A new NN-PSO hybrid model for forecasting Euro/Dollar exchange rate volatility

An accurate estimation of exchange rate return volatility is an important step in financial decision making problems. The main goal of this study is to enhance the ability of GARCH-type family models in forecasting the Euro/Dollar exchange rate volatility. For this purpose, a new neural-network-based hybrid model is developed in which a predefined number of simulated data series generated by the calibrated GARCH-type model along with other explanatory variables is used as input variables. The optimum number of these data series and other parameters of the network are tuned by an efficient particle swarm optimization algorithm. Using two datasets of real Euro/Dollar rates, how the proposed hybrid model could reasonably enhance the results of GARCH-type models and the traditional neural network in terms of different performance measures is demonstrated. We also illustrate how the respective simulated data series as the input variable to the network could contribute to improve the accuracy of volatility forecasting.