This thesis proposes a Threshold-Switching GARCH model which allows an asymmetric reaction of the conditional volatility to the arrival of news. In the Threshold-Switching GARCH model the method of capturing an asymmetric reaction is induced by both the slope and intercept of the equation of news impact curve to be changeable through the signs of past shocks. By an analysis of parameters of the equation of the news impact curve, it can be shown that the Threshold-Switching GARCH model can not only capture the different impacts of unexpected error terms but also the behavior of the inversion of asymmetry and news size bias effects. Furthermore, this thesis discusses the statistical characteristics, including the formulas of the unconditional second order moment, the fourth order moment, kurtosis of the innovation and the proof of the stationary process. In this thesis I use the data of Taiwan stock price to evaluate the validity of the Threshold-Switching GARCH model. The sample period is from January 6, 1987 to December 31, 1997, the number of which is 3151. With the application of the Threshold-Switching GARCH model, the extensive Threshold-Switching GARCH model and some asymmetry GARCH models to study the volatility of Taiwan stock price, the empirical results are shown that the Threshold-Switching GARCH model is better than some traditional asymmetric GARCH models in capturing the behavior of volatility. Moreover, the empirical results of volatility of Taiwan stock price indicate that the behaviors of volatility have GARCH effects, asymmetric reaction and news size bias effect. In order to test the robustness of the Threshold-Switching GARCH model, the subsample from October 11, 1989 to December 31, 1997 is chosen to discuss the structural change of the volatility behavior between the bear and the bull periods. The empirical result indicates that the level and persistence of volatility in the bear period are different from those in the bull period. The level of the volatility in the bear period is higher than that in the bull period. The volatility in the bull period is more persistent than that in the bull period. The test of the robustness in the subsample indicates that the fitness of the Threshold-Switching GARCH model is still better than that of the traditional GARCH models.