Forecasting with Vector Autoregressions versus a Univariate ARIMA Process: An Empirical Example with U.S. Hog Prices

Efforts to improve the accuracy of forecasts have turned toward more sophisticated prediction approaches. The use of vector autoregressions (VAR) as an alternative to more simple univariate time series processes reflects, in part, this move. The VAR approach incorporates the information from more than one economic time series into the forecasting process. Out-of-sample quarterly forecasts of hog prices were generated over the 1976-1982 period via a VAR and an univariate ARIMA process. The results indicated no improvement in forecasting ability by the more complex VAR procedure based on several measures of performance.