Genetic TSA

Financial time series analysis by Genetic Algorithms (GA) is characterised by large datasets with many non-linear dependencies. The default assumption that all genes are non-linearly linked leads to very large search times and intractability. Linkage learners can improve performance by identifying non-linear linkage, and dividing the search space where linkage does not occur. However, these learners are slow. The first paper below presents a novel perturbation-based linkage-learning algorithm that focuses on the semantics of composability, with consequential speed improvements, and the second paper applies linkage detection to financial time series:

  • gLINC: Identifying Composability using Group Perturbation (Coffin & Clack, in Proceedings Genetic and Evolutionary Computing Conference (GECCO'06), pp 1133-1140, ACM, 2006)
  • Nonlinearity linkage detection for financial time series analysis (Chiotis & Clack, in Proceedings Genetic and Evolutionary Computation Conference (GECCO'07), pp 1179-1186, ACM, 2007)

Christopher D. Clack
Department of Computer Science
UCL
Gower Street
London
WC1E 6BT