馬可維茲的均值方差理論是資產配置最初的架構，但是用這個架構可能會面對到一些挑戰像是對於最佳權重的估計誤差太大，為了要解決估計誤差的問題，我們透過馬可夫轉換模型去改善輸入資產的報酬，更確切來說，這個方法是用當期機率 (filtering probabilities) 和 Clarke & de Silva (1998) 去改善輸入資產的報酬。最後我們利用元大投信在台灣發行的四檔有名股票型基金去做回測並算出了考慮交易成本的策略統計量，結果顯示當資料有厚尾分佈和 GARCH 性質時，誤差項服從學生t分佈的馬可夫轉換 GARCH 模型有最好的投資組合報酬相較於其他方法。;Markowitz mean-variance framework is a debut for modern portfolio allocation, but using it may encounters several challenges such as the optimal weight is sensitive to the estimation errors of the model. To overcome the problem of estimation errors, we improve inputted log-returns using regime switching models. This method provides a way to estimate portfolio weight using filtering probabilities and Clarke and de Silva (1998). Finally, we conduct a backtesting study using four famous ETFs in Taiwan issued by Yuanta Securities Co., Ltd. and report the portfolio′s strategy summary statistics after accounting for transaction costs. The results show that when the data has fat tail distribution and GARCH effect, a regime switching GARCH(1,1) model with Student′s t innovations dominates the others methods in terms of our portfolio allocation.
