Univariate and multivariate ARIMA versus vector autoregression forecasting

The purposes of this study are two: 1) to compare the forecasting abilities of the three methods: univariate autoregressive integrated moving average (ARIMA), multivariate autoregressive integrated moving average (MARIMA), and vector autoregression (both unconstrained — VAR — and Bayesian — BVAR) and 2) to study the idea that one advantage of vector autoregressions is that the models can easily and inexpensively be reestimated after each additional data point. All of these methods have been shown to provide forecasts that are more accurate than many econometric methods, which require more resources to implement. ; These methods were applied to seven economic variables: real GNP, annual inflation rates, unemployment rate, the money supply (Ml), gross private domestic investment, the rate on four- to six-month commercial paper, and the change in business inventories. The major results of this study are: 1) on average, the method that performs best in terms of the root mean square error (RMSE) is the multivariate ARIMA model; 2) the univariate ARIMA and BVAR methods perform approximately the same on average; 3) reestimating the VAR model after each data point increases the accuracy of this method; 4) reestimating the BVAR model after each data point becomes available decreases the accuracy of this method; and 5) the VAR method using reestimation is approximately as accurate as the BVAR method.