本研究嘗試應用類神經網路於臺灣股市發行量加權股價指數報酬之預測，探討統 合考量基本面與技術面資訊的類神經網路模式（統合模式）的預測績效。實証結果發現類神 經網路模式中的網路架構、連結加權初始值、學習速率與資料會交互作用，影響模式的預測 結果，至於各模式的預測績效，整體而言，類神經網路模式中，統合模式優於其它只分別考 量基本面資訊或技術面資訊的兩種模式。而以統合模型與同樣納入基本面與技術面預測變數 的無母數迴歸模式相比較，則前者優於後者。

　　　　　This paper uses backpropagation neural networks to make oneperiod ahead prediction of the returns of Taiwan Stock Index. The performance of the networks with both fundamental and technical indicators (the integrated model) is compared with that of the networks with either fundamental indicators (the fundamental model) or technical ones (the technical model). A nonparametric regression model is also used in the comparison to understand if the neural network has better predicting power than the statistical model. The empirical results show that architecture, initial random value, and learning rate have interactive influence on the performance of a network, but no systematic rule of the interaction is found. As for performance, the integrated model as a whole is superior to the other three models when gauging against prediction errors, market timing ability and abnormal returns generation.