Economic Forecasting

Let us begin a more detailed discussion of these challenges with economic forecasting and the econometric model. Its importance as part of David's activities at NIESR can be measured by the fact that modelling and forecast­ing for the economy absorbed half the Institute's budget.

Before describing the history of NIESR's modelling and forecasting, it is appropriate to comment on David's fundamental approach to measurement, modelling and forecasting in economics as a field which he lays out in his paper, “Is Progress in Economic Science Possible?” (Worswick 1972). He first observes that economic variables are not like scientific entities which have clear and particular meanings like specific gravity. Economic variables such as tons of steel are ultimately proxies for value or utility in the minds of consum­ers. Workers are proxies for hours of human labour which may vary in differ­ent situations. Also, relationships between economic variables, such as the consumption function in which income determines consumption, result from human decision-making which varies over time and circumstances. Therefore, the attempt to describe the economy in terms of its inter-related variables through the use of statistical techniques such as econometrics, and to make projections about the future of path of the economy based on econometric modelling, is fraught with difficulty and ambiguity from the start.

As an example of the false accuracy of econometric relationships, David cites the case of the Phillips curve. Using data from 1861 to 1913, Phillips estimated a single equation relating the change in wage rates in the UK to the level of unemployment. This relationship was then used to incorporate later data and form a prediction that 2.5% unemployment could stop inflation. This was then linked to another idea that a higher level of unemployment would be favourable to economic growth.

When David came to the National Institute, he encouraged its Executive Committee to come to a decision that not more than one half of its resources be devoted to the regular quarterly forecast of the British economy and the accompanying analysis. This was important in that it prevented the Institute from being drawn into the development of ever more complicated and costly econometric modelling at a time when such activity and its seemingly endless demands were coming into their own. It also allowed time and resources to be devoted to other lines of research previously outlined in Section 3.2.

When David took over the Directorship of NIESR, the forward estimates of the main components of GDP were not yet the result of an econometric model per se. Individual equations describing specific relationships were relied on, but these equations were not joined together in a simultaneous model (see Jones 1998: Chapter 4). During the 1960s, the building of a complete econo­metric model gradually took place. But the need to linearise the individual equations was difficult since many of the successful forecasting relationships were non-linear and did not perform as well when transformed into the linear context of the simultaneous model. However, in August 1969, a suitable sim­ulation program was developed for a non-linear model with eleven equations which generated forecasts.

The job of improving forecasts was the focus of a large amount of work by the NIESR research team over a wide range of subjects as the scope and capac­ity of computers increased. David was involved as a member of the editorial board in overseeing the development of forecasting during his years as Director. Jones noted in 1998 that:

Today, the [NIESR's forecasting] model can be described as having Keynesian features in the short term, but with classical long-run properties such that out­put is determined by the size of the labor force and the state of technology. Recent research has continued to refine the model along a number of different lines, each combining empirical validity with theoretical rigour (ibid.: 34).

In 1971, David wrote an extraordinarily clear and honest introduction to a book by M.J.C. Surrey called The Analysis and Forecasting of the British Economy which laid out the methods used at that time by the National Institute to produce quarterly forecasts. The book encompassed a discussion of all the variables and equations used in the Institute’s econometric model. In his introduction, David discussed the various different contexts for viewing and understanding forecasting methods and outcomes. The idea, he says, is to be completely open about the methodology of forecasting so that the student or researcher can reproduce for herself the Institute’s forecasts based on the information in the book. The quarterly estimates should be consistent in two ways: first, the rules of accounting should be maintained within each time period, and second, the relationships between different variables within and between periods should be consistent with the postulated structural equa­tions. He notes that, in its essentials, the forecasting process has changed rela­tively little:

The first step is to make estimates of the probable changes in certain “exoge­nous” variables, notably investment, exports, import prices and public expendi­ture, and to derive the remaining “endogenous” variables, such as consumption and the volume of imports, by using a model which is, in its essentials, a lagged multiplier combined with an “accelerator” for stock-building (Worswick 1971: 4).

He then points to the increasing importance of computers in obtaining forecasts rapidly from changing one or more exogenous variables (ibid.).

The next question is to ask whether the forecasts are any good. But now you have to ask what exactly is being tested? The reason for this second question is that the judgment of the (human) forecaster may be used to adjust forecasts in the light of special knowledge not reflected in the equations. This judgment is important in improving the accuracy of the forecast. But to test its accuracy the actual forecast must be used and the number of available forecasts may be too few for very exacting tests. At quarterly intervals, one would still hardly be satisfied with twelve such observations, still a small number, and certainly not just one. In particular, trends may be hard to detect.

Another difficulty arises when account is taken of the fact that forecasts are made on the basis of “unchanged policies”. It may be that policies are changed within the forecast period in which case it would be perverse to compare the actual outcome directly with the original forecast. Of course, the econometric model could be re-estimated with the policy change included, or with other measurement changes in some of the variables over the forecast period. But now it is not clear at all what the meaning of the accuracy of the original fore­cast is. It is comparing apples with oranges.

Finally, David addresses the National Institute model itself as described in Surrey's book and notes that it is small. Many of the equations which com­prise it are non-linear and all have to be constantly maintained in the light of data and policy changes. (He notes that a larger and more comprehensive linear econometric model was tried, but it performed badly.) He draws atten­tion to the Phillips curve relationship which was used to predict unemploy­ment but notes that the relationship had broken down, necessitating a change in the model.

The idea that the National Institute should undertake economic forecasting originated with economists within the Treasury, and there had been some movement of economists between the two institutions.

Two newspaper articles by economic correspondents attest to the appropri­ateness of David's and the National Institute forecasting team's approach to forecasting. The first was by Peter Jay in The Times on 25 November 1971. He noted that the Institute's quarterly forecasts were central to the reputation of conventional national income (or “Keynesian”) projections. These forecasts were widely published and reported, but often faced a lack of public under­standing of what the forecasters were trying to do. Jay refers to David's ‘fasci­nating, totally intelligible and elegant introduction to M.J.C. Surrey's book on forecasting' (Jay 1971: 25), discussed above. He describes David as ‘a rare economist who throughout a long and distinguished academic career has combined a superior mathematical proficiency with an unquenchable skepti­cism about the ability of econometrics to displace political economy and sea­soned judgement in the management of national economic affairs' (ibid.).

The second article, from The Sunday Telegraph on 8 September 1972, is by Patrick Hutber who also concurs with the National Institute's approach to forecasting and policy-making. He refers to a recent Institute forecast as a ‘prediction of what may happen if things go on as they are' (ibid.: 21). The case discussed shows ‘just how damaging the effects of the current inflation are liable to be. Left unchecked, accelerating inflation next year would mean that much of the higher consumer spending would be swallowed up in rising prices, so that demand would be lower, production rise less and unemploy­ment stay painfully high' (ibid.). Hutber then traces out further outcomes of the forecast which he claims he has been saying himself ‘until my voice gets hoarse and my typing fingers ache' (ibid.). Such approval from the press about National Institute forecasts was not infrequent.

7