Applying Genetic Algorithm to Integrate Technical Indicators for Investment Decision

• Main Authors: Charles Chang, 張佑瑋
• Other Authors: T.P. Liang
• Format: Others
• Language: zh-TW
• Published: 1999
• Online Access: http://ndltd.ncl.edu.tw/handle/23892941037747273124

碩士 === 國立中山大學 === 資訊管理學系 === 87 === In this research, we apply genetic algorithms to integrate technical indicators to support investment decisions. The objectives are to examine whether genetic algorithms can perform well in investment and what are the major design factors that may affect the performance of the genetic algorithms. A prototype system was developed. We trained and tested the system using the daily trading data of the Taiwan Stock Exchange Weighted Price Index, from January 1991 to December 1998. Four major design factors have been studied in the research: (1) restricting the matching of technical indicators or not, (2) using the sharing fitness scaling schema or not, (3) the length of training period, and (4) the number of generations evolved. The results show that genetic algorithms outperform the strategies of buy-and-hold and using single technical indicator. Concerning the design factors, only “length of training period” has been found significant. The interaction between “restricting the matching” and “using the sharing fitness scaling schema” is also significant. We conclude that genetic algorithms are very promising to produce better investment performance if properly configured, especially under the following conditions: (1) To restrict the matching of technical indicators, (2) To use the sharing fitness scaling schema, and (3) Longer training period.