Trading Strategy Mining with Gene Expression Programming
Changhao Huang, Chang‐Biau Yang, Hung‐Hsin Chen · 2013
In the paper, we study the investment on Tai- wan Stock Exchange Capitalization Weighted Stock Index (TAIEX), which is assumed to be tradable. We apply the gene expression programming (GEP) to mining profitable trading strategies in the training phase. GEP is a good tool for evolving formulas since the logical view of its chromosome is a tree structure and the physical implementation is a linear string. In the testing phase, we find out some template intervals from historical data series which are similar to the leading interval. Then the trading strategies extracted from the template intervals are invoked to determine the trading signal. To keep stability of investment return, we perform a simple majority vote on a set of trading strategies. The testing period contains TAIEX starting from 2000/9/14 and ending on 2012/1/17, more than eleven years. In our experiments, the lengths of training intervals are 60, 90, 120, 180, and 270 trading days. The best cumulative return 236.25% and the best annualized return 10.63% occur when each training interval has 180 days, which are higher than the cumulative return 0.96% and annualized return 0.08% of the buy-and- hold strategy. The experimental results show that our method with higher voting threshold can usually make profitable trading decisions.