Creating an agent-based artificial market

Master's thesis by Per Berséus

Abstract

This thesis is a study within econophysics, a research field where financial problems are investigated using methods from statistical physics.

An artificial stock market was implemented in Java and used to assess some assertions made in empirical studies by J.-P. Bouchaud and collegues. The model was able to reproduce several characteristics of high-fequency data from financial markets, e.g. a leptocurtic return distribution and a nearly constant response function. Much of the emphasis of this report is made on the methodology itself, since many modelling issues remain to be solved in the agent-based computational economics field. However, the market was also used to simlulate trading scenarios, where the impact of different trading strategies was analysed.

More specifically, it was shown in simulation that an overall diffusive price development can result from a competition between two populations of traders: one of liquidity providers, who create persistence in the prices, and another of liquidity takers, who create anti-persistence. It was also shown that a slowly decaying trade sign autocorrelation function could result from similar trading criteria, but not in any obvious way from liquidity provider's division of large market orders into several smaller ones.

Keywords

agent-based modelling, ABM, artificial markets, complex systems, diffusion, econophysics, liquidity takers, liquidity providers, order book.

Full text

Contact

For any comments or requests, plase send an e-mail to info@berseus.se.

This page was last updated on April 16 2007 -- webmaster@berseus.se