Are Global Systematic Risk and Country-Specific Idiosyncratic Risk Priced in the Integrated World Markets? Abstract Empirical evidence showing significant effects of local factors on international equity returns while failing to find significant effects from global systematic risk seems counter-intuitive in today’s integrated world markets. This paper uses the conditional second moments estimated from an asymmetric dynamic conditional correlation model to measure the time-varying world beta and country-specific idiosyncratic risks, and tests the relationship between country-level index returns and world beta risk conditioned on positive and negative world market returns. The results show that the conditional dynamic world beta risks significantly predict the crosscountry variation in expected index returns, while country-specific risk is not significantly priced. JEL Classification: G11, G12, G15 Keywords: World beta risk; Country-specific idiosyncratic risk; Dynamic conditional correlation model. i 1. Introduction Most would agree that financial markets have become increasingly global in the past decades. The recent financial crisis provides evidence of the strong interlinkages among international capital markets. Therefore, one would presume that globalization has led to national stock markets moving more closely together and that internationally traded capital assets should be globally priced. Nonetheless, a large amount of academic research devotes itself to recognizing whether or not the world market risk and country-specific idiosyncratic risk are priced and continues to point to the significance of the effects of local factors while failing to find significant effects from global systematic risk [e.g., Cumby and Glen (1990), Ferson and Harvey (1994), Harvey and Zhou (1993), and Karolyi and Stulz (2003)]. The empirical evidence is summarized well in a recent study by Bali and Cakici (2010), who provide a general cross-sectional test of global capital market integration in an international capital asset pricing model (ICAPM). Using 37 country-level index data and a global market risk factor, they show that there is a positive and significant relationship between expected index returns and country-specific idiosyncratic risk, but the relationship between a global-wide systematic risk and individual country's expected returns is flat. They conclude that the "finding that the differences in countries' stock market returns can be explained by the differences in country-specific risks is . . . consistent with the view that global stock markets are not fully integrated." This conclusion has important implication to international investment because it shows that substantial risk-reduction can be created from diversification. 1 The purpose of this paper is to challenge the above findings and provide more accurate evidence on global capital market integration. The contribution of the paper is twofold. First, on the observed weak global systematic effect, this paper argues that the literature on conditional risk-return relationship stimulated by Pettengill et al. (1995) provides a possible explanation. It states that the relationship between market beta and realized returns is conditional on the market return. In up markets, high-beta securities should be rewarded for bearing risk with higher returns than low-beta securities, but in down markets high-beta securities experience lower returns than low-beta securities. Thus, it is necessary to partition the data into up market and down market periods based on the sign of the realized market excess return. Empirical studies using data from various countries mostly confirm a significant direct relationship between beta and returns in up markets and a significant inverse relationship between beta and returns in down markets. Among them Fletcher (2000) and Tang and Shum (2003) are of particular interest in that they use country-level index data and find a significant conditional relationship between index returns and a world market beta. However, Fletcher uses a CAPM assuming full integration and does not discuss the country-specific idiosyncratic risk, while Tang and Shum only consider exchange rate risk as the country-specific risk. This paper uses a partial integration ICAPM and considers both world beta risk and country-specific idiosyncratic risk. The second contribution of the paper is on the significance of the effects of countryspecific risk on international equity index returns, which is seldom challenged in the literature. One possible challenge worth exploring is the validity of the measure of idiosyncratic risk. This issue has been raised in the literature of firm-level idiosyncratic risk, where monthly idiosyncratic volatility is often measured by realized volatility in the previous month using 2 daily data. 1 Spiegel and Wang (2005), Brockman et al. (2009), and Fu (2009) provide empirical evidence supporting that, compared to the realized monthly idiosyncratic volatility calculated from daily data, the conditional idiosyncratic volatility estimated from GARCH models using monthly data is a more accurate proxy for expected future idiosyncratic volatility.2 These studies, however, all focus on the measure of idiosyncratic volatilities and do not apply time-varying models to measure the dynamics of the systematic market risk. Bali et al. (2012), on the other hand, apply the dynamic conditional correlation (DCC) model of Engle (2002) to construct dynamic conditional market betas and investigate the significance of the conditional betas in predicting the cross-sectional variations in expected returns using firmlevel data. 3 But they do not discuss the idiosyncratic risk. This paper is the first in the literature to consider both dynamic idiosyncratic risk and dynamic systematic risk, and the first to apply the DCC model to country-level equity indices and to the world market integration issue. 1 This literature particularly concerns the idiosyncratic volatility puzzle raised by Ang et al. (2006). Measuring monthly idiosyncratic volatility by realized volatility in the previous month using daily data, they find that stocks with high idiosyncratic volatilities in the previous month have abysmally low average monthly returns. 2 Fu (2009) argues that the idiosyncratic volatility puzzle is caused by the use of the realized idiosyncratic volatility, which is not a good predictor of the expected idiosyncratic volatility. Spiegel and Wang (2005), using monthly data and EGARCH models, also find that stock returns are increasing with the level of idiosyncratic volatility. Brockman et al. (2009) apply Fu’s EGARCH model to another set of international index data and make the same conclusion. On the other hand, Hueng and Yau (2013) show that realized idiosyncratic volatilities work as well as conditional idiosyncratic volatilities in predicting international index returns. 3 You and Daigler (2010) apply the DCC model to estimate the conditional correlations among international equity markets and show that the correlations change over time. 3 In sum, this paper revisits the relationship among world market risk, country-specific risk, and expected returns in international stock markets by proposing the following two improvements over the previous literature. First, this paper uses conditional second moments, instead of lagged realized second moments, to measure the systematic and idiosyncratic risks. The Asymmetric Dynamic Conditional Correlation Multivariate EGARCH (A-DCC-MVEGARCH) model introduced by Capiello et al. (2006) is used to model the time-varying conditional world beta risk and to derive the conditional country-specific idiosyncratic risk. Second, when running the Fama-MacBeth cross-sectional regressions, this paper takes into account the conditional relationship between the global systematic risk and the index returns by partitioning the data into up market and down market periods based on the sign of the realized world market returns. The next section discusses the regression models that are used to test the significance of the systematic and idiosyncratic risks in international asset pricing. The A-DCC-MV- EGARCH model used to estimate the conditional second moments is also specified in this section. Section 3 describes the data and examines the relevant sample statistics. The crosscountry risk-return analyses are shown in Section 4. Section 5 concludes the paper. 2. The Model If international markets are completely integrated, the expected market returns for country i ( Ri ,t ) depend on the covariance with the world market index returns [Dumas and Solnik (1995)]: Et −1 Ri ,t = λt −1Covt −1 ( Ri ,t , Rw,t ) , (1) 4 where λt is the expected world price of risk and Rw,t is the world return. For completely segmented markets, only the variance of the index returns of a country affect the expected returns of a country’s index returns: Et −1 Ri ,t = λi ,t −1Vart −1 ( Ri ,t ) . (2) where λi ,t is the expected price of country i's idiosyncratic risk. Bali and Cakici (2010) generalize (1) and (2) and consider the following partialintegration model: Et −1 Ri ,t = λt −1Covt −1 ( Ri ,t , Rw,t ) + λi ,t −1Vart −1 ( Ri ,t ) . (3) To test the cross-sectional predictive power of world market risk and country-specific risk under this partial integration model, they use the following Fama and MacBeth (1973) regressions across countries in each month t: Ri ,t = γ 0,t + γ 1,t Betai ,t −1 + γ 2,t IVOLi ,t −1 + ε i ,t , (4) where γ 1,t is the price of world beta risk and γ 2,t is the price of country-specific risk. Country i's monthly world market beta ( Betai ,t ) is obtained by regressing country i's daily market portfolio index returns ( Ri ,d ,t ) on the daily world market portfolio returns ( Rw,d ,t ): Ri ,d ,t = µi ,t + Betai ,t ⋅ Rw,d ,t + ri ,d ,t , for d = 1, 2, . . ., Dt , (5) where Dt is the number of trading days in month t and ri ,d ,t is the daily idiosyncratic return. The country-specific idiosyncratic volatility in month t is defined as the realized monthly 5 standard deviation of the daily idiosyncratic returns: IVOLi ,t = Dt ∑ (r d =1 i , d ,t − ri , d ,t ) . That is, they 2 run a two-stage regression. In the first stage, the daily data are used in (5) to obtain the estimates of the monthly observations for Betai ,t and IVOLi ,t . Then in the second stage, they use lagged realized risk measures [ Betai ,t −1 and IVOLi ,t −1 ] to proxy for the expected risks [ Et −1 ( Betai ,t ) and Et −1 ( IVOLi ,t ) ] and run regression (4). Using country-level aggregate market index data from 37 countries and a world market portfolio index, Bali and Cakici find that the time-series average of the estimated effect of IVOLi ,t −1 on Ri ,t (i.e., γˆ2,t ) is positive and statistically significant, while the relationship between average returns on countries' stock market indices and world market beta (i.e., γˆ1,t ) is positive but insignificant.4 This indicates that the country-specific risk is priced and the price of this risk is the same across countries, but the systematic world beta risk is not priced. When analyzing the firm-level data in the U.S. stock market, Pettengill et al. (1995) argue that the Fama-MacBeth methodology cannot directly test the expected risk-return relationship implied by the CAPM. The CAPM shows that since the expected market return must be greater than the risk-free rate (otherwise all investors would hold the risk-free assets), the expected return of any risky asset must be a positive function of beta. However, the FamaMacBeth methodology utilizes the realized returns to proxy for the expected returns. Pettengill et al. argue that there must be a non-zero probability that the realized market return is smaller 4 Bali and Cakici (2010) also replace lagged idiosyncratic volatility with lagged country- specific total volatility and reach the same conclusions on integration/segmentation. 6 than the risk-free rate (otherwise no investor would hold the risk-free assets). To solve this problem, they partition the market into an up market and a down market based on the sign of the realized market excess return. A positive risk-return relationship should exist in the up market and an inverse relationship should exist in the down market. In addition, to test for a positive risk and return tradeoff, two necessary conditions need to hold, namely a positive excess market return on average and a symmetric risk premium in up and down markets. Fletcher (2000) and Tang and Shum (2003) adopt Pettengill et al.'s idea and use country-level index data to test the conditional relationship between index returns and a world market beta: Ri ,t = γ 0,t + γ 1,+tδ t ⋅ Eˆ t −1 Betai ,t + γ 1,−t (1 − δ t ) ⋅ Eˆ t −1 Betai ,t + ε i ,t , (6) where δ t = 1 if Rw,t > 0 (an up world market) and δ t = 0 if Rw,t <0 (a down world market). Fletcher assumes that the world beta is constant over time ( Betai ,t = Betai ) and uses monthly data for the whole sample period (1970-1998) to estimate Et −1 Betai ,t for each country. Tang and Shum assume that the world beta is constant within a year and is estimated by using the data over the past five years in rolling regressions. Both find evidence confirming Pettengill et al.'s argument. But apparently they do not consider country-specific idiosyncratic risk.5 This paper generalizes (4) and (6) and considers the following model: 5 Tang and Shum (2003) use country indices denominated in domestic currencies and consider exchange rates as an idiosyncratic risk factor. They include exchange rate in the first step of the regressions when estimating the world beta to remove the country-specific exchange rate effects. But when testing the risk-return relationship in the second stage, they do not consider the idiosyncratic risk. 7 Ri ,t = γ 0,t + γ 1,+t δ t ⋅ Eˆ t −1 ( Betai ,t ) + γ 1,−t (1 − δ t ) ⋅ Eˆ t −1 ( Betai ,t ) + γ 2,t Eˆ t −1 ( IVOLi ,t ) + ε i ,t . (7) Unlike the aforementioned studies, this paper uses autoregressive conditional second moments to estimate the expected world beta and country-specific risk. Spiegel and Wang (2005), Brockman et al. (2009), and Fu (2009) all provide empirical evidence showing that the conditional idiosyncratic volatility estimated from GARCH models using monthly data is a more accurate proxy for expected future idiosyncratic volatility than the realized monthly idiosyncratic volatility calculated from daily data. The same argument can apply to the covariance between individual country index returns and the world market returns [You and Daigler (2010) and Bali et al. (2012)]. Therefore, I construct the measures of the expected world market beta and idiosyncratic volatility by using the Asymmetric Dynamic Conditional Correlation Multivariate EGARCH (A-DCC-MV-EGARCH) model suggested by Cappiello et al. (2006). Cappiello et al.'s asymmetric DCC model generalizes Engle's (2002) DCC model by allowing conditional asymmetries in correlation dynamics. This modeling strategy involves regressions in two stages. In the first stage a univariate GARCH model is estimated for each asset. In the second stage, the transformed residuals resulting from the first stage are used to estimate a conditional correlation estimator. Specifically for the purpose of this paper, it is assumed that country i's index return ( Ri ,t ) and the world market index return ( Rw,t ) are conditionally bivariate normal with zero hi ,t hiw,t expected value and covariance matrix H t ≡ hiw,t . In the first stage, I use the following hw,t univariate EGARCH model to estimate the conditional variances hi ,t and hw,t : 8 ln h j ,t = κ j + α j ⋅ ln h j ,t −1 + β j ⋅ R j ,t −1 h j ,t −1 +γ j ⋅ R j ,t −1 h j ,t −1 j = i , w. , (8) This EGARCH model, proposed by Nelson (1991), takes account for the importance of the asymmetric responses of conditional volatilities to positive and negative news. In the second stage, let ε j ,t = R j ,t h j ,t be the standardized returns with h j ,t estimated from the first stage and ε t = ε i ,t ε w,t ' . Express the covariance matrix as H t = Ct ⋅ Pt ⋅ Ct , where hi ,t Ct = 0 0 . Then Pt = Ct−1 ⋅ H t ⋅ Ct−1 is the correlation matrix with ones on the diagonal hw,t and the off-diagonal element less than or equal to one in absolute value. Engle (2002) suggests that the diagonal element of the correlation matrix can be constructed by estimating another univariate GARCH process. This computational advantage makes the DCC model more attractive than other multivariate GARCH specifications such as the VEC model and the BEKK representation [see Engle and Kroner (1995)]. q11,t Qt ≡ q12,t q11,t q12,t * is symmetric and Q = t q22,t 0 Let Pt = Qt*−1 ⋅ Qt ⋅ Qt*−1 , where 0 , the matrix version of the DCC model q22,t is given by Qt = ( P − a 2 P − b 2 P ) + a 2ε t −1ε t' −1 + b 2 ⋅ Qt −1 , where P = E (ε t ε t' ) and a and b are parameters to be estimated. To allow for conditional asymmetric responses in correlation dynamics to positive and negative news, Cappiello et al. generalize the evolution of the correlation to: Qt = ( P − a 2 P − b 2 P − g 2 N ) + a 2ε t −1ε t' −1 + g 2 nt −1nt' −1 + b 2 ⋅ Qt −1 , 9 (9) where nt = I [ε t < 0] ε t and N = E (nt nt' ) . The indicator function I [⋅] equals one if the argument is true and zero otherwise, and is the Hadamard product. To estimate the model, P and N are proxied by their sample analogues and Q0 is set to equal P . A sufficient condition for Qt to be positive definite is a positive semi-definite ( P − a 2 P − b 2 P − g 2 N ) . A necessary and sufficient condition for this to hold is a 2 + b 2 + ξ ⋅ g 2 < 1, where ξ is the maximum eigenvalue of P −1/2 ⋅ N ⋅ P −1/2 . Conditioning on the parameters estimated in the first stage, a, b, and g can be estimated by maximizing the likelihood: − 1 log ( Pt ) + ε t' Pt −1ε t . ∑ 2 t Once Pt is estimated in the second stage, combining Ct estimated from the first stage yields the estimate of H t . Then the time-varying conditional correlation is hiw ,t hi ,t hw ,t , the h conditional world market beta is measured as Eˆ t −1 ( Betai ,t ) = iw,t , and the idiosyncratic return is hw,t constructed as ri ,t = Ri ,t − Eˆ t −1 ( Betai ,t ) ⋅ Rw,t . The conditional idiosyncratic volatility Eˆt −1 ( IVOLi ,t ) is the squared root of the conditional variance from another univariate EGARCH process (8) for ri ,t . Once Eˆ t −1 ( Betai ,t ) and Eˆt −1 ( IVOLi ,t ) are obtained from the A-DCC-MV-EGARCH model in the first stage, they are plugged into the cross-sectional regression (7) in the second stage. 3. Data I use the same data as those in Bali and Cakici (2010), whose data end in September 2006, but update their data to May 2012. The data are obtained from Datastream Global indices and 10 include stock market indices for 37 countries plus the world market portfolio, all denominated in U.S. dollars. There are 23 developed markets and 14 developing or emerging markets.6 Panel (A) of Table 1 shows the summary statistics of the monthly market index returns for each country and the world market, including the means, standard deviations, and constant correlations with the world market index returns. The starting month for each country is shown in the second column. The sample ends in May 2012 for all countries. Compared to those reported in Bali and Cakici, the updated data show slightly lower means and standard deviations. More interestingly, the constant correlations with the world are higher in most of the countries with the new data added, indicating that the global markets are getting more integrated after their study.7 Comparing across countries, the summary statistics are in general very similar to those in Bali and Cakici in that the emerging markets exhibit higher average returns and higher standard deviations of returns compared to the developed markets. In Panel (B) of Table 1, the first four columns show the means, standard deviations, maximum values, and minimum values of the time-varying correlations with the world estimated from the A-DCC-MV-EGARCH model. As can be seen for most countries, the 6 Ince and Porter (2006) point out several data problems in Datastream. Since most of the problems identified in their paper are concentrated in the smaller size firms, to make sure that the potential problems do not change my conclusions, I also do the analyses using an alternative data source - the MSCI large- and mid-cap price returns, available from January 1980. This dataset also avoids the problem from using index returns. These results do not change the conclusions in the paper and are available from the author upon request. 7 Even though here I follow Bali and Cakici (2010) and preliminarily use the average correlations with the world market as an alternative measure of integration, note that it is wellknown in the literature that correlation is not a good measure of integration. See, for example, Dumas et al. (2003) and Carrieri et al. (2007). 11 variations of the correlations are significant. When regressing the dynamic correlations on a constant and a linear time trend, I find that the slope against the time trend (reported in the fifth column) is positive and statistically significant for all countries except for Finland (an insignificantly positive slope) and Japan (an insignificantly negative slope). This indicates that the degree of integration with the world is increasing over time in most of the markets. Panels (A) and (B) of Table 2 report the summary statistics for the realized world market beta, Betai ,t −1 , and the conditional measure of world market beta, Eˆ t −1 ( Betai ,t ) , respectively. The realized beta, used in Bali and Cakici, is estimated by using daily data within each month to run time-series regression (5). In general the realized beta in Panel (A) shows high volatilities, with the standard deviation as high as the mean in most of the markets. The firstorder autocorrelation of the realized beta is mostly lower than 0.5, indicating that past realized market beta is not a good predictor of future market beta. On the other hand, in Panel (B), the conditional measure of beta estimated from the A-DCC-MV-EGARCH model is more stable, with the standard deviation much lower than the mean in most of the markets. In addition, the conditional measure of beta is very persistent, with the first-order autocorrelation coefficient mostly higher than 0.9. Furthermore, the last column of Table 2 shows that the correlations between the realized beta and the conditional beta are mostly lower than 0.5. Therefore, it is expected that using the conditional beta would yield different results than those in the previous studies that use the realized beta. Panels (A) and (B) of Table 3 report the summary statistics of the realized idiosyncratic volatility, IVOLi ,t −1 , and the conditional measure of idiosyncratic volatility, Eˆ t −1 ( IVOLi ,t ) , respectively. The realized monthly idiosyncratic volatility is calculated from the daily 12 idiosyncratic returns estimated in equation (5). Compared to those reported in Bali and Cakici, the realized country-specific volatilities are in general lower with the updated data added. But the cross-country comparison is in general consistent with those reported in Bali and Cakici in that the country-specific idiosyncratic volatility is much higher in the emerging markets than in the developed markets. The first-order autocorrelations are around 0.5 for most countries, which again indicates that past realized idiosyncratic volatility is not a very good predictor of future idiosyncratic volatility. The conditional idiosyncratic volatility estimated from the ADCC-MV-EGARCH model, on the other hand, generally has a higher mean and is more stable compared to the realized measure. The conditional measure is also very persistent. Finally, the last column of Table 3 shows that the correlations between the realized measure and the conditional measure are mostly lower than 0.5. Therefore, it is important to replace the realized measures with the conditional measures in order to test the risk-return relationship. 4. Empirical Results The empirical work starts by revisiting Bali and Cakici's (2010) results in equation (4), where the Fama and MacBeth (1973) regression is used for the extended sample period to examine the cross-country relationship between expected returns and the realized measures of risks. Following Bali and Cakici, I include two more control variables, the earnings-to-price ratio and the dividends-to-price ratio, in the regressions. The first row of Panel (A) in Table 4 reports the time-series averages of the estimated coefficients, their p-values based on the Newey and West (1987) heteroscedasticity- and autocorrelation-adjusted t-statistics, and the time-series averages of the R-squared. The newly added data do not alter Bali and Cakici's conclusions: the country-specific risk is priced and its effect on index returns is statistically 13 significant. The effect of the systematic risk on index returns, on the other hand, is highly insignificant. The second row of Panel (A) shows the results from using the conditional measures of the world market beta and idiosyncratic risks. The conditional measures fit the model better by raising the R-squared by 1.7%. More importantly, the effect of the idiosyncratic risk on the index returns is lowered and only statistically significant at the 7.7% level. Recall that the world markets are more integrated after Bali and Cakici's study. Therefore, one would expect the significant effect of the country-specific risk on returns found in their paper to decline in the new sample. Using the conditional measure, the results of this paper are consistent with this expectation.8 Next I turn the attention to the systematic risk. Consistent with previous studies, the results in Panel (A) of Table 4 show that there is no significant unconditional relationship between expected returns and the world market beta. However, these results from the FamaMacBeth regressions are apparently subject to Pettengill et al.'s (1995) critics and cannot directly test the expected risk-return relationship implied by the CAPM. To conduct a valid test, the Fama-MacBeth regressions should be conditioned on positive and negative world market returns. The results of the conditional regression (7) in the up and down world markets are shown in Panel (B) of Table 4. Consistent with Fletcher (2000) and Tang and Shum 8 I also run a regression with conditional measures using only data prior to the recent financial crisis (data up to 2007). The conclusions do not change. The estimated coefficient on the idiosyncratic risk is higher (0.105) in the pre-crisis period than that (0.094) in the whole sample period, and is again only marginally significant at the 6.7% level. This result supports the argument that the world market is more integrated during the financial crisis. 14 (2003), when the sample is partitioned into up and down world markets, there is a significant positive relationship between index returns and the world beta in up market months and a significant and negative relationship in down market months. Therefore, high beta indices outperform low beta indices when the realized world market return is positive, but incur higher losses when the realized world market return is negative. To test for a positive systematic risk and return tradeoff, the last column of Panel (B) reports the test statistics of the two-sample t-test for a symmetric risk premium in up and down market months. The symmetric relationship cannot be rejected at the conventional significance level. Combined with the positive mean world return, the evidence supports the expectation of a positive reward for bearing systematic risk. Along with the results in Panel (A), after adding the most recent data when the world markets are highly integrated, the world market beta remains significant but the country-specific risk is only marginally significant. Further support from a different sample The empirical results so far provide more intuitive evidence that country-specific risk plays a lesser role in determining international equity returns in a more integrated world capital market. To further support this argument, I consider a more recent sample period when the global market is increasingly more integrated. Studies have provided evidence supporting that the degree of world financial market integration has increased since the 1990s. For example, Carrieri et al. (2007) use an integration index to show that emerging markets have become 15 more integrated with other global markets after the early 1990s.9 Hardouvelis et al. (2006) find that European markets became fully integrated in the second half of the 1990s due to the formation of the European Union. You and Daigler (2010) use time-varying correlations to show that the correlations of international index markets have increased since the late 1990s. Therefore, I consider data starting from the 1990s to see whether the effect of country-specific risk on international equity returns is insignificant in a more integrated world capital market. A larger pool of international index data is available starting from June 1994. The MSCI Investable Market Indices cover all investable large, mid, and small cap securities across 24 developed markets, 21 emerging markets, and 26 frontier markets, a total of 71 markets plus the world market portfolio. In Appendix, Tables A1-A3 report the summary statistics of the data for the developed markets, the emerging markets, and the frontier markets, respectively. Since the data on the earnings-to-price ratio and the dividends-to-price ratio are not available for half of the countries, these two variables are not included in the regression. Since the results in Section (II) of Table 4 show that these two variables do not have significant effects on returns when the conditional measures are used, it should not be too unreasonable to assume that ignoring these two variables would not change the conclusions.10 Table 5 reports the estimation results. There is a significant and positive relationship between index returns and the world beta in up market months and a significant and negative 9 Bekaert and Harvey (1995, 2000) imply that liberalization of the financial markets of emerging countries causes further integration of international stock markets. Bekaert, Harvey, and Lumsdaine (2002) show that significant break points exist for factors that indicate world financial market integration for several emerging markets after the liberalization. 10 Fletcher (2000) and Tang and Shum (2003) do not include these two variables in their analyses, either. 16 relationship in down market months. The relationships in the up and down markets are symmetric. Therefore, the systematic risk is priced. On the other hand, I find no evidence of a priced country-specific idiosyncratic risk. The relationship between index returns and idiosyncratic risk is highly insignificant. This result indicates that the global capital markets are highly integrated so that the country-specific idiosyncratic risk is not important in pricing international equity indices. 5. Conclusions Empirical evidence pointing to the significant effects of local factors on global equity returns while failing to find significant effects from global systematic risk seems counter-intuitive in today’s integrated world markets. This paper revisits this empirical issue by providing more accurate evidence. Specifically, an asymmetric dynamic conditional correlation multivariate EGARCH model is used to estimate the time-varying conditional world market beta risk and to derive the conditional country-specific idiosyncratic risk. In addition, when using the Fama and MacBeth (1973) methodology to test the significance of the relationship between world beta risk and country-level index returns, this paper takes into account the conditional relationship between the global systematic risk and index returns by partitioning the data into up market and down market periods based on the sign of the realized world market returns. Using 37 country-level index data and a global market risk factor from 1973 to 2012, and a larger pool of data from 71 countries for the period 1994-2012, this paper shows that the conditional world beta risks significantly affect country-level index returns, while countryspecific risk factors are not significantly priced. The results, therefore, support international financial integration. 17 References Ang, A., Hodrick, R. J., Xing, Y., and Zhang, X. (2006). The Cross-Section of Volatility and Expected Returns. Journal of Finance 61, 259-299. Bali, T. G., Engle, R. F. and Tang, Y. (2012). Dynamic Conditional Beta is Alive and Well in the Cross-Section of Daily Stock Returns. Fordham University Schools of Business Research Paper No. 2089636. Available at SSRN: http://ssrn.com/abstract=2089636. Bali, T.G. and Cakici, N. (2010). World Market Risk, Country-Specific Risk and Expected Returns in International Stock Markets. Journal of Banking and Finance 34, 11521165. Bekaert, G. and Harvey, C. (1995). Time-Varying World Market Integration. Journal of Finance 50, 403-444. Bekaert, G. and Harvey, C. (2000). Foreign Speculators and Emerging Equity Markets. Journal of Finance 55, 565-613. Bekaert, G., Harvey, C. and Lumsdaine, R. (2002). Dating the Integration of World Equity Markets. Journal of Financial Economics 65, 203-247. Brockman, P., Schutte, M.G. and Yu, W. (2009). International Evidence. Working Is Idiosyncratic Risk Priced? paper. Available at The SSRN: http://ssrn.com/abstract=1364530. Cappiello, L. Engle, R. F. and Sheppard, K. (2006). Asymmetric Dynamics in the Correlations of Global Equity and Bond Returns. Journal of Financial Econometrics 4, 537-572. 18 Carrieri, F., Errunza, V. and Hogan, K. (2007). Characterizing World Market Integration through Time. Journal of Financial and Quantitative Analysis 42, 915-940. Cumby, R. E. and Glen, J. D. (1990). Evaluating the Performance of International Mutual Funds. Journal of Finance 45, 497-522. Dumas, B., Harvey, C., Ruiz, P. (2003). Are Correlations of Stock Returns Justified by Subsequent Changes in National Outputs? Journal of International Money and Finance 22, 777-811. Dumas, B., and Solnik, B. (1995). The World Price of Foreign Exchange Risk. Journal of Finance 50, 445–479. Engle R. F. (2002). Dynamic Conditional Correlation - A Simple Class of Multivariate GARCH Models. Journal of Business and Economic Statistics 20, 339-350. Engle R. F. and Kroner, K. (1995). Multivariate Simultaneous GARCH. Econometric Theory 11, 122-150. Fama, E. and MacBeth, J. (1973). Risk, Return, and Equilibrium: Empirical Tests. Journal of Political Economy 81, 607-636. Ferson, W. F. and Harvey, C. (1994). Sources of Risk and Expected Returns in Global Equity Markets. Journal of Banking and Finance 18, 775-803. Fletcher, J. (2000). On the Conditional Relationship between Beta and Returns in International Stock Returns. International Review of Financial Analysis 9, 235-245. 19 Fu, F. (2009). Idiosyncratic Risk and the Cross-Section of Expected Stock Returns. Journal of Financial Economics 91, 24-37. Hardouvelis, G., Malliaropulos, D. and Priestley, R. (2006). EMU and European Stock Market Integration. Journal of Business 79, 365-392. Harvey, C. and Zhou, G. (1993). International Asset Pricing with Alternative Distributional Specifications. Journal of Empirical Finance 1,107-131. Hueng, J. and Yau, R. (2013). Country-Specific Idiosyncratic Risk and Global Equity Index Returns. International Review of Economics & Finance, 25, 326-337. Ince, O. S. and Porter, R. B. (2006). Individual Equity Return Data from Thomson Datastream: Handle with Care! Journal of Financial Research, 29, 463–479. Karolyi G. A. and Stulz, R. M. (2003). Are Financial Assets Priced Locally or Globally? Handbook of the Economics of Fnance. Vol 1B: 975-1020 Newey, W. K. & West, K. D. (1987). A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix. Econometrica 55, 703-708. Nelson, D. (1991). Conditional Heteroskedasticity in Asset Returns: a New Approach. Econometrica 59, 347-370. Pettengill, G., Sundaram, S. and Mathur, I. (1995). The Conditional Relation Between Beta and Returns. Journal of Financial and Quantitative Analysis 30, 101-116. 20 Tang, G. Y. N., & Shum, W.C. (2003). The Conditional Relationship between Beta and Returns: Recent Evidence from International Stock Markets. International Business Review 12, 109-126. You, L. & Daigler, R. T. (2010). Is International Diversification Really Beneficial? Journal of Banking and Finance 34, 163-173. 21 Table 1: Summary Statistics of International Market Indices The data are from Datastream Global indices. The sample ends in May 2012 for all countries. The dynamic correlations with the world are estimated from the A-DCC-MV-GARCH model. The asterisk * indicates statistical significance at the 5% level, and ** at the 10% level. Country Data start Argentina Australia Austria Belgium Brazil Canada Chile China Denmark Finland France Germany Greece Hong Kong India Ireland Italy Japan Korea Malaysia Mexico Netherlands New Zealand Norway Philippines Poland Portugal Singapore South Africa Spain Sweden Switzerland Taiwan Thailand Turkey UK US WORLD Aug-93 Jan-73 Jan-73 Jan-73 Jul-94 Jan-73 Jul-89 Jul-93 Jan-73 Mar-88 Jan-73 Jan-73 Jan-90 Jan-73 Jan-90 Jan-73 Jan-73 Jan-73 Sep-87 Jan-86 May-89 Jan-73 Jan-88 Jan-80 Nov-88 Mar-94 Jan-90 Jan-73 Jan-73 Mar-87 Jan-82 Jan-73 May-88 Jan-87 Jun-89 Jan-73 Jan-73 Jan-73 (A) Market Index Returns Const. Std. Mean Corr. w/ Dev. WORLD 0.552 9.257 0.521 1.103 7.242 0.659 0.965 6.789 0.527 0.990 5.886 0.684 1.525 10.760 0.689 0.945 5.550 0.766 1.572 6.740 0.484 1.541 11.121 0.429 1.115 5.944 0.624 0.987 8.673 0.680 1.101 6.780 0.731 0.939 6.066 0.721 0.673 10.312 0.494 1.399 9.893 0.533 1.294 10.647 0.367 1.074 7.262 0.677 0.860 7.636 0.581 0.757 6.180 0.701 1.143 11.043 0.567 1.282 8.627 0.446 1.605 8.608 0.609 1.057 5.607 0.829 0.900 6.407 0.629 1.188 8.034 0.677 1.236 9.020 0.480 0.831 10.805 0.623 0.490 6.165 0.671 1.033 8.422 0.641 1.320 8.262 0.568 0.834 6.619 0.776 1.326 7.357 0.754 1.024 5.172 0.728 0.905 10.780 0.452 1.535 10.665 0.529 2.291 16.557 0.383 1.076 6.513 0.741 0.928 4.484 0.823 0.868 4.527 1.000 22 (B) Dynamic Correlation with World Trend Mean Std Max Min Slope x103 0.573 0.051 0.751 0.390 0.093** 0.660 0.181 0.956 0.144 0.862* 0.488 0.206 0.917 0.132 0.887* 0.652 0.146 0.924 0.271 0.493* 0.709 0.062 0.886 0.358 0.354* 0.752 0.089 0.892 0.491 0.342* 0.496 0.235 0.793 -0.171 2.387* 0.440 0.213 0.780 -0.051 2.051* 0.592 0.133 0.894 0.316 0.755* 0.682 0.000 0.682 0.682 0.000 0.736 0.137 0.964 0.364 0.752* 0.672 0.211 0.970 -0.021 1.014* 0.516 0.253 0.940 -0.094 2.431* 0.583 0.200 0.911 -0.330 0.596* 0.354 0.284 0.806 -0.297 3.260* 0.682 0.100 0.874 0.452 0.479* 0.572 0.213 0.960 0.133 1.144* 0.714 0.057 0.845 0.527 -0.021 0.565 0.230 0.906 -0.104 2.218* 0.526 0.126 0.803 0.213 0.573* 0.643 0.232 0.929 0.029 2.329* 0.806 0.072 0.951 0.624 0.349* 0.611 0.115 0.838 0.338 1.047* 0.666 0.140 0.908 0.348 0.897* 0.482 0.065 0.664 0.299 0.240* 0.626 0.149 0.934 -0.088 0.940* 0.621 0.108 0.854 0.387 1.024* 0.646 0.148 0.923 0.036 0.457* 0.553 0.137 0.885 0.348 0.739* 0.750 0.091 0.900 0.461 0.606* 0.723 0.174 0.958 0.170 1.214* 0.710 0.118 0.914 0.252 0.326* 0.494 0.048 0.759 0.361 0.110* 0.500 0.146 0.817 0.010 0.696* 0.406 0.227 0.748 -0.017 2.588* 0.787 0.112 0.975 0.489 0.658* 0.822 0.114 0.964 0.420 0.120* --------------------- Table 2: Summary Statistics of the Realized and Conditional Market Betas The realized beta is estimated by equation (5) using daily data. The conditional beta is estimated from the A-DCC-MV-GARCH model using monthly data. Country Argentina Australia Austria Belgium Brazil Canada Chile China Denmark Finland France Germany Greece Hong Kong India Ireland Italy Japan Korea Malaysia Mexico Netherlands New Zealand Norway Philippines Poland Portugal Singapore South Africa Spain Sweden Switzerland Taiwan Thailand Turkey UK US Mean 0.790 0.555 0.607 0.693 1.275 0.788 0.561 0.729 0.604 1.146 0.903 0.886 0.726 0.653 0.405 0.727 0.755 0.960 0.682 0.455 1.000 0.888 0.471 0.920 0.397 0.980 0.671 0.541 0.733 0.989 1.038 0.733 0.512 0.575 0.869 0.929 0.955 (A) Realized Beta Std. Auto corr. 0.735 0.160 0.560 0.352 0.578 0.572 0.464 0.445 0.701 0.356 0.396 0.452 0.472 0.312 0.711 0.323 0.569 0.394 0.662 0.360 0.507 0.479 0.508 0.529 0.729 0.123 0.676 0.266 0.676 0.358 0.571 0.454 0.706 0.363 0.636 0.658 0.732 0.141 0.605 0.412 0.716 0.179 0.451 0.509 0.484 0.151 0.621 0.464 0.673 0.162 0.762 0.365 0.497 0.509 0.546 0.247 0.742 0.480 0.483 0.544 0.610 0.487 0.433 0.342 0.738 0.194 0.913 0.263 1.274 0.357 0.453 0.357 0.359 0.663 (B) Conditional Beta Corr. between Realized Mean Std. Auto corr. and Conditional Betas 1.181 0.272 0.932 0.216 1.038 0.303 0.947 0.288 0.672 0.354 0.951 0.554 0.839 0.229 0.919 0.304 1.633 0.389 0.747 0.131 0.942 0.201 0.963 0.518 0.719 0.341 0.965 0.307 1.045 0.584 0.915 0.359 0.779 0.210 0.954 0.339 1.295 0.238 0.907 0.170 1.094 0.236 0.942 0.150 0.899 0.318 0.942 0.409 0.999 0.479 0.900 0.303 1.144 0.482 0.802 0.133 0.755 0.717 0.976 0.469 1.078 0.199 0.855 0.226 0.938 0.319 0.959 0.357 0.991 0.219 0.951 0.557 1.227 0.495 0.912 0.272 0.859 0.289 0.881 0.363 1.148 0.340 0.884 0.310 1.002 0.171 0.945 0.463 0.839 0.178 0.909 0.155 1.167 0.293 0.959 0.432 0.925 0.218 0.915 0.121 1.443 0.414 0.536 0.033 0.833 0.172 0.937 0.518 1.113 0.382 0.929 0.065 1.025 0.272 0.952 0.399 1.048 0.200 0.833 0.422 1.178 0.301 0.905 0.430 0.824 0.141 0.897 0.060 1.101 0.256 0.885 -0.035 1.115 0.372 0.907 0.345 1.371 0.738 0.981 0.348 1.063 0.218 0.926 0.165 0.824 0.162 0.940 0.574 23 Table 3: Summary Statistics of the Realized and Conditional Idiosyncratic Volatilities The realized idiosyncratic volatility is calculated from equation (5) using daily data. The conditional idiosyncratic volatility is estimated from the DCC-MV-GARCH model using monthly data. Country Argentina Australia Austria Belgium Brazil Canada Chile China Denmark Finland France Germany Greece Hong Kong India Ireland Italy Japan Korea Malaysia Mexico Netherlands New Zealand Norway Philippines Poland Portugal Singapore South Africa Spain Sweden Switzerland Taiwan Thailand Turkey UK US (A) Realized IVOL Mean Std. Auto corr. 6.225 3.898 0.526 4.848 2.210 0.508 3.709 1.772 0.608 3.690 1.578 0.460 6.267 3.266 0.608 2.896 1.314 0.523 4.178 1.853 0.510 7.261 3.676 0.622 4.205 1.993 0.407 5.707 2.987 0.686 4.067 1.788 0.531 3.597 1.543 0.426 6.691 3.318 0.605 6.309 3.857 0.587 6.595 3.524 0.428 4.424 1.988 0.427 4.981 2.393 0.606 4.003 2.048 0.635 8.086 5.025 0.719 5.078 4.327 0.671 5.211 3.253 0.631 3.420 1.581 0.576 4.651 2.105 0.462 5.322 2.336 0.532 6.064 3.157 0.408 6.887 3.698 0.581 3.857 1.543 0.500 4.941 2.877 0.607 5.781 2.693 0.455 4.008 1.839 0.480 4.874 2.204 0.524 3.480 1.418 0.476 7.432 3.484 0.678 7.171 3.756 0.621 11.153 5.810 0.579 3.792 1.838 0.656 2.471 1.496 0.599 (B) Conditional IVOL Mean Std. Auto corr. 7.743 2.136 0.920 5.024 1.398 0.978 4.998 1.846 0.915 4.148 0.877 0.944 7.337 2.417 0.927 3.425 0.275 0.980 5.721 1.373 0.963 9.312 3.487 0.918 4.499 0.893 0.936 6.056 1.493 0.962 4.277 1.187 0.967 3.998 0.895 0.957 7.782 3.399 0.930 7.183 3.392 0.902 9.127 2.435 0.945 5.160 1.472 0.946 5.849 1.548 0.970 4.230 0.503 0.949 8.142 3.447 0.916 6.668 2.966 0.940 6.359 2.041 0.976 3.056 0.404 0.953 4.566 0.595 0.984 5.602 1.195 0.966 7.609 1.536 0.961 7.832 2.249 0.962 4.542 0.650 0.951 5.855 2.161 0.913 6.643 1.238 0.967 4.097 0.749 0.937 4.810 1.197 0.892 3.550 0.675 0.922 8.689 2.634 0.982 8.747 2.022 0.974 14.340 3.240 0.988 3.777 1.348 0.958 2.399 0.556 0.933 24 Corr. between Realized and Conditional IVOLs 0.438 0.186 0.468 0.413 0.465 -0.014 0.487 0.405 0.259 0.540 0.399 0.355 0.564 0.596 0.395 0.333 0.490 0.004 0.565 0.594 0.526 0.359 0.327 0.358 0.373 0.483 0.367 0.537 0.324 0.421 0.320 0.313 0.565 0.501 0.520 0.519 0.279 Table 4: Cross-Sectional Regressions In Panel (A), (I) reports the cross-sectional regression results from: Ri ,t = γ 0,t + γ 1,t Betai ,t −1 + γ 2,t IVOLi ,t −1 + γ 3,t EPi ,t −1 + γ 4,t DYi ,t −1 + ε i ,t , where the monthly Betai ,t is obtained by regressing country i's daily market returns on the daily world market returns [Equation (5)] and the idiosyncratic volatility IVOLi ,t is the realized monthly standard deviation of the daily idiosyncratic returns obtained from the regression. The control variable EPi ,t is the natural logarithm of the earnings-to-price ratio, and DYi ,t is the natural logarithm of the dividends-to-price ratio in month t. Rows (II) and (III) of Panel (A) reports the cross-sectional regression results from: Ri ,t = γ 0,t + γ 1,t Eˆ t −1 ( Betai ,t ) + γ 2,t Eˆt −1 ( IVOLi ,t ) + γ 3,t EPi ,t −1 + γ 4,t DYi ,t −1 + ε i ,t , where the monthly observations of Eˆ t −1 ( Betai ,t ) and Eˆ t −1 ( IVOLi ,t ) are obtained from the regression results of the A-DCC-MVEGARCH model [Equations (8) and (9)]. In Panel (B), (I) reports the cross-sectional regression results from: Ri ,t = γ 0,t + γ 1,+t Dt ⋅ Betai ,t −1 + γ 1,−t (1 − Dt ) ⋅ Betai ,t −1 + γ 2,t IVOLi ,t −1 + γ 3,t EPi ,t −1 + γ 4,t DYi ,t −1 + ε i ,t , and (II) and (III) report the cross-sectional regression results from: Ri ,t = γ 0,t + γ 1,+t Dt ⋅ Eˆ t −1 ( Betai ,t ) + γ 1,−t (1 − Dt ) ⋅ Eˆ t −1 ( Betai ,t ) + γ 2,t Eˆ t −1 ( IVOLi ,t ) + γ 3,t EPi ,t −1 + γ 4,t DYi ,t −1 + ε i ,t . The average intercepts, average slope coefficients, and average R2 are presented. The numbers in parentheses are P-values calculated based on Newey and West (1987) t-statistics. A P-value of 0.000 indicates that the P-value is nonzero, but smaller than 0.0005. (A) Model (I) Realized Measures (II) Conditional Measures γ0 γ1 γ2 (B) γ3 γ4 2.525 0.096 0.118 0.677 -0.015 (0.001) (0.549) (0.036) (0.001) (0.940) 1.102 0.230 0.094 0.277 0.123 (0.186) (0.511) (0.077) (0.246) (0.509) 25 Average R2 0.329 0.346 γ 1+ γ 1− 0.701 -0.881 (0.000) (0.001) 1.926 -2.522 (0.000) (0.000) H 0 : γ 1+ = γ 1− P-Value 0.594 0.343 Table 5: Cross-Sectional Regressions for 71 Markets This table reports the cross-sectional regression results from: Ri ,t = γ 0,t + γ 1,+t Dt ⋅ Eˆ t −1 ( Betai ,t ) + γ 1,−t (1 − Dt ) ⋅ Eˆ t −1 ( Betai ,t ) + γ 2,t Eˆ t −1 ( IVOLi ,t ) + ε i ,t . The average intercepts, average slope coefficients, and average R2 are presented. The numbers in parentheses are p-values calculated based on Newey and West (1987) t-statistics. A p-value of 0.000 indicates that the p-value is nonzero, but smaller than 0.0005. γ0 γ 1+ γ2 γ 1− 0.187 2.664 -3.049 0.024 (0.237) (0.000) (0.000) (0.710) 26 Average R2 0.199 H 0 : γ 1+ = γ 1− p-value 0.509 Appendix: The data in Tables A1-A3 are from MSCI Investable Market Indices. The sample ends in May 2012 for all countries. The dynamic correlations with the world are estimated from the ADCC-MV-GARCH model. A value of 0.000 indicates that it is nonzero but smaller than 0.0005. Table A1: Summary Statistics of International Market Indices (Developed Markets) (A) Market Index Returns Constant Data start Country Mean Std Correlation with World AUSTRALIA May-94 0.675 6.224 0.844 AUSTRIA May-94 0.431 6.930 0.736 BELGIUM May-94 0.415 6.141 0.778 CANADA May-94 0.875 6.069 0.861 DENMARK May-94 0.866 5.880 0.795 FINLAND May-94 0.810 9.033 0.748 FRANCE May-94 0.456 6.066 0.892 GERMANY May-94 0.492 6.743 0.881 GREECE May-94 -0.024 10.02 0.606 HONG KONG May-94 0.509 7.464 0.698 IRELAND May-94 0.440 6.728 0.795 ISRAEL May-94 0.643 7.003 0.637 ITALY May-94 0.201 7.053 0.774 JAPAN May-94 -0.088 5.494 0.653 NETHERLANDS May-94 0.492 6.015 0.886 NEW ZEALAND May-94 0.352 5.985 0.692 NORWAY May-94 0.847 7.750 0.799 PORTUGAL May-94 0.232 6.544 0.691 SINGAPORE May-94 0.457 7.591 0.724 SPAIN May-94 0.505 6.894 0.799 SWEDEN May-94 0.917 7.458 0.849 SWITZERLAND May-94 0.648 4.911 0.779 UNITED KINGDOM May-94 0.421 4.758 0.901 USA May-94 0.626 4.680 0.955 WORLD May-94 0.418 4.614 1.000 27 (B) Dynamic Correlation with World Mean Std Max Min 0.803 0.644 0.737 0.843 0.747 0.719 0.851 0.827 0.568 0.709 0.751 0.661 0.733 0.645 0.854 0.627 0.765 0.635 0.704 0.774 0.810 0.744 0.874 0.943 --- 0.073 0.179 0.112 0.024 0.077 0.091 0.104 0.166 0.208 0.055 0.091 0.000 0.150 0.058 0.059 0.118 0.110 0.095 0.084 0.069 0.094 0.125 0.059 0.036 --- 0.939 0.906 0.899 0.888 0.867 0.863 0.963 0.958 0.914 0.874 0.896 0.661 0.958 0.820 0.938 0.875 0.917 0.823 0.890 0.876 0.931 0.901 0.952 0.979 --- 0.633 0.168 0.429 0.784 0.497 0.441 0.334 0.020 -0.015 0.469 0.373 0.661 0.276 0.522 0.633 0.304 0.420 0.404 0.554 0.537 0.501 0.256 0.675 0.834 --- Table A2: Summary Statistics of International Market Indices (Emerging Markets) Country BRAZIL CHILE CHINA COLOMBIA CZECH REPUBLIC EGYPT HUNGARY INDIA INDONESIA KOREA MALAYSIA MEXICO MOROCCO PERU PHILIPPINES POLAND RUSSIA SOUTH AFRICA TAIWAN THAILAND TURKEY Data start May-94 May-94 May-94 May-94 May-95 May-96 May-94 May-94 May-94 May-94 May-94 May-96 May-97 May-95 May-94 May-94 May-96 May-94 May-94 May-94 May-94 (A) Market Index Returns (B) Dynamic Correlation with World Constant Mean Std Correlation Mean Std Max Min with World 1.388 11.260 0.665 0.665 0.132 0.854 0.316 0.770 6.679 0.600 0.602 0.052 0.706 0.468 0.347 10.469 0.518 0.556 0.189 0.809 0.123 1.049 8.313 0.415 0.388 0.151 0.674 0.052 1.033 8.244 0.575 0.507 0.259 0.846 -0.308 0.992 9.300 0.419 0.360 0.119 0.847 0.165 1.088 10.780 0.695 0.642 0.121 0.858 0.362 0.772 9.202 0.552 0.495 0.192 0.820 0.048 0.835 12.773 0.505 0.520 0.137 0.822 0.184 0.801 11.661 0.605 0.615 0.179 0.868 0.195 0.414 8.722 0.453 0.504 0.132 0.792 0.200 1.073 7.732 0.793 0.783 0.073 0.895 0.506 0.437 5.607 0.281 0.278 0.116 0.636 -0.199 1.112 8.305 0.524 0.518 0.000 0.518 0.518 0.241 9.409 0.466 0.448 0.085 0.619 0.219 0.640 11.288 0.629 0.601 0.185 0.897 0.050 2.188 14.589 0.572 0.589 0.096 0.817 0.403 0.848 7.872 0.709 0.682 0.116 0.867 0.354 0.190 8.558 0.561 0.513 0.118 0.752 0.317 0.373 10.980 0.540 0.515 0.098 0.798 0.171 1.949 15.451 0.521 0.520 0.130 0.722 0.234 28 Table A3: Summary Statistics of International Market Indices (Frontier Markets) Country ARGENTINA BAHRAIN BANGLADESH HERZEGOVINA BOTSWANA BULGARIA CROATIA ESTONIA GHANA JAMAICA JORDAN KENYA KUWAIT LITHUANIA MAURITIUS NIGERIA OMAN PAKISTAN QATAR ROMANIA SLOVENIA SRI LANKA TOBAGO TUNISIA UAE ZIMBABWE (A) Market Index Returns (B) Dynamic Correlation with World Constant Data start Mean Std Correlation Mean Std Max Min with World Nov-10 -5.976 10.196 0.488 0.486 0.000 0.486 0.486 May-02 -0.122 6.596 0.378 0.298 0.057 0.483 0.234 Nov-10 -3.363 12.306 -0.222 -0.139 0.000 -0.139 -0.139 Nov-10 -1.534 7.109 0.470 0.493 0.000 0.493 0.493 Nov-10 -0.299 5.807 0.570 0.605 0.000 0.605 0.605 Nov-10 -3.361 9.550 0.681 0.687 0.000 0.687 0.687 Nov-10 -2.074 5.584 0.618 0.619 0.000 0.619 0.619 Nov-10 -0.977 8.102 0.822 0.816 0.000 0.816 0.816 Nov-10 -2.123 6.254 0.122 0.283 0.000 0.283 0.283 Nov-10 0.751 5.343 0.306 0.167 0.136 0.603 0.052 May-96 0.487 5.592 0.277 0.213 0.171 0.681 0.006 Nov-10 -1.081 7.223 0.639 0.571 0.000 0.571 0.571 May-02 0.594 6.629 0.499 0.394 0.123 0.596 0.216 Nov-10 -0.454 5.470 0.793 0.589 0.008 0.607 0.578 Nov-10 -0.401 3.332 0.383 0.421 0.000 0.421 0.421 Nov-10 -1.164 5.989 0.612 0.628 0.102 0.823 0.399 May-02 0.760 6.250 0.482 0.332 0.169 0.780 0.106 Nov-10 -0.200 5.854 0.381 0.396 0.000 0.396 0.396 May-02 1.626 10.809 0.345 0.268 0.223 0.740 -0.044 Nov-10 -0.630 11.447 0.873 0.868 0.000 0.868 0.868 Nov-10 -2.717 6.222 0.861 0.829 0.000 0.829 0.829 Nov-10 -2.892 4.034 0.109 0.149 0.000 0.149 0.149 Nov-10 1.374 3.500 0.159 0.108 0.000 0.108 0.108 Nov-10 -1.028 5.587 -0.107 -0.004 0.000 -0.004 -0.004 May-02 1.342 11.782 0.386 0.312 0.190 0.721 0.042 Nov-10 -0.082 4.965 0.000 0.005 0.000 0.005 0.005 29