Sunteți pe pagina 1din 18

International Business Review 12 (2003) 109126 www.elsevier.

com/locate/ibusrev

The conditional relationship between beta and returns: recent evidence from international stock markets
Gordon Y.N. Tang a,b,, Wai C. Shum b
a

International Graduate School of Management, University of South Australia, Adelaide SA 5001, Australia b Department of Finance and Decision Sciences, Hong Kong Baptist University, Kowloon Tong, Kowloon, Hong Kong Received 11 April 2001; received in revised form 30 October 2001; accepted 12 July 2002

Abstract The riskreturn relationship is one of the fundamental concepts in nance that is most important to investors and portfolio managers. Finance theory argues that the beta or systematic risk is the only relevant risk measure for investors. However, many studies have showed that betas and returns are not related empirically, no matter in domestic markets or in international stock markets. This paper examines the conditional relationship between beta and returns in international stock markets for the period from January 1991 to December 2000. After recognizing the fact that while expected returns are always positive, realized returns could be positive or negative, we nd a signicant positive relationship between beta and returns in up market periods (positive market excess returns) but a signicant negative relationship in down market periods (negative market excess returns). The results are robust for both monthly and weekly returns and for two different proxies of the world market portfolio. Our ndings indicate that beta is still a useful risk measure for portfolio managers in making optimal investment decisions. 2003 Elsevier Science Ltd. All rights reserved.
Keywords: Conditional relationship; Beta and returns; International markets

Corresponding author. Tel.: +852-3411-5585; fax: +852-3411-7563. E-mail address: gyntang@hkbu.edu.hk (G.Y.N. Tang).

0969-5931/03/$ - see front matter 2003 Elsevier Science Ltd. All rights reserved. doi:10.1016/S0969-5931(02)00090-2

110

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

1. Introduction

The riskreturn relationship is one of the fundamental concepts in nance that is most important to investors and portfolio managers. One of their tasks is to estimate the investment risk. The famous Capital Asset Pricing Model, CAPM (Black, 1972; Lintner, 1965; Sharpe, 1964) argues that beta or systematic risk is the only relevant risk measure for investment and a positive trade-off between beta and expected returns should exist. Because of its importance and relevance to all investors, it is one of the most extensively tested nancial models in the literature. The CAPM states that the expected return of an asset is a positive function of three variables: beta (the covariance of asset returns and market returns divided by the variance of the market returns), the risk-free rate and the expected market return. The major focus of the tests on CAPM is to check whether returns are statistically positively related to betas. Since in reality only realized asset/market returns are available, average realized returns are used to proxy the expected returns and returns on security index are used to proxy the market returns. Empirical tests in 1970s (e.g. Fama & MacBeth, 1973) support the validity of the CAPM. However, empirical evidence in 1990s (e.g. Fama and French, 1992, 1996; Jegadeesh, 1992) indicates that betas are not statistically related to returns, and so some researchers conclude that beta is dead and suspect the validity of beta in measuring risk. Previous empirical tests are mainly based on the Fama and MacBeth (1973) methodology. However Isakov (1999) argued that this methodology does not leave much opportunity for beta to appear as a useful risk measure in the riskreturn relationship for two reasons. First, the model is expressed in terms of expected returns; however, tests can only be performed on realized returns. Secondly, the realized market excess return does not behave as expected since it is too volatile and is often negative. Recent studies (e.g. Pettengill, Sundaram, & Mathur, 1995 for the US market; and Isakov, 1999 for the Swiss market) suggested an alternative approach to assess the reliability of beta as a measure of risk. The alternative approach is that when the realized market returns exceed the risk-free rate of interest (i.e. the realized market excess returns are positive), the betas and realized excess returns should have a positive relationship. Similarly, when the realized market excess returns are negative, the betas and realized excess returns should have a negative relationship. This paper extends this approach and applies it to 13 international stock markets. The results show that beta is still a good measure of risk and is signicantly related to realized returns in both up and down market situations. Hence, beta is still a useful measure of risk for investors in making optimal investment decisions. The rest of the paper is organized as follows: Section 2 presents the theoretical framework used in this paper. Section 3 provides a literature review of the relationship between beta and returns. Section 4 describes the data and methodologies used. Section 5 reports our empirical results and Section 6 concludes the paper.

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

111

2. Theoretical framework The CAPM states a simple positive relationship between asset expected return and its beta that is widely tested in domestic markets. However, the international equilibrium riskreturn relationship is more complex. The expected return of an international stock market is not only a function of its covariance with the world market portfolio, but, in the presence of exchange rate risks, additional risk premiums must be added to the riskreturn relation to reect the covariance of the international market returns with the various exchanges rates (the currencies betas). If there are k 1 international stock markets, there will be k additional currency risk premiums. Hence, Solnik (2000, p. 166) stated that the expected return of an international stock market is the sum of the market risk premium plus various currency risk premiums E(Ri) Rf biw RPw gi1 RP1 gi2 RP2 gik RPk, (1) where Rf is the risk-free interest rate, biw the sensitivity of stock market, i, to the world market movements, RPw the world market risk premium equals to E(Rw) Rf, gi1 to gik the sensitivity of stock market, i, to the currencies 1 to k, and RP1 to RPk are the risk premiums on currencies 1 to k. As the CAPM assumes a positive riskreturn trade-off, the expected market return must be greater than the risk-free rate of return. Since the expected market risk premium (i.e. [E(Rm) Rf]) must be positive, the expected return to any risky portfolio is a positive function of beta. The previous tests on the validity of CAPM are based on this positive relationship between beta and returns. The same argument can also be applied to the international CAPM for the relationship between market betas and the expected individual market returns. However in reality, the realized market return can be lower than the risk-free rate, so the previous test results of the CAPM should be modied. On an average, the realized market return should be greater than the risk-free rate and investors realize that there is a non-zero probability that the market return will be less than the riskfree rate. If the realized market return must be greater than the risk-free rate, no investor will hold the risk-free asset. In addition, as beta is a risk measure, an asset with a high beta should have a higher risk than an asset with a low beta. Since realized return can be favorable or unfavorable, the returns between high-beta and low-beta assets should be different. If the realized return is favorable, high-risk asset (i.e. high-beta asset) should have a higher return than low-risk asset (i.e. low-beta asset) and vice versa. Isakov (1999) argued that this condition could be recognized by the sign of the realized market excess return. A favorable outcome exists if market risk premium is positive and an unfavorable outcome occurs if market risk premium is negative. This means that when the realized market excess return is positive, highbeta asset should have a larger return than low-beta asset and similarly when the realized market excess return is negative, high-beta asset should have a smaller return than low-beta asset. If this result can be observed in the data, then beta can be regarded as a useful measure of risk. The above discussion presents a segmented relationship between beta and realized returns: a positive relation in up market (positive risk premium) periods and a nega-

112

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

tive relation in down market (negative risk premium) periods. If the realized market returns are seldom less than the risk-free rate, this conditional relationship between beta and returns will only have a minor impact on testing the relationship between beta and returns. However, Isakov (1999) and Pettengill, Sundaram, and Mathur (1995) showed that this conditional relationship occurs frequently, indicating previous empirical studies that tested the unconditional relationship between beta and returns are biased. This paper tests the segmented relationship between beta and returns in international stock markets and nds a signicant impact of beta on realized returns.

3. Literature review Previous literature examined the validity of CAPM by testing the existence of a positive linear relationship between beta and returns. Although the model postulates a positive trade-off between beta (risk) and expected returns, researchers, in general, always found a weak but positive relationship between beta and returns over the sample period. Hence, they claimed that the results are inconsistent with the positive linear relationship between beta and returns as prescribed by the CAPM and the validity of the CAPM is in question. The ndings of these major studies are as follows. Fama and MacBeth (1973) tested the validity of the CAPM using a three-step approach. In the rst period, individual stocks betas are estimated and portfolios are formed according to these estimated betas. In the second period, betas of portfolios that are formed in the rst period are estimated. In the nal step, using data from a third time period, portfolio returns are regressed on portfolio betas (obtained from the second period) to test the relationship between beta and returns. They found a signicant average market excess return of 1.30% per month and on an average, for the period 1935 through 1968, a positive relationship exists between beta and monthly returns. They concluded that the results support the CAPM in the US stock market. However, Schwert (1983) suggested that Fama and MacBeth (1973) only provided a very weak support for a positive riskreturn trade-off since the positive riskreturn relationship found is not signicant across sub-periods. Furthermore, when considering the seasonal behavior of their results, the t-statistic for the 19351968 sample period becomes highly suspect and the basic riskreturn trade-off virtually disappears. Reinganum (1981) found that the cross-sectional differences in portfolio betas and the differences in average portfolio returns are not reliably related, i.e. the returns on high-beta portfolios are not signicantly higher than the returns on low-beta portfolios, casting doubts on the empirical content of CAPM. Tinic and West (1984) found that January has a larger risk premium than the other months and further that the signicant relationship between risk and expected returns only exists in January. When data for the January months are excluded from the analysis of the riskreturn trade-off, the estimates of risk premiums are not signicantly different from zero. Thus, they concluded that their results reject the validity

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

113

of CAPM. Lakonishok and Shapiro (1986) examined the monthly returns of all stocks traded on the New York Stock Exchange (NYSE) and found that return on individual security is not specically related to its degree of systematic risk, but is signicantly related to the market capitalization values. They concluded that the traditional (beta) as well as the alternative (residual standard deviation) risk measure is not able to explain the cross-sectional variation in return; only size can signicantly explain it. Haugen and Baker (1991) examined the risk and return characteristics of 1000 US stocks that have largest market capitalization over all US exchanges and markets between 1972 and 1989. They found that the market portfolio is not efcient because low-risk stocks seem to have abnormally high returns, contradicting the relationship between beta and returns as prescribed by CAPM. Fama and French (1992) studied the monthly returns of NYSE stocks and found an insignicant relationship between beta and average returns. They concluded that the CAPM cannot describe the last 50 years of average stock returns and only market capitalization and the ratio of book value to market value have signicant explanatory power for portfolio returns. The previous ndings are against beta as a useful measure of risk. However, Pettengill et al. (1995) developed a conditional relationship between beta and realized returns by separating periods of positive and negative market excess returns. Using US stock market data in the period 1936 through 1990, they found a signicant positive relationship between beta and realized returns when market excess returns are positive and a signicant negative relationship between beta and realized returns when market excess returns are negative. This signicant relationship is also found when data are divided by months in a year. Furthermore, they found support for a positive riskreturn relationship. Isakov (1999) followed the approach of Pettengill et al. (1995) and examined the Swiss stock market for the period 19831991. He found supporting results that beta is statistically signicant related to realized returns and has the expected sign. Hence, Isakov (1999) concluded that beta is a good measure of risk and is still alive. In summary, previous empirical studies on the unconditional relationship between beta and returns found that the CAPM only provides an inadequate explanatory power for the riskreturn relationship observed in both domestic and international stock markets. However, results from empirical studies on the conditional relationship between beta and returns support the model and found a signicant conditional relationship in domestic stock markets. A logical question followed is whether the conditional relationship between beta and returns can also be applied to international stock markets. To the best of our knowledge, no study (except one) has investigated this issue. Fletcher (2000) examined the conditional relationship between beta and returns in international stock markets between January 1970 and July 1998 using the approach of Pettengill et al. (1995). Using monthly returns of the MSCI equity indices of 18 countries and the MSCI world index, Fletcher (2000) found that a consistent result exists. There is a signicant positive relationship between beta and returns in periods when the world market excess returns are positive and a signicant negative relationship in periods when the world market excess returns are negative. Besides, this relationship is symmetric and there is a positive mean excess return on the world

114

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

index on an average. Fletcher (2000) also found that the signicant conditional relationship in January exists only in periods of positive market excess returns and the relationship is insignicant in periods of negative market excess returns. The results differ from those obtained from Pettengill et al. (1995) on the US market data. Although Fletchers (2000) study supported the conditional relationship between beta and returns in international stock markets, the empirical evidence is still limited. This paper aims to examine the conditional relationship between beta and returns using more recent data of international stock markets. Furthermore, our paper differs from Fletchers paper in at least six areas. First, the period covered in this study is from January 1991 to December 2000, representing a more recent data set compared with Fletchers study. Secondly, we use raw country indices in domestic currencies instead of MSCI country indices in US dollars to examine the conditional relationship of riskreturn trade-off. Hence, our study separates the exchange rate movements from the stock market returns. Thirdly, we use a value-weighted world index (i.e. MSCI world index) as well as an equally weighted world index, separately, to proxy the world market index and examine the results between these two indices. Fletcher (2000) used only MSCI world index as the world market index. Fourthly, we use both monthly and weekly returns to examine the conditional relationship between beta and returns while Fletcher used only monthly returns. Hence, our study can be regarded as more comprehensive. Fifthly, we use the model of international CAPM in nding betas of international stock markets. This model deals with exchange rate risks explicitly by adding additional risk premiums to the riskreturn relationship, and so no assumption is required. Fletcher (2000) used the model of domestic CAPM in an international setting, and so additional assumptions (e.g. capital markets are integrated and purchasing power parity holds) are required. Our model can be regarded as less restrictive. Sixthly, Fletcher (2000) used the whole sample period in estimating the betas of each country, which is different from Fama and MacBeths (1973) method. It needs the strong assumption that betas are constant over the whole sample period. Also, the method does not link the historical betas to expected returns (proxy by the next period return). In this paper, we follow the Fama and MacBeth methodology in examining the conditional relationship between beta ands returns. Our results should provide more reliable empirical evidence and help portfolio managers in dealing with investment decisions.

4. Data and methodology 4.1. Data The data collected from DataStream consist of monthly and weekly index returns of 13 countries, the Morgan Stanley Capital International (MSCI) world index (a value-weighted index) and an equally weighted world index. Our sample period covers from January 1991 to December 2000, producing a total of 120 data points

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

115

of monthly returns and 522 data points of weekly returns. The 13 countries are France, Germany, Netherlands, UK, Japan, Canada, and US (the G7), Belgium, Denmark, and Switzerland (three European countries), Hong Kong, Singapore and Taiwan (three Asian countries with open stock markets). Information of the sample countries is shown below.

Summary of data used Country France Germany Netherlands UK Japan Canada US Belgium Denmark Switzerland Hong Kong Singapore Taiwan Country index CAC 40 Country risk-free rate

1-Month interbank offered rate DAX KURS 1-Month interbank offered rate AEX 1-Month interbank offered rate FTSE 100 1-Month interbank offered rate Nikkei 225 stock average 1-Month interbank offered rate Toronto SE 300 composite 1-Month Canada T-bills S&P 500 3-Month US T-bills BEL 20 1-Month interbank offered rate Copenhagen SE general 1-Month interbank offered rate Swiss market 1-Month interbank offered rate Hang Seng index 1-Month interbank offered rate Straits times 1-Month interbank offered rate Taiwan SE weighted 30-Day money market

4.2. Test of a systematic relationship between beta and returns Following Pettengill et al. (1995), we have two purposes: (1) to test for a systematic and conditional relationship between beta and realized returns; and (2) to test for a positive long-run riskreturn trade-off. The original systematic and unconditional relationship between beta and realized returns is also examined for comparison. To test for a systematic relationship between beta and returns, we apply the threestep approach of Fama and MacBeth (1973) with a minor change. Since country

116

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

indices are well-diversied portfolios, the rst-step of Fama and MacBeth (1973) approach can be skipped. The sample period can then be divided into two equal 5year sub-periods, which are country beta estimation period and a test period of 5 years. In the country beta estimation period, betas are estimated using the 5-year data with each sub-sample by regressing country index returns against the world market returns plus 12 additional currency risk premiums. The estimation of beta utilizes the equation (Solnik, 2000) as shown below Rit biw RPwt gi1 RP1t gi2 RP2t gik RPkt
it

(2)

where Rit is the realized excess return of a country index, i, in period t(Ri Rf), biw the sensitivity of country index, i, to the world market, RPwt the world market risk premium equals to realized world market return minus the domestic risk-free rate (Rw Rf), gi1 to gik the sensitivity of asset, i, to the currencies 1 to k, and RP1t to RPkt are the risk premiums on currencies 1 to k in period t. The country betas in 1996 are estimated from the period 19911995. Similarly, the country betas in 1997 are estimated from the period 19921996, and so on. The test for a positive riskreturn trade-off utilizes the following equation: Rit g0 g1 bit
it

(3)

Eq. (3) estimates the market risk premium using realized excess returns of country index and the country-estimated beta. Under the CAPM assumption, betas in the estimation period are good proxy for betas in the test period. Hence, Eq. (3) tests a systematic unconditional relationship between beta and realized returns. If the value of g1 is greater than zero with a signicant t-statistic, a systematic unconditional relationship between beta and realized returns is supported. For testing the systematic, conditional relationship between beta and realized returns, the following equation is used: Rit g0 g1 d bit g2 (1 d) bit
it

(4)

where d 1 if RPwt 0 (i.e. when world market excess returns are positive), and d 0 if RPwt 0 (i.e. when world market excess returns are negative). The above relationship is examined for each month in the test period 19962000 by estimating either g1 or g2, depending on the sign of the world market excess returns. We, then, follow Pettengill et al. (1995) to set the hypotheses to test the signicance of g1 and g2. Since g1 is estimated in periods of positive market excess returns, the expected sign associated with it is positive. Hence, the following hypothesis is tested: H0: g 1 H1: g 1 0, 0.

Since g2 is estimated in periods of negative market excess returns, the expected sign associated with it is negative. Hence, the following hypothesis is tested: H0: g 2 0,

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

117

H1: g 2

0.

If the null hypothesis is rejected in both situations, a systematic conditional relationship between beta and realized returns is supported. The data are also divided into months of a year to test for the seasonality effect (observed in previous studies, e.g. Tinic & West, 1984, of unconditional relationship between beta and realized returns). We examine whether seasonality exists in the studies of conditional relationship between beta and realized returns. In this study, besides a value-weighted world index (i.e. MSCI world index), we also use an equally weighted world index to test the unconditional and conditional relationship between beta and returns. We then examine whether the results of using these two indices as proxy for the market index are different. Furthermore, monthly data as well as weekly data are used to the whole testing procedure discussed above to examine the results between using monthly and weekly returns. 4.3. Test of the positive riskreturn trade-off The second goal of the study is to determine whether a systematic relationship between beta and returns translates into a positive reward for holding risk (i.e. do high-beta assets, on an average, earn higher returns than low-beta asset?). Pettengill et al. (1995) suggested that if a systematic conditional relationship between beta and returns exists, two conditions are required for a positive reward for holding beta risk. The rst condition is that market excess returns are, on an average, positive. The second condition is that the riskreturn relationship is symmetrical between periods of positive and negative market excess returns. The average market excess return for the test period is calculated to test for the rst condition. The t-test is used to determine whether market excess returns are, on an average, positive. For the second condition, the risk premiums during up and down markets are compared to test for symmetry. Since the expected signs of risk premiums during up and down markets differ, the sign for risk premiums in down markets is reversed. After the adjustments, the following hypothesis is tested by using the MannWhitney U-test, which has the advantage that populations are not required to be normally distributed and with equal variances H0: m1 H1: m1 m2, m2.

If the null hypothesis cannot be rejected, there will be no difference between the risk premiums during up and down markets. Hence, the riskreturn relationship is symmetrical during periods of positive and negative market excess returns. 5. Empirical results 5.1. Beta vs. realized returns Table 1 presents the results for the unconditional relationship between beta and realized returns using monthly returns. Employing MSCI world index returns to

118

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

Table 1 Estimates of slope coefcients Name of market index (a) Monthly data Rit MSCI Equally weighted (b) Weekly data Rit MSCI Equally weighted g0 g1 g1 bit 0.2068 0.1577 g1 bit 0.0067 0.0238
it

T-statistic

P-value

0.438 0.346
it

0.6613 0.7296 0.9641 0.8563

g0

0.045 0.181

proxy world market returns, the mean value of g1 ( 0.21) cannot reject the null hypothesis of no relationship between risk and returns at the 5% level. Employing an equally weighted world index returns, the mean value of g1 ( 0.16) is also insignicantly different from zero at the 5% level. Table 1 also presents the same results using weekly returns. For the MSCI world returns, the mean value of g1 ( 0.01) cannot reject the null hypothesis of no relationship between risk and returns at the 5% level. For the equally weighted world returns, the mean value of g1 ( 0.02) also cannot reject the null hypothesis at the 5% level. The results for unconditional relationship between beta and realized returns are, as expected, not signicant. The results are consistent across two different world market indices and across two different investment horizons. The biased results presented above are due to the combination of positive and negative market excess returns, as suggested by Pettengill et al. (1995). Given conditional relationship between beta and realized returns, positive market excess returns and negative market excess returns should be segregated. Table 2 presents results after the segregation effect is considered (tested by Eq. (4)). It presents the slope coefcients for up and down markets using monthly returns. Employing MSCI world
Table 2 Estimates of slope coefcients for up markets and down markets Name of market index Panel A. Up markets Panel B. Down markets

g1 (a) Monthly data Rit g0 MSCI 6.1414 Equally 6.7721 weighted (b) Weekly data Rit g0 MSCI 3.3497 Equally 3.0288 weighted

T-statistic g1 d bit 36.547 30.862 g1 d bit 68.837 68.902

P-value

g2

T-statistic

P-value

g2 (1 d) bit it 0.0001 3.7814 0.0001 4.3653 g2 (1 d) bit it 0.0001 1.9814 0.0001 1.8170

12.494 13.034

0.0001 0.0001

19.831 20.463

0.0001 0.0001

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

119

returns to proxy world market returns, the mean value of g1 during up markets (positive market excess returns) is 6.14 and the mean value of g2 during down markets (negative market excess returns) is 3.78. Both values are signicant at the 5% level. For the equally weighted world returns, the mean value of g1 during up markets is 6.77 and the mean value of g2 during down markets is 4.37, which are both signicant at the 5% level. Table 2 also presents the slope coefcients for up and down markets using weekly returns. For the MSCI world returns, the mean value of g1 during up markets is 3.35 while that of g2 during down markets is 1.98, which are both signicant at the 5% level. For the equally weighted world returns, the mean value of g1 during up markets is 3.03 while that of g2 during down markets is 1.82, which are both signicant at the 5% level. The results in Table 2 are consistent with previous results on conditional relationship using the US and Swiss market data. There is a signicant positive (negative) relationship between beta and returns during periods of positive (negative) market excess returns. Our ndings provide strong evidence that highrisk stock markets outperform low-risk stock markets when the realized world market excess return is positive and similarly the high-risk stock markets incur higher losses when the realized world market excess return is negative. 5.2. Seasonality in riskreturn relationship To test for the seasonality effect, country index returns are separated according to the months in a year and the riskreturn relationship is reexamined using Eqs. (3) and (4). Table 3 presents the regression coefcients from Eq. (3) using monthly MSCI world returns. The results cannot reject the null hypothesis of no riskreturn relationship for all months at the 5% level. Table 3 presents the same results using equally weighted world monthly returns. The results are consistent with those of monthly MSCI world returns that none of the 12 months show a signicant linear relationship between risk and returns at the 5% level. Table 4 presents the regression coefcients from Eq. (3) using MSCI world and equally weighted weekly returns. The results are consistent with those using monthly returns to examine the seasonality effect. It, thus, shows that there is no difference in using monthly or weekly data in examining the seasonality effect. Our results nd no January effect in the unconditional relationship between beta and returns, which is different from Fletcher (2000) who found a signicant positive riskreturn relationship in January. The reason may be due to the fact that different models are used (international CAPM instead of domestic CAPM is used in our study). Table 5 presents the slope coefcients for up markets and down markets using MSCI world monthly returns. In up-market months, the null hypothesis of no relationship between beta and returns is rejected at the 5% level for all months except June. In down-market months, the null hypothesis of no relationship between beta and returns is rejected at the 5% level for all months. Table 5 also presents the same results using the equally weighted world monthly returns. In up-market months, the null hypothesis of no relationship between beta and returns is rejected at the 5%

120

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

Table 3 Estimates of slope coefcients from monthly data Sample period g1


it

T-statistic

P-value

(a) MSCI world index Rit g0 g1 bit All Months 0.2068 January 0.9128 February 0.0226 March 1.8522 April 0.5685 May 1.5881 June 0.3895 July 0.8287 August 0.5359 September 3.5674 October 2.1912 November 1.2794 December 1.3166 (b) Equally weighted world index Rit g0 All months 0.1577 January 0.1296 February 0.1738 March 0.1058 April 0.1228 May 0.1177 June 0.1447 July 0.1936 August 0.1675 September 0.2964 October 0.1619 November 0.1408 December 0.1376

g1 bit

0.438 0.820 0.025 1.520 0.582 1.428 0.503 0.660 0.229 1.898 0.978 0.774 1.102
it

0.6613 0.4156 0.9798 0.1335 0.5630 0.1581 0.6168 0.5120 0.8195 0.0623 0.3320 0.4417 0.2748 0.7296 0.9104 0.8481 0.9061 0.9223 0.9042 0.8370 0.8806 0.9464 0.8663 0.9495 0.9140 0.9024

0.346 0.113 0.192 0.118 0.098 0.121 0.207 0.151 0.068 0.169 0.064 0.108 0.123

level for all months except June. In down-market months, the null hypothesis of no relationship between beta and returns is rejected at the 5% level for all months. The insignicant slope coefcients found in June in Table 5 are due to the absence of negative market excess returns in down-market periods. Since the monthly market excess returns in June are all positive throughout the period 19962000, we cannot separate the effect of positive and negative market excess returns in the conditional relationship between beta and returns. Eq. (4) includes an additional dummy variable and the summation of the dummy variables associated with up-market and downmarket periods is always equal to 1. This situation, known as dummy-variable-trap (a case of perfect collinearity) in econometrics, causes the regression coefcients untrustworthy and the interpretation of the slope coefcients collapses. This situation will be improved if higher frequency data are used in the analysis since both positive and negative returns should exist with higher frequency data. Table 6 presents the slope coefcients for up-market and down-market months using the MSCI world and equally weighted world weekly returns. The results show

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

121

Table 4 Estimates of slope coefcients from weekly data Sample period g1


it

T-statistic

P-value

(a) MSCI world index Rit g0 g1 bit All months 0.0067 January 0.0894 February 0.0703 March 0.2289 April 0.0745 May 0.1888 June 0.0061 July 0.0061 August 0.0445 September 0.5164 October 0.2086 November 0.2268 December 0.1106 (b) Equally weighted world index Rit g0 All months 0.0238 January 0.0293 February 0.0183 March 0.0142 April 0.0168 May 0.0195 June 0.0268 July 0.0242 August 0.0377 September 0.0285 October 0.0260 November 0.0236 December 0.0212

g1 bit

0.045 0.176 0.175 0.456 0.163 0.452 0.021 0.014 0.088 0.987 0.327 0.360 0.189
it

0.9641 0.8601 0.8611 0.6486 0.8703 0.6517 0.9832 0.9885 0.9298 0.3247 0.7442 0.7190 0.8501 0.8563 0.9519 0.9616 0.9740 0.9674 0.9541 0.9226 0.9474 0.9416 0.9447 0.9654 0.9666 0.9621

0.181 0.060 0.048 0.033 0.041 0.058 0.097 0.066 0.073 0.069 0.043 0.042 0.048

a signicant positive relationship between beta and returns in up-markets for all months and a signicant negative relationship between beta and returns in downmarkets for all months at the 5% level, no matter which proxy for the world market is used. Furthermore, no seasonality is found as all months show consistent results. 5.3. Risk vs. return: a test for a positive trade-off Table 7 presents the average monthly and weekly excess returns for the sample period 19962000. The average annualized MSCI world monthly excess return is 8.78% while that of equally weighted world monthly excess return is 7.91%. Both values are signicantly positive at the 5% level. The table also shows that the average annualized MSCI world weekly excess return is 8.80% and that of equally weighted world weekly excess return is 7.73%. Again, both values are signicantly positive at the 5% level. Hence, the rst condition for a positive riskreturn trade-off is met. The second condition of a positive riskreturn trade-off requires a consistent

122

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

Table 5 Estimates of slope coefcients for up markets and down markets from monthly data Sample period Panel A. Up markets g1 T-statistic P-value Panel B. Down markets g2 T-statistic P-value

(a) MSCI world index Rit g0 g1 d bit g2 (1 d) bit it All months 6.1414 36.547 0.0001 3.7814 January 5.9214 25.234 0.0001 4.9286 February 3.2965 9.780 0.0001 2.4514 March 6.6962 15.666 0.0001 4.6353 April 3.6711 10.135 0.0001 1.7730 May 5.2394 11.600 0.0001 2.5371 June 0.3895 0.503 0.6168 July 7.1819 19.832 0.0001 1.0526 August 8.1525 8.759 0.0001 5.6904 September 8.9394 22.304 0.0001 3.2962 October 8.4249 10.361 0.0001 4.7651 November 8.7427 23.922 0.0001 7.2085 December 4.9864 14.899 0.0001 2.7872 (b) Equally weighted world index Rit g0 g1 d bit g2 (1 d) bit All months 6.7721 30.862 0.0001 4.3653 January 7.0393 13.813 0.0001 6.2094 February 4.2935 11.666 0.0001 2.8182 March 4.8883 10.386 0.0001 3.7388 April 6.1649 13.700 0.0001 3.8217 May 5.6149 12.146 0.0001 1.5596 June 0.1447 0.207 0.8370 July 6.6979 11.308 0.0001 2.4043 August 12.6480 16.373 0.0001 6.2325 September 9.0306 17.320 0.0001 4.6105 October 9.8553 7.362 0.0001 6.1431 November 7.5255 11.322 0.0001 6.6406 December 4.3287 7.294 0.0001 2.8138

12.494 13.307 3.992 6.720 2.877 3.997 2.252 3.291 5.224 3.128 12.497 4.448
it

0.0001 0.0001 0.0002 0.0001 0.0055 0.0001 0.0279 0.0017 0.0001 0.0027 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0059 0.0026 0.0001 0.0001 0.0038 0.0001 0.0028

13.034 8.596 5.052 5.780 5.578 2.849 3.142 5.442 5.967 3.009 7.048 3.112

relationship between beta and returns during up and down markets. This condition is tested using MannWhitney U-test. Table 8 presents the results of the Mann Whitney U-test using monthly market excess returns. There are 438 (474) months with positive market excess returns and 342 (306) months with negative market excess returns for the MSCI (equally weighted) world index. The null hypothesis of no difference between the risk premiums during up and down markets is rejected at the 5% level in the MSCI case but not in the case of the equally weighted index. Thus, that the riskreturn relation is symmetrical between periods of positive and negative market excess returns is partially supported using monthly returns. Table 8 also reports the results of the MannWhitney U-test using weekly market excess returns. There are 1979 (2029) weeks with positive market excess returns and 1414 (1364) weeks with negative market excess returns for the MSCI (equally weighted) world index. The null hypothesis of no difference between the risk pre-

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

123

Table 6 Estimates of slope coefcients for up markets and down markets from weekly data Sample period Panel A. Up markets g1 T-statistic P-value Panel B. Down markets g2 T-statistic P-value

(a) MSCI world index Rit g0 g1 d bit g2 (1 d) bit it All months 3.3497 68.837 0.0001 1.9814 19.831 January 3.6824 23.000 0.0001 2.4905 7.846 February 2.8150 21.046 0.0001 1.9692 7.508 March 3.3888 19.468 0.0001 2.1729 6.188 April 3.0670 20.881 0.0001 1.9299 6.436 May 2.8253 22.134 0.0001 1.7853 6.778 June 1.8838 20.062 0.0001 1.1581 6.008 July 3.0082 23.919 0.0001 1.4756 5.807 August 3.4286 24.011 0.0001 1.8171 6.166 September 3.2885 18.588 0.0001 1.9001 5.283 October 4.0843 18.371 0.0001 1.6544 3.727 November 4.6305 19.620 0.0001 3.3736 7.587 December 3.7866 18.911 0.0001 2.0689 5.039 (b) Equally weighted world index Rit g0 g1 d bit g2 (1 d) bit it All months 3.0288 68.902 0.0001 1.8170 20.463 January 3.4217 23.002 0.0001 2.0938 6.956 February 2.5315 18.159 0.0001 1.8196 6.733 March 3.3475 22.834 0.0001 2.4598 8.803 April 3.0647 23.326 0.0001 2.1046 8.204 May 2.0633 17.081 0.0001 0.9550 3.909 June 1.6704 16.992 0.0001 0.9756 4.883 July 2.7254 25.316 0.0001 1.6784 7.762 August 3.3947 20.852 0.0001 1.5092 4.625 September 2.6695 17.747 0.0001 0.8794 3.061 October 3.8002 18.420 0.0001 1.9429 4.642 November 4.9520 24.327 0.0001 4.0639 11.398 December 3.3497 22.091 0.0001 2.3188 7.898

0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0024 0.0001 0.0001 0.0001

miums during up and down markets is rejected at the 5% level in both cases. Thus, a symmetrical relationship of market excess returns in up and down markets does not exist in the weekly returns. The symmetrical relationship only exists in using equally weighted world monthly excess returns to proxy market excess returns. This result, in addition to the nding of a signicant and positive average monthly market excess return, strongly supports the expectation of a positive reward for holding risk. However, for both MSCI world and equally weighted world weekly returns, only the rst condition is satised. Thus, a symmetrical riskreturn trade-off does not exist for these cases. The reason may be due to the different magnitudes of market excess returns in up and down markets. There is a large market excess return in up markets than in down markets. Combined with longer periods of positive market excess returns than periods of negative market

124

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

Table 7 Average monthly/weekly market excess returns Using MSCI world index as a proxy for market index Monthly Annualized mean excess returns Mean excess return Variance T-statistic P-value 8.7801% 0.7038% 0.0018 4.6177 0.0001 Weekly 8.8041% 0.1624% 0.0004 4.6408 0.0001 Using equally weighted world index as a proxy for market index Monthly 7.9062% 0.6361% 0.0023 3.7364 0.0002 Weekly 7.7333% 0.1433% 0.0004 4.1529 0.0001

Table 8 Statistics of MannWhitney U-test Using MSCI world index as a proxy for market index Using equally weighted world index as a proxy for market index

(a)Monthly data Months of positive excess returns Months of negative excess returns Z-statistic P-value (b) Weekly data Weeks of positive excess returns Weeks of negative excess returns Z-statistic P-value

438 342 4.9288 0.0001 1979 1414 2.0912 0.0365

474 306 0.2960 0.7672 2029 1364 6.1554 0.0001

excess returns, it, thus, shows that there is a difference of market risk premium between up-market and down-market periods.

6. Conclusions Previous studies on testing unconditional relation between beta (systematic risk) and returns found a weak and insignicant relationship. However, when taking into account of the conditional relationship between beta and returns, a signicant relationship is found. Our study examines this conditional relationship in 13 international stock markets for the period 19912000. Like the studies of conditional relationship between beta and returns on the US market, there is a signicant positive relationship in up markets and a signicant negative relationship in down markets. The results are the same no matter the MSCI world index or an equally weighted

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

125

world index returns are used to proxy world market returns. Also, our results are robust for both monthly and weekly returns. Unlike previous studies, no January seasonal effect is found in the unconditional relationship between beta and returns. However, a signicant positive relationship in up markets and a signicant negative relationship in down markets are found for all months in a year. Furthermore, this paper nds that both monthly and weekly market excess returns are, on an average, positive but the symmetry of riskreturn relationship in up markets and down markets is weak, which is inconsistent with previous studies of conditional relationship of CAPM. The main reason may be due to the different models used. Overall, this paper shows that high-beta countries do capture higher returns in up markets and poorer returns in down markets than low-beta countries. Our results are clearly useful for international investment. Since signicant conditional relationship is observed in the model of international CAPM, the paper supports the continuous use of beta as a relevant risk measure. Beta is still a useful tool in explaining the cross-sectional differences in country returns and for portfolio management (e.g. for market-timing strategies). Hence, international investors and portfolio managers should note our ndings in making their optimal investment decisions. Acknowledgements The authors thank the comments from three anonymous referees on an earlier version of this paper. References
Black, F. (1972). Capital market equilibrium with restricted borrowing. Journal of Business, 45, 445455. Fama, E., & French, K. (1992). The cross-section of expected stock returns. Journal of Finance, 47, 427465. Fama, E., & French, K. (1996). The CAPM is wanted, dead or alive. Journal of Finance, 54, 19471958. Fama, E., & MacBeth, J. (1973). Risk, return, and equilibrium: empirical tests. Journal of Political Economy, 81, 607636. Fletcher, J. (2000). On the conditional relationship between beta and return in international stock markets. International Review of Financial Analysis, 9, 235245. Haugen, R. A., & Baker, N. L. (1991). The efcient market inefciency of capitalization-weighted stock portfolios. Journal of Portfolio Management, 17, 3540. Isakov, D. (1999). Is beta still alive? Conclusive evidence from the Swiss stock market. The European Journal of Finance, 5, 202212. Jegadeesh, N. (1992). Does market risk really explain the size effect? Journal of Financial and Quantitative Analysis, 27, 337351. Lakonishok, J., & Shapiro, A. C. (1986). Systematic risk, total risk and size as determinants of stock market returns. Journal of Banking and Finance, 10, 115132. Lintner, J. (1965). The valuation of risk assets and the selection of risky investments in stock portfolios and capital budget. Review of Economics and Statistics, 47, 1337. Pettengill, G. N., Sundaram, S., & Mathur, I. (1995). The conditional relation between beta and returns. Journal of Financial and Quantitative Analysis, 30, 101116. Reinganum, M. (1981). A new empirical perspective on the CAPM. Journal of Financial and Quantitative Analysis, 16, 439462.

126

G.Y.N. Tang, W.C. Shum / International Business Review 12 (2003) 109126

Schwert, G. (1983). Size and stock returns, and other empirical regularities. Journal of Financial Economics, 12, 312. Sharpe, W. F. (1964). Capital asset prices: a theory of market equilibrium under conditions of risk. Journal of Finance, 19, 425442. Solnik, B. H. (2000). International investments. Reading, MA: Addision-Wesley. Tinic, S., & West, R. (1984). Risk and return: January vs. the rest of the year. Journal of Financial Economics, 13, 561574.