XIV CONGRESO INTERNACIONAL DE LA ACADEMIA DE CIENCIAS ADMINISTRATIVAS A.C. (ACACIA) TÍTULO:

Transcription

1 XIV CONGRESO INTERNACIONAL DE LA ACADEMIA DE CIENCIAS ADMINISTRATIVAS A.C. (ACACIA) TÍTULO: Financial Market Liberalization and the Pricing of Idiosyncratic Risk CAPÍTULO: FINANZAS Y ECONOMÍA AUTORES: Biqing Huang, John Wald, and Rodolfo Martell * Angelo State University University of Texas Barclay s Global Investors Contacto: john.wald@utsa.edu Monterrey Nuevo León, del 7 al 30 de Abril de 010 * Huang is at Angelo State University, Wald is at the University of Texas at San Antonio, and Martell is at Barclay s Global Investors. Please contact john.wald@utsa.edu with questions or comments. The authors wish to thank Onur Bayar, Esther Eiling, Ron Rutherford, Tim Simin, Andrey Ukhov, and seminar participants at Concordia University and at the 009 FMA meetings for comments on earlier drafts.

2 Abstract We test the impact of idiosyncratic risk on stock returns for emerging markets that experienced a financial market liberalization. We expect that idiosyncratic risk should be positively associated with returns prior to financial market liberalization, but that liberalization should diminish this effect. Moreover, prior to liberalization, we expect that the number (concentration) of stocks available in the market should be negatively (positively) correlated with the degree to which idiosyncratic risk is priced. On average, the empirical evidence supports these hypotheses. These findings suggest that the decrease in the cost of capital found around liberalizations can be explained by decreases in the pricing of idiosyncratic risk. Additionally, the degree to which idiosyncratic risk is priced may be useful as a measure of the success of financial market liberalization.

3 Financial Market Liberalization and the Pricing of Idiosyncratic Risk November 10, 009 Abstract We test the impact of idiosyncratic risk on stock returns for emerging markets that experienced a financial market liberalization. We expect that idiosyncratic risk should be positively associated with returns prior to financial market liberalization, but that liberalization should diminish this effect. Moreover, prior to liberalization, we expect that the number (concentration) of stocks available in the market should be negatively (positively) correlated with the degree to which idiosyncratic risk is priced. On average, the empirical evidence supports these hypotheses. These findings suggest that the decrease in the cost of capital found around liberalizations can be explained by decreases in the pricing of idiosyncratic risk. Additionally, the degree to which idiosyncratic risk is priced may be useful as a measure of the success of financial market liberalization. 3

4 I. Introduction If the number of available assets in an economy is small, idiosyncratic risk may be positively priced as investors are unable to fully diversify their risks. We consider emerging markets which have over time introduced (and sometimes reversed) financial market liberalization as a natural experiment for this hypothesis. Thus, we expect that idiosyncratic risk will be priced before liberalization, but that the degree to which such risk is priced should diminish after liberalization. Moreover, we expect the impact of idiosyncratic risk to be largest in markets where the number of available stocks is smallest or the concentration of stocks is highest. Hughes, Liu, and Liu (007) provide a convenient APT-based model which illustrates these effects, and we frame the discussion below using their notation. We test our hypotheses using data before and after financial market liberalization and deliberalization events from 19 emerging market economies. We estimate a stock s idiosyncratic risk as the variance of the residual from a four factor model, where the factors capture local market returns, world market returns, size, and book-to-market (see Rouwenhorst (1999) for an application of Fama-French factors to emerging markets), and we consider several alternative factors for robustness. Overall, our results support our hypotheses. Idiosyncratic risk appears to be positively priced prior to financial market liberalization, but this effect diminishes after liberalization. Moreover, we find that prior to liberalization the 4

5 effect of idiosyncratic risk is, on average, larger for countries with fewer stocks or countries with more concentrated market capitalization. On the other hand, we find limited evidence that deliberalization impacts the degree to which idiosyncratic risk is priced, and this may reflect the smaller sample of deliberalization events. A number of papers address the impact of financial market liberalization on returns, the cost of capital, and investment behavior. For instance, Kim and Singal (000) and Henry (000a) show that financial market liberalization is associated with positive stock returns. Bekaert and Harvey (000) show that financial market liberalization is associated with a decrease in the cost of capital of between 5 and 75 basis points. Our analysis suggests that this decrease in the cost of capital may be explained by a decrease in the pricing of idiosyncratic risk. Bekaert and Harvey (1997) show that financial market liberalization is typically associated with a reduction in market volatility, while Henry (000b) shows that liberalization is associated with an increase in investment. Further, Chari and Henry (004) show that the impact of financial market liberalization on an individual stock depends on the relative covariance between the stock, the local market, and the world market. Bae, Bailey, and Mao (006) provide evidence that the information environment becomes more open following financial market liberalization. A related literature also examines which factors are associated with developing country market segmentation; see for instance Bekaert, Harvey, Lundblad, and Siegel (008) and Carrieri, 5

6 Chaieb, and Errunza (009). A separate literature addresses the impact of idiosyncratic risk on asset returns. Whereas traditional capital asset pricing models argue that only systematic risk factors should be incorporated into asset prices, a number of papers (see Lehmann (1990); Goyal and Santa-Clara (003); Brown and Ferreira (004); Malkiel and Xu (006); and Fu (008)) find a positive relation between idiosyncratic risk and returns. On the other hand, Ang, Hodrick, Xing, and Zhang (006 and 008) find a negative relation between idiosyncratic risk and return, while Wei and Zhang (005), Bali, Cakici, Yan, and Zhang (005), and Jiang and Lee (006) find no robust relation between idiosyncratic risk and returns. We add to the literature on idiosyncratic risk and financial market liberalization by providing a straightforward test of the degree to which idiosyncratic risk is priced in an environment where the financial markets are closed, then open, then, in some cases, closed again. Our basic results largely support the theoretical implications. Moreover, the degree to which idiosyncratic risk is priced could help measure how effectively a particular financial market liberalized. That is, the variation between countries in the degree that financial market openings impact the pricing of idiosyncratic risk suggests that certain liberalizations (and deliberalizations) were more effective than others. Section II reviews the theoretical basis for our tests. Section III describes 6

7 our empirical methods and data. Section IV presents our empirical results, and section V concludes. II. Theoretical Foundations In a general APT model idiosyncratic risk should not be priced in a large economy regardless of whether investors have access to private information. 1 However, for a small closed economy, idiosyncratic risk can be priced under an APT formulation. Hughes, Liu, and Liu (007) provide the following APT formulation of the risk premium for the case with no information asymmetry and a finite number of assets: E v R p f A x (1) F M Where v is an N x 1 vector of the mean of asset payoffs, F is the NxN covariance matrix of the factors, F, is the NxN diagonal covariance matrix of the residuals in the pricing equation, is an N x K constant matrix of factor loadings, R f is the risk-free rate, p is an N x 1 vector of the average price, A is investors absolute risk aversion coefficient, M is the total number of investors, and x is an N x 1 vector of the average supply of risky assets. The first term inside the parentheses refers to the component of returns 1 See also the discussion regarding the pricing of idiosyncratic risk in APT models in Varian (199), p Malkiel and Xu (006) also show that if some investors do not hold the market portfolio, idiosyncratic risk will be priced. With a more stringent set of assumptions, CAPM holds, and under CAPM idiosyncratic risk is not priced regardless of the number of available assets. See Berk (1997) for a formal discussion of the conditions necessary for CAPM to hold. A finding that idiosyncratic risk is priced is consistent with the suggestion that some of the restrictive assumptions necessary for CAPM do not fully hold. 7

8 from idiosyncratic risk, and the second term in the parentheses refers to the component from systematic risks. As the number of available assets increases to infinity, the impact of idiosyncratic risk in the Hughes et al. (007) formulation drops to zero. Given their assumptions, Hughes et al. find similar results even if there is asymmetric information in the economy. That is, regardless of whether investors have asymmetric information, idiosyncratic risk will be priced in an economy with a limited number of assets, but this impact diminishes as the number of assets increases. 3 By examining the pricing of stocks around liberalization and deliberalization events, we test a basic implication visible in the Hughes et al. (007) model. Before liberalization, the local financial market is closed to outside investors, and thus the local assets can only be priced by local investors. Local investors cannot fully diversify the idiosyncratic risk associated with local stocks before liberalization, particularly if the number of available stocks is small. 4 Additionally, if a country s market is dominated by a few large stocks, this greater concentration may limit diversification opportunities even if the total number of stocks is relatively large. After liberalization, foreign investors enter the economy, and we expect 3 More specifically, Hughes, Liu, and Liu (007) show that if investors have information about idiosyncratic risk in large economies, this information should have no impact or expected returns. If investors have information about systematic risk, Hughes et al. show that this information should only impact expected returns by changing loadings on risk factors. On the other hand, for small economies, idiosyncratic risk should be priced regardless of whether there is asymmetric information. 4 In most of these countries, local investors were also not allowed to own foreign assets prior to liberalization. Of the set of countries we consider, foreign investment by local investors is permitted by Mexico, Taiwan, and Venezuela prior to liberalization. 8

9 that the price of local assets to be at least partly determined by the investment choices of these foreigners. As foreign investors have many more investment options, the effective number of assets available to the marginal investor after liberalization is large, and thus the incremental impact of idiosyncratic risk should decline to zero. Similarly, while the number or concentration of assets available in the economy may impact the pricing of idiosyncratic risk prior to liberalization, this relation should not hold once markets are open. Conversely, if the market at some latter time closes to foreign investors (a deliberalization event) we would expect that idiosyncratic risk would again be positively priced. More formally, our first hypothesis addresses the impact of financial market liberalization or deliberalization on the pricing of idiosyncratic risk: H1: Before liberalization, firm specific volatility (idiosyncratic risk) impacts stock returns positively; After liberalization, the impact of idiosyncratic risk is reduced. After deliberalization, the impact of idiosyncratic risk increases. Our second hypothesis addresses the relation between the number or concentration of securities available in the economy and the pricing of idiosyncratic risk. H: Before liberalization, idiosyncratic risk has a larger impact on returns if 9

10 the number (concentration) of securities is smaller (larger). After liberalization, the impact of idiosyncratic risk should be independent of the number or concentration of securities in the local economy. III. Data and Empirical Method III.A. Data We measure the pricing effect of idiosyncratic risk on stock return in 19 emerging markets. We use monthly stock returns and market returns in local currencies from the Standard and Poor's Emerging Markets Database (EMDB) from January 198 to March 003 to test our hypotheses. To avoid the transition period around liberalization, we drop the data six months before and after the liberalization date. We use liberalization and deliberalization dates as detailed by Bae et al. (006) and Bekaert et al. (00). Because of data limitations described below, we are only able to examine liberalizations in 17 countries, and deliberalizations in three countries. We calculate stock returns at time t as: r t log(( Pt Div t ) Pt 1 ), where P t is the stock price at time t, P t 1 is the stock price at time t-1, and Div t is the dividend at time t. We purge the data of observations where either the price or lagged price equals zero. We present results with local currency returns as asset prices should be determined locally prior to liberalization. We also adjust for changes in the local currency (several countries changed their units 10

11 of currency during the period we analyze). For our analysis around liberalization events, we consider only those stocks which were available both before and after the liberalization event, and similarly we limit ourselves to the same stocks before and after deliberalizations. This ensures that our results are not caused by new listings, mergers, or other changes in the types of securities available in the market around these events. 5 We gather local risk free rates from International Financial Statistics and from the Taiwan Central Bank web site. We use the local Treasury bill as the risk-free rate if it is available, otherwise we use other short-term rates as substitutes. Because the local risk-free rate is annualized, we convert it into a monthly rate by using the following equation: r 1 r ) 1. Here, log( f, f, m y r, represent the monthly risk-free rate, and f y f m r, represent the annualized risk-free rate. Local market index returns are available as part of the EMDB data, and we obtain monthly MSCI world market return data from Bloomberg. As the world return data is in U.S. dollars, we convert it into local currency using historical exchange rates. We obtain exchange rates from International Financial Statistics, with the exception of Argentina s exchange rates which are downloaded from Wikipedia. Additionally, we winsorize the return data at the 0.1% level. 6 As an additional robustness check, we consider an analysis based on daily stock return data from EMDB and daily market return data from Bloomberg. 5 The results are stronger if we do not impose this restriction. 6 We also winsorize the upper and lower 0.5% and find similar results. 11

12 The daily data is only available from 1990 for the countries in our sample, and this greatly reduces the sample of liberalization events we can consider. III.B. Estimating Idiosyncratic Risk We estimate idiosyncratic volatility using a procedure similar to that in Spiegel and Wang (005); the idiosyncratic risk equals the variance of the return innovation beyond what investors expect given returns on the market and other factors. Whereas the prior literature typically uses a three factor model with the local market and Fama-French size and book-to-market factors, we include the world market as an additional factor. A number of studies have shown that particularly after liberalization, the world market is an additional factor for these countries. For example, Bailey and Lim (199) and Bekaert and Urias (1996) suggest that after liberalization, correlations between emerging market returns and world market returns increase. Bekaert and Harvey (1997) find increasing influence of world factors on volatility after liberalization consistent with increased market integration, and Patro and Wald (005) find that emerging market firms have increased exposure to the world market and decreased exposure to the home market following liberalization. The idiosyncratic risk we estimate equals the variance of the residual in: R LCMKT WLDMKT Rf R Rf ) ( R Rf ) i, t i, m i, m LC, t i, m WLD, t t ( SMB HML SMB i, m HML i, m i, () 1

13 For each stock i, in market m, at time t, we estimate this model based on a 40 month rolling window. 7 X is the estimated loading on factor X, i, m R i, m, is t the return on stock i in market m at time t, R LC, is the return of the local market m at time t, R WLD, t is the return of the world market m at time t, t Rf is the risk free rate at time t, SMB, is the return on the smallest 1/3 rd minus the m t biggest 1/3 rd capitalization stocks for market m, and HML, is the return on highest 1/3 rd book-to-market minus lowest 1/3 rd book to market stocks for market m. We calculate the idiosyncratic risk for the period immediately following our estimation window ( i, ) as the GARCH(1,1) estimate of the ending period m t variance. 8 That is, we specify the following process for the residual: ~ N (0, hi, m, ) i, t h h (3) i, 0 1 i, 1 i, 1 The last equation in (3) describes the evolution of the conditional variance of i,. As in Spiegel and Wang (005), we use the end-of-period variance ( h i, i, ) as our estimate of the idiosyncratic risk for the next period. 9 7 The results are similar, albeit slightly weaker, with a 30-period window. Additionally, we eliminate windows where half of the stock returns equal zero, as, most likely, these are infrequently traded securities with stale prices. 8 Whereas Spiegel and Wang (005) and Fu (008) use an EGARCH(1,1) model, we use a GARCH(1,1). We use the GARCH model because, for our data, the estimated volatility from GARCH is closer to the realized next-period volatility than that estimated by either the EGARCH or OLS models. Kat and Heynen (1994) also document that GARCH outperforms EGARCH for some financial time series. Using ordinary least squares or EGARCH estimates produces similar, albeit somewhat weaker, results. 9 Using the standard deviation gives similar results. 13

14 III.C. Robustness A possible source of misspecification arises if our basic pricing model (equation ) misses an important factor or includes a useless factor. As the world market factor may not be priced prior to liberalization, we also test whether excluding it impacts our results. We find that excluding the world market factor has little impact on our conclusions. A second possibility is that an exchange rate factor as in Adler and Dumas (1983), should be included in the analysis. We therefore generate an exchange rate factor of unexpected deviations from purchasing power parity equal to: exch t _ log( s t s t 1 ) t bill t rf t, where t s is the spot exchange rate at time t, s t 1 is the spot exchange rate at time t-1, t _ bill t is the U.S. Treasury bill at time t, and rf t is the local risk-free rate at time t. However, as exchange rates for these developing countries were highly volatile over this time period, adding an exchange rate factor adds considerable noise to the analysis. We therefore exclude the exchange rate analysis from the final tables (if we winsorize the data at more than 0.1%, the overall results with the exchange rate factor are similar to those presented). Bekaert, Harvey, and Lundblad (009) consider whether a liquidity factor (see, for instance, Amihud and Mendelson (1986)) is priced in emerging markets. They find that their liquidity measures, either the proportion of stocks with zero returns (ZR) or the predicted price pressure (PP), predicts future returns. We therefore construct these additional factors with our data. 14

15 Our results are strengthened using the ZR factor, and somewhat weakened using the PP factor. We discuss these alternative results in more detail below. III.D. Empirical Method 1 Before liberalization, we expect idiosyncratic risk to be positively priced according to our hypothesis I. Thus, we could consider idiosyncratic risk as a separate factor in addition to the four well-known factors. This corresponds to the impact of idiosyncratic risk suggested by equation (1) from Hughes et al. (007). In order to estimate idiosyncratic risk, we first calculate the abnormal return which is the residual from the four factor model using our rolling regressions. Thus, we estimate equation (), where m, i, t is the abnormal return. We then estimate the average impact of idiosyncratic risk and financial liberalization on abnormal returns with: * * Liberalize * * Liberalize (4) m, i, t 1 m, i, t 1 3 m, i, t The large number of firm-specific coefficients makes estimating all of the betas and the average idiosyncratic volatility effect in one equation infeasible. Thus we use this partial regression approach, where we first control for firm-specific factor effects, then estimate the average impact of idiosyncratic volatility. In this model, the first coefficient on 1 represents the average 15

16 pricing effect of idiosyncratic risk before liberalization. The second coefficient on * m, i, t 1 Liberalize represents the average changes in the pricing effect of idiosyncratic risk after liberalization. We include Liberalize to control for other overall changes in returns following liberalization (see Kim and Singal, 000, Bekaert and Harvey, 000, and others). We expect that before liberalization, idiosyncratic risk ( ) has a positive effect on abnormal returns; after liberalization, this effect should decrease. Note that we estimate equation (4) using our entire sample of firms with observations before and after liberalization, whereas equation () is estimated firm-by-firm. Thus, we estimate individual firm b s, but an average impact of idiosyncratic risk q. In all cases, we provide robust standard errors adjusted for clustering by firm (see Petersen (008) for discussion). Similarly, we consider the impact of deliberalization for the three countries in our sample with deliberalization events. In this case, our estimate of equation (4) applies to firms with observations from before and after deliberalization events, and Liberalize refers to the liberalized (pre-deliberalization) time period. We are also interested in testing whether the number of stocks has any relation with the pricing effect of idiosyncratic risk prior to financial market liberalization. Our second hypothesis states that the pricing effect of idiosyncratic risk diminishes with the number of stocks before liberalization. We therefore use the following model to estimate the effect of the number of 16

17 securities on the pricing effect of idiosyncratic risk: m, i, t * 1 * Liberalize * * NStock 3 * Liberalize 4 * NStock * NStock 5 * * Liberalize 6 * NStock * * Liberalize 7 m, i, t (5) Here NStock, is the total number of stocks (as captured by EMDB) trading in m t market m at time t. 10 We focus on which captures the impact of the 5 number of stocks on the pricing effect of idiosyncratic risk, and on which 6 captures the additional effect of the number of stocks on the pricing effect of idiosyncratic risk after liberalization. Prior to liberalization, we expect that markets with fewer securities would have a larger pricing effect of idiosyncratic risk, and thus should be negative. After liberalization, we expect no 5 relation between the number of stocks in the local market and the impact of idiosyncratic risk, thus we expect to be positive and roughly of the same 6 magnitude as 5. Alternatively, we consider a Herfindahl-Hirschman Index (HHI) to estimate the relation between the concentration of different equities available in the local market and the degree to which idiosyncratic risk is priced. That is, if there are only a few large stocks in the market, this concentration may again provide reduced diversification opportunities. We calculate the HHI of a market at time t as: HHI, i m t m / M ) (, where i m m, is the market t 10 In unreported regressions, we also consider the log of the number of stocks, and this transformation somewhat weakens the results. 17

18 capitalization of firm i in market m at time t, and M, is the total market m t capitalization (see, for instance, Carlton and Perloff (005)). We use the following model to estimate the relation between HHI, idiosyncratic risk, and abnormal returns: m, i, t * 1 * Liberalize * * HHI 3 * Liberalize 4 * HHI * HHI 5 * * Liberalize 6 * HHI * * Liberalize 7 m, i, t (6) We again focus on which captures the impact of the market concentration 5 on the pricing effect of idiosyncratic risk, and on 6 which captures the additional effect of the market concentration on the pricing effect of idiosyncratic risk after liberalization. Prior to liberalization, more concentrated markets will have a higher HHI, and we expect idiosyncratic risk to have a larger impact in these markets. Thus we expect to be positive. After 5 liberalization, we expect no relation between the number of stocks (also HHI) in the local market and the impact of idiosyncratic risk; thus we expect to 6 be negative and roughly of the same magnitude as 5. III.E. Empirical Method We also examine the degree to which idiosyncratic risk is priced in returns using a factor loading method. We follow Ang et al. (006 and 008) in using Fama and MacBeth (1973) regressions. Thus, for each stock, we regress the firm s stock returns on idiosyncratic volatility along with the factor loadings from 18

19 the previous 40 months. We repeat this regression using before and after liberalization data to estimate two sets of parameters. The Fama-MacBeth cross-sectional regressions take the form: x (7) R i, c i, m m, i, t 1 i, m i, i, where is the idiosyncratic volatility estimated up to the previous month, i, 1 is a vector of factor loadings estimated over months t-39 to t, and i, is x i, the residual. 11 We focus on explaining the coefficients i, m, which estimate the degree to which idiosyncratic volatility is priced for each stock before and after liberalization or deliberalization. We formally test (adjusting for clustering of the estimates by firm) whether average g differs before versus after liberalization or deliberalization events. We are also interested in testing whether γ can be explained by the number of stocks available in the local market prior to financial market liberalization. Our second hypothesis states that the number of stocks should be negatively related to γ. We therefore estimate: i, m * a b1 NStock m b Liberalize NStock b3 Liberalize i, m (8) The first coefficient, b1 on NStock,, represents the average impact of the m t number of stocks on the pricing effect of idiosyncratic risk before liberalization. The second coefficient, b on Liberalize NStock, *, represents the additional m t impact of the number of stocks on the pricing effect of idiosyncratic risk after liberalization. We include Liberalize by itself to control for other changes in 11 We also estimate the model using lag betas. The results are similar, albeit slightly weaker. 19

20 returns following liberalization. Before liberalization, we expect countries with fewer securities to have a higher impact of idiosyncratic risk. After liberalization, we expect no relation between the number of stocks in the local market and the impact of idiosyncratic risk, thus we expect b to be negative and roughly of the same magnitude as b 1. As above, we consider market concentration rather than the raw number of securities on the pricing of idiosyncratic risk. Thus we estimate: i, m * (9) a b1 HHI b Liberalize HHI b3 Liberalize i, m The first coefficient, b 1 on HHI,, represents the average impact of m t market concentration on the pricing effect of idiosyncratic risk before liberalization. The second coefficient, b on Liberalize HHI, *, represents m t the additional impact of market capitalization on the pricing effect of idiosyncratic risk after liberalization. Before liberalization, we expect countries with more concentrated markets to offer fewer diversification opportunities and thus to have a higher impact of idiosyncratic risk. Thus we expect b 1 to be positive before liberalization. After liberalization, we expect no relation between the number of stocks in the local market and the impact of idiosyncratic risk, thus we expect b to be negative and roughly of the same magnitude as b 1. IV. Empirical Results Table I presents the sequence of liberalizations and deliberalizations from 0

21 Bae et al. (006) and Bekaert et al. (00); see Bekaert et al (00) for a detailed description of these events. There are a total of 19 countries in our data with financial market liberalizations, and we use the Official Liberalization event dates for our liberalization analysis except for China and Poland, for which we use the cross-listing or country fund listing date as the liberalization date. Although Malaysia and Turkey have liberalization dates, the EMDB data does not have sufficient pre-liberalization data for our analysis, thus we are left with only 17 liberalization events. The three countries with deliberalization events for which we have sufficient data are Malaysia, Thailand, and Turkey. 1 Table II presents summary statistics on the variables used in our analysis. Panel A includes the 17 countries with data before and after financial market liberalization, and there are 833 firms with data available both before and after liberalization. Panel B includes those three countries that also had a deliberalization, and there are 41 firms with data available both before and after deliberalization. In Panel A of Table II, we report that the means of all returns, SMB, HML, idiosyncratic risk, and HHI decrease after liberalization, whereas the number of stocks increases. After deliberalization, except for world return, excess world return, and the number of stocks, the means of these variables also decreases as reported in Panel B of Table II. The 1 For the daily data analysis, we only have observations around liberalizations for three countries: Chile, Korea, and Philippines. For the daily deliberalization analysis, we have data for Turkey, Argentina, Chile, Malaysia, and Thailand. 1

22 number of stocks in the local economy typically increases over time, thus there are more stocks after liberalization and after deliberalization. Note that Bae et al. (006) find that idiosyncratic risk increases after financial market liberalization, whereas we find that idiosyncratic risk for the same stocks declines. If we instead consider the idiosyncratic risk for all stocks, not just those with data available both before and after the event, and exclude data from the same range around liberalizations as Bae et al., we also find that average idiosyncratic risk increases after liberalizations. IV.A. Method I Results Table III shows the results from using method 1 based on equation (4) to measure the effect of idiosyncratic risk on returns. The estimated coefficients for equation (4), the number of observations, the number of firms available with before and after liberalization data, and the adjusted R-squared are reported. Column (1) estimates equation (4) with data before and after liberalization for the 17 countries with available data. Column () estimates equation (4) with data before and after deliberalization for the three countries with available data. The results support Hypothesis I; in column (1), the significantly positive coefficient on shows that before liberalization greater idiosyncratic risk implies significantly higher excess stock returns. The significant negative coefficient on Liberalize shows that after liberalization, the pricing effect * of idiosyncratic risk decreases. Note that the magnitudes of these effects are

23 similar, thus we observe a significant impact of idiosyncratic risk prior to liberalization but little effect after. In column () of Table III, we examine the impact of idiosyncratic risk on returns around deliberalizations. We find no evidence that idiosyncratic risk increases after deliberalization. In terms of economic impact, the estimated coefficient on idiosyncratic risk in the first column of Table III (0.089) implies a sizeable change in expected returns for a given change in risk. Specifically, a one standard deviation change in idiosyncratic risk (0.097) would correspond to an average change of 86 basis points per month, or a 10.9% change in annualized returns. We can also interpret the change in the cost of capital from the decrease in the pricing of idiosyncratic risk. Specifically, the average idiosyncratic risk before liberalization is 0.03, and this suggests that the average cost of capital would decrease by 8 basis points per month (0.03 times 0.089) on average. This change in the cost of capital is consistent with the decrease found by Bekaert and Harvey (000) of between 5 and 75 basis points per month around liberalization events. We next estimate the results from Equation (5), which includes interactions between the number of stocks, the idiosyncratic risk, and whether the market has liberalized. However, as the estimated coefficients on these interactions are not significant, we do not include them in a separate table. We estimate the results from equation (6) in Table IV. As expected, we find a positive coefficient on HHI before liberalization, and this * 3

24 coefficient is significant at the 10% level. Markets with greater concentration (higher HHI) appear to be more affected by idiosyncratic risk than those with lower concentration. We find a negative coefficient on Liberalize, and this suggests that after liberalization, the degree * HHI * to which market concentration impacts the pricing effect of idiosyncratic risk decreases. Column () of Table IV provides the estimates around deliberalization events; however, the estimated coefficients are not significantly different from zero for the smaller deliberalization sample. Thus, supporting our hypothesis II, we observe some evidence of a positive relation between market concentration and idiosyncratic risk prior to liberalization, but little after liberalization. As a further robustness test, we rerun the basic analysis given in Table III for the three countries with available daily return data around liberalizations and the five countries with daily data around deliberalizations. However, for this small sample of countries, we find no significant results using method I. We also redo our analysis using either the ZR or PP liquidity factors described in Bekaert et al. (009). Including ZR as an additional factor strengthens our results, whereas including PP weakens it; the results with these additional factor are presented in the first appendix table, A.1. IV.B. Method II Results Table V presents our estimated γ coefficients from equation (7) where we 4

25 regress returns on estimated variance and other factor loadings. The γ coefficients measure the degree to which idiosyncratic risk is priced in the market. We test the differences in γ by regressing the estimated parameters on a dummy variable equal to one if the observation occurs after liberalization and adjusting for clustering by firm. This firm-level clustering adjustment is critical as the γ estimates are autocorrelated a simple t-test produces p-values which are several orders of magnitude smaller. Comparing before and after liberalization for the 17 countries, we see that γ Before Lib is positive and significantly different from the γ After Lib, indicating that the pricing effect of idiosyncratic risk estimated from equation (7) decreases significantly after liberalization. Moreover, the average coefficient on idiosyncratic risk of 1.77 before liberalization is much larger than the 0.13 coefficient estimated by Fu (008) in similar regressions for the U.S. Thus, our estimated impact of idiosyncratic risk in the pre-liberalization developing markets is economically much more significant than that found in the U.S. 13 Again, these results support our hypothesis I for the 17 countries with data before versus after liberalization. 14 In the after liberalization versus after deliberalization sample presented in Table V for the three countries with deliberalization events, the estimated γ 13 We cannot reject the alternative hypothesis that after liberalization the coefficient on idiosyncratic risk in these markets equals the 0.13 estimated coefficient in Fu (008). Thus, there may still be a small positive impact of idiosyncratic risk after financial market liberalization which our data is insufficient to precisely detect. 14 Whereas our first method provides an estimate of the decrease in the cost of capital consistent with Bekaert and Harvey (000), this second method suggests a decrease larger than their estimate. 5

26 coefficients become significantly positive after deliberalization, again consistent with the notion that closed markets should have larger impacts of idiosyncratic risk. Table VI presents our estimated coefficients from equation (8) on the relation between the number of stocks in the local market and the degree to which idiosyncratic risk is priced. Column (1) estimates equation (8) with data before and after liberalization for the 17 countries. Column () estimates equation (8) with data before and after deliberalization for the three countries with such events. In column (1) the estimated coefficients have the expected signs. Before liberalization, the number of stocks has a negative relation with the pricing effect of idiosyncratic risk; after liberalization, the coefficient on the number of stocks is positive, and these are significant at the 10% level. These results suggest that before liberalization, idiosyncratic risk has a larger impact on returns if the number of securities is smaller. However, as expected, financial market liberalization decreases the effect of the number of securities on the pricing of idiosyncratic risk. Note that the magnitudes of these effects are almost identical. Thus we observe that the number of securities has a significant negative effect on the pricing of idiosyncratic risk prior to liberalization, but effectively none after. In column () of Table VI we provide estimated coefficients from before and after deliberalization for the three countries with such events. The relation between the number of stocks and the pricing effect of idiosyncratic risk is not 6

27 statistically significant for this small sample of deliberalizations. In Table VII, we estimate the effect of local market concentration on the pricing of idiosyncratic risk using equation (9). Column (1) estimates equation (9) with data before and after liberalization for the 17 countries. Column () estimates equation (9) with data before and after deliberalization for the three countries with such events. In column (1) the estimated coefficients have the expected signs; before liberalization, HHI has a significantly positive relation with the pricing effect of idiosyncratic risk; after liberalization, the additional effect of HHI is negative. These coefficients are significant (at the 1% level and 5% levels, respectively) again demonstrating that our market concentration measure, HHI, appears to capture this effect more than just the raw number of stocks. These results demonstrate that before liberalization idiosyncratic risk has a larger impact on returns if the market concentration is higher. Further, financial market liberalization decreases the effect of market concentration on the pricing of idiosyncratic risk. Again, the magnitudes of these effects are almost identical; thus we observe a significant negative effect of HHI on the pricing of idiosyncratic risk prior to liberalization, but effectively none after. These results around liberalizations further support our hypothesis II. The deliberalization results are again not significant. As a robustness test, we also consider this analysis using daily data. Similar to our results in Table V, we find a significant decrease in the pricing of idiosyncratic risk after liberalizations for the three countries with available daily 7

28 data. Examining deliberalizations, we find that the three countries we analyze with monthly data (Malaysia, Thailand, and Turkey) also have a significant increase in the pricing of idiosyncratic risk using daily data. However, using daily data there was no evidence of an increase in the pricing of idiosyncratic risk for the other two countries with deliberalizations (Argentina and Chile). For robustness, we also consider the analysis using either the ZR or PP liquidity factors. Our basic finding, that the pricing of idiosyncratic risk is reduced after financial market liberalization holds using either of these measures, and the statistical significance increases when using these additional factors. These results are presented in the appendix table A.. IV. C. By-Country Estimation Tables VIII and IX provide estimates of equations (4) and (7), our methods I and II, on a by-country basis. As with prior studies that examine the impact of financial market liberalization, the effects are somewhat mixed. Nevertheless, 13 out of the 17 countries considered in Table VIII have negative coefficients on the interaction between Liberalize and idiosyncratic risk, and six of these negative coefficients are significant. The six countries with significantly negative coefficients are Argentina, Jordan, Korea, Mexico, Nigeria, and Pakistan. On the other hand, only ten out of the 17 countries in Table IX have decreasing g, and these changes are only significant for two countries, Argentina and Nigeria. Thus, whereas both methods provide 8

29 significant evidence supporting the decrease in the impact of idiosyncratic risk on average, the results by-country are more varied. We also consider the deliberalizations on a by-country basis; however, these are generally not significant. If a financial market liberalization is successful in allowing foreign portfolio investment, we would expect to see a decrease in the pricing of idiosyncratic risk as well as increases in returns during liberalization. Figure 1 graphs the relation between the returns during liberalization and the change in the pricing of idiosyncratic risk. On average, we find that greater decreases in the pricing of idiosyncratic risk are more likely to occur in countries which had higher returns during liberalizations, suggesting that some liberalizations (such as Argentina or Nigeria) were more successful than others (such as India or Poland). However, this cross-country relation between returns during liberalization and changes in the pricing of idiosyncratic risk is not statistically significant. V. Conclusion We test whether idiosyncratic risk is priced in emerging markets before and after financial market liberalization, and before and after deliberalization. We consider two methods one treats idiosyncratic risk as an additional factor and the other uses Fama-MacBeth factor loading regressions. Using both methods, we find a significant positive impact of idiosyncratic risk on returns 9

30 before liberalization, and this impact is an order of magnitude larger than that found by Fu (008) for the U.S. The degree to which idiosyncratic risk is priced decreases significantly after financial market liberalization. This decrease in the pricing of idiosyncratic risk can explain the reduction in the cost of capital found by Bekaert and Harvey (000) around financial market liberalizations. We also find some evidence that the pricing of idiosyncratic risk increases after deliberalizations using our factor loadings methodology, and this more limited result may reflect the relatively small sample (three countries) of deliberalizations available for this analysis. We test whether the number of stocks or the concentration of stocks available in the economy impacts how idiosyncratic risk is priced prior to liberalizations. The factor loading regressions provide some evidence that before liberalization, idiosyncratic risk has a greater impact on returns in markets with fewer available stocks. The results with market concentration are somewhat stronger; both empirical methods indicate that idiosyncratic risk is priced more in closed markets which are also more concentrated, and that this effect dissipates after financial market liberalization. Overall, the impact of idiosyncratic risk on returns agrees with the theoretical implications for small economies that open their financial markets. Additionally, this analysis raises the possibility that changes in how idiosyncratic risk is priced could be used to measure the effectiveness of financial market liberalizations. 30

31 Reference Alder, Michael. and Bernard Dumas, 1983, International Portfolio Choice and Corporate Finance: A Synthesis, The Journal of Finance 38, Amihud, Y. and H. Mendelson, 1986, Asset Pricing and the Bid-Ask Spread, Journal of Financial Economics 17, Ang, Andrew., Robert. J. Hodrick, Yuhang Xing, and Xiaoyan Zhang, 006, The Cross-Section of Volatility and Expected Returns, The Journal of Finance 61, Ang, Andrew., Robert. J. Hodrick, Yuhang Xing, and Xiaoyan Zhang, 008, High Idiosyncratic Volatility and Low Returns: International and Further U.S. Evidence, Journal of Financial Economics, forthcoming. Bae, Kee-Hong., Warren Bailey, and Connie X. Mao, 006, Stock Market Liberalization and the Information Environment, Journal of International Money and Finance 5, 1-5. Bailey, Warren., and Joseph Lim, 199, Evaluating the Diversification Benefits of the New Country Funds, Journal of Portfolio Management 18, Bali, Turan G., Nusret Cakici, Xuemin Yan, and Zhe Zhang, 005, Does Idiosyncratic Risk Really Matter? The Journal of Finance 60, Bekaert, Geert., and Campbell R. Harvey, 1997, Emerging Equity Market Volatility, Journal of Financial Economics 43, Bekaert, Geert., and Campbell R. Harvey, 000, Foreign Speculators and Emerging Equity Markets, The Journal of Finance 55, Bekaert, Geert., Campbell R. Harvey, and Robin L. Lumsdaine, 00, Dating 31

32 the Integration of World Equity Markets, Journal of Financial Economics 65, Bekaert, Geert., Campbell R. Harvey, and Christian Lundblad, 009, Liquidity and Expected Return: Lessons from Emerging Markets, Review of Financial Studies, forthcoming. Bekaert, Geert., Campbell R. Harvey, Christian Lundblad, and Stephan Siegel, 008, What Segments Equity Markets? NBER Working Paper. Bekaert, Geert., and Michael S. Urias, 1996, Diversification, Integration, and Emerging Market Closed-end Funds, The Journal of Finance 51, Berk, Jonathan B., 1997, Necessary conditions for the CAPM, Journal of Economic Theory 73, Brown, David., and Miguel A. Ferreira, 004, Information in the Idiosyncratic Volatility of Small Firms. University of Wisconsin Working Paper. Carlton, Dennis W., and Jeffrey M. Perloff, 005, Modern Industrial Organization. 4th ed. New York: Addison Wesley. Carrieri, Francesca, Ines Chaieb, and Vihang Errunza, 009, Do Implicit Barriers Matter for Investibility and Globalization. McGill University Working Paper. Chari, Anusha., and Peter Blair Henry, 004, Risk Sharing and Asset Prices: Evidence from a Natural Experiment, The Journal of Finance 59, Fama, Eugene F., and Kenneth R. French, 199, The Cross-Section of 3

33 Expected Stock Returns, The Journal of Finance 47, Fama, Eugene F., and James D. MacBeth, 1973, Risk, Return, and Equilibrium: Empirical Tests, Journal of Political Economy 71, Fu, Fangjian, 008, Idiosyncratic Risk and the Cross-Section of Expected Stock Returns, Journal of Financial Economics, forthcoming. Goyal, Amit., and Pedro Santa-Clara, 003, Idiosyncratic Risk Matters! The Journal of Finance 3, Henry, Peter Blair., 000a, Stock Market Liberalization, Economic Reform, and Emerging Market Equity Prices, The Journal of Finance 55, Henry, Peter Blair., 000b, Do Stock Market Liberalizations Cause Investment Booms? Journal of Financial Economics 58, Heynen, Ronald C., and Harry M. Kat, 1994, Volatility Prediction: A Comparison of the Stochastic Volatility, GARCH(1,1), and EGARCH(1,1) Models, The Journal of Derivatives, Hughes, John S., Jing Liu, and Jun Liu, 007, Information Asymmetry, Diversification, and Cost of Capital, The Accounting Review 8, Jiang, Xiaoquan., and Bong-Soo Lee, 006, The Dynamic Relation between Returns and Idiosyncratic Volatility, Financial Management 35, Kim, E Han., and Vijay Singal, 000, Stock Market Openings: Experience of Emerging Economies, The Journal of Business 73, Lehmann, Bruce N, 1990, Fads, Martingales, and Market Efficiency, Quarterly Journal of Economics 105,

34 Malkiel, Burton G. and Yexiao Xu, 006, Idiosyncratic Risk and Security Returns, Princeton University Working Paper. Petersen, Mitchell A. 008, Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches, Review of Financial Studies, forthcoming. Patro, Kilip K., and John K Wald, 005, Firm Characteristics and the Impact of Emerging Market Liberalizations, Journal of Banking & Finance 9, Rouwenhorst, K. Geert., 1999, Local Return Factors and Turnover in Emerging Stock Markets, The Journal of Finance 54, Spiegel, Matthew I., and Xiaotong Wang, 005, Cross-sectional Variation in Stock Returns: Liquidity and Idiosyncratic Risk, Yale University Working Paper. Varian, Hal, 199, Microeconomic Analysis. W.W. Norton & Co., New York. Wei, Steven X., and Chu Zhang, 005, Idiosyncratic Risk Does Not Matter: A Re-examination of the Relationship between Average Returns and Average Volatilities, Journal of Banking & Finance 9,