Some Empirical Findings on Neural Network-Based Forecasting When Subjected to Autoregressive Resampling

This chapter investigates the challenge of forecasting stationary autoregressive data by using a neural network approach, when data has been subjected to different resampling strategies relevant to autocorrelated data. In particular, the implementation and performance of the moving block bootstrap as well as the stationary bootstrap on stationary data is considered, and empirical findings are observed on the inferential aspects and accuracy of these often considered bootstrap approaches particularly within an AR(1) scenario. This computational investigation illustrates the impact and relevance of using a neural network-based approach in the forecasting of autoregressive models, and highlights the drawbacks when relying on the usual bootstrap approach.