Summary and Conclusions

The many changes that occurred in the format of the monetary sector of the Treasury model in the 1980s are indicative of the difficulties which are inherent in the modelling of asset demand equations. The multi-asset portfolio approach, although theoretically appealing, has not as yet proved to be empirically tractable for most sectors. The limitations of the available aggregate time-series data, and the complications involved in the simultaneous estimation of large systems of equations, have prevented any large-scale development of empirical work on financial systems along these lines. Much of the difficulty in estimating financial models can be traced to the high degree of multicollinearity among the variables. Although it is not hard to find asset demand equations which fit the data well, the coefficients on individual variables cannot be reliably estimated. In the Treasury model, the problems are to some extent alleviated by the imposition of specific separability restrictions on the asset aggregation scheme adopted, and the consequent development of a sequential decision-tree framework within which to analyse the allocation process. Such an approach increases the number of asset allocation decisions which have to be considered, but reduces the number of variables that are relevant to each stage of the decision process. However, it is doubtful whether the multi-equation portfolio approach has yielded empirical results that have really been significantly different from, or superior to, those obtained by single-equation studies. One solution to this problem may lie in the use of ‘mixed’ estimation techniques, which use prior information about the model coefficients to augment the sample data. This prior information can be obtained from a variety of sources: theoretical calculations, cross-section studies, previous time-series studies on different data, or even practical experience. Such an approach was adopted by Backus et al. (1980) in a study of the US financial system, using the method pioneered by Smith and Brainard (1976), while Honohan (1980) has used similar techniques to analyse UK life assurance companies.