Portfolio allocation using XCS experts in technical analysis, market conditions and options market

Schulenburg (2000) first proposed the idea to model different trader types by supplying different input information sets to a group of homogenous LCS agent. Gershoff (2006) investigated this idea further with XCS agent. This paper takes an extra step to build a trading system that not only adopts the multi-XCS agent idea, but also utilizes knowledge from discretization theory, modern portfolio theory, options theory and methods of combining multiple models. In comparison to previous work, a wider range of input data were used including technical analysis, general market conditions and options market conditions. Secondly, quantization of continuous financial series was achieved using entropy-based discretization and histogram equalization. Thirdly, subtle investment strategies can now be generated as a result of taking stock price magnitude into account. Finally, multiple agents' predictions were combined using a variant of stacking. Empirical results show the best-performing XCS agents always outclass benchmark agents in every stock examined. Variance is reduced after combining predictions from multiple models. The technical analysis XCS agent was able to replicate a well known technical trading rule widely used in the 60s.

[1]  Leo Breiman,et al.  Bagging Predictors , 1996, Machine Learning.

[2]  Ron Kohavi,et al.  Supervised and Unsupervised Discretization of Continuous Features , 1995, ICML.

[3]  Yoav Freund,et al.  Experiments with a New Boosting Algorithm , 1996, ICML.

[4]  David H. Wolpert,et al.  Stacked generalization , 1992, Neural Networks.

[5]  Prakash P. Shenoy,et al.  MODELING FINANCIAL PORTFOLIOS USING BELIEF FUNCTIONS , 2002 .

[6]  L. Marengo,et al.  A learning-to-forecast experiment on the foreign exchange market with a classifier system , 1997 .

[7]  Tung Wan Cheng,et al.  Strategy of futures trading mechanism using extended classifier system , 2004, 2004 2nd International IEEE Conference on 'Intelligent Systems'. Proceedings (IEEE Cat. No.04EX791).

[8]  Peter Ross,et al.  Strength and Money: An LCS Approach to Increasing Returns , 2000, IWLCS.

[9]  Stewart W. Wilson.  Classifier Fitness Based on Accuracy , 1995, Evolutionary Computation.

[10]  G. Kendall,et al.  Learning with imperfections - a multi-agent neural-genetic trading system with differing levels of social learning , 2004, IEEE Conference on Cybernetics and Intelligent Systems, 2004..

[11]  An-Pin Chen,et al.  Applying Extending Classifier System to Develop an Option-Operation Suggestion Model of Intraday Trading - An Example of Taiwan Index Option , 2005, KES.

[12]  W. Andrew,et al.  LO, and A. , 1988 .

[13]  Stewart W. Wilson,et al.  Learning Classifier Systems, From Foundations to Applications , 2000 .

[14]  Chiu-Che Tseng Piotr J. Gmytrasiewicz Chris Ching.  Refining Influence Diagram For Stock Portfolio Selection , 2001 .

[15]  An-Pin Chen,et al.  Using Extended Classifier System to Forecast S&P Futures Based on Contrary Sentiment Indicators , 2005, 2005 IEEE Congress on Evolutionary Computation.

[16]  Ian Witten,et al.  Data Mining , 2000 .

[17]  Fabio Roli,et al.  Stock Market Prediction by a Mixture of Genetic-Neural Experts , 2002, Int. J. Pattern Recognit. Artif. Intell..

[18]  P. P. Shenoy,et al.  Bayesian Network Models of Portfolio Risk and Return , 2000 .

[19]  Peter Ross,et al.  Explorations in LCS Models of Stock Trading , 2001, IWLCS.

[20]  Christopher Stone,et al.  Foreign Exchange Trading Using a Learning Classifier System , 2008, Learning Classifier Systems in Data Mining.

[21]  Peter Ross,et al.  An Adaptive Agent Based Economic Model , 1999, Learning Classifier Systems.