RDP 7901: Estimation and Statistical Evaluation of an Economic Model 6. Concluding Remarks

This paper looks at several issues relating to full system estimation of a linearised version of a non-linear model. A series of criteria are used for model evaluation, since there are no formal procedures for this type of model. However the problem of conflict from different econometric criteria often occurs. In this case subjective evaluation, which the author has tried to avoid in this paper, would be necessary if firm conclusions are to be drawn. The analysis also illustrates that it is often difficult to separate out econometric issues in a model-building project from specification issues.

One point that can be made from the study of linearisation issues is that even though the non-linear control solution reported in this paper may be unreliable because of the problems mentioned in section 4, the results suggest that the linearised model may not provide a close approximation to the simulation properties of the non-linear model. Thus it may be necessary to investigate the use of a non-linear estimator for the RBA76 model. Another point is that if linearisation of a model such as RBA76 is appropriate, the results of section 4.1 suggest that the use of alternative plausible methods to calculate the weights for use in linearisation does make differences to the properties of the model, but these are perhaps not very important. It is difficult to draw firm conclusions on this because of the conflicting evidence of different econometric criteria, and the difficulty of deciding on the possible importance of apparently small differences in parameter estimates, autocorrelation properties, etc.

Section 5 tried to evaluate various aspects of the RBA76 estimation technique: some of the conclusions are ambiguous. The omission of the second averaging stage changes the parameter values and, in particular, reduces the apparent rate of the partial equilibrium adjustment speeds. The OLS estimates also have this feature, and in addition the size and significance of the parameters for monetary disequilibrium are reduced. This is the same result as that found by Bacon and Johnston (1977), and could arise either from misspecification of the monetary dynamics of RBA76, or from simultaneous equation bias in the OLS results.

The use of the COSMA transformation reduces serial correlation in the structural residuals of the model, but a considerable amount of residual serial correlation remains. Work is continuing on improving the specification of RBA76 and the data set used for estimation in an attempt to reduce serial correlation in the residuals of the estimated model.