Evolutionary System for Generating Investment Strategies Rafal Dre˙ zewski, Jan Sepielak Department of Computer Science AGH University of Science and Technology, Krak´ ow, Poland drezew@agh.edu.pl Abstract. The complexity of generating investment strategies problems makes it hard (or even impossible), in most cases, to use traditional techniques and to find the strict solution. In the paper the evolutionary system for generating invest- ment strategies is presented. The algorithms used in the system (evolutionary al- gorithm, co-evolutionary algorithm, and agent-based co-evolutionary algorithm) are verified and compared on the basis of the results coming from experiments carried out with the use of real-life stock data. 1 Introduction Investing on the stock market requires analyzing of the great number of strategies (which security should be chosen, when it should be bought or sold). Accurate anal- ysis is important during predicting and choosing the optimal investment strategy. It plays an important role in a future success. Majority of the investment decisions are based on present and historical data. The trend anticipation depends on many assump- tions, parameters and conditions. Consideration of so many assumptions, combinations of parameters and their values leads to the comparison of the great number of graphs. The evaluation of parameters of many securities is difficult and time consuming for the investor and the analyst. As a result, the investor or the analyst is able to analyze only the small subset of the possible strategies, so the optimal investment strategy is usually not found [9]. The set of the strategies which consists of indicator function is infinite because the complexity of the strategy can be unlimited. Formulas of the given strategy are functions of hundreds (or thousands) of parameters. Complexity of the problem makes it impossible to use direct search methods and instead of it a heuristic approach must be used. For example, it is possible to apply here evolutionary algorithms because there exist many solutions to the problem and finding optimal solutions is not necessary— suboptimal solutions are usually sufficient for an investor. Evolutionary algorithms are optimization and search techniques, which are based on the Darwinian model of evolutionary processes [2]. One of the branches of evolutionary algorithms are co-evolutionary algorithms [7]. The general difference between them is the way in which the fitness of the individual is evaluated. In the case of evolutionary al- gorithms the fitness of the individual depends only on how “good” is the solution of the given problem encoded within its genotype. In the case of co-evolutionary algorithms