, a value of 0.04 between Germany (host) and the United States (source) implies that 4% of the total trade of the United States is with Germany. Chan et al. (2005) note that by pursuing goods and services of foreign companies, investors of source countries are able to obtain information about these firms. All the above mentioned variables capture the level of crosscountry information accessibility between home and host nations of the investors and measure the potential barriers (costs) that foreign investors encounter when seeking more information about foreign markets (for detailed description see Chan et al., 2005; Fidora et al., 2007). 2.3. Control variables In examining the implications of AI on equity portfolio holdings of international investors, we also control for the possible effects of other factors that are known to affect such holdings. First, a benchmark describing the country asset allocation of foreign investors, if all countries were equal along the information cost dimension, is needed. The ICAPM suggest that, in the absence of deadweight costs, international investors should hold equities of each country in proportion to the country’s share in the world market capitalization. As such, the ICAPM provides a benchmark for equity portfolio holdings. The ICAPM benchmark (Mij,t) is computed as:

, M ij;t ¼ MC ij;t

n X

! MC j;t ;

i–j

ð2Þ

j¼1

where MCij,t is the market capitalization of the recipient or host country j at time t when the source country is i. Since we model the foreign portfolio holdings, the benchmark allocations do not include the source country’s market capitalization. Second, Chan et al. (2005) show that stock market size/development positively affects international portfolio investments. We control for this effect by the ratio of stock market capitalization to GDP as a measure of equity market development (see also Tsoukas, 2011). These measures are obtained from the World Development Indicator (WDI) of the World Bank. 7

Erel et al. (2012) show that more cross-border mergers and acquisition deals occur between the countries with closer geographic proximity, sharing common language, and having higher bilateral trade, indicating that these factors are suitable proxies of cross-country information and affect the flow of funds.

2257

Third, Keim and Madhavan (1995) conjecture that transaction costs (a measure of the state of market microstructure in a country) are important in determining the investment performance of foreign investment and may significantly offset the expected return generated by an otherwise good investment strategy. The consumption and portfolio choice framework of Rowland (1999) demonstrates that as the extent of transaction costs rise, the benefits of portfolio diversification decrease. This implies that despite the advantages of international diversification, foreign investors may underweight those countries with higher transaction costs or a poorer market microstructure. We control for the market microstructure effects of a country with country level (aggregate) trading costs maintained by Elkins/McSherry (E/M) and reported in the annual Global Stock Market Factbook of S&P. This measure aggregates the estimates of country level fees, commission and market impact8 for an average transaction in the country. Fourth, Solnik and McLeavey (2009) note that the total risk (standard deviation) of a foreign investment is composed of two components – the volatility due to movement in local market return (local equity market volatility), and the volatility due to movement in foreign exchange rate (exchange rate risk). They show the following relationship:

rf 6 r þ rs

ð3Þ

where rf is the volatility of the foreign asset (i.e. variability of returns from investing in foreign assets) measured in foreign currency, r is the volatility of returns in local currency and rs is the volatility of the exchange rate return. We use a measure of local equity market volatility (r) by including a 3-year moving average, cross-sectional standard deviation of the equity returns. This measure is constructed using the total monthly returns index of S&P/ IFC denominated in local currency. Klau and Fung (2006) correctly show that the trade weighted effective exchange rate is a better metric of the macroeconomic effects of exchange rates than a single bilateral rate and Carrieri et al. (2006) argue in favour of real effective exchange rates against nominal effective exchange rates. Therefore, we measure the exchange rate risk (rs) by the standard deviation of rolling 36 monthly observations of trade weighted real effective exchange rate (REER) obtained from the Bank of International Settlement (BIS). Carrieri et al. (2006) conjecture that ‘‘. . .using changes in the real exchange rate helps overcome possible complications due to fixed exchange rate regimes or discrete changes in nominal exchange rates due to devaluations or currency peg management’’ (p. 514). They argue that with the use of nominal exchange rates, it is virtually impossible to ascertain whether the nominal exchange rate risk measure incorporates deviation from the PPP or other macroeconomic factors. However, the REER captures the real exchange rate risk not taken into account by changes in inflation rate differential.9 Fifth, to account for the level of economic development of the host nations we include the log of per capita GDP of each country, sourced from the WDI of the World Bank. Sixth, to allow for the time varying degree of financial liberalization (market integration) in the model, we include the capital control intensity measure (inward investment restriction) prescribed by Chan et al. (2005). We use the capital control measure of the Economic Freedom Network. This index ranges from 0 to 10 with 0 representing absolutely closed markets and 10 fully open markets for foreign investments. As noted earlier, inward investment restriction also accounts for any time effect in the financial liberalisation/integration process (see de Jong and de Roon, 2005; 8 Market impact refers to the difference between the actual price paid and the average of the stock’s high, low, opening and closing prices during a particular trading day. Hence, it signifies the difference between the actual price paid by the investor and the price the investor would have paid had s/he not traded the security. 9 Also, see Kodongo and Ojah (2011) and Moerman and van Dijk (2010).

2258

C. Thapa et al. / Journal of Banking & Finance 37 (2013) 2255–2267

Bali and Cakici, 2010). We downloaded this data from the website http://www.freetheworld.com. For further details, see Chan et al. (2005). Seventh, studies (see for instance, Bohn and Tesar, 1996; Richards, 2005) also suggest that investors may favour countries with recent higher historical returns, commonly referred to as the feedback trading hypothesis. With a view to capturing the possibility of transactions based on this phenomenon, we include a 3-year moving average domestic market return (momentum in equity return) in the model. Eighth, La Porta et al. (2000) show that a common law system provides better legal protection rights for minority shareholders, while a civil (code) law system offers the least. Such a difference in investor protection is likely to affect investors’ choice. Therefore, we control for its effect by including a dummy variable (minority shareholder protection) which takes the value of one for common law countries and zero for others. Finally, to capture the effect of diversification, we also include the correlation coefficient, for each year, between the pair-country equity returns constructed using the country level daily total return index of S&P/IFC.10 3. The methodology and sample features 3.1. The panel data models To capture the wide cross-sectional and temporal variations in our data set we employ panel data regression models. A panel data set-up allows us to utilize both cross-section and time series information in our estimations. A number of studies have highlighted the benefits of using panel data. Hsiao (1985) and Solon (1989), among others, note that compared to purely cross-section or time series, panel data control for unit specific effects suffer significantly less from multicollinearity problems, yield reliable and more efficient estimates and provide internal instrumental variables. However, panel data models also have their own unique problems, particularly intra-clustering heteroscedasticity and parameter heterogeneity. We deal with these issues as far as possible, and produce estimates that are unbiased and efficient. In our analysis, each of the host countries (j) receives an allocation from each of the source countries (i) for the 9-year period. This implies that each cross-section unit is a cluster of matched pairs and hence the ‘‘clustered errors’’, i.e. the observations within each group (i, j), may be correlated, inducing correlation in the error term within each panel. We address the issue of within panel clustering, the key source of heteroscedasticity, by using the robust variance–covariance matrix estimator which corrects the standard errors for intra-cluster correlation (standard errors adjusted for over 500 clusters in the cross-section units). Wooldridge (2002) suggests that the corrected variance–covariance matrix is robust to any kind of intra-cluster correlation and arbitrary heteroscedasticity, provided N (cross-section units) is large relative to the number of units in each cluster. In our sample, N is over 500 and the number of units within each cluster (i, j) is up to a maximum of nine observations (i.e. 9 years of country allocation). In all our random effect GLS estimations, we correct the standard errors using the cluster robust variance–covariance matrix (see equation 7.26 of Wooldridge, 2002, pp. 152 and 330). The correction not only yields efficient cluster robust standard errors, but the unrestricted O matrix estimator also allows for arbitrary serial correlation. These variance–covariance estimates provide much more correct coverage rates than panel-level heteroscedasticity, i.e. they are more robust to any type of correlation within the observations of 10 We thank the anonymous referee for drawing our attention to this important factor.

each panel (i, j). Such correction significantly improves the statistical efficiency of our estimates, producing more efficient standard errors (for further details see Wooldridge, 2002, pp. 329–330). Although the choice of employing panel data models is motivated by the increase in degrees of freedom and supply of a higher level of information producing more reliable estimates, the estimated magnitude of the slope coefficients is assumed to be identical across countries, i.e., bj = b. If this assumption holds in its pristine form, it will generate the most precise estimates. However, as our data are at country level, a homogeneous parameter is a very strong and unrealistic assumption. As such, the slope coefficients reported are most probably not identical across countries/time and represent an average slope coefficient, allowing us to comment on the average predictive relationship between the variables. In Section 4.7, we show that slope coefficients do vary across countries, depending on the quality of country level macro-institutions. Given that Geographic proximity and Common language are timeinvariant, in most of our estimations we use the more efficient random effect model. However, we also use the fixed effect model by including the time-varying variables only (i.e. trade based information in robustness checks) controlling for any unit specific effects. Furthermore, in our fixed effect estimation, we correct the standard errors for heteroscedasticity arising from cross-sectional dependency using the Driscoll and Kraay (1998) standard errors, which are robust to general forms of cross-sectional and temporal dependence. 3.2. The sample features Since our data set incorporates cross-section and temporal variations in foreign equity portfolio holdings (year-end) by 33 source countries in 36 host countries, we use Eq. (4) to compute the average value of each factor (fi,j,t) for each of the 36 host countries.

Av erage ¼

T X K 1 X fi;j;t K; T t¼1 i¼1

ð4Þ

where K is number of source countries i.e. 33, and T is the number of time periods, i.e. 9 years (2001–2009) and f represents the four variables of interest, i.e. bilateral portfolio allocation (wi,j,t), Geographic proximity, Common language and Trade based information. The results are reported in Table 1. Column 2 shows that the investors based in 33 source countries invest most of their funds in developed countries (mainly in United States, United Kingdom, France, Germany, and Japan) and the smallest proportions in emerging markets (Peru, Argentina, Chile, Philippines and Thailand). This clearly indicates that investors favour developed countries more than the emerging markets.11 In terms of Geographic proximity, as reported in miles in column 3 of Table 1, on average, the five countries which are furthest from the 33 source countries are Australia, New Zealand, Taiwan, Argentina and Chile, whereas Germany, Austria, Denmark, Belgium, and the Czech Republic are closest to the source countries. An observation of portfolio allocation against the distance suggests that investors tend to favour investing in countries that are geographically closer to their home country. Column 4 records the extent of common language shared between the host and source countries. The figures in column 4 demonstrate that countries such as the United States, United Kingdom, Canada, New Zealand and India share a common language (i.e. English) and rank among the highest for sharing common language with the 33 investor countries, along 11 Literature also suggests that the differences in information accessibility between home markets and foreign markets is responsible for home bias, i.e. the tendency to over-invest in home markets relative to the prescription of ICAPM (see French and Poterba, 1991; Nieuwerburgh and Veldkamp, 2009). To get a better sense of the data we report the average home bias per source country in Appendix B.

2259

C. Thapa et al. / Journal of Banking & Finance 37 (2013) 2255–2267 Table 1 Descriptive statistics. Countries

Foreign allocation (%)

Geographic proximity (in miles)

Common language (0–1)

Trade based information (0–1)

Argentina Australia Austria Belgium Brazil Canada Chile China Czech Republic Denmark Finland France Germany Greece Hungary India Indonesia Italy Japan Malaysia Mexico New Zealand Norway Peru Philippines Poland Portugal Russia South Korea Sweden Switzerland Taiwan Thailand Turkey United Kingdom United States

0.426 1.459 1.016 1.564 0.476 1.379 0.096 0.711 0.532 0.490 1.401 11.249 9.613 0.500 0.785 0.623 0.609 2.715 7.341 0.648 0.583 0.458 0.759 0.048 0.058 0.856 0.814 0.911 1.520 2.221 4.071 0.031 0.212 0.798 13.483 37.101

6913.410 6951.632 1041.975 964.877 6172.379 3944.883 6898.210 4314.132 907.302 811.133 1383.948 1179.580 954.905 1826.867 1237.174 4463.307 5628.104 2020.181 4888.455 5195.878 5855.803 8501.110 1483.158 6237.012 5097.118 1487.978 1835.198 3981.991 4642.285 2085.449 1770.013 7325.880 4772.857 2818.696 1265.105 5183.295

0.009 0.394 0.159 0.352 0.029 0.582 0.059 0.000 0.000 0.294 0.029 0.288 0.188 0.000 0.000 0.494 0.000 0.029 0.000 0.000 0.059 0.414 0.000 0.086 0.294 0.000 0.009 0.000 0.294 0.001 0.176 0.000 0.234 0.000 0.494 0.494

0.002 0.006 0.005 0.018 0.006 0.037 0.002 0.043 0.003 0.004 0.005 0.030 0.075 0.002 0.003 0.006 0.004 0.027 0.031 0.006 0.004 0.001 0.003 0.001 0.002 0.004 0.002 0.009 0.014 0.009 0.010 NA 0.005 0.005 0.032 0.083

Averages of the main variables of interest are calculated. Portfolio allocation is the average proportion of funds invested by the investors based in 33 source countries in 36 host countries (see Appendix, Table A1) over the sample period (2001–2009). The portfolio holdings are sourced from the IMF. Geographic proximity is the average distance (in miles) between the host and host nations. The measure of Common language represents the degree to which the investors in source and host countries share the same common language. Higher value suggests that the host country shares a common language with a higher number of source countries. The data on Geographic proximity and Common language are obtained from www.nber.org/wei/data.html. Trade based information is measured by the ratio of the pair’s total export and import scaled by the source country’s total trade (export plus import) for the year. The figures denote the degree to which the source countries share bilateral trade with the 36 host countries. Data on this variable are sourced from International Financial Statistics published by the IMF.

Table 2 Portfolio allocation, the ICAPM benchmark, and access to information.

ICAPM Benchmark

Model 1

Model 2

Model 3

Model 4

0.896*** (27.41)

0.882*** (16.84) 0.924*** (13.45)

0.833*** (17.88) 0.911*** (14.52) 1.045*** (6.13)

5.273*** (7.52)

1.353*** (13.00)

1.082*** (13.29)

0.816*** (16.94) 0.661*** (7.76) 0.858*** (5.35) 0.346*** (5.37) 1.998*** (9.38)

0.44 0.08 7602

0.46 0.02 7602

0.56 0.10 6875

Geographic proximity Common language Trade based information Constant Overall R2 Incremental R2 Number of observations

0.36 – 7925

Using specification (5) [wi,j,t = a + Mi,j,t + AIi,j,t + ei,j,t] foreign equity portfolio allocation of international investors based in country i in the equity shares of country j for time t (wi,j,t) is regressed against the ICAPM benchmark (Mij,t) and the three measures of access to information (AIi,j,t) in a nested regression format. The equations are estimated using a random effect panel data model. T-statistics are in parentheses and standard errors are corrected for intra-cluster correlation in which the resulting variance–covariance matrices are robust to intra-cluster correlation and arbitrary heteroscedasticity.  Denote statistical significance at 10%.  Denote statistical significance at 5%. *** Denote statistical significance at 1%.

2260

C. Thapa et al. / Journal of Banking & Finance 37 (2013) 2255–2267

with Australia, Denmark and Belgium. These statistics suggest that investors are more inclined to invest in countries sharing a common language possibly because it facilitates better flow/accessibility of information. Estimates in column 5 reveal that countries such as the United States, United Kingdom, Germany, France, Canada and China share the highest average bilateral trade with the source countries and are also the highest recipients (except China) of portfolio allocation from investors in the source countries. Most of the emerging countries score lower on this measure, implying that investors are inclined to invest more in countries with a significant bilateral trade with their home nations. 4. The results Models are estimated to examine whether variations in the degree of access to information (AI) available about foreign (host) markets affect portfolio allocation decisions of international investors. We begin by estimating specification (5) in a nested regression form whereby the first regression of portfolio allocation is regressed against the ICAPM benchmark (Mi,j,t), followed by incremental additions of the three main measures of AI (AIi,j,t), a is the intercept.

wi;j;t ¼ a þ Mi;j;t þ AIi;j;t þ ei;j;t

ð5Þ

The results reported in Table 2 show that coefficients of the benchmark portfolio (Mi,j,t) in all four models are statistically significant and bear expected signs (positive). The coefficients of all three measures of access to information (AIi,j,t) are also statistically significant with expected signs. The negative coefficients of Geographic proximity (models 2–4) suggest that investors tend to invest less in markets that are further away from home. The positive coefficients of the Common language (models 3 and 4) and Trade based information (model 4) also confirm that if home and host markets share a common language and have close trade relations, investors are likely to have a greater degree of information about the host markets and are willing to invest more in such countries. Overall, these estimates suggest that investors’ portfolio choice is significantly affected by the level of AI between home and host markets – the higher the level of information access about a market, the higher the investment in risky assets. The R2 of the four nested-models shows that the explanatory power increases with the inclusion of each of the measures of AI. The incremental explanatory power of Geographic proximity and Trade based information are highest while that of Common language12 the least. Overall, the estimates suggest that international investors’ portfolio allocation is not only affected by the benchmark portfolio, as articulated by the ICAPM, but also by the differences in the level of AI about host markets that potentially create deadweight cost. Although the signs and statistical significance of the estimates reported in Table 2 establish the expected association between foreign equity portfolio holdings and the differences in AI, the estimates may suffer from the omitted variable problem. This may lead to biased and inefficient estimates in the absence of other variables in models that could potentially affect the portfolio allocation choice of international investors. In the following sub-sections we address the issue of model specification and the choice of estimation methods. 4.1. Controlling for omitted variable bias For the reasons described earlier (Section 2.3) we include all control variables and run the random effect regressions using specification (6). In addition, we include a dummy variable that takes 12 The standardized betas from a simple pooled OLS are: Geographic proximity = 0.226, Common language = 0.099 and Trade based information = 0.265.

the value of one for the years 2007–2009 and zero otherwise, to account for the possible effects of the 2007–2009 financial crisis on investors’ portfolio choices.

wi;j;t ¼ a þ M i;j;t þ AIi;j;t þ Controlsj;t þ ei;j;t

ð6Þ

The estimates of specification (6) are reported in Table 3. The results show that the overall explanatory power (R2) of specification (6) is higher than that of specification (5) (Table 2), indicating the importance of control variables. Although the size of all the estimates of AI measures change their relevance on equity portfolio allocation, they still remain statistically significant. The coefficient of Geographic proximity (0.593) signifies that foreign investors are inclined to under-invest in countries that are further away from their home countries. Similarly, the positive and significant coefficient of the Common language (0.767) suggests that investors invest more in countries that share the same language, as it is relatively easier to obtain more information about the market. Finally, the positive and significant coefficient of Trade based information (0.697) also confirms that investors are able to extract more information about a foreign country’s investment environment if the home and host countries have substantial trade (goods and services) relations. This evidence supports the view that investors are more confident in

Table 3 Factors affecting the portfolio allocation of international investors. All controls ICAPM Benchmark Geographic proximity Common language Trade based information Equity market development Market microstructure Local equity market volatility Exchange rate risk Economic development Inward investment restriction Momentum in equity return Minority shareholder protection Bilateral return correlation 2007–2009 financial crisis dummy Constant Overall R2 Number of observations

0.735*** (11.92) 0.593*** (8.96) 0.767*** (5.97) 0.697*** (7.60) 0.104*** (3.33) 0.287* (1.96) 0.497*** (3.66) 0.541 (0.55) 0.189** (2.30) 1.278*** (3.14) 0.303* (1.95) 0.435** (2.30) 0.0447 (1.22) 0.0261 (1.06) 3.707*** (3.63) 0.64 6535

Using specification (6) [wi,j,t = a + Mi,j,t + AIi,j,t + controlsj,t + ei,j,t] foreign equity portfolio holdings of international investors based in country i in the equity shares of country j for time t (wi,j,t) regressed against the ICAPM benchmark (Mij,t), the three measures of access to information (AIi,j,t) and various control factors that are known to affect the portfolio allocation of international investors. The equation is estimated using the random effect panel data model. T-statistics are in parentheses. The standard errors are corrected for intra-cluster correlation in which the resulting variance–covariance matrix is robust to any kind of intra-cluster correlation and arbitrary heteroscedasticity. * Denote statistical significance at 10%. ** Denote statistical significance at 5%. *** Denote statistical significance at 1%.

2261

C. Thapa et al. / Journal of Banking & Finance 37 (2013) 2255–2267 Table 4 Alternative specifications and the methods of estimation. Fixed effect model ICAPM Benchmark Geographic proximity Common language Trade based information Equity market development Market microstructure Local equity market volatility Exchange rate risk Economic development Inward investment restriction Momentum in equity return Minority shareholder protection Bilateral return correlation 2007–2009 financial crisis dummy Constant Overall R2 Number of observations

***

0.620 (6.23) NA NA

0.225*** (4.03) 0.260*** (3.29) 0.247*** (3.15) 0.0318 (1.04) 0.580 (1.15) 0.371*** (2.77) 1.274*** (3.07) 0.284*** (2.66) NA 0.056* (1.75) 0.087 (0.12) 2.327*** (3.07) 0.18 (within) 6601

Endogeniety test ***

0.734 (11.96) 0.572*** (8.69) 0.771*** (5.81) 0.617*** (7.31) 0.103*** (3.49) 0.275*** (4.54) 0.0430* (1.65) 0.241 (1.03) 0.214** (2.40) 1.027** (2.09) 0.323* (1.74) 0.414** (2.27) 0.0178 (1.09) 0.00172 (0.03) 4.102*** (3.95) 0.64 5862

Excluding major financial centres ***

0.620 (3.27) 0.498*** (4.31) 0.589*** (4.23) 0.729*** (13.82) 0.344*** (5.29) 0.235* (1.85) 0.0381 (1.42) 0.269 (0.99) 0.382*** (3.90) 1.355*** (2.68) 0.211 (1.27) 0.654** (2.31) 0.0184* (1.89) 0.0830 (1.23) 1.372 (1.27) 0.63 5679

Free float benchmark 0.233*** (11.88) 0.592*** (8.97) 0.714*** (5.99) 0.599*** (7.91) 0.118*** (4.49) 0.257*** (3.94) 0.0391 (1.32) 0.503 (1.02) 0.208** (2.49) 0.213*** (3.54) 0.293* (1.90) 0.442** (2.36) 0.0468 (1.23) 0.0406 (0.88) 4.827*** (4.58) 0.64 6535

In all specifications, foreign equity portfolio holdings of international investors are regressed against the ICAPM benchmark, measures of access to information (Fixed effect model includes only the time varying measure of access to information (i.e. trade based information) and other factors that are known to affect the portfolio allocation of international investors. The second model (Endogeniety test) includes 1 year lagged measures of access to information. The third model (Excluding major financial centres) excludes the United States, United Kingdom and Japan as source nations. Finally, (Free float benchmark) replaces the standard ICAPM benchmark with the free float benchmark. Except the Fixed effect model, all are estimated using the random effect model. T-statistics are in parentheses. The fixed effect model uses the Driscoll and Kraay (1998) standard errors that are robust to general forms of cross-sectional and temporal dependence. The standard errors of the random effect models are corrected for intracluster correlation in which the resulting variance–covariance matrices are robust to intra-cluster correlation and arbitrary heteroscedasticity. * Denote statistical significance at 10%. ** Denote statistical significance at 5%. *** Denote statistical significance at 1%.

holding stocks of foreign companies whose goods and services are well known to them. Among the control variables, most of them bear the expected signs and are statistically significant. The implications of equity market development, market microstructure, local equity market volatility, economic development, inward investment restrictions, minority shareholders’ rights and bilateral equity returns’ correlations are generally consistent in the current and all the subsequent specifications we employ in this study. The exchange rate risk, however, does not seem to exert any significant effect on investors’ portfolio allocation.13 Overall, the estimates in Table 3 corroborate that although other control factors also affect the international portfolio choice of investors, the allocation is significantly affected by the level of AI across the markets. This clearly signifies that investors tend to invest more in the markets on which they hold a higher level of information.

4.2. The fixed effect estimation We further test the robustness of the above findings using alternative specifications. The results are reported in Table 4. 13

Possibly the effect of the exchange rate volatility is captured by local equity market risk as they are highly correlated (with a correlation coefficient of 0.64).

Although the use of the random effect model ensures our estimates are as efficient as possible, the coefficients may still be suspected of bias because of time-invariant pair-country bilateral and country-specific effects, such as common colonial history, special bilateral treaty, favourite partner nation status, and the difference between emerging and developed markets effects. If any of these individual effects exist and are correlated with any of the regressors, particularly with our key variables of interest, the estimates may be biased. We use the fixed effect model (specification 7) to address this issue. Although fixed effect estimation has its own limitations, particularly when rarely changing variables are in the model, it eliminates the country-specific and bilateral cross-country time invariant effects. However, we are only able to include the time varying Trade based information variable in the estimation, as the other two measures of information accessibility (Geographic proximity and Common language) are time invariant. To ensure the efficiency of the estimates, we estimate the fixed effect model using the Driscoll and Kraay (1998) standard errors, which are robust to general forms of cross-sectional and temporal dependence. The results are reported in the second column (Fixed effect model) of Table 4.

wi;j;t ¼ ai;j þ Mi;j;t þ Trade based informationi;j;t þ Time v arying controlsj;t þ FEi;j þ ei;j;t

ð7Þ

2262

C. Thapa et al. / Journal of Banking & Finance 37 (2013) 2255–2267

The estimates show that the coefficient of Trade based information is statistically significant even after controlling for all time invariant unit fixed effects. However, it should be noted that the size of the coefficient declines to 0.225 from 0.697 in Table 3. This is not surprising because the fixed effect model only exploits the ‘within’ variation in the data set, ignoring the cross-country variations. Furthermore, because of the loss in efficiency, the ‘within’ R2 is only 18% when compared with the explanatory power (R2) of other models.

market capitalization is freely floated and is available to foreign investors. Although this assumption may apply to a considerable extent for most developed nations, there could be severe restrictions on foreign ownership in emerging markets. We re-compute the benchmark portfolio using the S&P/IFC’s freely investable market capitalization indices, which correct for foreign ownership restriction (see Edison and Warnock, 2003).14 We estimate specification (9), which is similar to specification (6) except that it now includes the free float benchmark (FL M i;j;t ).

4.3. Endogeniety

wi;j;t ¼ a þ FL Mi;j;t þ AIi;j;t þ Controlsj;t þ ei;j;t

Errunza (2001) conjectures that with their increasing stake in the portfolio of foreign markets, investors, primarily institutional investors, may demand more corporate and macro-information, which may itself help mitigate the adverse effect of information costs. If this argument holds, then our estimates are likely to suffer from endogeniety problems arising from reverse causality (i.e. a higher allocation of investors’ funds in a country may lead to an improvement in accessibility of information about a foreign market). Following Gelos and Wei (2005), we address this issue using 1 year lagged values of bilateral information accessibility measures, as in specification (8).

wi;j;t ¼ a þ Mi;j;t þ AIi;j;t1 þ Controlsj;t þ ei;j;t

ð9Þ

The estimate reported in column 5 (Free float benchmark) of Table 4 shows that although the positive coefficient of the benchmark portfolio remains statistically significant, it is smaller than the coefficient of the original measure of the benchmark portfolio. Moreover, the coefficients of all the measures of AI remain qualitatively similar in size, as in previous models, and are correctly signed and statistically significant. Overall, our earlier finding that investors’ portfolio allocation is significantly affected by cross-country information accessibility holds, even after controlling for possible restrictions in foreign equity ownership. 4.6. Economic relevance: An example

ð8Þ

The estimates in Table 4 (column 3 – endogeniety test) show that the coefficients of the lagged values of all the measures of AI (Geographic proximity, Common language and Trade based information) are statistically significant and have theoretically plausible signs. Moreover, the size of the coefficients and explanatory power of the model remain qualitatively similar to those of the random effect model (Table 3), indicating that the AI is a significant factor in deciding investors’ portfolio allocations. 4.4. Tradability in major financial centres Solnik and McLeavey (2009) suggest that the benefits of international diversifications are higher when transacting directly in foreign markets. This is particularly important for foreign securities that have lower international exposure. However, investors need not trade directly in foreign markets to gain exposure to international equities because in major financial centres such as the United States, United Kingdom and Japan, investors can trade in international stocks, exchange traded funds (ETFs) and/or depository receipts, predominantly for companies that have significant international operations. The impact of information (deadweight) costs arising from information accessibility may be significantly lower for trading in these centres and if so, our estimates may be biased. To address for this possibility we exclude the United States, United Kingdom and Japan as investor/source countries from our sample and re-run specification (6) as the investors from these countries may not have to invest directly in foreign countries to obtain the benefit of international portfolio diversification. The results reported in column 4 of Table 4 reveal that the coefficients of all information access measures are statistically significant, even when the possible effects of the tradability in major financial centres are isolated through sampling. Clearly, the results indicate that the variations in information costs arising from information accessibility have a significant influence on portfolio allocation decisions of international investors. 4.5. Free float benchmark Finally, we address the issue of the availability of equity shares for foreign investors. The benchmark allocation used so far in the models above (Mi,j,t) is based on the assumption that the entire

The above discussions have been based on the expected signs and statistical significance of the coefficients of AI and control factors. In this sub-section, we illustrate the economic relevance of the trade based information (TBI), which is the only time varying measure of bilateral information access in our data set. We first generate two pair-country dummies with the United States being the source country and Canada and Argentina as host countries.15 We interact the two pair-country dummies with the bilateral trade based information variable and include them (TBI USA trade partner; i.e. TBI USA Canada and TBI USA Argentina) in the regression (10). The estimates are presented in Table 5.

wi;j;t ¼ a þ M i;j;t þ TBIi;j;t þ TBI USA trade partner þ Controlsj;t þ ei;j;t

ð10Þ

Table 5 shows that the sum of the coefficients of the variables relevant for the United States–Canada is 0.701 (0.398 + 0.303) suggesting that a one percentage point increase in the proportionate bilateral trade between the United States and Canada boosts the equity portfolio allocation of a US based investor by 0.70% in Canada. Similarly, the sum of the coefficients pertinent to the United States–Argentina is 0.44 (0.398 + 0.042) which implies that a unit percentage point increase in the proportionate bilateral trade between the United States and Argentina increases a US investors’ portfolio allocation in Argentinian stocks only by 0.44%. The difference (0.701% vs. 0.44%) could clearly be attributed to the substantial differences in the bilateral trade between the US and Canada and the US and Argentina. In our sample, the average proportionate bilateral trade figure between the United States–Canada is 0.18 whereas the same figure for the United States–Argentina is 0.0035. The difference between the two interactive variables also suggests that if the information difference (for example, the bilateral trade proportion) between the US and Argentina becomes the same as that between the US and Canada, ceteris paribus, US portfolio holders should increase the Argentinian stocks’ weight by approx14 These indices are only available for emerging markets. By implication, we assume that the full value of market capitalization in developed countries is available to foreign investors. 15 Given the very large number of possible combinations of home and host countries in our data set we use the cases of these two diverse host countries to illustrate the economic relevance of bilateral trade based information from the perspective of a US investor.

C. Thapa et al. / Journal of Banking & Finance 37 (2013) 2255–2267 Table 5 Economic significance: an example. USA_trade_partner ICAPM Benchmark Trade based information (TBI) TBI_USA_Canada TBI_USA_Argentina Geographic proximity Common language Equity market development Market microstructure Local equity market volatility Exchange rate risk Economic development Inward investment restriction Momentum in equity return Minority shareholder protection Bilateral return correlation 2007–2009 financial crisis dummy Constant Overall R2 Number of observations

0.734*** (11.91) 0.398*** (7.58) 0.303** (1.96) 0.042*** (3.13) 0.592*** (8.92) 0.769*** (6.00) 0.404*** (3.33) 0.372* (1.85) 0.496** (2.16) 0.544 (0.55) 0.190** (2.30) 1.279*** (3.14) 0.303* (1.95) 0.437** (2.32) 0.0350 (0.17) 0.0265 (1.06) 3.698*** (3.61) 0.63 6535

Using specification (10) [½wi;j;t ¼ a þ Mi;j;t þ TBIi;j;t þ TBI USA trade partner þ Controlsj;t þ ei;j;t ] foreign equity portfolio holdings of international investors are regressed against the ICAPM benchmark, each measure of access to information and other factors that are known to affect the portfolio allocation of international investors. Each regression also includes an interactive variable (TBI_USA_trade_partner) representing the product of the two dummies (United States–Canada and United States–Argentina pair countries) and Trade based information. The equations are estimated using the random effect panel data model and the standard errors are corrected for intra-cluster correlation in which the resulting variance–covariance matrix is robust to intra-cluster correlation and arbitrary heteroscedasticity. t statistics are reported in parentheses. * Denote statistical significance at 10%. ** Denote statistical significance at 5%. *** Denote statistical significance at 1%.

imately 59% (0.701/0.44–1) relative to the weight currently held by US investors. For example, in our sample the average current allocation from US investors into Argentinean stocks constitutes 0.04% and by applying the 59% increment, provided the information difference is same, ceteris paribus, the new average weight should be 0.063%. Such a pair-country comparative analysis clearly shows that the importance of the analysis of AI bears substantial economic significance for investors. 4.7. Portfolio allocation and the quality of macro-institutions It is possible that the perceived risk of investing in a country owing to the paucity of information could be mitigated, to some extent, by an effective legal environment in which investors’ rights are protected. Similarly, availability of information alone may not reduce investors’ risk if the law enforcing the contacts is weak. Therefore, it is important to examine whether the quality of legal (or macro) institutions interacts with the bilateral information

2263

access variables in shaping the foreign equity portfolio allocation of international investors. We address this issue by incorporating a composite measure of the quality of macro-institutions for the host countries. The measure is sourced from the World Bank Governance Indicator and is comprised of two broader aspects of the regulatory/macro-environment. The first is the regulatory quality based on a scale of 0–100, capturing the perceptions of the local government’s ability to formulate and implement sound policies that are effective for private sector development. The second, also measured on a scale of 0–100, is the rule of law and captures the extent to which local agents show confidence and follow the rules of society, especially the quality of contract enforcement, property rights, the police and the courts. To create a composite index, we add both indices (the regulatory quality and the rule of law) and scale the sum by 0.5. The value of the resulting composite index ranges from 0 to 100 and is referred to as ‘macro-institutions’ in our models. Higher ratings indicate a better quality of macro-institutions and therefore greater confidence by foreign investors. We allow for the interaction between the macro-institutions and the three bilateral measures of the level of AI by incorporating ‘AI_macro institutions’ in the model as in specification (11). Furthermore, we also generate a single information access factor (CBIF) from the three measures of bilateral information access using the principal component analysis method. CBIF captures more than 73% of total variations in the three measures of bilateral information access measures and has a positive correlation coefficient of 0.68 with Common language, 0.71 with Trade based information and a negative coefficient of 0.56 with Geographic proximity. We also incorporate an interaction between macro-institutions and CBIF and re-estimate specification (11). These estimates are reported in Table 6.

wi;j;t ¼ a þ M ij;t þ AIijt þ AI Macro institutionsi;j;t þ Controlsj;t þ ei;j;t ð11Þ Wooldridge (2003, p. 191) notes that in regression with interactive terms, we cannot evaluate the implications of AI and AI_macro institutions estimates separately, in terms of their size and statistical significance. For example, if we simply evaluate the coefficient on Geographic proximity, we run the risk of incorrectly concluding its effect on portfolio holdings because with the interaction term, the coefficient supposedly shows the effect when the value of macro-institution = 0, which does not make sense. In such cases, Wooldridge (2003) suggests that inferences should be drawn from the F-test of whether the sum of the coefficients of AI measures and AI_macro institutions variables (AI + AI_macro institutions) is significantly different from zero. Estimates in Table 6 (in the box and highlighted in bold) show that the F-tests of the joint hypothesis (i.e. the sum of the coefficients of AI and AI_macro institution is equal to zero) are statistically significant in all models. These estimates suggest that the impact of the measures of information access on foreign equity portfolio holdings of international investors is dependent on the strength of macro-institutions of the host country. This implies that for similar levels of information accessibility, countries with stronger macro-institutions are attracting higher proportions of the portfolios of international investors. Moreover, from a policy implications point of view, it also suggests that countries planning to attract more international equity portfolio investments need to improve their legal/macro-institutional environments. Such improvements are likely to mitigate the adverse effects of being located away from the home country of the investors and of speaking a different language. Further to the above discussion, to illustrate the relevance of the macro-institutions we estimate the effect of information accessibility on portfolio holdings of investors by plugging-in ac-

2264

C. Thapa et al. / Journal of Banking & Finance 37 (2013) 2255–2267

Table 6 Access to information, macro-institutions and portfolio allocation. Geographic proximity

Common language

***

ICAPM Benchmark Access to information (AI) AI_macro institution AI + AI_macro institution (F-test) macro-institution Equity market development Market microstructure Local equity market volatility Exchange rate risk Economic development Inward investment restriction Momentum in equity return Minority shareholder protection Bilateral return correlation 2007–2009 financial crisis dummy Constant

***

0.815 (18.08) 0.409*** (5.10) 0.392** (2.12) 0.017*** (3.98) 2.679* (1.90) 0.248*** (3.23) 0.0971* (1.77) 0.124* (1.83) 0.0756 (0.07) 0.138 (1.34) 0.957** (2.05) 0.338** (2.15) 0.609*** (2.58) 0.0502*** (2.61) 0.0212 (0.51) 1.70*** (3.99)

Overall R2 Number of observations

0.874 (15.62) 0.227*** (6.14) 0.636*** (3.23) 0.863*** (7.45) 0.427* (1.65) 0.240*** (3.05) 0.138*** (3.07) 0.0993 (1.45) 0.180 (0.17) 0.224** (2.49) 1.172** (2.54) 0.324** (2.12) 0.0560 (0.21) .0859*** (3.36) 0.0353 (0.90) 1.653* (1.77)

0.62 7218

0.61 7218

Trade based information ***

CBIF

0.683 (8.97) 0.414*** (3.74) 0.523* (1.89) 0.937** (3.91) 1.010** (2.43) 0.385*** (3.48) 0.0590* (1.67) 0.0384* (1.81) 0.453 (0.49) 0.224** (2.12) 1.012** (2.38) 0.341** (2.30) 0.328 (1.59) 0.065*** (3.15) 0.0365 (0.82) 2.412** (2.45)

0.745*** (12.67) 0.278*** (7.68) 0.307* (1.83) 0.585*** (8.12) 0.130 (1.55) 0.115*** (4.54) 0.0565** (2.19) 0.0572 (0.79) 0.642 (0.66) 0.157 (1.50) 1.192*** (3.71) 0.330** (2.20) 0.381* (1.88) 0.0249 (1.12) 0.0117 (0.29) 3.172*** (3.67)

0.63 6535

0.67 6535

Foreign equity portfolio holdings of international investors are regressed against a measure of access to information (in each equation), the ICAPM benchmark, macroinstitution, and other factors that are known to affect the portfolio allocation of international investors using specification (11) ½wi;j;t ¼ a þ M i;j;t þ AIi;j;t þ AI Macro institutionsi;j;t þ Controlsj;t þ ei;j;t . CBIF is generated from the three measures of access to information (Geographic proximity, Common language and Trade based information) using the principal component analysis. Each regression also includes the interactive variable (access to information_macro institutions) representing the product of macro-institutions and information access. The equations are estimated using the random effect panel data model and the standard errors are corrected for intra-cluster correlation in which the resulting variance–covariance matrix is robust to intra-cluster correlation and arbitrary heteroscedasticity. T-statistics are reported in parentheses. Note: the coefficients of ‘Access to Information + AI_macro institution’ and F-statistics (highlighted in bold) are estimated outside the regression equation. * Denote statistical significance at 10%. ** Denote statistical significance at 5%. *** Denote statistical significance at 1%.

tual values of the quality of macro-institutions. The estimates for countries with two different ends of macro-institutional ratings (Russia and Denmark), using a sample period average (2001–

Table 7 Effects of access to information and macro-institutions: an example. Country

Denmark Russia

Quality of macroinstitutions

Coefficient of ‘Access to information + AI_macro institution’ Geographic proximity

Common language

Trade based information

CBIF

98.21 27.5

0.024 0.301

0.852 0.402

0.928 0.558

0.580 0.362

The effect of access to information in the context of the quality of macro-institution is examined. Among the sample countries Demark has the highest quality of macroinstitutions (98.21) and Russia has the lowest (27.5) on a scale of 0–100. These figures (in all the regressions this variable is scaled by 100 to report the coefficients as elasticity) are incorporated in the coefficient function of ‘information access + AI_macro institution’ demonstrating the effect of access to information on the portfolio holdings of international investors. As reflected in the size of the coefficients, the positive effect of access to information is higher (lower) in countries with higher (lower) quality of macro-institutions.

2009), are given in Table 7.16 These estimates confirm that the size of coefficients varies with variations in the ratings of macro-institutions; the coefficient for Geographic proximity for Denmark is 0.024 while that for Russia is 0.301. Note that the average coefficient of Geographic proximity for both countries is the same i.e. (0.017) but when we input the macro-institution rating of 27.5 (scaled by 100) for Russia in ‘AI + AI_macro institutions’ the negative coefficient becomes materially bigger. The cases of Denmark and Russia confirm that the effect of Geographic proximity also depends on the quality of country level macro-institutional infrastructure. Similarly, for different quality levels of macro-institutions, the coefficients of all the other three measures of information accessibility are different for Denmark than for Russia. These values reiterate the findings that the impact of information accessibility on investors’ portfolio allocation varies with the quality of macro-institution in the host country.

16 The quality of macro institution is rated highest for Denmark (98.21) and lowest for Russia (27.46). In order to facilitate the use of the coefficient as elasticity, the variable representing macro institution is scaled by 100. Similarly, the values plugged-in ‘AI + AI_macro institution’ equations are also scaled by 100.

2265

C. Thapa et al. / Journal of Banking & Finance 37 (2013) 2255–2267 Table 8 Portfolio allocation, financial crisis and access to information. Geographic proximity ICAPM Benchmark Access to information (AI) AI_2007_2009 AI + AI_2007_2009 (F-test) Equity market development Market microstructure Local equity market volatility Exchange rate risk Economic development Inward investment restriction Momentum in equity return Minority shareholder protection Bilateral return correlation Constant Overall R2 Number of observations

Common language

***

***

Trade based information ***

CBIF

0.905 (18.26) 0.891*** (13.36) 0.069* (1.95) 0.960*** (13.49) 0.250*** (3.22) 0.127* (1.68) 0.128* (1.85) 0.0271 (0.27) 0.166* (1.91) 1.090*** (2.63) 0.343** (2.28) 0.676*** (3.03) 0.500** (2.53) 1.493*** (6.04)

0.861 (15.45) 0.888*** (5.15) 0.075 (0.84) 0.963*** (5.45) 0.222*** (2.72) 0.202* (1.77) 0.0812 (1.14) 0.0396 (0.41) 0.293*** (4.45) 1.390*** (3.24) 0.259* (1.74) 0.122 (0.47) 0.815 (1.50) 1.246 (1.47)

0.559 (8.45) 0.623*** (12.78) 0.436*** (6.28) 1.059*** (13.89) 0.135*** (3.16) 0.146 (1.24) 0.0192 (0.24) 0.0510 (0.60) 0.299*** (3.86) 1.416*** (3.62) 0.272* (1.89) 0.428** (2.05) 0.892* (1.84) 0.869 (0.97)

0.733*** (12.92) 0.145*** (19.80) 0.628*** (3.55) 0.773*** (21.83) 0.106*** (3.41) 0.092* (1.78) 0.0548 (0.71) 0.582 (0.59) 0.180** (2.14) 1.301*** (3.23) 0.304** (1.97) 0.425** (2.26) 0.0564 (1.28) 3.059*** (3.81)

0.59 7218

0.55 7218

0.60 6535

0.57 6535

Using specification (12) [wi;j;t ¼ a þ M i;j;t þ AIi;j;t þ AI 2007 2009 þ Controlsj;t þ ei;j;t ] foreign equity portfolio holdings of international investors are regressed against the ICAPM benchmark, a measure of access to information (one in each equation), a dummy representing the turbulent market of 2007–2009, and other factors that are known to affect the portfolio allocation of international investors. Each regression also includes an interactive variable (AI_2007_2009) representing the product of the 2007–2009 financial crisis dummy and information accessibility variables. The equations are estimated using the random effect panel data model and the standard errors are corrected for intra-cluster correlation in which the resulting variance–covariance matrix is robust to intra-cluster correlation and arbitrary heteroscedasticity. t statistics are reported in parentheses. Note: The coefficients of ‘AI + AI_2007_2009’ and F-statistics (highlighted in bold) are estimated outside the regression equation. * Denote statistical significance at 10%. ** Denote statistical significance at 5%. *** Denote statistical significance at 1%.

4.8. Impact of the 2007–2009 financial crisis As noted earlier (Section 1), it is possible that investors withdraw their investments more from less transparent markets when faced with uncertain/volatile market conditions. Our sample includes periods of both tranquil (2001–2006) as well as turbulent market conditions (2007–2009) around the world. This allows us to examine whether the portfolio allocation of international investors is dependent on market conditions, given the varying level of AI. We are particularly interested in examining whether the relationship between portfolio allocation and information accessibility that we find in previous estimations holds, becomes less, or more prominent during tranquil/turbulent market conditions. We follow the methodological approach discussed in Section 4.6, and include a dummy variable representing the financial crisis that started in 2007 (2007–2009 financial crisis dummy) and its interaction with measures of AI in specification (12).

wi;j;t ¼ a þ Mi;j;t þ AIi;j;t þ AI 2007 2009 þ Controlsj;t þ ei;j;t

ð12Þ

The estimates, along with the F-statistics, testing for the hypothesis that the sum of coefficients of information accessibility and its interaction with the crisis period dummy are zero (AI + AI_2007_2009 = 0), is reported in Table 8. In all cases the sum of the coefficients of the two regressors (AI and AI_2007_2009) is larger than the coefficient of information accessibility, indicating the effect of the crisis period on portfolio allocation. A much larger shift (from 0.623 to

1.059) is observed in the coefficient of Trade based information, indicating that trade partnership between home and host country becomes more important at the time of the crisis. In other words, these changes in the estimates, along with their statistical significance, confirm that the effect of information cost becomes more pronounced when the market is turbulent. Investors seem to allocate higher proportions of their portfolio in countries against which they have a greater propensity for acquiring information during the period of market turbulence. When faced with higher uncertainty, investors tend to minimise their holdings in countries that are not so well known to them. This is possibly driven by the fear of the contagion effect which accentuates the adverse effect of information immobility (i.e. enhances the cost of acquiring information) on foreign equity portfolio holdings. 5. Conclusions Extant literature documents that relative to the theoretical suggestion of ICAPM investors do not optimally invest in foreign equities to fully exploit the benefits of international portfolio diversification. Studies argue that such disproportionate investments are possibly due to the deadweight costs of holding foreign equities, including the costs of accessing information about foreign markets and securities. Consequently, the proportion of funds invested in foreign countries should be associated with the degree of cross-country accessibility and flow of information. This paper

2266

C. Thapa et al. / Journal of Banking & Finance 37 (2013) 2255–2267

examines this prediction using an extensive data set, alternative measures of information accessibility, and rigorous econometric techniques. The results that are robust to alternative definitions of access to information and regression specifications reveal three important findings. First, we find a significant and positive association between information accessibility about foreign markets and foreign equity portfolio allocations. Investors tend to allocate more funds to countries that are geographically closer to home, share a common language, and have higher bilateral trade. Second, the effect of information accessibility is not symmetric across countries but also depends on the quality of legal/macro-institutions of the host country. Given the same level of information, international investors prefer to invest more in countries that have a higher quality of macro-institutions, characterised by stronger regulatory and law enforcement provisions. This suggests that countries aiming to attract foreign equity portfolio investors can mitigate the adverse effects of higher information costs (e.g. language differences, geographic distance and lower bilateral trade) by improving the quality and efficiency of domestic macro-institutions. Finally, the results also demonstrate that the relevance of cross-country information accessibility on portfolio allocation becomes more prominent when markets are turbulent, which implies that the availability of, and access to, information is more important when markets are turbulent than when they are tranquil. Consequently, during the turbulent periods of financial crisis, international investors tend to favour those markets for which they are able to access and obtain more information.

Appendix A (continued)

Source countries

Host countries

Philippines Poland Russia South Korea Thailand Turkey

Mexico Peru Philippines Poland Russia South Korea Taiwan Thailand Turkey

Appendix B Home allocation, ICAPM benchmark and home bias. This appendix provides estimates of the extent of home bias in sample countries over the period sample period (2001–2009). Column one presents average home allocations estimated by deducting the total inward foreign portfolio investment from the home country market capitalization. Column two is the average proportion of the country’s share in world market capitalization and column 3 is a simple average measure of home bias constructed by deducting the benchmark from the home allocation. Source countries

Home allocation by source countries

ICAPM benchmark weight

Home bias (home allocation – benchmark)

Argentina Australia Austria Belgium Brazil Canada Chile Czech Republic Denmark Finland France Germany Greece Hungary India Indonesia Italy Japan Malaysia Mexico New Zealand Norway Philippines Poland Portugal Russia South Korea Sweden Switzerland Thailand Turkey United Kingdom United States

0.97 0.82 0.68 0.77 0.81 0.81 0.96 0.87 0.78 0.48 0.71 0.68 0.82 0.67 0.91 0.87 0.72 0.83 0.93 0.71 0.81 0.75 0.89 0.89 0.61 0.91 0.79 0.73 0.62 0.86 0.88 0.67 0.84

0.001 0.023 0.003 0.007 0.010 0.036 0.002 0.001 0.005 0.006 0.051 0.048 0.004 0.001 0.012 0.002 0.021 0.106 0.003 0.004 0.001 0.005 0.001 0.002 0.002 0.010 0.012 0.011 0.027 0.002 0.002 0.083 0.417

0.967 0.792 0.681 0.764 0.804 0.776 0.958 0.873 0.777 0.471 0.659 0.637 0.818 0.674 0.898 0.867 0.701 0.723 0.925 0.710 0.804 0.749 0.894 0.891 0.603 0.897 0.773 0.722 0.588 0.855 0.873 0.588 0.425

Appendix A A.1. Sample source and host countries Source countries

Host countries

Developed markets Australia Austria Belgium Canada Denmark Finland France Germany Greece Italy Japan New Zealand Norway Portugal Sweden Switzerland United Kingdom United States

Australia Austria Belgium Canada Denmark Finland France Germany Greece Italy Japan New Zealand Norway Portugal Sweden Switzerland United Kingdom United States

Emerging markets Argentina Brazil Chile Czech Republic Hungary India Indonesia Malaysia Mexico

Argentina Brazil Chile China Czech Republic Hungary India Indonesia Malaysia

C. Thapa et al. / Journal of Banking & Finance 37 (2013) 2255–2267

References Bali, T.G., Cakici, N., 2010. World market risk, country-specific risk and expected returns in international stock markets. Journal of Banking and Finance 34, 1152–1165. Bohn, H., Tesar, L.L., 1996. U.S. equity investment in foreign markets: portfolio rebalancing or return chasing? American Economic Review 86, 77–81. Carrieri, F., Errunza, V., Majerbi, B., 2006. Does emerging market exchange risk affect global equity prices? Journal of Financial and Quantitative Analysis 41, 511– 540. Chan, K., Covrig, V., Ng, L., 2005. What determines the domestic bias and foreign bias? Evidence from mutual fund equity allocations worldwide. Journal of Finance 60, 1495–1534. Claessens, S., Djankov, S., Lang, L., 2000. The separation of ownership and control in East Asian corporations. Journal of Financial Economics 58, 81–112. Cooper, I., Kaplanis, E., 1986. Costs to crossborder investment and international equity market equilibrium. In: Edwards, J., Franks, J., Mayer, C., Schaefer, S. (Eds.), Recent Developments in Corporate Finance. Cambridge University Press, New York. Cooper, I., Kaplanis, E., 1994. Home bias in equity portfolios, inflation hedging, and international capital market equilibrium. Review of Financial Studies 7, 45–60. De Jong, F., de Roon, F.A., 2005. Time-varying market integration and expected returns in emerging markets. Journal of Financial Economics 78, 583–613. Driscoll, J.C., Kraay, A.C., 1998. Consistent covariance matrix estimation with spatially dependent panel data. The Review of Economics and Statistics 80, 540–560. Edison, H.J., Warnock, F.E., 2003. A simple measure of the intensity of capital controls. Journal of Empirical Finance 10, 81–103. Engelen, P., van Essen, M., 2010. Underpricing of IPOs: firm, issue and countryspecific characteristics. Journal of Banking and Finance 34, 1958–1969. Erel, I., Liao, R.C., Weisbach, M.S., 2012. Determinants of cross-border mergers and acquisitions. Journal of Finance 67, 1045–1082. Errunza, V.R., 2001. Foreign portfolio equity investments, financial liberalization, and economic development. Review of International Economics 9, 703–726. Fernandes, N., Ferreira, M.A., 2008. Insider trading laws and stock price informativeness. Review of Financial Studies 22, 1845–1887. Fidora, M., Fratzscher, M., Thimann, C., 2007. Home bias in global bond and equity markets: the role of real exchange rate volatility. Journal of International Money and Finance 26, 631–655. French, K.R., Poterba, J.M., 1991. Investor diversification and international equity markets. American Economic Review: Papers and Proceedings 81, 222–226.

2267

Gelos, R.G., Wei, S.-J., 2005. Transparency and international portfolio holdings. Journal of Finance 60, 2987–3020. Hsiao, C., 1985. Benefits and limitations of panel data. Econometrica 4, 121–174. Keim, D.B., Madhavan, A., 1995. Anatomy of trading process empirical evidence on the trading behaviour of institutional traders. Journal of Financial Economics 37 (3), 371–398. Klau, M.A., Fung, S.S., 2006. The new BIS effective exchange rate indices. BIS Quarterly Review 3, 51–65. Kodongo, O., Ojah, K., 2011. Foreign exchange risk pricing and equity market segmentation in Africa. Journal of Banking and Finance 35, 2295–2310. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R.W., 2000. Investor protection and corporate governance. Journal of Financial Economics 58, 3–27. Lau, S.T., Ng, L., Zhang, B., 2010. The world price of home bias. Journal of Financial Economics 97, 191–217. Moerman, G.A., van Dijk, M.A., 2010. Inflation risk and international asset returns. Journal of Banking and Finance 34, 840–855. Nenova, T., 2003. The value of corporate voting rights and control: a cross-country analysis. Journal of Financial Economics 68, 325–351. Nieuwerburgh, S.V., Veldkamp, L., 2009. Information immobility and the home bias purzzle. Journal of Finance 64, 1187–1215. Portes, R., Rey, H., 2005. The determinants of cross-border equity flows. Journal of International Economics 65, 269–296. Richards, A., 2005. Big Fish in small ponds: the trading behaviour and price impact of foreign investors in Asian emerging equity markets. Journal of Financial and Quantitative Analysis 40, 1–27. Rowland, P.F., 1999. Transaction costs and international portfolio diversification. Journal of International Economics 49, 145–170. Solnik, B., McLeavey, D., 2009. International Investments, sixth ed. Pearson Addison Wesley, New York. Solon, G.S., 1989. The value of panel data in economic research. In: Kasprzyk, D., Duncan, G.J., Kalton, G., Singh, M.P. (Eds.), Panel Survey. John Wiley, New York, pp. 486–496. Subramanian, A., Wei, S.-J., 2005. The WTO promotes trade, strongly but unevenly. Journal of International Economics 72, 151–175. Tsoukas, S., 2011. Firm survival and financial development: evidence from a panel of emerging Asian economies. Journal of Banking and Finance 35, 1736–1752. Wooldridge, J.M., 2002. Econometric Analysis of Cross-Section and Panel Data. MIT Press, Cambridge, MA. Wooldridge, J.M., 2003. Introductory Econometrics: A Modern Approach. Thomson, South Western USA. Zingales, L., 1994. The value of the voting right: a study of the Milan stock exchange experience. Review of Financial Studies 7, 125–148.