An XCS approach to forecasting financial time series

Preen, Richard

Abstract

This paper extends current LCS research into financial time series forecasting by analysing the performance of agents utilising mathematical technical indicators for both environment classification and in selecting actions to be executed in the environment. It compares these agents with traditional models which only use such indicators to classify the environment and exit at the close of the next day. It is proposed that XCS agents utilising mathematical technical indicators for exit conditions will not only outperform similar agents which close the trade at the end of the next day, but also result in fewer trades and consequently lower commissions paid. The results show that in five out of six assets, agents using indicator exit conditions outperformed those exiting at the close of the next day, before commissions were factored in. After commissions are factored in, the performance gap between the two agent classes further widens. Additionally, the agent's best results are continuously able to outperform a buy and hold strategy.

• Presentation Conference Type: Conference Paper (unpublished)
• Conference Name: GECCO '09: Proceedings of the 11th Annual Conference Companion on Genetic and Evolutionary Computation Conference: Late Breaking Papers
• Start Date: Jun 8, 2009
• End Date: Jun 8, 2009
• Publication Date: Jan 1, 2009
• Peer Reviewed: Peer Reviewed
• Pages: 2625-2632
• Book Title: Proceedings of the 11th annual conference companion on Genetic and evolutionary computation conference - GECCO '09
• DOI: https://doi.org/10.1145/1570256.1570372
• Keywords: computational finance, learning classifier systems, XCS
• Public URL: https://uwe-repository.worktribe.com/output/1005183
• Publisher URL: http://dx.doi.org/10.1145/1570256.1570372
• Additional Information: Title of Conference or Conference Proceedings : 11th Annual Conference on Genetic and Evolutionary Computation (GECCO '09)