Refining the Breeding of Hybrid Strategies

Albert E. Marks and David F. Midgley - University of South Wales and Lee G. Cooper - University of California at Los Angeles

In earlier work (Management Science, March 1997) we used a genetic algorithm and a market-response model to coevolve the actions (prices and promotional activities) of artificial brand managers in an iterated oligopoly. Despite the simplicity of the artificially bred brand manager strategies, the best of these strategies performed well in comparison with the actual historical actions of brand managers for major brands of canned ground coffee in a regional U.S. market. This paper reports the consequences of relaxing various restrictions (e.g. single population evolution) imposed on the artificially bred brand manager strategies. In particular, it is shown how profit performance improves as we allow brand specific responses with separately evolved populations of agents corresponding to separate brands.

Scheduled for Session 2.6 Agent - Based Computational Economics - I