On the Utility of Trading Criteria Based Retraining in Forex Markets

This research investigates the ability of genetic programming (GP) to build profitable trading strategies for the Foreign Exchange Market (FX) of three major currency pairs (EURUSD, USDCHF and EURCHF) using one hour prices from 2008 to 2011. We recognize that such environments are likely to be non-stationary. Thus, we do not require a single training partition to capture all likely future behaviours. We address this by detecting poor trading behaviours and use this to trigger retraining. In addition the task of evolving good technical indicators (TI) and the rules for deploying trading actions is explicitly separated. Thus, separate GP populations are used to coevolve TI and trading behaviours under a mutualistic symbiotic association. The results of 100 simulations demonstrate that an adaptive retraining algorithm significantly outperforms a single-strategy approach (population evolved once) and generates profitable solutions with a high probability.