資料挖掘是一種智慧型資料分析的方法，但是應用於證券投資領域上的研究相當有限。本研究利用關聯規則配合技術分析的方法來挖掘進出場規則，並以台灣股市驗證，發現在訓練期間內確實可以找出具高報酬的進出場規則。 資料挖掘只找出許多符合最小支持度和最小信心度的規則，缺乏一套有系統的方式去選擇在測試期應用的規則和驗證方式，此外，支持度或信心度高的規則並不保證高獲利。本研究利用遺傳演算法來挑選在測試期應用的規則，以證券投資問題為例，就是將挖掘出的進出場規則，挑選組合成最適的證券交易策略，並以移動視窗的訓練方式做長期的驗證。 目前遺傳演算法於證券交易策略的研究中，遺傳演算法雖然以其強大的解答搜尋能力成功的解決了部分的問題，然而由於先天上缺乏彈性的字串結構框架，必須對證券交易策略的架構有所限制，雖然透過遺傳程式規劃，在結構上提供的強大彈性，允許以組合或分解的方式，產生前所未有的新技術指標，可以改善必須對證券交易策略的架構有所限制，但是它所產生的證券交易策略卻令人難以理解。本研究先以關聯規則找出該股票適合的進出場規則，再以遺傳演算法組合成最適的證券交易策略，則可以發掘出易懂且具彈性的證券交易策略。 Data mining is an intelligent data analysis method whose applications research on investment are very few. We use association rules to find entry and exit rules which are then used to conduct experiments on real dataset obtained from Taiwan Stock Exchange (TSE). The experimental results show high returns in training and testing periods. Data mining can find many rules with minimum support and minimum confidence, but there are no systematic methods to select rules for use in testing period. Besides, rules with high support or confidence can’t make high return absolutely. We use genetic algorithms to select rules for constructing an optimal trading strategy. Finally, we use sliding windows for long-term verification. Although, genetic algorithms can solve some problems with its searching capability. However, it uses fixed bitstrings that induce constraint trading strategy’s framework. Genetic programming uses tree structure to generate new technical indicators that provide flexible trading strategy’s framework. However, it is hard to understand. We use genetic algorithms to select rules for constructing an optimal trading strategy which is understandable and flexible.
