Hybrid Evolutionary Techniques for FX Arbitrage Prediction

This paper discusses the need for a missing value technique to fill in gaps in time series representing foreign exchange (FX) prices and assist in the observation of potential arbitrage opportunities. It highlights the requirement for prediction methods to establish the persistence of these opportunities (latency). Naieve missing value and prediction techniques are investigated and then compared with Kalman Filtration, Ensemble Kalman Filtration, Regression and Neural Network techniques. A technique not known to be applied in this domain before, namely NeuroEvolution using Augmented Topologies (NEAT), is then examined in order to asses its ability in filling in missing values and the prediction of arbitrage opportunities in comparison to these other more established techniques. Hybrid functions, incorporating the most successful of the techniques, are constructed in order to ascertain whether combinations of techniques are more successful than their constituents. Data from for various data providers for three markets is used taken over periods representing different levels of market activity (liquidity)