A direct test of the information content of the OECD growth forecasts

We develop a regression test to evaluate forecasts relative to the naı̈ve 'last observation' benchmark forecast. To this end, the current year forecast is decomposed into the last observation and the information content that forecasters add. Similarly, the 1-year-ahead forecast can be viewed as the sum of the last available realization, the information content of the current year forecast and the information content added for next year. Evaluation regressions provide a test of the statistical significance of the components of forecasts. The test is applied to the current and 1-year-ahead growth forecast produced by the OECD for 21 countries. The results show that when the evaluation statistics indicate a poor quality of forecasts, occasionally they do contain information. The information content of the 18-month horizon forecasts is low. As the horizon shortens, the quality of the forecasts increases. Alternative naı̈ve forecasts such as the last realization or the current year forecast as a 1-year-ahead forecast do not, generally, perform better.