Forecasting using nonlinear long memory models with artificial neural network expansion

Abstract

We compare a number of models for obtaining h-step ahead minimum mean square error forecasts for nonlinear long memory processes. The forecasts from a proposed approximate nonlinear long memory, Fractionally Integrated Artificial Neural Network (FI - ANN) model, are compared to pure long memory models, e.g., ARFIMA(1,d,0) and LocalWhittle, pure nonlinear, i.e., Artificial Neural Network (ANN), and high order autoregressivemodel. Consider several nonlinear specifications in nonlinear long memory processes, the one- or two-step ahead forecasts of FI - ANN model generally perform better than ANN and other alternative models in the Monte Carlo simulation. The model is used to forecast series of inflation. © 2013 Springer-Verlag Berlin Heidelberg.