台灣加權股價指數投資組合之基因演算法建構模型

Abstract

一個自由經濟的投資市場必然是由投機 ，避險與套利三大模組所 組成，三者缺一不可。然而為使此三大功能的運作順利，一個成功的指數 投資組合便成為投資領域上一門不可或缺的分析技術。在投機方面，指數 基金的推出使廣大的小額投資人也能享有全面投資的便利; 在避險功能上 ，此投資組合是調整風險值 B 的最佳標的；至於套利方面，搭配指數期 貨的基差套利更少不了一組足以代表現貨市場的指數投資組合。 故本研 究針對指數投資組合的建構方式，嘗試提出一個不同於以往線性迴歸下之 B 值分析模式，而代之以基因工程為基礎的人工智慧投資選股系統。經實 證結果，本研究在以台灣加權股價指數為標的物的模擬績效上，基因演算 法所構建的投資組合在組成個股支數少於五十支時，與大盤指數的相關係 數高達 97%，此結果較傳統之『未分層市場加權模型』及『分層市值加權 模型』更接近於大盤，顯示基因演算法的強大空間搜尋能力，的確在傳統 的投資群組理論外，為金融財務界提供了一套不同於以往的思考模式，而 以此角度出發，當能發展出一套更具效率與實用性的系統。 The purpose of this study is to use the powerful searching ability ofgenetic algorithm to construct a stock selection model for Taiwan stockindex portfolio. The genetic model will be compared with the Capitalization-Weighted-Nonstrtified Model and Capitalization-Weighted-Stratified Model,which are purposed by Meade and Salkin. From the data analysis, it was discovered that GA stock selection model is stable when its fitness fuction return value reaches a asymptotic limit.It shows that the GA stock selection model is reliable. In the results of comparism amount these models, the portfolio constructed from GA stock selection model is better than the other two to track the Taiwan stock index.In addition, the GA stock selection model has a great characteristic, thestability, no matter how many stocks the portfolio contents. The more stocksthe portfolio contents, the better tracking results are. All of the threemodel exhibit that it is exactly match the portfolio theorem. After the dataanalysis of GA stock selection model effective period and forecasting powerare considered, it shows that 288 days is the information cycle of Taiwan stock market. However, when the stock elements increase, the longer leaningdata-area is better. In summary, the powerful space-aearching ability of genetic algorithm is a really feasible manner to develop a constructing model in portfolio kingdom.