We are interested in the generic dynamical properties of the FX market. For this purpose, we have developed a computer microsimulation of an artificial market. Our modeling is built around the behavior of the real actors of the market: dealers and market makers. These actors are represented by computerized agents which interact together according to the rules of trading in effect in the FX market. The reason of this micro-level approach is the belief that the complex behavior of the FX market emerges as the result of the collective action of many ``nonlinear'' agents in mutual interaction, rather than due to the complicated and unknown nature of each individual. This view is a typical concept in the science of complexity.
In order to extract the fundamental properties of a FX market and identify the key ingredients, we first consider a simple model. Our agents work only with one pair of currencies. Their behavior is directed by non evolving strategies, but all of them can be different. The dealers use charting methods to predict the evolution of the market. As the dealers act at different time steps and do not all use the same tools of technical analysis, they can obtain different conclusions on this evolution. Similarly, the market makers can adopt different strategies. They may decide to protect themselves, finding immediately someone who takes the opposite position, or may speculate as dealers do. This heterogeneity of decisions is certainly an important point of the global behavior of the market.
From this simulation, several questions (currently under investigation) can be addressed: how well does this model reproduce the dynamics of the real FX market? and how do the agents act on the model to give this result? which are the critical parameters the agents must have to mimic the real actors of the market? To answer these points, we propose to compare our model with the real market on the basis of an analysis of the statistical distribution of quotes and volatility, for which many results and properties are known.
Our FX market model is being developed on a parallel computer (IBM SP2) to obtain fast simulation, even with a large number of agents. We also provide a graphical interface which gives the possibility to examine the state of the market and of the actors, turning our simulation environment into a tool for performing financial numerical experiments.