股票之預測模型，可幫助投資人在作投資決策時，可降低投資風險，本文鑑於人工智慧技術的開發，類神經網路的出現，解決了許多傳統人力不能解決的問題，且中國大陸每年經濟成長率約為8.8-14.25%，居亞太之冠，於是本研究的目的是將類神經網路應用於上海、深圳B股指數的預測，而作一實證的研究，給投資者作為參考。 本研究嘗試以類神經網路為研究方法，將它應用於中國大陸上海、深圳B股指數的預測，並與統計上的線性迴歸模式比較，以探討類神經網路模式預測績效是否優於線性迴歸模式。預測期間是上海、深圳B股於1997年10月至1998年12月的股票空頭市場；輸入變數方面，以相關分析法與逐步回歸法來選取類神經網路的輸入變數，複迴歸只以逐步回歸法選取最佳自變數組合，並設計積極交易法則與保守交易法則兩種來模擬買賣股票，以投資報酬率作為評估、比較。研究結果發現為： 1、 類神經網路模式與複迴歸模式中，所得到的投資報酬率與買入持有策略的結果比較，類神經網路模式與複迴歸模式本身都具備了預測能力。 2、只考慮預測漲跌方向性，複迴歸模式比類神經網路模式可得到的較佳結果，若考慮預測漲跌幅度，則類神經網路模式比複迴歸模式較佳結果。 3、類神經網路模式交易法則中，在空頭市場中，保守交易法則比積極交易法則可得到較佳報酬率。 4、類神經網路在選取輸入變數的比較，上海B股以逐步回歸法選取輸入變數可獲致較佳結果，深圳B股以相關分析法選取輸入變數可得較好結果，這顯示選取輸入變數的方法與股票預測模式存有某種關係，也就是類神經網路在預測時應以網路來自己來選取輸入變數，才能得到較佳結果。 Artificial neural network method is used in the paper to forecast of Shanghai and Shenzhen B shares. The forecast period used in this research is from October 1997 to December 1998. A stepwise regression is chosen to compare with the result of the correlation method and neural network method. In our research, investment return rate is used to measure performance on neural network model and multiple regression model. The result of our finding is as follows: 1、The comparison of investment return rates on buy-and-sell hold strategy between artificial neural network and multiple regression have proof the ability to predict. 2、As far as the direction of the stock will go, multiple regression have a better result than neural network. For measure the variability of the stock, the artificial neural network has better result than the multiple regression. 3、When the artificial neural network is used on the conservative mode have shown better investment return rate than the active mode. 4、For choosing variables the stepwise regression shown have better result on the Shanghai B shares. On Shenzhen B shares correlation method have shown better result than stepwise regression. In our model it have shown a relationship between the methods of choosing variables and the model used to forecast stock. In our conclusion we find the artificial neural network has learning ability which allow to improve the model performance than multiple regression. Our research is only serve as pioneer study to understand the neural network model and stock market on Shenzhen B shares and Shanghai B shares. A more extensive information and study is needed in the future research.
