Evolution of Bargaining Strategies for Double Auction Markets Using Genetic Programming

In market-based control systems ideas from microeconomic theory are used to solve resource allocation problems, e.g., routing in telecom networks. Resource allocation is achieved in a decentralised fashion where selfish software agents buy and sell the resources and hence no central planning is necessary. The performance of market-based systems generally depends on the bargaining strategies that the agents use and the market structure. Genetic programming is a method for automatic construction of computer programs by means of artificial evolution. A population of solutions, represented as actual computer programs, competes and the best solutions are then selected for reproduction and subjected to different evolutionary operators, e.g., mutation and crossover. This continues until a solution that performs well enough has been evolved. This thesis discusses how genetic programming can be used to evolve bargaining strategies for double auction markets, which is the most commonly used market structure in market-based control systems. We also compare the evolved strategy with a strategy described in the literature in two different problems. The first is a problem previously described and the second a more complex problem with a more dynamic market. In both problems both strategies perform nearly identical with regards to speed and trading price. In the more dynamic market the evolved strategy performs slightly better than the strategy from the literature, regarding the variation of the actual trading price from the theoretical equilibrium price. 1 This thesis corresponds to 20 weeks of full-time work. i