RDP 2015-12: Modelling the Australian Dollar 7. Conclusion

This paper has presented an error correction model of the Australian dollar real exchange rate that has displayed robust explanatory power for a number of decades, as well as a range of augmented versions of this model. The rolling error correction model and Markov-switching variations have examined whether the relationship between the real exchange rate and the baseline model's existing explanatory variables have changed significantly over time. A second set of variations have examined whether two recent important macroeconomic developments – namely, the resources boom and foreign central banks' unconventional monetary policy actions – may have helped to identify additional medium-term determinants of the exchange rate.

Taken as a whole, while the results from these augmented models support the notion that there have been some additional influences on the real exchange rate in recent years, they do not fully account for the behaviour of the exchange rate during the period. Moreover, it is difficult to quantify these influences definitively, as even though each of the model specifications considered can be justified a priori based on theory, they produce a range of estimated equilibrium paths for the real exchange rate. Further, objective measures of in- and out-of-sample fit do not suggest that any single model's explanatory power is clearly superior to the others, or even to that of the baseline model, over the entire sample period.

This exemplifies the imprecise nature of exchange rate modelling, which is well established in the literature. Choices regarding the modelling framework, the explanatory variables and even the sample period will all influence the models' results. Consequently, it is difficult to conclude that any single model is the ‘true’ model of the exchange rate, or even the ‘best’ model of the exchange rate, and therefore to conclude that any estimated level of the exchange rate is the ‘appropriate’ level. Instead, such assessments will inevitably require a degree of judgement, incorporating a broader analysis of a range of other economic variables.