Computational Intelligence in Financial Forecasting and Agent-Based Modeling: Applications of Genetic Programming and Self-Organizing Maps

Created by W.Langdon from gp-bibliography.bib Revision:1.4202

  author =       "Michael Kampouridis",
  title =        "Computational Intelligence in Financial Forecasting
                 and Agent-Based Modeling: Applications of Genetic
                 Programming and Self-Organizing Maps",
  school =       "School of Computer Science and Electronic Engineering,
                 University of Essex",
  year =         "2011",
  address =      "UK",
  month =        nov,
  keywords =     "genetic algorithms, genetic programming",
  URL =          "",
  URL =          "",
  size =         "180 pages",
  abstract =     "This thesis focuses on applications of Computational
                 Intelligence techniques to Finance and Economics. First
                 of all, we build upon a Genetic Programming (GP)-based
                 financial forecasting tool called Evolutionary Dynamic
                 Data Investment Evaluator (EDDIE), which was developed,
                 and reported on in the past, by researchers at the
                 University of Essex. The novelty of the new version we
                 present, which we call EDDIE 8, is its extended
                 grammar, which allows the GP to search in the space of
                 the technical indicators in order to form its trees. In
                 this way, EDDIE 8 is not constrained into using
                 pre-specified indicators, but it is left up to the GP
                 to choose the optimal ones. Results show that, thanks
                 to the new grammar, new and improved solutions can be
                 found by EDDIE 8. Furthermore, we present work on the
                 Market Fraction Hypothesis (MFH). This hypothesis is
                 based on observations in the literature about the
                 fraction dynamics of the trading strategy types that
                 exist in financial markets. However, these observations
                 have never been formalised before, nor have they been
                 tested under real data. We therefore first formalize
                 the hypothesis, and then propose a model, which uses a
                 two-step approach, for testing the hypothesis. This
                 approach consists of a rule-inference step and a
                 rule-clustering step. We employ GP as the rule
                 inference engine, and apply Self-Organising Maps (SOMs)
                 to cluster the inferred rules. After running
                 experiments on real datasets, we are able to obtain
                 valuable information about the fraction dynamics of
                 trading strategy types, and their long and short term
                 behaviour. Finally, we present work on the Dinosaur
                 Hypothesis (DH), which states that the behavior of
                 financial markets constantly changes and that the
                 population of trading strategies continually co-evolves
                 with their respective market. To the best of our
                 knowledge, this observation has only been made and
                 tested under artificial datasets, but not with real
                 data. We formalise this hypothesis by presenting its
                 main constituents. We also test it with empirical
                 datasets, where we again use a GP system to infer rules
                 and SOM for clustering purposes. Results show that for
                 the majority of the datasets tested, the DH is
                 supported. Thus this indicates that markets have
                 non-stationary behaviour and that strategies cannot
                 remain effective unless they continually adapt to the
                 changes happening in the market.",
  notes =        "

Genetic Programming entries for Michael Kampouridis