运用演化式计算在选股决策之研究.doc

L- PAGE 58 2002年管理創新與新願景研討會 2002年管理創新與新願景研討會 L- PAGE 51 運用演化式計算在選股決策之研究 The Study of Chose Stocks Decision Making by Evolutionary Computation 洪朝富　楊啟良　林木花 真理大學資訊管理學系 台北縣淡水鎮真理街32號 TEL:02#5531 E-Mail: chfong@.tw 摘要 傳統中所探討解決多準則決策的研究中，大部份以單目標最佳化模型追求多準則決策問題解為多，即為將多個準則合併成單一目標函數，並賦予每個準則權重，再加以實證而得出理想的最佳解。運用此類模型的一個最主要缺點是可能將原先在多準則空間上的兩個不相同解視為單目標空間上的同一個解。 本研究目的和單目標模型不同的是在於提出一個決策組合的模式，運用遺傳程式規劃、遺傳演算法解決多目標決策的問題，因此將以股票的投資組合為研究標的。由於在Hitoshi Iba的論文中證實了遺傳程式規劃有60%以上正確的預測能力，首先以遺傳程式規劃導出較準確的預測函數，其次在股票的排列組合的可行方案產生部份，採用遺傳演算法進行染色體的亂數產生、選擇、交換、突變等工作，根據獲利的Fitness評估決定染色體的存活機率，找出較佳的組合。最後以三十個交易日和兩支股票為限，求出的組合的確較一般的單一股票的買賣的獲利為高。 關鍵詞：多準則決策、遺傳程式規劃、遺傳演算法、預測函數、投資組合。 ABSTRACT In the traditional researches of the Multi-Objective Decision Making, the most of them were to use the Single-Objective optimized models to try to accomplish. That makes several criteria combine to be the Single-Objective function and weighting, each criterion prove them to figure out what is the ideal optimization. One of main weaknesses for the Single-Objective optimized models is to properly regard previous two different solutions in Multi-Objective space as the same one in the Single-Objective space. The difference between our studying purpose and the Single-Objective model is to address a decision combination, and to apply Genetic Programming, Genetic Algorithm, and the Multi-Objective decision theory. Due to the sixty-percent predictability of Genetic Programming proved in Hitoshi Iba’s essay, the prediction function is resulted through Genetic Programming. First, we use Genetic Programming to find most precise prediction function. The next, the sections of producing feasibility program in stack’s portfolios, adapting to random product, Section, Crossover, Mutation and so on evaluate percentages of chromosome through Fitness. Final, we use thirty trading date and two stacks to prove our portfolio’s earn profits better than single stack. Keywords: Multi-Objective Decision Making, Genetic Programming, Genetic Al