Three essays in financial market prediction.

摘要

The dissertation comprises three essays, each of which addresses a specific problem in financial market prediction. The first essay proposes to apply a multi-step optimization method, Maximization by Parts (MBP), to estimate the Copula-GARCH models. The Copula-GARCH models allow very flexible joint distributions by splitting the marginal behaviors from the dependence relation. The Inference Functions for Margins (IFM) method is broadly adopted to estimate the Copula-GARCH models. This paper will show that the IFM method is subject to small-sample biases. I propose to apply the MBP method to estimate the models. The efficiency gain of the MBP method is supported by both simulation and empirical studies. The procedures described here are applied to the daily returns of U.S. and Canadian stock markets.; The second essay (co-authored with Qi Zhu) introduces a new estimator for the predictive regressions. The method of least squares is subject to small-sample biases in predictive regressions with highly persistent regressors. This paper presents a bias-reduced estimator that minimizes the weighted sum of squared autocorrelations of the fitted residuals. Consistency and asymptotic normality of this estimator are established under the assumption of serially uncorrelated innovations. The Monte Carlo studies demonstrate that the finite sample performance of our estimator is better than the existing methods. Our method provides weak evidence on the predictability of stock returns during the post-war period.; The third essay (coauthored with Richard Luger) suggests a two-stage method for the Value-at-Risk (VaR) estimation. The estimation of the VaR uses the square root of the variance and is subject to the non-linear transformation bias, due to a Jensen's inequality effect. This paper proposes a two-stage estimation procedure to reduce the VaR estimation bias in conventional GARCH models. The first-stage model forecasts the standard deviation directly to avoid the non-linear transformation bias. We further construct a second-stage model via quantile regression by including selected instrument variables. We illustrate the use of this two-stage model in international stock indices, foreign exchange rates, and individual stocks. The empirical results support the effectiveness of this two stage model.