Title: Forecasting long memory via a VAR model
Authors: Luc Bauwens - Universite catholique de Louvain (Belgium)
Guillaume Chevillon - ESSEC Business School (France) [presenting]
Sebastien Laurent - AMU (France)
Abstract: A large dimensional vector autoregressive (VAR) model can generate long memory in its components under conditions which restrict the VAR parameters. We compare the forecasting performance of univariate ARFIMA and HAR models, a VAR estimated by ML under the CHL constraints, and a VAR estimated by MCMC. The latter is based on a Gaussian prior density that incorporates the CHL restrictions through the prior mean of the VAR parameters, while the prior variances control the tightness of the restrictions. The forecast comparisons are done on simulated and real data.