Discovering Financial Technical Trading Rules Using Genetic Programming with Lambda Abstraction

Abstract

We applied genetic programming with a lambdaabstraction module mechanism to learn technical tradingrules based on S&P 500 index from 1982 to 2002. Theresults show strong evidence of excess returns overbuy-and-hold after transaction cost. The discoveredtrading rules can be interpreted easily; each rule usesa combination of one to four widely used technicalindicators to make trading decisions. The consensusamong these trading rules is high. For the majority ofthe testing period, 80percent of the trading rules givethe same decision. These rules also give hightransaction frequency. Regardless of the stock marketclimate, they are able to identify opportunities tomake profitable trades and out-perform buy-and-hold.