The new development of econometrics and its applications in financial markets.

摘要

The dissertation consists of three chapters in current developments of financial econometrics. The common theme of these chapters involves in applying these new econometrics techniques in addressing real problems in financial markets.;The first chapter explores the new Bayesian econometrics technique and its application in mutual fund analysis. The flexibility of Bayesian econometric framework allows us to take mutual fund family strategies into consideration when evaluating performance of individual mutual fund. More specifically, we introduce an informative prior to accommodating different fund family strategies. In empirical test, the alphas from our Bayesian model are more persistent through time than traditional OLS alphas.;Chapter two introduces a new approach to replicating hedge fund returns. Specifically, we use low-frequency (monthly) models to forecast high-frequency (daily) hedge fund returns. This approach addresses the common problem that confronts investors who wish to monitor their hedge funds on a daily basis---disclosure of returns by funds occurs only at a monthly frequency, usually with a time lag. To fit monthly hedge fund returns with investable assets or factors, we apply a dynamic style analysis model, which provide us time-varying beta positions through time. Using hedge fund indexes, we show that our replication approach closely forecasts the actual daily returns of the indexes. We illustrate how our simple replication approach can be used to (1) hedge daily hedge fund risk; and (2) estimate and control value-at-risk.;Chapter three applies a basket of econometrics models to test different hypotheses on China's B-share discount puzzle. We first establish our hypotheses explaining the puzzle based on empirical evidence and theoretical arguments. Then, each hypothesis is associated with an explanatory variable and run through a fixed-effect panel estimation model. In addition to increasing degrees of freedom and reducing the collinearity among explanatory variables, panel-methods can improve the precision of estimates of model dynamics in short time-series. The significant explanatory variables are test further with Granger causality model. iv