MARKET STRUCTURE, PRICE DISCOVERY AND NEURAL LEARNING IN AN ARTIFICIAL FX MARKET

In this paper, we simulate a decentralized multiple dealership market using agent based model. Risk averse dealers receive order flow from customers, which can not be observed by the other dealers. Then dealers trade among themselves. Neural net-works are used to represent a decision model for each dealer. In the course of several different experiment design, we investi-gate a number of features of our agent-based model: informa-tional efficiency of the market and the impact of changing of market structure on this efficiency. Our simulated market is able to replicate some features of the experiments with human subject regarding informational efficiency.