Below one of the chancellor's wise men, Martin Weale, asks if economic forecasting owes more to luck than to skill.

The Independent has identified the National Institute, to which I belong, as the most accurate macroeconomic forecaster of 1996. This is naturally gratifying, but such competitions do raise serious questions. Is forecasting a matter of skill or chance, and is the rigour of the models on which economic forecasts are based an important factor?

The institute's models have a clearly specified theoretical structure. As Cambridge Nobel Prizewinner Richard Stone observed, "Economics is neither a pure subject like mathematics, of which one does not ask that the theories should be applicable to actual phenomena, nor is it a collection of facts, like the objects on a junk heap, of which one does not ask how they are related."

For the problem faced by a modeller is that there are any number of statistical relationships that might explain the data. Theory is needed to identify sensible restrictions that can first be tested on the data and then imposed as a tool for improving model specification.

³ÉÈËVRÊÓƵ

There is a strong element of judgement involved in forecasting. If the equation forecasting consumer spending has been underpredicting, how much of the error should be carried forward into the forecasts? There are no clear principles explaining this. This element of uncertainty is one reason why a model which fits the data well does not always give a good forecast. A second reason is that economic models do not treat the whole of the economic environment in isolation. A model of the UK has to take a view as to how fast world trade is likely to grow. And it would be most peculiar if a forecast produced on the assumption that the oil price would stay constant proved correct in a year when the oil price doubled. A forecaster who is right for the wrong reasons is not a good forecaster but simply lucky.

For such reasons academic forecasters do not have a better track-record at short-term forecasting than do City forecasters, whose models sometimes have no clear theoretical basis to them. One sometimes hears non-academics suggesting that almost everything can be explained in terms of a single variable such as the money stock. Such a proposition is theoretically doubtful and has no sound statistical basis to it. But it may occasionally deliver good forecasts of the economy.

³ÉÈËVRÊÓƵ

Academic models can, however, fulfil a role only rarely attempted by private-sector modellers. Their models, because of their theoretical coherence, can be used for purposes of policy analysis. What, for instance, is the long-term cost of government borrowing? To be able to address this question needs a model in which proper stock-flow accounts are kept and in which interest payments are cumulated, so that the implications of current borrowing for future taxation can be assessed. One also needs a coherent view of how consumer behaviour is affected by the stock of national debt and of any implications of tax rates for labour supply. It is important to know whether the true social cost of government borrowing is similar to or different from that faced in the gilts market.

There are other important policy issues which have been addressed only by the publicly funded modellers. Any economic forecast must have a margin of error surrounding it. One obvious area where an analysis of uncertainty is very important is in the matter of inflation targeting. The government has an annual inflation target of 2.5 per cent. This probably means that, at any time, the inflation rate is as likely to be above as below 2.5 per cent. But how often should we expect it to be above 3 per cent per annum, or below 2 per cent? The answer depends on the policy rule that the Treasury uses when setting interest rates and tax rates. Publicly funded modellers are able to investigate the performance of particular policy rules while City forecasters have not started to address this question. Perhaps privately funded research should not be expected to address issues predominantly only of public interest.

There are, as the chancellor of the exchequer has observed, many divergent forecasts, and there may be no obvious reason for believing that any one is going to be better than any other. But commentators should look beyond these forecasts at the range of analysis made possible by well-researched models if they want to be able to reflect knowledgeably on the underlying state of the economy.

Martin Weale is directorof the National Institute ofEconomic and Social Research. Macroeconomic modelling at the National Institute issupported by the Economicand Social Research Council as part of a wider ESRC programme.

³ÉÈËVRÊÓƵ