Evolving Marketplace Designs by Artificial Agents

I Declaration This dissertation is submitted to the University of Bristol in accordance with the requirements of the degree of Master of Science in the Faculty of Engineering. It has not been submitted for any other degree or diploma of any examining body. Except where specifically acknowledged, it is all work of the Author. Labs-Bristol). Some work had been done in the evolution of market mechanism: A continuous space of auction-types was introduced in an intelligent-agents-based E-market. Experiments have been done not only optimizing the parameters for trading agents, but also the style of market mechanism in which the trading agents operate. A new hybrid market mechanism was found, which differs from (never found before) the real world auctions [Cliff, 2001b]. The aim of this dissertation is to replicate these experiments to verify that Cliff's results are reliable. First I tested the some results from [Cliff, 2001a] to verify that my GA works well in finding the optimal 8-dimential parameters for fixed-mechanism market. I got results which are significantly better than the experiment based on educated guess parameters. In replicating the evolving mechanism experiment, three markets with different S-D schedules were tested based on GA. I got very similar results like Cliff's did in Market 2 and Market 3. Because of the limitation of computing facilities, I have reduced generation number and number of individuals within a population. The low-performance may cause failing to find some good ones in Market 1. However, the success of Market 2 and Market 3 has proved that Cliff's results on evolving market mechanism are reliable. And there is not a major difference of using different GAs. The evolved market mechanism is an internal character of the market, which is independent from the optimization techniques used. III Acknowledgements This dissertation would not have come into being without the guidance of my thesis supervisor, Dr Tim Kovacs. Tim has been always supportive in this project, he helps me to get this project from HP labs-Bristol (I can still remember that raining day, we lost our ways to HP), and the weekly meeting and feedback has greatly improved my work. And thank for his patience with my slow progress in this project. Labs-Bristol. I would like to thank Dave Cliff and Andrew Byde for all their valuable suggestions on this project, and especially for Dave's insightful explanation on the difference between our results of the experiments. I …