Financial time series forecasting using agent based models in equity and FX markets

We investigate the application of machine learning Agent Based Modelling (ABM) techniques to model and forecast various financial markets including Foreign Exchange and Equities, especially models that could reproduce the time-series properties of the financial variables. We model the economy by considering non-equilibrium economics. We adopt the features that are required for modelling non-equilibrium economics using ABMs and replicate the non-equilibrium nature of the financial markets by considering a set of bounded rational heterogeneous agents, with different strategies that are ranked according to their performance in the market. We consider markets where there are different agents interacting among themselves and forming some sort of patterns. For example, the patterns are equity prices or exchange rates. While the agents have been interacting in the artificial market, the generated patterns (price dynamics) they co-produce would match with the real financial time-series. In order to get the best fit to the real market, we need to search for the best set of artificial heterogeneous agents that represents the underlying market. Evolutionary computing techniques are used in order to search for a suitable set of agent configuration in the market. We verify the forecasting performance of the artificial markets by comparing that with the real financial market by conducting out-of-sample tests.