RDP 7904: Some Aspects of RBA76 and RBF1 1. Introduction
September 1979
This paper analyses parameter instability and simulation characteristics of two econometric models. The two models are: a version of RBA76 which is denoted RBA76(T), and a version of RBFI with an endogenous monetary sector which is denoted RBF1(M). The models differ in several ways with the most important differences being: size, continuous versus discrete time specification, estimation procedure, and history and philosophy of development.[1] The main reasons for this analysis are:
- a natural extension of model validation, and
- a basis for isolating problem areas in the models and using this information as a guide for further research. This reason can be broadened to a desire to show that keeping a macroeconomic model in good order is a difficult task which may require considerable resources.
A further reason can be found in the recent debate on the alleged existence of a “consensus model”. A paper by the Institute of Applied Economic and Social Research[2] has suggested the existence of such a model and that RBF1 is an example of the consensus model. They further suggest that the consensus model is useful for analysis of the current economic situation. If this is so, then there should be little or no evidence of mis-specification or poor simulation performance with this model when data for the mid-1970s are used.
The “scientific method” used in this study needs to be emphasised. The results presented are derived in a mechanical fashion. Particular versions of each model are estimated over varying sample periods and tests of parameter stability and simulation characteristics are carried out. No judgemental adjustments or respecifications are made in general to either the simulation results or the specifications of the models. This is the case even where the results suggested mis-specification etc. This would not be the procedure followed if these models were being used in practical applications. However, the procedure of successively updating parameter estimates by lengthening the sample period when new data become available is common and could be viewed as the minimum maintenance for a macroeconomic model.
In the analysis an attempt was made to hold constant as many factors as possible. While this may limit the scope of the paper it may allow greater confidence to be given to the causes of various results.
The remainder of the paper is split into three sections. The first looks at parameter stability and simulation characteristics of RBA76(T). The second considers similar aspects of RBFI(M) while the last section draws some conclusions.
While similar tests are conducted for both models, the analysis is specific to each model. That is, the results are not used to judge whether one model is “better” than another but can be considered part of a procedure of reviewing each model.
In general, reviews of econometric models are not frequent but the results of such reviews are often instructive and sobering. The following quote from Tobin[3] seems relevant:
“The pressure to produce current forecasts, preferably accurate ones, has a number of consequences which are unfavourable for the improvement of the science and the art. First, the model builders frequently resemble generals refighting the last battle. They are always respecifying their models in ad hoc ways that would have avoided recent errors and to avoid outlandish future numbers … Second, the obsession with forecasting performance tends to make models sluggish … The models are caught by surprise when the real world economy does move sharply and cumulatively.”