Forecasting tourism demand with denoised neural networks

The automated Neural Network Autoregressive (NNAR) algorithm from the forecast package in R generates sub-optimal forecasts when faced with seasonal tourism demand data. We propose denoising as a means of improving the accuracy of NNAR forecasts via an application into forecasting monthly tourism demand for ten European countries. Initially, we fit NNAR models on both raw and denoised (with Singular Spectrum Analysis) tourism demand series, generate forecasts and compare the results. Thereafter, the denoised NNAR forecasts are also compared with parametric and nonparametric benchmark forecasting models. Contrary to the deseasonalising hypothesis, we find statistically significant evidence which supports the denoising hypothesis for improving the accuracy of NNAR forecasts. Thus, it is noise and not seasonality which hinders NNAR forecasting capabilities.