Multi-agent based order book model of financial markets

Supervisor:  D. Grebenkov
Affiliation: Laboratory of Condensed Matter Physics, Ecole Polytechnique
Email: denis.grebenkov@polytechnique.edu

The price development of assets on financial markets are 
determined by the superposition of the actions of market 
participants creating offer and demand for a financial 
asset. The resulting price time series exhibit significant
autocorrelations for very small intra day scales and a 
"fat-tailed" distribution of asset returns.  Memory 
effects and non-Gaussian features of price time series 
are of great importance for automated trading, especially
for high-frequency trading. Fractional Brownian motion 
and Levy processes are now classical models that may 
partly account for these features.

Recently, an alternative, statistical-physics-oriented 
approach was proposed [1,2]. This approach consists in 
simulating an order book, in which many agents trade one 
asset at a virtual exchange.  Depending on the current
asset price, each agent can randomly choose to buy or cell 
the asset according to simple probabilistic rules. This
action determines the new price.  In this way, the asset 
price time series is generated.

The internship aims at studying multi-agent-based order 
book models and their applicabity for generating futures 
price series.  After implementation of the model, the
generated time series have to be statistically analyzed 
and compared to futures historical data. The goal is to 
determine the appropriate probabilistic rules for agents 
that would reproduce the main features of historical 
price series. 


References 

[1] T. Preis, S. Golke, W. Paul, J. J. Schneider,
"Multi-agent-based Order Book Model of financial markets",
EuroPhys. Lett. 75, 510 (2006).

[2] Samanidou, Zschischang, Stauffer, Lux,
"Agent-based models of financial markets",
Rep. Prog. Phys. 70, 409 (2007).