Neural network architures for efficient modeling of FX futures options volatility

The importance of volatility modeling is evidenced by the voluminous literature on temporal dependencies in financial market assets. A substantial body of this literature relies on explorations of daily and lower frequencies using parametric ARCH or stochastic volatility models. In this research we compare the model performance of alternate neural network models against that of the (G)ARCH framework when applied to hourly volatility of FX futures options. We report that the results obtained from the application of a closed-form Bayesian regularization radial basis function neural network are considerably more efficient than those produced by alternate neural network topologies and the (G)ARCH model formulation.