Three essays in agent-based market analysis

This Thesis applied the agent-based methodology to the evaluation and the design of limit order market. Chapter 1 studies the consequences of different behavioral assumptions about agents’ trading strategies and of pre-trade quote disclosure on market performance. Information disclosure is not always beneficial, depending on the trading behavior. Chapter 2 studies the continuous double auction from the point of view of market engineering: we tweak a resampling rule often used for this exchange protocol and search for an improved design. We consider two families of resampling rules and obtain the following main result: a resampling rule based on a price band around the best quotes is superior. Chapter 3 proposes and computationally applies a genetic algorithm to test if learned agents’ behavior converges to the same equilibrium strategy that is derived analytically in the paper by Foucault et al. (2005). As a second aim, we propose a model that relaxes one of the assumptions of that model: traders are allowed to cancel a placed order. We study how strategies are affected by the new rule. Results with and without cancellation are also compared.