Agent-Based Modelling of Stock Markets Using Existing Order Book Data

We propose a new method for creating alternative scenarios for the evolution of a financial time series over short time periods. Using real order book data from the Chi-X exchange, along with a number of agents to interact with that data, we create a semi-synthetic time series of stock prices. We investigate the impact of using both simple, limited intelligence traders, along with a more realistic set of traders. We also test two different hypotheses about how real participants in the market would modify their orders in the alternative scenario created by the model. We run our experiments on 3 different stocks, evaluating a number of financial metrics for intra- and inter-day variability. Our results using realistic traders and relative pricing of real orders were found to outperform other approaches.