Forecasting Accuracy of Alternative Techniques: A Comparison of Alabama Forecasts

Abstract

Macroeconomists have always been interested in developing a tool to forecast macroeconomic aggregates and to perform policy analyses. The most notable models are Box-Jenkins, autoregressive-integrated-moving-average (ARIMA), and vector autoregression (VAR) models. The first is a purely time series model and does not rely on any economic theory. The VAR and ARIMA models, on the other hand, can be derived by specific economic theory, and VAR is by far the most successful in capturing the notion of general equilibrium at the aggregate level. Proponents of VAR claim that the VAR models are superior to the structural models. In an effort to evaluate these claims, this paper develops a vector autoregression (VAR) model and a structural forecasting model for the Alabama economy. The forecasting errors of the two models for different forecasting horizons are then used to compare the models’ accuracy.