本文旨在探討在資本資產定價模型架構之中應用八種不同的β係數估計模型對於預測台灣個股系統性風險的能力以及對於台股投資者建立避險策略的有效性。研究樣本對象為1991年1月至2021年7月之所有台灣上市及上櫃共1,754檔股票。β係數的估計方法大致分別為最小平方法、貝葉斯收縮法、縮尾處理法以及衰減參數縮尾處理法,再利用前期估算的β係數對未來估算的β係數進行迴歸預測分析,從而篩選出預測性能優異的估計模型並將之用於建立市場避險投資組合,並回測模型成效。 實證結果顯示,已實際應用於其他已開發國家股市的β係數估計方法同樣可應用於台灣股市。而使用縮尾處理與衰減參數能有效提升β係數之預測準確度,至於使用貝葉斯收縮後所得出的預測準確度亦位於前列。在資料頻率使用方面,日報酬率在估算β係數上的實用性明顯高於月報酬率。最後,在建立市場避險投資組合的應用中,我們發現上述預測表現優異的β係數模型能有效降低投資組合報酬的波動率。總體而言,本研究驗證的β係數預測模型可供台股投資者作為評估系統性風險之指標,同時投資者應妥善衡量並考慮其它投資因子,以期達到更佳的投資組合風險管理成效。
This thesis studies the predicting performances of eight CAPM beta estimators and their effectiveness in hedging strategies for Taiwanese investors. Our sample ranges from 1991 to 2021, covering 1,754 listed companies. We use a combination of ordinary least squares, Bayesian shrinkage, slope winsorization, and decay parameter method to estimate CAPM betas of individual stocks. By running predictive regression, we test whether these ex-ante beta estimates can reliably predict their ex-post counterparts. Finally, we use these betas to form market-neutral stock portfolios and evaluate their hedging performances. Our analysis brings four major conclusions. First, the beta estimators applied in other stock markets of developed countries are also effective in the Taiwanese stock market. Second, slope winsorization and decay parameters can significantly improve the predictive accuracy of beta estimates, while Bayesian shrinkage estimators also perform well. Third, betas estimated from daily data are better than those estimated from monthly data in predicting future betas. Forth, beta estimates with good predictive performances can effectively reduce variances of beta-hedged portfolios. Overall, this thesis provides a general guideline for improving the measurement of stock-level systematic risk through the lens of CAPM betas in Taiwan. Taiwanese investors can better manage their portfolio risks by monitoring market betas and alternative investment factor exposures.