On the impact of streaming interface heuristics on GP trading agents: an FX benchmarking study

摘要: Most research into frameworks for evolving trading agents emphasize aspects associated with the evolution of technical indicators and decision trees / rules. One factors that drives development such is non-stationary, streaming nature task. However, it heuristics used to interface evolutionary framework data which potentially have most impact on quality resulting agents. We demonstrate including a validation partition has significant determining overall success Moreover, rather than conduct continuous basis, only retraining when changes in are detected also yields advantages. Neither these widely recognized by agent frameworks, although both relatively easy add current frameworks. Benchmarking over 3 year period EURUSD foreign exchange supports findings.