RDP 8612: A Weekly Model of the Floating Australian Dollar 4. Conclusions

The Australian dollar at first sign appeared not to be a random walk. This would mean that profits could be made by so-called “technical” analysis. When account was taken of the appropriate distribution under the null of unit roots and of the existence of heteroscedasticity (and the implied non-normality), a random walk with drift could not be rejected. This result bears out the conclusion of Lowe and Trevor (1986) that exchange rate forecasters did not do better than a random walk for one-step predictions.

There appeared to be significant differences between the first and second years of the float. While parameters were not significantly different, the exchange rate process in the first year was normal and not heteroscedastic, with an opposite result for the second year. This conclusion is consistent with the dramatically increased variability of the exchange rate in the second year. When monetary targetting was abandoned in 1985, evidently exchange rate targetting did not take its place; indeed, monetary policy was conducted on the basis of a checklist of key economic variables.

The first-differenced monetarist approach to flexible exchange rates did not (variance) encompass the univariate time series model, and the data evidence produced insignificant parameters. Dropping purchasing power parity, and using the monetary model in levels did give results that encompassed the univariate model. Higher (expected) U.S. money and lower output tended to significantly increase the rate of depreciation of the Australian dollar. The insignificant effects of Australian M3 can be attributed to the ever increasing difficulty in forecasting this variable over the sample. The process of de-intermediation and re-intermediation introduced a great deal of uncertainty (and thereby lack of faith) associated with recent observations of this variable. The exchange rate does not appear to have been significantly affected by Australian money and output, a result which is consistent with the Trevor and Donald (1986) conclusion that the trade-weighted exchange rate index appears independent of Australian interest rates.

The general conclusion is that there appears to be a structural model which will dominate the random walk. The structural results are conditioned by the very restrictive assumptions made about the generation of the expected future values of the predetermined variables. These restrictions were necessary to permit these early results. The results of this paper will give encouragement to structural exchange rate model builders when sufficient data has accumulated to undertake a less restrictive study. In particular, the simultaneous modelling of the current account and the exchange rate is bound to improve the explanation of the data generation process.