本論文探討整合型情緒指標對台股報酬的預測能力，並分析投資人情緒相較於傳統總體經濟指標的預測表現。本文以台灣加權股價指數超額報酬率為預測標的，樣本期間為2004年5月至2025年12月，評估情緒與總經指標之樣本內預測與樣本外預測之表現，比較等權重加權平均法 (EW)、主成分分析法 (PCA) 與偏最小平方法 (PLS) 之預測能力。實證結果顯示：第一，在樣本內預測表現方面，投資人情緒指標對股票報酬的預測力顯著優於傳統總體經濟指標；第二，樣本內預測結果顯示 PLS 方法的預測能力優於 PCA 與 EW 方法；第三，以均方差 (root mean squared error， RMSE) 評估樣本外預測表現之結果顯示，PLS 所建構的情緒指標並未優於 PCA 與 EW 方法。;This thesis evaluates the predictive ability of aggregate sentiment index on excess returns of Taiwan stock index， and compares the forecasting performance of investor sentiment relative to traditional macroeconomic fundamental indicators. Utilizing the excess returns of the Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX) as the forecast target， covering the period from May 2004 to December 2025， we evaluate in-sample and out-of-sample forecast performances of investor sentiment and macroeconomic indicators， among approaches of equal-weighted average (EW)， principal component analysis (PCA)， and partial least squares (PLS). The empirical findings are three-fold: First， in terms of in-sample forecast performance， the aligned investor sentiment index exhibits significantly outperforms the aligned index of macroeconomic fundamental indicators. Second， regarding the in-sample forecast performance of the PLS method is better than those of PCA and EW methods. Third， the out-of-sample forecast evaluation based on the root mean squared error (RMSE) shows the aligned investor sentiment constructed via PLS does not outperform both PCA and EW approaches.
