本研究是要評量可觀察的總體經濟變數是否可以預測股票市場未來的走向，在此 研究對象是臺灣股票指數。在美國、西歐、日本股票市場相關研究中，各別有其總體經濟變 數具其解釋能力。本研究中採用一般動差法 (GMM)，來對臺灣股票指數不同期間報酬作時間 序列迴歸；並用 Kalman 準則預測法，以不同的總體經濟變數，來對不同期間的年複利超額 報酬率作檢測。 結果顯示在 GMM 模型中，有一些總體變數具有解釋能力，而且年複利報酬 率所採用的期間愈長，解釋能力愈佳；此外，若用年複利成長率所採用的期間愈長，解釋能 力也愈佳。而且在 Kalman 準則預測法中，顯示所投入的總體變數對於超額報酬的走向預測 具有 50 ％以上的統計顯著水準。這對於投資專家和基金經理人而言是相當有用的發現，因 為當預測未來超額報酬為負時，他們可以事先將資產轉移至債券市場；當預測未來超額報酬 正時，他們可以事先將資產轉移至股票市場。

　　　　　This study examines whether observable macroeconomic variables can be used to forecast the direction of different horizon returns of the Taiwan index. Different macroeconomic state variables similar to those that have shown to have some explanatory power in the U.S., Western Europe, and Japanese stock markets are constructed. Time series regressions of various horizon excess returns of the Taiwan index are performed against macroeconomic state variables using Generalized Method of Moments GMM. Out of sample Kalman filter forecasts for various horizon ahead continuously compounded annualized excess returns of the Taiwan index using various macroeconomic state variables as input are also generated. The results of this study suggest that the macroeconomic state variables examined tend to provide more explanatory power for longer horizon returns. In addition, the macroeconomic state variables constructed using longer horizon growth rates tend to provide more explanatory power than the macroeconomic state variables constructed using shorter horizon growth rates. Finally, out-of-sample results show that Kalman filter forecasts using several of the macroeconomic state variables as inputs are able to correctly predict the sign of the excess return more than 50 ％ of the time at a statistically significant level.