Title: Forecasting with VARs with time-variation in the mean
Authors: Andries van Vlodrop - VU Amsterdam (Netherlands) [presenting]
Marta Banbura - European Central Bank (Germany)
Abstract: A vector autoregressive model allowing for time variation in the mean and the variance is proposed. The unobserved time-varying mean is assumed to follow a random walk and we also link it to long-run Consensus forecasts, similar in spirit to so called democratic priors. The changes in variance are modelled via stochastic volatility. We propose a Bayesian methodology for inference in the model. The proposed Gibbs sampler allows the researcher to use a large cross-sectional dimension in a feasible amount of computational time. This is in contrast to standard time-varying-parameter VARs, which typically are not used beyond a cross-sectional dimension of three or four variables. Furthermore, standard simulation smoothing methods can be used to perform inference on the latent process of the local mean. The slowly changing mean can account for a number of secular developments such as changing inflation expectations, slowing productivity growth or demographics. We compare the forecasting performance of our model to time invariant VARs with Minnesota and democratic priors and standard time-varying-parameter VARs for several countries. The model performs well against the alternatives. In particular, incorporating survey information through our local mean approach improves the long run forecasting performance of VAR models.