Title:

\"Robust Statistical Inferences for Stock Market Prices:

An Exact Small-sample F-test and Prediction Analysis\"

Speaker:

Guo-Liang Tian, Ph.D

University of Maryland at Baltimore, USA

Time：

Dec. 20(Wednesday), 2006, 13:00-14:00

Location：

Room202, Guanghua Building

Abstract:

The variance ratio (VR) statistic is commonly used to test for the random walk (RW) hypothesis against non-RW alternative. Many applications using VR statistics employ asymptotic normal approximations to conduct statistical inferences, which are unreliable in finite samples and are often misleading. In this paper, we first derive the exact small-sample distribution of the VR statistic, which is a beta distribution under the iid Gaussian null hypothesis, and show that the resulting F-test is robust in the class of elliptical distributions. We then present several prediction issues of interest and then develop robust prediction bounds using frequentist approaches in the framework of parametric statistics. Non-parametric prediction approaches are also presented. Simulations are preformed to evaluate the empirical type I error rates (or the size) and powers for the exact F-test and the asymptotical normal test. Three data sets from Shanghai and Shenzhen Stock Exchange Indexes and Hong Kong Hang-Seng Index are used to illustrate the proposed methodologies.