本研究主要是以台灣股票市場為對象，建立與時變動ß 係數的估計模型，並對估計模型進行比較以期獲得最佳的結果。由於ß 係數會因時間、報酬型態、估計期間的不同而變動，所以CAPM模型中假設ß 係數為固定值，是不斷遭到質疑與否定的。本研究主要是運用一般化自我迴歸異質變異數模型(GARCH model)、Schwert and Seguin 的延伸市場模式及卡爾曼瀘嘴法(Kalman filter approach)來建 ß 係數的估計模型，再對模型進行比較以期獲得最佳的估計結果，並進一步將類股報酬區分成多頭市場與空頭市場，討論類股系統性風險在夕空市場中的差異。經由實證結果可以發現，對多數類股而言，Schwert and Seguin模式在估計隨時間變動 ß 係數的績效上最佳，其預測績效優於GARCH模式及卡爾曼瀘嘴法，且在所有模型中，皆顯示分類股報酬在空頭市場的系統性風險大於多頭市場。
The purpose of this paper is to estimate a better time-varying model of beta for Taiwan stock market. As beta coefficient may change owing to different time, return and estimating period, a key assumption that beta in CAPM is constant is suspicions and denied. We use the GARCH, Schwert and Seguin model and Kalman filter approach to estimate time-varying betas, further examining the performance for each model to obtain a best result. And we break the market return into bull and bear market to discuss the differences of systematic risk in bull and bear market. Our empirical results show that the performance is best when using the Schwert and SEguin model to estimate for most classified indices, which is better than that of GARCH and Kalman filter approach. And the systematic risk is higher in bear market than in bull market in all models.
