Front Matter Page

Research Department

Table of Contents

• Summary

• I. Introduction

• II. A Rational Expectations Restriction on Forecasting Optimality

• III. The Forecasting Accuracy of the World Economic Outlook

  • 1. The overall period, 1971-91

  • 2. Comparison of business cycles and recession years

• IV. Time Series Forecasts

  • 1. Forecasting accuracy of time series models

  • 2. Comparison of forecasts

  • 3. An adjustment method

• V. Concluding Remarks

• Tables

  • 1. Forecast Error Statistics for Output Growth

  • 2. Forecast Error Statistics for Inflation

  • 3. Pooled Forecast Error Statistics for Output Growth Over the Business Cycles

  • 4. Pooled Forecast Error Statistics for Inflation Over the Business Cycles

  • 5. Time Series Models of Output Growth

  • 6. Time Series Models of Inflation

  • 7. Forecast Error Statistics for Time Series Models

  • 8. Comparison of Forecast Errors

• Charts

  • 1. Forecast Errors in World Economic Outlook Projections for Output Growth

  • 2. Forecast Errors in World Economic Outlook Projections for Inflation

• References

Summary

A useful discussion about forecasting accuracy should provide a qualitative assessment of the way in which various forms of inefficiency in. a projection are related. These inefficiencies may be due to the way forecast errors are used to make current projections or because the economic model is not the minimum variance model. If the nature of the relation between the various kinds of inefficiencies can be established, an adjustment method can be constructed to improve the accuracy of a forecast. Most of the previous studies on forecasting accuracy evaluate the accuracy of a projection based on the characterization of the regression error. A major difficulty with this criterion is that it provides only a joint test of unbiasedness and efficiency. However, if a distinction is made between the different properties of a projection, an appropriate adjustment factor can be derived to improve the accuracy of the projection.

This paper presents a simple criterion to evaluate the accuracy of a forecast based on the properties of the forecast error. The optimality conditions of a simple optimization problem under rational expectations are used to show that the standard efficiency conditions are necessary but not sufficient. A key feature of the method is its characterization of how different kinds of inefficiency are related. Moreover, the criterion provides simple adjustment factors that reduce the inefficiency of the forecast.

The proposed criterion is used to examine the accuracy of projections of output growth and inflation in the World Economic Outlook for the group of seven major industrial countries over the 1971-91 period and during each of the business cycles in this period. Time-series models are estimated and the accuracy of the forecasts generated by these models is examined relative to the projections of the World Economic Outlook.

Empirical test results suggest that the current-year forecasts for growth and inflation in the World Economic Outlook are unbiased for 1971-91, but year-ahead projections overstate growth and understate inflation, on average, by ½ of 1 percentage point in 1971-82. After 1982, year-ahead projections for both growth and inflation are unbiased. In the 1990-91 period, the projection errors are lower than in the two previous cyclical downturns, and the projections are generally unbiased.

However, both current and year-ahead forecasts of growth and year-ahead projections of inflation are inefficient. By contrast, time-series models provide more accurate forecasts of growth and inflation than the World Economic Outlook in 1971-91, suggesting that the accuracy of the projections in the World Economic Outlook could be improved by the use of such model-based methods. The criterion presented in this paper provides an alternative for improving the performance of judgmental projections.