Time-varying beta and the Asian financial crisis: Evidence from the Asian industrial sectors

Japan and the World Economy 22 (2010) 228–234

Contents lists available at ScienceDirect

Japan and the World Economy journal homepage: www.elsevier.com/locate/jwe

Time-varying beta and the Asian financial crisis: Evidence from the Asian industrial sectors§ Taufiq Choudhry *, Lin Lu, Ke Peng School of Management, University of Southampton, Highfield, Southampton SO17 1BJ, UK

A R T I C L E I N F O

A B S T R A C T

JEL classification: G1 G12 G15

This paper empirically investigates the effects of the Asian financial crisis of 1997–98, and the period immediately afterwards, on the time-varying beta of four industrial sectors (chemical, finance, retail and industry) of Indonesia, Singapore, South Korea, and Taiwan. We apply daily data from 1992 to 2002 and the bivariate MA-GARCH model (BEKK) to create the time-varying industrial betas. Results provide evidence of the influence of the Asian financial crisis, and the period after, on the time-varying industrial betas of these countries. These results may have implications for investors who are interested in portfolio risk management. ß 2010 Elsevier B.V. All rights reserved.

Keywords: Time-varying beta GARCH BEKK model Asian financial crisis Volatility

1. Introduction It is now considered an empirical fact that the beta of a risky asset or portfolio is time varying and not constant (Fabozzi and Francis, 1978). Movement in the time-varying beta maybe due to both macroeconomic and/or microeconomic factors (Bos and Newbold, 1984). These factors could include operational changes in the company, or changes in the business environment peculiar to the company, the rate of inflation, general business conditions and expectations about relevant future events.1 This paper investigates the effects of the Asian financial crisis of 1997–98 on the time-varying beta of industrial sectors of four countries of the Far East. To our knowledge no other study investigates the influence of the Asian crisis on the time-varying beta of Asian industrial sectors.2 During the Asian financial crisis of § We would like to thank an anonymous referee for several useful comments and suggestions. We also thank the seminar participants at the All China Economics International Conference, Hong Kong 2009 for several useful comments on an earlier version of the paper. Any remaining errors and omissions are our responsibility alone. * Corresponding author. Tel.: +44 2380599286; fax: +44 2380593844. E-mail address: [email protected] (T. Choudhry). 1 See Rosenberg and Guy (1976a,b) for a detailed discussion of the factors. 2 Maroney et al. (2004) explore risk and return relations in six Asian equity markets affected by the 1997 Asian financial crisis. Their results show national equity betas increased and average returns fell substantially after the start of the crisis, and they think this is due to leverage increases. Choudhry (2005) provides a study of the affect of the crisis on the beta of Malaysian and Taiwanese firms. Results indicate a rise in the beta during the crisis, especially in the case of Malaysian firms.

0922-1425/$ – see front matter ß 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.japwor.2010.06.003

1997–98, the exchange rates and the stock prices of some of the Far East countries fell dramatically in value. There was also a dramatic decrease in the capital inflow to the Far East countries from $93 billion in 1996 to $9.4 billion in 1998. The average daily changes in the stock markets were negative for most Asian stock markets during 1996–98, indeed during this period 10 percent daily changes in the Asian stock markets became common (Kaminsky and Schmukler, 1999). The crisis slowed worldwide economic growth; increased risk premiums in debt markets; stock markets became more volatile; and confidence indicators fell in many countries (Choudhry, 2005). Given the current on-going global financial crisis, it is of empirical interest to investigate the potential effect of a past financial crisis on the time-varying beta of risky assets. We know that the Asian crisis of 1997–98 had a drastic negative effect on the economy and the financial markets of the Far East region, but less is known of the effects of this crisis on the beta of Asian industrial sectors. Studies have investigated the effect of the Asian crisis at the general market and firm level but not yet at the industrial level despite the fact that portfolio managers do often take into consideration industrial level stock in financial portfolios. Therefore, the focus of this paper is to investigate the effects of a financial crisis at the industrial level. As stated by Chen and So (2002) during the Asian financial crisis the observed volatility of the financial markets around the world also increased.3 The rise in the volatility during the crisis of 1997–98 could be due to two factors: firstly, inflammatory statements by 3

Market volatility peaked in the months of October 1997 through January 1998.

T. Choudhry et al. / Japan and the World Economy 22 (2010) 228–234

government officials; and secondly, by the introduction (or elimination) of restrictions on financial market transactions (Kaminsky and Schmukler, 1999). The higher implied volatility during the crisis period indicates that investor uncertainty about future stock market returns had increased: a jump in the stock market volatility may be perceived by the investors as an increase in the risk of equity investment. According to Chen and So (2002) when stock returns are more volatile, all other things being equal, market risk (beta) is expected to be larger, with greater stock price responses for firms that have greater exposure to the source of the risk. If so, investors may shift their funds to less risky assets, such as bonds. This reaction would tend to raise the cost of funds to firms issuing stock. 2. The (conditional) CAPM and the time-varying beta One of the assumptions of the capital asset pricing model (CAPM) is that all investors have the same subjective expectations on the means, variances and covariances of returns.4 According to Bollerslev (1988), economic agents may have common expectations on the moments of future returns but these are conditional expectations and therefore random variables rather than constant.5 The CAPM that takes these conditional expectations into consideration is sometimes known as conditional CAPM. The conditional CAPM provides a convenient way to incorporate the time-varying conditional variances and covariances (Bodurtha and Mark, 1991).6 An asset’s beta in the conditional CAPM can be expressed as the ratio of the conditional covariance between the forecast error in the asset’s return, the forecast’s error of the market return and the conditional variance of the forecast error of the market return. The following analysis relies heavily on Bodurtha and Mark (1991). Let Ri,t be the nominal return on asset i (i = 1, 2, . . ., n) and Rm,t the nominal return on the market portfolio m. The excess (real) return of asset i and market portfolio m over the risk-free asset return is presented by ri,t and rm,t, respectively. The conditional CAPM in excess returns may be given as ðr i;t jIt1 Þ ¼ biIt1 Eðr m;t jIt1 Þ

(1)

where

biIt1 ¼

covðRi;t ; Rm;t jIt1 Þ covðr i;t ; r m;t jIt1 Þ ¼ varðRm;t jIt1 Þ varðr m;t jIt1 Þ

(2)

and E(jIt1) is the mathematical expectation conditional on the information set available to the economic agents in the previous period (t  1), It1. Expectations are rational based on Muth’s (1961) definition of rational expectation where the mathematical expected values are interpreted as the agent’s subjective expectations. According to Bodurtha and Mark (1991) asset i’s risk premium varies over time due to three time-varying factors: the market’s conditional variance, the conditional covariance between the asset’s return and the market’s return, and/or the market’s risk premium. If the covariance between asset i and the market portfolio m is not constant then the equilibrium returns Ri,t will not be constant.7 4 See Markowitz (1952), Sharpe (1964) and Lintner (1965) for details of the CAPM. 5 According to Klemkosky and Martin (1975) betas will be time-varying if excess returns are characterized by conditional heteroscedasticity. 6 Hansen and Richard (1987) have shown that omission of conditioning information, as is done in tests of constant beta versions of the CAPM, can lead to erroneous conclusions regarding the conditional mean variance efficiency of a portfolio. 7 In this paper we apply the domestic CAPM and not the international CAPM. The international CAPM takes into account both the systematic risk and the exchange rate risk. The only way to use the domestic CAPM in the international context is by making two assumptions, (i) PPP holds at any point in time and (ii) consumption baskets are the same for all investors.

229

3. The data, the bivariate MA-GARCH model and testing the effects of the crisis 3.1. Data In this paper, we employ daily industrial stock indices from Indonesia, Singapore, South Korea, and Taiwan. The crisis affected Indonesia and South Korea more heavily than Singapore and Taiwan (Neiss, 2009).8 The range of the data is from 1 January 1992 to 30 December 2002.9 For the empirical analysis the total period is further broken into three smaller sub-periods: pre-crisis period (January 1992 to June 1997), crisis period (July 1997 to June 1998), and post-crisis period (July 1998 to December 2002). The total period January 1992 to December 2002 is also applied in the empirical tests. Four industrial sectors10 (chemical, finance, retail and industry) are included for each country.11 The chemical sector includes companies in specialty chemicals (e.g. adhesives, coatings, additives), products derived from life sciences (e.g. pharmaceuticals), agricultural chemicals (e.g. pesticides, fertilizers) and consumer care products (e.g. detergent, bleach, cosmetics). The industry sector index is composed of aerospace industries, building supplies, industrial-building products, business equipment, chemicals, machinery (both light and industrial), metals fabrication, paper and packaging, and photo equipment. The definition of the financial sector used here includes banks, insurance, investments, real-estate investment, real estate, and savings and loans. Finally, the retail sector includes apparel, department stores, food stores, and miscellaneous shops. The market index is represented by the general stock market index of each country: Indonesia, Jakarta SE composite index; South Korea, Korea SE composite index; Singapore, Straits Times industrial index; Taiwan, Taiwan SE weighted index. The return on the risk-free asset applied is the 3-month deposit return for each country.12 Stock returns are created by taking the first difference of the log of the stock index, and the excess stock return is calculated as the nominal industrial stock return minus the return on the risk-free asset. We also include a global factor in the empirical tests, to test the effect of the international volatility. The global factor is represented by the return in the MSCI world index. The massive capital flow, or subsequent cease, across the borders had a major impact on the boom, bust of the Asian stock markets and the real economic deterioration in the region. By including the global factor in our tests we aim to capture the potential effects of the financial markets outside the Far East region on the industrial timevarying betas under study. All data are obtained from Datastream. The basic statistics for all excess returns (including the market return) are found to be skewed and have excess kurtosis and thus are found to be non-normal by means of the J–B tests. These results are not presented to save space but are available on request. 3.2. Bivariate GARCH (BEKK) model The following bivariate GARCH(p,q) model may be used to represent the returns from asset i and the market portfolio (m). 8 We select countries with the different level of the crisis effect are for comparison purposes. 9 We kept the data to 11 years so that pre- and post-crisis periods would be approximately the same length. 10 We chose industries based on the availability of the data. We need the same length data for the same industries from all four countries. 11 An industrial sector index is termed sector-based. These indices represent companies from one industrial sector of the economy. 12 Once again availability of the data determined the risk-free asset return.

T. Choudhry et al. / Japan and the World Economy 22 (2010) 228–234

230

This presentation is termed by Engle and Kroner (1995) the BEKK model; the conditional covariance matrix is parametrized as yt ¼ m þ et  uet1

(3)

et  Nð0; Ht Þ Vt1

(4)

vechðHt Þ ¼ C 0 C þ

q p K X K X X X A0Ki eti e0ti Aki þ B0K j Ht j Bk j K¼1 i¼1

(5)

K¼1 i¼1

where yt = (rit, rm,t) is a (2  1) vector containing returns from asset i and the market portfolio (m), m is a 2  1 vector of constant, Aki, i = 1, . . ., q, k = 1,. . . K, and Bkj j = 1, . . ., p, k = 1, . . ., K are all N  N matrices. This formulation has the advantage over the general specification of the multivariate GARCH in that conditional variance (Ht) is guaranteed to be positive for all t (Bollerslev et al., 1994). The BEKK model is sufficiently general to include all positive definite diagonal representations, and nearly all positive definite vector representations. The moving average (MA) term uet1 is included to capture the effect of nonsynchronous trading. According to Susmel and Engle (1994), non-synchronous trading induces negative serial correlation, and the MA term allows for autocorrelation induced by discontinuous trading in the asset. The following presents the BEKK bivariate GARCH(1,1), with K = 1. Ht ¼ C 0 C þ A0 et1 e0t1 A þ B0 Ht1 B  " 2 A11 A12 0 e1;t1 Ht ¼ C 0 C þ A21 A22 e2;t1 e1;t1     B11 B12 B11 B12 0 Ht1 þ B21 B22 B21 B22

# e1;t1 e2:t1  A11 A12  A21 A22 e22;t1

where C is a 2  2 lower triangular matrix with intercept parameters, A and B are 2  2 square matrices of parameters. The BEKK parameterization requires estimation of only 11 parameters in the conditional variance–covariance structure, and guarantees Ht positive definite. Importantly, the BEKK model implies that only the magnitude of past returns innovations is important in determining current conditional variances and covariances. The parameters in A reveal the extent to which the conditional variances of the two variables are correlated with past squared errors. The off-diagonal elements represent how the past squared error of one variable affects the conditional variance of another variable. The parameters in B depict the extent to which the current levels of conditional variances are correlated with past conditional variances. The off-diagonal elements indicate the extent to which the conditional variance of one variable is correlated with the lagged conditional variance of another variable. High values of off-diagonal elements support a correlation between volatility of two variables. The time-varying beta (b) for asset i is calculated as Hˆ 12;t Hˆ 22;t

The following OLS regression is applied to investigate the effect of the crisis and the period after the crisis on the time-varying beta:

bit ¼ a0 þ a1 CV it1 þ a2 MV t1 þ a3 GV it1 þ et

(7)

where bit is the individual industrial time-varying beta as defined in Eq. (6), CVit is the conditional volatility of the individual industrial sector, MVt is the market conditional volatility, GVt is the global factor conditional volatility, and et is the random error term with the standard assumptions.13 The parameters a1, a2, and a3, measure the effects of the conditional volatility of the individual industrial sectors, the local market, and the global factor on the beta of the four industrial sectors of each country, respectively. If the sign on these parameters is positive (and significant) then a rise in the volatility of the industry, local market, and/or the global factor should increase the beta of the firm. If investors perceive a rise in the volatility of the industrial sector and/or the local stock market as an increase in the risk of equity investment, then the conditional volatility of the individual industrial market (CV) and/ or the conditional volatility of the local market (MV) should impose a direct effect on the beta of the industry. Similarly, an increase in the global factor volatility seen as an increase in risk (GV) should have a direct relationship with the beta. Application of the three sub-periods of pre-crisis, crisis and post-crisis makes it possible to investigate and compare the changes in the effects of the three volatilities on the beta. 4. Empirical results

(5a)

bi;t ¼

3.3. Testing the effects of the Asian financial crisis

(6)

where Hˆ 12;t is the estimated conditional covariance between the specific asset returns and market portfolio returns, and Hˆ 22;t is the estimated conditional variance of the market portfolio returns from the bivariate BEKK model. Given that conditional covariance and conditional variance are time-dependent, the stock beta will also be time-dependent. We apply the time-varying beta defined in Eq. (6).

The bivariate BEKK GARCH results are quite standard. Given this, and the lack of space, they are not presented in this paper but are available on request. In summary, the ARCH coefficients are all positive and significant implying volatility clustering in both the industrial return and the market return. All ARCH coefficients are less than unity in size. Some evidence of a significant MA (u) coefficient is found for both the industrial sectors and the market. The Ljung-Box Q statistics on the standardized (normalized) 1=2 residuals (et/Ht ) and standardized squared residuals (et/Ht2 ) are applied to specify adequacy of the first two conditional moments. All series are found to be free of serial correlation (at the 5% level). The absence of serial correlation in the standardized squared residuals implies the lack of need to encompass a higher order ARCH process (Giannopoulos, 1995). The normality test also fails to indicate non-normal standardized and/or standardized squared residuals. These results are available on request. Figs. 1 and 2 present the time-varying betas of the four industrial sectors of Korea and Indonesia, respectively. These figures clearly show that during and after the 1997–98 crisis the beta of these industries was higher and more intense than during the pre-crisis period. Taiwan and Singapore figures are not presented to save space but are available on request. They also presented similar conclusions but on a smaller scale (as expected). Tables 1–4 present estimations of Eq. (7) for all industrial sectors of each country for the four periods. Eq. (7) is estimated sixty four times by means of OLS regression. By means of the Cochran–Orcutt method, all regressions are corrected for serial correlation. The results are also adjusted for heteroscedasticity.14 Table 1 shows the results from Indonesia. The constant term is significant in all tests. For Indonesia, there is little evidence of conditional volatilities affecting the beta. In a few of the tests the 13 The conditional volatilities of the sectors, local market and global factor are obtained from the BEKK model. 14 The regression is estimated by using least square but then a consistent estimate of the covariance matrix allowing for heteroscedasticity is computed.

[(Fig._1)TD$IG]

T. Choudhry et al. / Japan and the World Economy 22 (2010) 228–234

[(Fig._2)TD$IG]

Fig. 1. Korean time-varying betas.

231

Fig. 2. Indonesian time-varying betas.

coefficients are negative but less than unity in absolute value. Thus, the magnitude of the effect is small. The industrial sector seems to be the most influenced and the retail sector is the least affected. Conditional volatility of the industries is the most influential. The market and global conditional volatilities had little effect. In the handful of significant cases, the size of the coefficient (in absolute value) is less than unity. The Indonesian results are somewhat of a surprise given the magnitude of the crisis effect. The R2 range from

0.185 to 0.292 and the Durbin–Watson statistics are quite satisfactory.15 The lack of any significant effect from the crisis could be because Indonesia during the crisis and earlier periods was also experiencing major political crisis, with heightened 15 To check for multicollinearity the variance inflation factor (VIF) for multicollinearity is applied. The VIF is equal to 1/(1  R2). A value of 5 or more indicates potential multicollinearity. For tests involving Indonesia the highest VIF is 1.41.

T. Choudhry et al. / Japan and the World Economy 22 (2010) 228–234

232 Table 1 Indonesia. Time

Constant

Chemical 1992–97 1997–98 1998–2002 1992–2002

0.8821*** 0.9984*** 0.9435*** 0.9185***

Finance 1992–97 1997–98 1998–2002 1992–2002

Rho

R2

DW

(0.8030) (1.1516) (1.4671) (0.6251)

0.4672*** (19.9322) 0.5444*** (10.2740) 0.512*** (20.487) 0.535*** (33.927)

0.222 0.293 0.277 0.280

1.98 1.89 1.95 1.93

(1.4700) (0.0069) (1.3000) (1.0814)

1.1193* (1.8380) 5.0500 (0.9668) 0.4526 (0.5086) 0.5311 (0.7528)

0.531*** (23.661) 0.483*** (8.812) 0.4426*** (16.409) 0.4571*** (10.5641)

0.284 0.229 0.192 0.230

1.95 2.01 1.94 1.98

(0.5522) (0.6200) (0.7390) (0.5755)

0.1966 (0.3569) 2.5800 (0.5119) 1.7067*** (2.2376) 0.9259 (1.4357)

0.512*** (22.465) 0.5489*** (10.4500) 0.4640*** (17.967) 0.5251*** (32.9990)

0.261 0.292 0.215 0.277

1.97 1.99 1.89 1.95

0.2076 (0.5070) 3.9961 (0.9270) 0.9017 (1.4434) 1.0226* (1.8877)

0.4570*** 0.4115*** 0.4322*** 0.4385***

0.207 0.185 0.188 0.194

1.96 2.07 2.00 2.05

R2

DW

CV

MV

GV

(190.78) (13.4350) (62.957) (95.143)

0.2865* (1.9237) 0.5181 (1.0490) 0.7777*** (4.0068) 0.2035 (1.5543)

0.0186 (0.0839) 0.6907 (0.9408) 0.8681*** (2.7174) 0.1820 (0.8900)

0.3574 4.4058 0.9698 0.3254

1.0620*** 1.1455*** 1.0094*** 1.0480***

(145.290) (12.945) (58.186) (91.548)

0.069 (0.300) 0.3091 (0.5383) 0.5480* (1.8618) 0.4284** (2.4216)

0.5076 0.0068 0.5765 0.3076

Retail 1992–97 1997–98 1998–2002 1992–2002

1.0507*** 1.1962*** 1.1240*** 1.0937***

(165.848) (12.008) (72.295) (93.827)

0.0150 0.1199 0.2664 0.1901

0.1493 0.5368 0.2706 0.1382

Industrial 1992–97 1997–98 1998–2002 1992–2002

0.9727*** 1.0885*** 1.0418*** 1.0117***

(233.769) (17.3389) (87.3351) (125.200)

0.3484*** (2.2778) 0.0561 (0.1029) 0.4802*** (2.1450) 0.2414 (1.6036)

(0.1040) (0.2480) (1.2078) (1.3791)

0.3052 (1.4016) 1.4622* (1.8541) 1.4697*** (4.6810) 0.0054 (0.0252)

(19.355) (7.1858) (16.289) (26.0171)

CV = conditional variance of the industrial sector. MV = conditional variance of the market. GV = conditional variance of the global market. * Significant at the 10%. ** Significant at the 5%. *** Significant at the 1%. Table 2 Korea. Time

Constant

CV

MV

GV

Chemical 1992–97 1997–98 1998–2002 1992–2002

1.0082*** 1.1407*** 1.0221*** 1.0260***

Finance 1992–97 1997–98 1998–2002 1992–2002

Rho

(346.8191) (33.8682) (160.160) (225.8040)

0.6253*** (4.1309) 1.2321** (2.3266) 0.5017*** (2.6186) 0.1288 (1.0215)

0.5580*** (3.1417) 1.9416** (2.4232) 0.1984 (0.8588) 0.3534** (2.2096)

0.3052 0.8634 0.2037 0.1432

(0.8561) (0.4763) (0.5500) (0.4984)

0.3400*** 0.5249*** 0.3826*** 0.4677***

(13.6564) (9.7651) (14.121) (28.294)

0.121 0.277 0.146 0.233

2.02 2.15 2.02 2.09

1.0626*** 1.2254*** 1.1041*** 1.0947***

(335.1655) (33.2089) (161.332) (219.833)

0.2330 (1.4012) 2.6307*** (5.0555) 0.7475*** (3.4014) 0.9819*** (7.0725)

0.2073 (0.9900) 3.5743*** (4.3960) 0.6108** (1.9799) 1.1141*** (5.7429)

0.0679 1.8364 0.3267 0.2762

(0.1621) (0.9529) (0.6932) (0.8051)

0.2924*** 0.5391*** 0.2818*** 0.4266***

(11.532) (10.1773) (10.0301) (25.2405)

0.084 0.339 0.089 0.204

1.98 2.25 2.03 2.11

Retail 1992–97 1997–98 1998–2002 1992–2002

0.8886*** 1.0761*** 0.8978*** 0.9091***

(211.684) (26.4130) (93.7108) (151.136)

0.5523*** (3.1521) 2.1450*** (4.1630) 0.2249 (0.9785) 0.2450* (1.7225)

1.0506*** (4.9634) 3.2006*** (3.8714) 0.2399 (0.7367) 0.5384* (2.7040)

0.1080 0.8442 0.1906 0.3080

(0.2060) (0.4088) (0.2915) (0.6994)

0.3276*** 0.5487*** 0.2869*** 0.3950***

(13.0867) (10.4227) (10.2116) (22.9890)

0.117 0.309 0.079 0.163

1.98 2.14 2.03 2.08

Industrial 1992–97 1997–98 1998–2002 1992–2002

1.0943*** 1.2269*** 1.0968*** 1.1078***

(426.223) (39.0835) (264.697) (281.6773)

0.3367** (2.1990) 2.5084*** (4.9394) 0.6726*** (3.5739) 1.1129*** (8.9250)

0.3360* (1.8075) 3.2687*** (4.1716) 0.8588*** (4.0943) 1.2988*** (8.3218)

0.3286 0.2863 0.0716 0.1152

(0.9803) (0.17166) (0.2969) (0.4940)

0.3015*** 0.5296*** 0.3809*** 0.4963***

(11.9337) (9.9224) (14.0490) (30.5728)

0.091 0.317 0.152 0.281

1.97 2.18 2.06 2.13

CV = conditional variance of the industrial sector. MV = conditional variance of the market. GV = conditional variance of the global market. * Significant at the 10%. ** Significant at the 5%. *** Significant at the 1%.

uncertainty and growing public unrest. As reported in Neiss (2009), there was also a severe drought causing a serious rice shortage, and there was a fall in oil prices. Tests applied in this paper do not take into consideration the effects of factors other than the financial crisis. Further, the crisis has not affected Indonesia in a homogeneous fashion across the country (Neiss, 2009). The crisis affected more heavily those industries producing many nontradable goods, where domestic demand is critical, imported raw materials are important, and credit or external financing is required. The Korean results (Table 2) indicate a high level of volatility influence, especially conditional volatilities of the industries and the local market. Almost all sectors are affected during all periods

considered. During the crisis (1997–98), the conditional volatility of the industry imposes a positive effect on the beta in all four sectors. Some of the significant coefficients are larger than unity. During this time the market volatility also imposes large size effects but of an inverse nature. During the other periods, the effect of the industry volatility is mostly positive and smaller. The market volatility effect is negative and in absolute value smaller. The global volatility imposes no effect at all during any period. The constant is again significant in all tests. The R2 ranges from 0.079 to 0.339. The largest R2 is for the crisis period tests. Once again, the Durbin–Watson statistics are also quite satisfactory.16 16

The largest VIF is 1.51 and thus no indication of multicollinearity in these tests.

T. Choudhry et al. / Japan and the World Economy 22 (2010) 228–234

233

Table 3 Singapore. Time

Constant

CV

Chemical 1992–97 1997–98 1998–2002 1992–2002

1.0115*** 1.0274*** 1.0004*** 1.0100***

(791.851) (147.9722) (877.079) (999.313)

Finance 1992–97 1997–98 1998–2002 1992–2002

1.0007*** 1.0176*** 1.0043*** 1.0066***

(1644.618) (283.553) (1317.379) (1928.485)

Retail 1992–97 1997–98 1998–2002 1992–2002

0.9554*** 0.9539*** 0.9571*** 0.9558***

Industrial 1992–97 1997–98 1998–2002 1992–2002

0.9930*** 1.0022*** 0.9895*** 0.9923***

R2

DW

(23.7522) (8.2049) (8.8065) (24.7336)

0.285 0.247 0.094 0.377

1.76 1.90 2.02 1.97

0.4733*** 0.5691*** 0.2359*** 0.3464***

(20.282) (11.0258) (8.0713) (19.5716)

0.226 0.311 0.372 0.584

1.85 2.17 1.97 2.00

0.3995*** 0.2835*** 0.1458*** 0.1639***

(16.4433) (4.6885) (4.8652) (8.7143)

0.166 0.088 0.510 0.512

1.92 2.03 1.95 1.97

0.5473*** 0.6157*** 0.2343*** 0.4494***

(24.6782) (12.4800) (8.1447) (26.8563)

0.301 0.370 0.230 0.658

1.75 2.19 1.98 2.00

MV

GV

0.0060 (0.1329) 0.5397*** (5.3441) 0.1296*** (4.0712) 0.1999*** (8.1696)

0.0238 0.4789 0.0322 0.0700

(0.2256) (1.1110) (0.3916) (0.9966)

0.5326*** 0.4623*** 0.2497*** 0.4199***

0.0220 (0.4546) 0.0602 (0.6919) 0.8941*** (28.9218) 0.7147*** (30.1995)

0.0274 (0.5668) 0.0601 (0.6775) 0.8989*** (27.4469) 0.7191*** (29.9900)

0.0343 (0.5857) 0.1175 (0.6737) 0.2192*** (3.9196) 0.2038*** (4.9012)

(1813.1712) (368.261) (881.353) (1675.113)

0.0416* (1.7304) 0.0715 (0.8757) 0.9360*** (37.0288) 0.7810*** (42.7171)

0.0353 (1.4673) 0.0450 (0.5438) 0.9754*** (33.2624) 0.7750*** (41.0700)

0.0372 (0.6409) 0.1401 (0.6338) 0.1994** (2.2322) 0.0790 (1.3368)

(1488.227) (296.602) (1800.634) (2006.972)

0.0237 (0.6513) 0.0220 (0.2863) 0.5641*** (19.0471) 0.3692*** (16.8557)

0.0191 (0.5240) 0.0207 (0.2629) 0.5786*** (19.0344) 0.3762*** (16.9829)

0.0027 (0.0590) 0.5657*** (5.8500) 0.1666*** (5.6142) 0.2183*** (9.1068)

Rho

0.0045 0.0277 0.0028 0.0053

(0.0840) (0.1916) (0.0683) (0.1632)

CV = conditional variance of the industrial sector. MV = conditional variance of the market. GV = conditional variance of the global market. * Significant at the 10%. ** Significant at the 5%. *** Significant at the 1%. Table 4 Taiwan. Time

Constant

CV

Chemical 1992–97 1997–98 1998–2002 1992–2002

0.9888*** 0.9945*** 0.9906*** 0.9902***

(437.618) (173.804) (334.207) (571.2691)

Finance 1992–97 1997–98 1998–2002 1992–2002

1.0290*** 1.0228*** 1.0101*** 1.0206***

Retail 1992–97 1997–98 1998–2002 1992–2002 Industrial 1992–97 1997–98 1998–2002 1992–2002

Rho

R2

DW

(0.5264) (1.1339) (1.0877) (0.2923)

0.2082*** (8.0345) 0.1439** (2.3076) 0.2518*** (8.9024) 0.2214*** (12.1457)

0.047 0.028 0.070 0.109

2.03 1.97 1.98 2.00

(0.1983) (0.0124) (1.1922) (0.7774)

0.2687*** 0.2179*** 0.2152*** 0.2467***

(10.512) (3.5639) (7.5331) (13.6027)

0.082 0.151 0.067 0.130

2.06 2.04 2.02 2.05

0.2503*** 0.3177*** 0.2588*** 0.2892***

(9.7223) (5.3663) (9.1084) (16.1410)

0.064 0.091 0.083 0.113

2.042 2.04 2.04 2.05

0.2684*** 0.3289*** 0.2300*** 0.2622***

(10.5073) (5.5209) (8.0672) (14.5415)

0.085 0.130 0.051 0.172

2.04 2.08 2.01 2.04

FV

GV

0.3188** (2.1354) 0.0965 (0.2879) 0.5329*** (3.7680) 0.4013*** (4.2884)

0.1577 (1.0862) 0.4292 (1.1630) 0.3338* (1.9369) 0.2819*** (2.6880)

0.1770 0.6675 0.2304 0.0488

(380.259) (165.653) (396.085) (561.1229)

0.0109 (0.0738) 2.1519*** (5.7566) 0.6075*** (4.0504) 0.4375*** (4.26061)

0.3349** (2.0164) 2.3773*** (5.8630) 0.2956* (1.7470) 0.1954* (1.7000)

0.0734 0.0072 0.2284 0.1315

0.8629*** 0.8360*** 0.8292*** 0.8466***

(296.528) (107.336) (237.509) (373.406)

0.0161 (0.1089) 0.2639 (0.6814) 0.4341*** (2.6755) 0.2354** (2.2700)

0.0869 (0.5481) 0.3102 (0.8774) 0.6390*** (3.8932) 0.2608** (2.4246)

0.8271** (2.0228) 0.3246 (0.5133) 0.4580* (1.8535) 0.1900 (0.9535)

0.9972*** 1.0014*** 1.0101*** 1.0029***

(507.690) (223.881) (566.693) (769.561)

0.4882*** (3.6502) 0.8357** (2.5010) 0.1720 (1.2059) 0.1680* (1.7828)

0.6968*** (5.1400) 1.0860*** (3.2775) 0.2313 (1.4589) 0.2540*** (2.5903)

0.0410 0.5726 0.0017 0.0088

(0.1527) (1.5976) (0.0129) (0.0749)

CV = conditional variance of the industrial sector. MV = conditional variance of the market. GV = conditional variance of the global market. * Significant at the 10%. ** Significant at the 5%. *** Significant at the 1%.

Singapore provides somewhat different results (Table 3). Results show that the chemical sector is the most affected and the remaining three sectors less so. Conditional volatility of most industries provides a negative effect, with the exception of for the chemical sector. No significant effect is found during the crisis period except for the chemical sector. The size of the significant coefficients is smaller than unity (in absolute value). Conditional volatility of the local market provides the opposite result. This time the significant effects are positive except for, again, the chemical sector. The size of the effects is smaller than unity. The global factor shows minimal significant effect. Notably the financial sectors imposed a positive effect during the post-crisis and over the total period. The size of the coefficients is less than unity. Like the previous two countries, the constant is significant in all tests. The

range of R2 is from 0.088 to 0.658 with the total period showing the largest ones for all sectors. The Durbin–Watson statistics are satisfactory.17 The Taiwanese results in Table 4 shows that the financial sector was affected the most and the retail sector the least. The significant effect of the conditional volatility of the industry is mostly positive and less than unity in size. The significant market volatility coefficients are mostly negative and smaller than unity in absolute value. Once again, the global effect is very little and all constant terms are significant. The range of the R2 is from 0.028 to 0.172. The Durbin–Watson is satisfactory.18 17 18

The highest VIF is 2.9. The highest VIF is 1.2.

234

T. Choudhry et al. / Japan and the World Economy 22 (2010) 228–234

We then conduct further tests to check for errors in the variables.19 The Hausman (1978) test and the Hansen (1982) test are employed to check for error-in-variables. The Hausman (1978) test is based on looking for a statistically significant difference between an efficient estimator under the null hypothesis of no misspecification, and a consistent estimator under the alternative hypothesis that misspecification is present. The Hansen (1982) test investigates whether or not the assumptions can be held in a sample test of the over-identifying restriction. Both the Hausman test and the Hansen test fail to show any error in the variables. Thus, results indicate very few cases of specification problems. These results are available on request. A comparison of results across the countries and industries indicates that Korea is the most affected and Indonesia the least.20 The global factor is an important factor to some extent for Indonesia and Singapore. During the crisis period (1997–98), all four Korean industries were affected by volatilities but in Indonesia and Singapore only one industry was affected. During the post-crisis period (1998–2002) all four countries were heavily affected.21 Excluding the Indonesian results, at least one of the volatilities influenced all industries of the remaining countries during all periods under consideration. For Indonesia, all industries were affected by one of the volatilities during the post-crisis period only. Our results show that along with betas of the firms in the Far East, the industrial sectors’ betas were also influenced during the Asian crisis of 1997–98. This result has implications for portfolio management involving industrial sector shares during financial crises. 5. Conclusion and implications This paper investigates whether the 1997 Asian financial crisis and the post-crisis period had any effect on the time-varying beta of the four selected industrial sectors (chemical, finance, industry and retail) of Indonesia, Singapore, South Korea, and Taiwan. We apply daily data ranging from 1 January 1992 to 30 December 2002, and create daily time-varying betas by means of the bivariate BEKK GARCH model. One main impact of the 1997–98 Asian crisis was an increase in the volatility of financial markets and capital flows around the world. Investors may perceive a rise in the stock market volatility as an increase in the risk of equity investment. We employ standard linear regression to investigate the effects of the financial crisis on the time-varying beta. Change in the beta is investigated by using the conditional volatility of the industry, the local market and the global factor as explanatory variables in the regression. The conditional volatilities are estimated by means of the BEKK bivariate GARCH model. We then further divide the total period into three smaller sub-periods, pre-crisis (1992–97), crisis (1997–98) and post-crisis (1998–2002). We conduct estimations for all three periods to investigate and compare the potential changes in the effects of volatility on the industrial beta from the pre- to the postcrisis period. Results show that during the crisis period (1997–98) there was a substantial increase in the positive effect of conditional volatility of the individual industry on the time-varying betas. There is also some evidence of an increase in the effect during the post-crisis period. Local market volatility mostly seems to impose a negative influence on the beta. Very little significant effect is found for the global factor. Of the four countries, Indonesia is the least affected.

19 If the explanatory variables are correlated with the error term and the least square estimate is biased and inconsistent, then the explanatory variables are measured with error. 20 The Korean result is understandable given it was one of the most effected countries but as explained earlier the Indonesian result is surprising. 21 Given the small size of the crisis, effect on Singapore the weak result is understandable.

Similar to the results presented in this paper, Choudhry (2005) and Maroney et al. (2004) also show increases in the beta of Asian companies and industrial sectors during the Asian crisis period. The increase in the market risk (beta) due to higher stock return volatility has implications for the decisions of investors in relation to portfolio risk management, as well as for the firms in financial operations (Chen and So, 2002). If market risk increases then investors will demand higher returns, or, alternatively, investors will form new investment portfolios to achieve their expected utility of wealth. The results presented also advocate further research in the area. Further research can be conducted using data from other Asian or non-Asian markets, data from individual firms, investigation of other financial crises, and perhaps also by means of a different method of estimation, etc. The current financial crisis does advocate a similar empirical investigation. References Bodurtha, J., Mark, N., 1991. Testing the CAPM with time-varying risk and returns. Journal of Finance 46, 1485–1505. Bollerslev, T., 1988. On the correlation structure for the generalized autoregressive conditional heteroscedastic process. Journal of Time Series Analysis 9, 121–131. Bollerslev, T., Engle, R., Nelson, D., 1994. ARCH Models. In: Engle, R.F., McFadden, D.L. (Eds.), Handbook of Econometrics, Vol. IV. Elsevier Science. Bos, T., Newbold, P., 1984. An empirical investigation of the possibility of stochastic systematic risk in the market model. Journal of Business 57, 35–41. Chen, C., So, R., 2002. Exchange rate variability and the riskiness of US multinational firms: evidence from the Asian financial turmoil. Journal of Multinational Financial Management 12, 411–428. Choudhry, T., 2005. Time-varying beta and the Asian financial crisis: investigating the Malaysian and Taiwanese firms. Pacific-Basin Finance Journal 13, 93–118. Engle, R., Kroner, K., 1995. Multivariate simultaneous generalized ARCH. Econometric Theory 11, 122–150. Fabozzi, F., Francis, J., 1978. Beta as a random coefficient. Journal of Financial and Quantitative Analysis 13, 101–116. Giannopoulos, K., 1995. Estimating the time-varying components of international stock markets risk. European Journal of Finance 1, 129–164. Hansen, L., Richard, S., 1987. The role of conditioning information in deducing testable restriction implied by dynamic asset pricing models. Econometrica 55, 587–614. Hansen, L., 1982. Large sample properties of generalized method of moments estimators. Econometrica 50, 1029–1054. Hausman, J., 1978. Specification tests in econometrics. Econometrica 46, 1251– 1272. Kaminsky, G., Schmukler, S., 1999. What triggers markets jitters? A chronicle of the Asian crisis. Journal of International Money and Finance 18, 537–560. Klemkosky, R., Martin, J., 1975. The adjustment of beta forecasts. Journal of Finance 30, 1123–1128. Lintner, J., 1965. The valuation of risk assets and the selection of risky investments in stock portfolios and capital budgets. Review of Economics and Statistics 47, 13–37. Markowitz, H., 1952. Portfolio selection. Journal of Finance 7, 77–91. Maroney, N., Naka, A., Wansi, T., 2004. Changing risk, return and leverage. The 1997 Asian financial crisis. Journal of Financial and Quantitative Analysis 39, 143– 166. Muth, J., 1961. Rational expectation and the theory of price movements. Econometrica 29, 1–23. Neiss, H., 2009. Conclusion. In: Carney, R. (Ed.), Lessons from the Asian Financial Crisis. Routledge, London. Rosenberg, B., Guy, J., 1976a. Prediction of the beta from investment fundamentals. Part 1. Financial Analysts Journal 32, 60–72. Rosenberg, B., Guy, J., 1976b. Prediction of the beta from investment fundamentals. Part 2. Financial Analysts Journal 32, 62–70. Sharpe, W., 1964. Capital asset prices: a theory of market equilibrium under conditions of risk. Journal of Finance 19, 425–442. Susmel, R., Engle, R., 1994. Hourly volatility spillovers between international equity markets. Journal of International Money and Finance 13, 3–25.