Assessing the forecasting accuracy of monthly vector autoregressive models: The case of five OECD countries

Abstract

Vector autoregressions have been proposed as good forecasting models in the recent past. This paper presents five different VAR specifications of the level of industrial production in five major industrial countries estimated on monthly data. The forecasting performance of the various specifications is analysed and compared with the accuracy of univariate time-series models. In general, the VAR forecasts perform better than the alternative procedures. Their performance in forecasting the calender outcome 1988:1–1988:12 (on an ex ante basis after the stock market crash in October 1987) reveals, however, that all VAR's have underestimated the growth rates in industrial production.