Learning Cancellation Strategies in a Continuous Double Auction Market

This paper deals with two different issues. On one side, it tries to determine if the equilibrium order placement strategies analytically derived in Foucault et al. (2005) are learnable by no-maximizing agents that update their strategies on the only base of their own past experience (via genetic algorithm). Results state outcome (but not strategic) equivalence. On the other side, it relaxes the assumption in the original model by Foucault for which cancellation is not allowed and evaluate market performance. Results are mixed; the introduction of a cancellation option turns out to be benecial dependently on the key determinants of the market dynamic (i.e., the arrival rate and the percentage of patient traders) and an additional setup variable: the initial level of order aggressiveness in the market.