Evaluation of Volatility Models: Evidence from Chinese Equity Markets

Abstract

This thesis aims to find the most appropriate model to estimate and forecast volatility in Chinese stock markets, and to investigate the differences between simple historical models and GARCH-type models.

The studied model collection includes seven models: Random Walk, RiskMetrics EWMA, GARCH, GARCH-in-mean, EGARCH, TGARCH and APARCH. The forecast performances of those models are then evaluated in seven different criteria including symmetric loss functions and asymmetric loss functions. Other measurements such as the forecast encompassing test is conducted to check whether GARCH-type models carry additional information than simple historical models. The whole evaluation process is conducted with two Chinese stock markets’ indices, namely the SSE composite index and the SZSE component index. The selected sample period with updated data spans from 04 March 2006 through 30 December 2016.

The empirical evidence shows that the Random Walk model has the worst performance among all studied models. Model Performance is highly sensitive to the choice of forecast error statistics. The asymmetric loss function suggests systematically over-prediction exists in the forecasts which might be caused by the choice of forecast period. GARCH models carry more information than the Random Walk model. But no significant evidence is found in this study to support that GARCH models carry additional information than the RiskMetrics EWMA model.