RDP 7904: Some Aspects of RBA76 and RBF1 4. Conclusion

It will be remembered that the results presented in sections 2 and 3 were derived in a mechanical fashion with no attention paid to the economic realism of the responses. For a statistician this is an appropriate testing procedure to employ. Nevertheless many economists, as distinct from statisticians, may conclude that at least some of the results illustrate less about the weaknesses of the models than the results first appear to show. Given these important caveats, there appear to be several conclusions from this study. These conclusions do not all necessarily apply to both RBA76(T) and RBF1(M).

First, the results suggest that when a model is re-estimated with additional data the properties of the re-estimated model should be closely investigated and compared with earlier versions. Further, it is possible that even alterations to one equation or a few crucial parameters of a model may alter the dynamic properties of that model.

Second, the response of a model to a shock can vary considerably depending in the assumed behaviour of policy instruments. For models with exogenous policy instruments sensitivity analysis with respect to instrument settings should be conducted before evaluating the results of a particular shock. For models with reaction functions care has to be taken in estimation as small changes in data can lead to a reaction function and hence a model with substantially different properties. A further implication of this is that for some forecasting exercises or counterfactual simulations a particular estimated reaction function may be inappropriate.

Third, from the various tests employed there is evidence of considerable parameter instability in both an absolute and relative sense. For some parameters there are noticeable trends. Information about these trends could and should be used to add to our understanding of how the economy works and the limitations of current specifications.

Fourth, there are considerable single equation errors which appear to increase with the lengthening of the sample period and are likely to be important in dynamic simulation performance.

Fifth, there are large simulation errors not only out of sample but also within sample.

Overall, the analysis even with its obvious limitations, suggests that either structural change has occurred or that the specification of the models is incorrect or appropriate only for particular parts of the sample period. Also the variability of parameter estimates indicates the potential importance of stochastic simulations in which parameters are allowed to vary.