⁽¹⁾ These annual post-mortems compare the forecasts made following the previous year's budget and published in the May Review with the outturn for the year just finished. They appear in the February or May Review following the year in question and also compare the National Institute's forecasts with the Treasury forecasts in the Financial Statement and Budget Report which accompanies the budget and the London Business School forecasts which are published in the Sunday Times.

⁽²⁾ See S. A. B. Page, ‘Errors in National Institute forecasts for the World Economy’, National Institute Economic Review, no. 68, May 1974, and R. L. Major and M. J. C. Surrey, ‘Errors in National Institute forecasts of the balance of pay ments’, National Institute Economic Review, no. 52, May 1970.

⁽³⁾ In the regular annual comparison of forecasts and out- turns published in the National Institute Economic Review, no. 75, February 1976, pages 98-100, there was a comparison of the end-year (1975) unemployment and year-on-year consumer price forecasts. These comparisons were included for the first time but will now be included on a regular basis.

⁽⁴⁾ J. C. K. Ash and D. J. Smyth, Chapter 6 of Forecasting the UK economy, Saxon House, 1973.

⁽⁵⁾ Defined as income from rent and self-employment, dividend and net interest receipts, and transfers to charities from companies.

⁽⁶⁾ Defined as employers' contributions to national insurance and health and pension funds less total national insurance and health contributions.

⁽⁷⁾ This is the implied price deflator of the constant price consumers' expenditure series. In much of the period con sidered the constant price basis was 1963 prices, but from 1974 onwards 1970 prices were used. This switch does not affect the predicted and actual rates of change in any significant way.

⁽¹⁾ The effect of changes in taxes and current grants will in general be unflattering to the National Institute‘s forecasts. Normally there will have been upward adjustments of specific taxes (also direct tax thresholds and current grants) in order to keep pace with inflation. But such ‘indexation’ is not allowed for by our definition of unchanged policies.

⁽¹⁾ See in particular; M. C. Kennedy, ‘How well does the National Institute forecast?’, National Institute Economic Review, no. 50, November 1969, and G. D. N. Worswick, ‘National Institute experience with econometric models’, Chapter 3 of G. A. Renton, ‘Modelling the economy’, Heinemann, 1975.

⁽²⁾ The concept of quadratic loss functions is broadly that the loss due to prediction error is proportional to the square of the error.

⁽¹⁾ For a discussion of both the Theil inequality coefficient and the quadratic loss concept see H. Theil, Applied economic forecasting, North Holland, 1966. Note that the Theil inequality coefficient is different from the earlier one given in H. Theil, Economic forecasts and Policy, North Holland, 1961.

⁽²⁾ The case, U=1, also occurs when P=2A.

⁽³⁾ This applies to the U statistics for the 1965-75 period. For the split periods there are a few examples of higher values (see table 4).

⁽⁴⁾ This error decomposition also follows Theil, op cit. The method of error decomposition has been criticised as being inappropriate for the predictive testing of an econometric model (see D. W. Jorgenson, J. Hunter, and M. I. Nadiri, ‘The predictive performance of econometric models of quarterly investment behaviour’, Econometrica, March 1970). However the correction recommended by P. J. Dhrymes et al (in ‘Criteria for evaluation of econometric models’, Annals of Economic and Social Measurement, July 1972), which allows for the component of prediction error due to imprecision in the estimate of β, is not appropriate because the forecasts examined here have been generated over a ten-year period during which the relevant equations have frequently changed. The Theil decomposition is presented here as a useful rough guide to the element of systematic error in the forecasts.

⁽⁵⁾ C. W. J. Granger and P. Newbold (in ‘Some comments on the evaluation of economic forecasts’, Applied Economics, 1973) suggest that the second decomposition is a more appropriate indicator of the type of error than the first.

⁽¹⁾ See J. A. Bispham, ‘The current inflation and short-term forecasting’, in M. Parkin and M. Sumner (editors), Incomes policy and inflation, Manchester, 1972.

⁽²⁾ One should note however that in the case of February 1974, because of the uncertainties involved with the oil crisis, the three-day week, and the miners' strike, there were two forecasts published in the Review. The optimistic forecast has been used for the analysis in this paper although in the case of some variables the pessimistic forecast proved more accurate. The pessimistic earnings forecast, for instance, was for a 12.5 per cent increase compared with the optimistic variant of 11.4 per cent.

⁽³⁾ Although it was due in part to the commodity price explosion, which virtually no forecaster predicted, coupled with the Stage 3 threshold agreements. Stage 3 was also ended prematurely in July 1974 when the Pay Board was abolished following the return of the Labour Government.

⁽¹⁾ See J. A. Bispham, op. cit., for a description of the procedures for forecasting the CPI. The accuracy of the fore casts of world prices has been analysed in S. A. B. Page, op. cit.

⁽²⁾ However there was also an explosion of subsidies in 1974 which was not allowed for in the unchanged policies assumption.

⁽³⁾ The elimination of the 1974 forecasts from the analysis of the errors in both the 1965-75 and 1970-75 periods sub stantially improves the performance for all the major variables considered (with the exception of the consumer price index, which deteriorates very slightly). For real personal disposable incomes the Theil inequality coefficient for the 1970-75 period improves from 1.163 to 0.391 for the February forecasts and from 0.873 to 0.173 for the May forecasts. There are similar improvements for the 1965-75 period.

⁽⁴⁾ The major exception is 1970, when the National Institute said comparatively little about unemployment in the Review.

⁽¹⁾ The unadjusted series are hardly revised at all. It is the seasonally adjusted series which tend to be revised rather more.

⁽²⁾ The record of the August and November Review forecasts is not so good but there are still very few incorrect forecasts of the direction of change.