Bootstrapping variance ratio test

Abstract

This paper proposes the use of bootstrap meths to improve the finite-sample properties of the variance ratio test. Based on Hall and Horowitz's (1996) adjustments. we show that the bootstrapped variance ratio test outperforms asymptotic tests by having relatively reliable rejection rates in the presence of serial dependence in the data. As an application, the bootstrap method is

applied to reexamine the predictability of stock returns in U.S. market. Us- ing monthly return data, we find positive serial correlation for short-horizon

returns, a result consistent with Lo and MacKinlay (1988) and Poterba and

summers (1988). Furthermore, the phenomenon is more evident for the equal- weighted index, which confirms the results of Keim and stambaugh (1986) and

Lo and MacKinlay (1988), and more significant in the post-war period, as doc- umented in Kim and Nelson (1998).