Confidence Intervals

The National Institute has a long history of analysing its forecasting record; recent Reviews have contained accounts of the accuracy of our forecasts of inflation (February 1984) and GDP (Savage 1983). Earlier studies include Brooks and Cuthbertson (1981) and Osborn and Teal (1979). Such studies ask two fundamental questions: how good have our forecasts been in the past, and how much confidence can we hold for our forecasts for the future? In the case of the first question the above-mentioned studies provide an almost complete answer. The second question can only be answered by such studies on the assumption that our past record will be maintained in the future. This note approaches the second question from an entirely different viewpoint. An estimate of confidence bands for the National Institute's Model 7 will be derived by the use of stochastic simulations of the model. Just as the earlier studies offer only a partial guide to the accuracy of our forecasting ability in the future, so this note is also only a partial answer to the question. This is because our forecasts do not consist simply of a mechanical model run; instead the basic model forecast is supplemented by much other information. A good example of this is the recent miners' strike which could not have been included in the forecast on the basis of a purely mechanical model forecast. We would therefore expect the model to have a larger error bound than our final published forecasts. The scope of this note, it must be stressed, is strictly only applicable to the model we use to forecast and not to the finally published forecasts.