Modelling market volatilities: the neural network perspective

This paper investigates the use of Artificial Neural Networks (NN) to forecast volatility. In particular, we assess the dynamic behaviour of market volatility by forecasting the volatility implied in the transaction prices of the Ibex35 index options. The use of the NN technique is done within the framework of a model building strategy that tries to capitalize on the well known feature of persistence in volatility series. We demonstrate that forecasting with non-linear NNs generally produces forecasts which, on the basis of out-of-sample forecast encompassing tests and mean squared error comparisons, ordinarily dominate forecasts from traditional linear methods. Better forecasting results for the NN are due to its flexibility to account for potentially complex non-linear relationships, which are not well captured by traditional linear methods. We test and reject the hypothesis that volatility changes are unpredictable on an hourly basis.