Macroeconomic real‐time forecasts of univariate models with flexible error structures

Abstract

This paper investigates the importance of flexible error structure specifications in two widely used univariate models, namely, autoregressive and unobserved component models, in fitting and forecasting 20 significant US macroeconomic variables. The in‐sample estimation reveals that the models with flexible error structures provide better in‐sample fit than the univariate models with homoscedastic errors. Furthermore, the density forecast analysis suggests that accommodating heavy tail, stochastic volatility, and serial correlation in error structures leads to significant improvements in short‐term forecasts. For most macroeconomic variables, the univariate models tend to yield more accurate one‐step‐ahead forecasts than the multivariate (vector autoregressive) models in terms of both point and density forecasts.