以遗传演算法建构股票交易决策函数-淡江大学机构典藏.doc

以遺傳演算法建構股票交易決策函數 吳盛富1 葉怡成2 1中華大學資訊管理學系 wishwind999@.tw 2淡江大學土木工程學系 140910@.tw 中華大學資訊管理學系 chiuden@.tw Building Stock Trading Decision Functions Using Genetic Algorithms Sheng-Fu Wu1 I-Cheng Yeh2 Deng-Yiv Chiu3 1Department of Information Management, Chung Hua University (wishwind999@.tw) 2Department of Civil Engineering, Tamkang University (140910@.tw) 3Department of Information Management, Chung Hua University (chiuden@.tw) 以遺傳演算法建構股票交易決策函數 摘要本探討以遺傳演算法建構股票交易決策函數。並以近台灣股市為例進行實證。007/5台灣股市011/1台灣股市依時序分成二部分：前段為訓練期間；後段為測試範例。實證的結論如下：以遺傳演算法建構股票交易決策函數關鍵字：股票市場、遺傳演算法、決策。 tock Trading Decision Functions Using Genetic Algorithms Abstract This paper explores building stock trading decision functions using genetic algorithms. To evaluate the effects, we used 18 years of Taiwans stock market data. The first stage start from 1992/2 to 2007/5 of Taiwans stock market; the second stage start from1995/6 to 2011/1. Both of them were divided into two subsequent periods: the front-end as the training period; the rest as the test period. Empirical findings are as follows: (1) The empirical results of the first stage showed that the model really increasingly become more and more intelligent during the training period, and its intelligence is still effective during the test period. (2) The empirical results of the second stage showed that although the model really increasingly become intelligent during the training period, its intelligence is not effective during the test period. (3) Both of the two stages showed that whether the trading volume grew up or not is more important than whether the price rose up, especially for the selling strategy. That is, the rule of thumb volume grows up before the price rises up is more useful than the rule of thumb market price has a tendency. (4) The empirical results of first stage and second stage are not consistent. This indicated that using genetic algorithm to construct stock decision function faces over-fitting predicament often happened in