An essential element in international economic policy coordination is the provision of forecasts, and particular interest attaches to the quality of forecasts produced by bodies like the OECD and IMF. These are produced by a mixture of `judgemental' and model-based methods. They incorporate ideas suggested by economic theory about relationships between economic aggregates as well as forward-looking information such as business survey data. In Discussion Paper No. 380, Research Fellow Michael Artis and Wenda Zhang describe the use of Bayesian Vector Autoregression (BVAR) techniques to produce forecasts, which they argue provide useful complements to those produced using more conventional methods. The authors construct quarterly Bayesian models to predict GDP growth, inflation and the balance of payments for each of the G7 economies and for the G5 economies as an ensemble.
In earlier work on the accuracy of the IMF's World Economic Outlook, Artis has identified the key role played by the sensitivity of forecast evaluation to the information available to forecasters. In the present paper, therefore, Artis and Zhang take particular care in `fine-tuning' their models so that they use information sets and assumptions that resemble as nearly as possible those of the IMF's forecasters.
They estimate their models using data for 1975-9 and then provide ex ante annual forecasts for one and for two years ahead over the period 1980-7. Having briefly demonstrated the superiority of their models to the cruder univariate autoregressive and unrestricted vector autoregressive techniques, they compare the performance of the BVAR forecasts against the actual out-turns with the World Economic Outlook's predictions for comparable forecast horizons. The BVAR models offered comparable quality forecasts in between a third and a half of the instances examined, though small changes in information assumptions could alter the relative performance of the different forecasts. The BVAR models did not appear to improve upon the conventional models' failure to predict turning points, but Artis and Zhang conclude that overall, they provide an impressively high and cheap standard of comparison.

Forecasts of the World Economy
M J Artis and W Zhang

Discussion Paper No. 380, February 1990 (IM)