RDP 8609: The Performance of Exchange Rate Forecasts 7. Conclusions

The evidence presented in this paper suggests, on the basis of published data, forecasters of the exchange rate have much about which to be modest. The vast majority of individuals providing the one week ahead $A/US$ hedge settlement rate forecasts published in the Australian Financial Review would have produced forecasts with a lower mean absolute error (or mean square error) had they supplied forecasts based on a simple rule such as no change. Notwithstanding this poor performance of the individual forecasts, the group mean forecast series performed relatively well, predicting the correct directional movement approximately two thirds of the time and outperforming all six benchmark models on the mean absolute error criterion. The group mean series, however, performed less well on the mean square error criterion having a larger mean square error than three of the benchmark models. of these benchmark models, the no change model and the time series model with one lag (the restricted model) produced the lowest errors.

Like the group mean forecasts of the $A/US$, the group mean forecasts of the US$/Yen rate produced a lower mean absolute error than the no change model. However, unlike the $A/US$ forecasts the group mean US$/Yen forecasts also outperformed the no change model on the mean square error criterion. Further, the average standard deviation of individual weekly forecasts was greater for the $A/US$ forecasts than for the US$/Yen forecasts. These results suggest that over the sample period the market for Australian dollars may have been characterised by greater uncertainty than was the US$/Yen market.