RDP 8609: The Performance of Exchange Rate Forecasts 5. A Comparison of the $A/US$ and US$/Yen Forecasts

This section compares the forecasting accuracy of the $A/US$ and US$/Yen forecasts. To facilitate comparison, the forecast errors have been converted to percentage forecast errors.

Table 4 reports the mean absolute percentage errors and the mean square percentage errors of the forecasts, as well as statistics on the directional performance of the two forecasted exchange rates. The table shows that the weekly percentage movements in the $A/US$ were, on average, larger than those in the US$/Yen rate (1.54 per cent compared to 1.22 per cent). In line with these larger movements in the $A/US$, the mean absolute percentage error was larger for the $A/US$ rate than for the US$/Yen rate. However, as a percentage of the no change forecast error, the mean absolute percentage error for the $A/US$ was marginally less than the mean absolute percentage error of the US$/Yen forecasts. This was the case for both the mean and median forecasts.

A comparison of the directional performance of the forecasts of the two exchange rates shows little difference between the forecasts. For both exchange rates, the group mean and median forecasts predict the exchange rate's moving in the correct direction just less than two thirds of the time.

While the above results from the MAE and direction of movement criteria suggest that there is little to distinguish between the performance of the $A/US$ and US$/Yen forecasts, this conclusion is not supported by a comparison of the mean square errors. Using the MSE criterion, the $A/US$ forecasts are less precise than the US$/Yen forecasts. While the MSE of the US$/Yen forecasts is some 5 per cent lower than that for the no change forecasts, the MSE of the $A/US$ forecasts is 12 per cent higher than that for the comparable no change forecasts. These results indicate that on average the $A/US$ was not significantly more difficult than the US$/Yen rate for the Australian foreign exchange dealers to forecast. However, the standard deviation of the forecast errors was substantially larger for the $A/US$ forecasts than for the US$/Yen forecasts. This suggests that the level of uncertainty may have been greater in the domestic market than it was in the US$/Yen market.

Further insight into the extent of relative uncertainty in the two markets can be gained by calculating the standard deviation of each week's individual forecasts and then averaging these standard deviations over the entire period. If a market is characterised by a high level of uncertainty the forecasts of a future exchange rate by individual forecasters should be relatively dispersed. In contrast, if there is little uncertainty as to the future exchange rate the individual forecasts should be relatively tightly distributed around the mean forecast.

Table 5 reports summary information regarding the standard deviation of the individual forecasts. The table shows that the average standard deviation of the $A/US$ forecasts was some 26 per cent higher than that of the US$/Yen forecasts (0.9264 for the $A/US$ compared to 0.7615 for the US$/Yen rate). In addition, of the 41 weeks for which forecasts were published, the standard deviation of the $A/US$ forecasts was greater than that of the US$/Yen forecasts on 28 occasions. These results again suggest that over the sample period, uncertainty with respect to future exchange rate movements may have been greater in the $A/US$ market than in the US$/Yen market.