Reviews of Important Papers on Forecasting
Before 1985 Reviews
Review of:

Robert Fildes and M. Desmond Fitzgerald (1983), “The use of information in balance of payments forecasting,” Economica, 50, 249-258.

Fildes examined the performance of three economists who made monthly forecasts of the U.K. balance of payments. The forecasts were for one month ahead for the period from July 1975 to December 1978. Some findings were consistent with prior evidence, for example the combined forecast of the three judgmental forecasts was better than the average forecaster (RMSE of 176 vs. 185). Fildes and Fitzgerald also examined an extrapolation (ARIMA) model and found that it was about equal in accuracy to the average forecaster (RMSE 184 vs. 185). The inaccuracy or, in economists’ terms, ‘irrationality’ of the judgmental forecasters was partially explained by the phenomenon of anchoring—they gave too much weight to the latest observation. Then they combined the extrapolation and judgmental forecasts and found little improvement over the combined forecasts of three judges (RMSE of 173 vs. 176). Finally, he concluded that bootstrapping models did not improve the forecast accuracy of any of the three judges. Apparently this is the only unchallenged failure in a long series of studies in bootstrapping. So I will challenge it. First, and most importantly, the sample size of three judges is too small to allow for a conclusion that bootstrapping was less accurate. Secondly, bootstrapping typically draws upon all of the information used by the judge. (In correspondence, Fildes added that a much wider information set was considered but without any success.)