Genetic algorithm learning and the chain-store game

In this paper the nature of predatory pricing is analyzed with genetic algorithms. It is found that, even under the same payoff structure, the results of the co-evolution of weak monopolists and entrants are sensitive to the representation of the decision-making process. Two representations are studied in this paper. One is the action-based representation and the other the strategy-based representation. The former is to represent a naive mind and the latter is to capture a sophisticated mind. For the action-based representation, the convergence results are easily obtained and predatory pricing is only temporary in all simulations. However, for the strategy-based representation, predatory pricing is not a rare phenomenon and its appearance is cyclical but not regular. Therefore, the snowball effect of a little crazinness observed in the experimental game theory wins its support from this representation. Furthermore, the nature of predatory pricing has something to do with the evolution of the sophisticated rather than the naive minds