投資人預測股價走勢的方法有很多,有人看技術面、有人分析基本面、也有人從國際股市的連動性來預測,而過去研究中,極少一起探討或比較總體因素、技術指標與外國股價指數,預測能力的優劣,因此本研究欲探討哪一類的變數對台股月報酬之預測能力最佳。本研究使用1988年8月到2007年6月之台股月資料結合總體經濟變數、技術指標變數、國外指數變數與全體變數,並利用倒傳遞類神經模型,結合移動窗格法來預測台股月報酬。研究結果顯示全體變數在訓練期間解釋能力、訓練期間預測方向的準確性、測試期間預測方向的準確性最好,其次依序為總體變數、指數變數、技術變數。而測試期間,平均絕對誤差百分比(MAPE)與誤差均方根誤差(RMSE)的結果並不一致。
Taiwan’s security market has been established more than 40 years. In this period have been many times bullish market and many times bear market. Everyone who buys stock knows the only way to make money in the stock market is buys stock at low price and sells stock at high. But investors don’t know when to buy stock and when to sell. This research used Back Propagation Neural Network to forecast the return of TAIEX, and used monthly data from August 1988 to June 2006 of the TAIEX, macroeconomic variables, Technical Indexes variables, Foreign Stock Indexes variables, all variables. From the result of the research, we have several findings. First, using the all variables to forecast has the best directional and explainable effect in training period and best directional effect in testing period. Second, the result of the Mean Absolute Percentage Error (MAPE) and root mean squared error (RMSE) are different in training period.