應用遺傳程式規劃於尋找台灣加權指數技術交易法則

Abstract

近年來遺傳程式規劃在財金領域上的應用備受重視，因其結構上的彈性特別適合解決在股票市場中尋找交易法則這類非線性的問題。 本論文在遺傳程式規劃的要素中引入技術指標，期望能在擴大解空間的優勢下使用代表台灣股市整體特性的台灣發行量加權股價指數以尋找適合應用於台灣股市的技術交易法則；在適存度函數的挑選上本研究使用悲觀性平均報酬率、平均報酬率以及報酬風險比當作引領演化的依據；在演化的過程中為了解決過適現象，本研究透過調整選擇方案的選擇壓力係數來緩和過適的發生；對於每個經由遺傳程式規劃所產生的技術交易法則則是藉由報酬風險比以及模型效率來評估其績效。在考量交易成本的狀況下，本研究發現使用此三種演化方式所得的技術交易法則大都能打敗單獨執行買進觀望策略所得的報酬。 當使用悲觀性平均報酬率為適存度函數時，發現產生的技術交易法則其模型效率相較於以其他兩種適存度函數進行演化所得到的技術交易法則更具有一致性；在使用以平均報酬率為適存度函數的實驗中，證實了使用以超額報酬做為評估適存度的方式並不適合台灣的股票市場；當觀察使用以報酬風險比產生的技術交易法則時，發現其風險與獲利評估標準在各應用期的皆展現較佳的一致性。

Recently, Genetic Programming has played major role on financial field. With its flexibility, the merit of tree structure, Genetic Programming has unparalleled ability to solve nonlinear problems, such as finding technical trading rules on stock markets. In this thesis, technical indicators are introduced into Genetic Programming for the purpose of expanding the search space. When finding technical trading rules on Taiwan stock market, we use TAIEX as the most representative raw data and choose PROM, ROM, and RRR as the fitness functions. In the progress toward evolution, we try to avoid overfitting by conditioning the selection pressure coefficient in the selection scheme. When evolution is over, RRR and model efficiency are used to evaluate the performance of each technical trading rule generated by Genetic Programming. Considering the transaction cost, we find that most technical trading rules derived from all of the fitness functions outperform the buy-and-hold strategy. First of all, when the evolution is led by PROM, the model efficiency of technical trading rules we gain are more consistent than those generated by the other fitness functions. Second, when the evolution is led by ROM, excess return adopted frequently in other research is not suitable on Taiwan stock market. The last but not the least, when the revolution is led by RRR, both the performance standard RRR and model efficiency show great consistence in each applied period.