股票市場是一個現代最普遍額外獲利的投資管道,但投資人往往無法即時且準確的判斷買賣時機,而導致資金虧損,因此如何幫助投資者掌握買賣點且獲得穩定甚至更多的利潤顯得格外重要,然而股市環境變化迅速,故必須建立一套即時股市交易決策系統來提供投資人額外的參考資訊。過去許多研究中,透過資料探勘技術來辨識買賣點都有不錯的獲利空間,但仍然無法準確的掌握適當的買賣時機,而導致獲利狀況不穩定。而本研究嘗試利用多重預測模型的結合改善無法精確辨識買賣點時機的問題,期望能達成穩定且較高的投資利潤。 本研究主要是利用線段切割法(Piecewise Linear Representation)來從歷史片段中獲得股價轉折之時機點,並結合支撐向量迴歸(Support Vector Regression)技術來學習每日的交易知識,且透過TS模糊規則(Takagi-Sugeno fuzzy rule-based )來學習股票買點與賣點之交易知識以加強控制交易訊號,期望透過模型間交互作用,嘗試改善買賣時機偏差的問題。實驗的結果顯示本研究提出之交易決策模型確實優於其他模型,成功預測台灣及美國股票的最適買賣時機,其獲利狀況相當穩定。
The daily stock turning point detection problems are investigated in this study. The support vector regression (SVR) model has been applied in various forecasting applications and proved to be with stable performances. In this research, SVR has been used to predict the trading signal since it could handle overall information effectively even under the complex environment of stock price variations. The trading signals from the historic database is derived from the application of piecewise linear representation (PLR) of stock price. Therefore, the temporary bottoms and peaks of stock price within the studied period are identified by PLR. TS fuzzy rules were applied to calculate the dynamic threshold which intersects the trading signal and provides the trading points. The fuzzy rules were trained and obtained from the trading signals generated by PLR during the training period. A collaborative trading model of SVR and TS fuzzy rule is used to detect the trading points for various stocks of Taiwanese and America under different trend tendencies. The experimental results show our system is more profitable and can be implemented in real time trading system.