주식시장간 상관계수의 시간불변성에 대한 검증

Knowledge about movements of correlation has important implications for econometric modeling and the empirical study of financial issues. In Bollerslev (1990), the constant correlation assumption allows a simple parameterization of the conditional covariance matrix in multivariate generalized autoregressive conditional heteroskedasticity (GARCH)model. In contrast to other multivariate GARCH models, such as Vech representation (Engle et al., 1984) and BEKK representation (Engle and Kroner,1995), the constant correlation model involves a relatively small number of parameters and, moreover, the conditions for positive definiteness of covariance matrix are easy to impose. However, many empirical results including Longin and Solnik (1995) show that the assumption of constant correlation does not hold for some financial data. The correlation constancy is also an important problem in financial studies. The correlation structure among different national stock returns is a crucial factor in determining the gains from international portfolio diversification, which was studied by Levy and Sarnat (1970) and Longin and Solnik (1995) among others. One of the inputs required for international investments is the ex-ante measure of the correlation, which is usually estimated by ex-post measures. But its reliability as proxies for ex-ante measures depends on whether the international correlation structure is intertemporally stable. The constant correlation also plays an important role in the context of hedge ratio estimation. Under the constant correlation assumption, the hedge ratio could be estimated by the ratio of two univariate conditional standard deviations multiplied by the correlation between the spot and futures returns. These observations indicate that a formal test of the constant correlation assumption will be a very useful tool for correctly specifying financial models. The objective of this paper is to extend Bera and Kim (2002)’s test for the constancy of correlations to the multi-correlation case and apply the extended test to the 5 Asian stock markets. To circumvent the blow-up of the tests when applied to the financial data following fat-tailed distributions,studentizing was used instead of asymptotic variance in the original test. The test results on the unconditional correlation between Asian markets in the period 1985~2010 rejected the constancy of correlation, which agrees with most previous empirical studies. In the pre-crisis and postcrisis periods, however, the null of constant correlations are not rejected in most markets. The tests on conditional correlations provides the almost same results: the inter-market correlations do change over time in the whole sample period, but do not show significant movements in the subsamples of pre-crisis and post-crisis period. These empirical results imply that the relationships between markets was stable at low level in the pre-crisis period, and increased to a higer level during the Asian crisis. The high correlation stays continued afterwards. Contrary to previous studies mostly denying the constant correlations, this paper states that the constancy of correlations depends on which sample period was chosen to be tested.