RDP 8701: The Australian Demand Function for Money: Another Look at Stability 6. Concluding Comments

This paper has sought to address in a systematic way the question of stability of the demand for money, in several conventional models. Given the uncertainties over the performance of the test procedures discussed earlier, interpretation of the results is perhaps not straightforward.

Some of the tests applied – the CUSUMSQ and homogeneity tests for example – are quite demanding. But on the other hand, a number of tests have been employed – a conclusion of evident instability does not depend exclusively on these demanding tests.

For one model (Sharpe and Volker), the data period is quite long, so that complete stability over that entire period may be too stringent a standard. However, this model showed evidence of instability in even its original, shorter sample period. Indeed, on the Wald test in Table 6, each of the four models is shown to be unstable in its original sample period at the 5 per cent level of significance.

On balance, our conclusion from the results reported here is that the conventional models cannot easily be accepted as empirically stable. The mid 1970s still represent a problem period for estimated equations.

This conclusion is not new. Indeed, the 1970s caused difficulties for many empirical relationships that may previously have been considered quite dependable. It may be that money demand equations are not any more unstable over this period than most other equations.

Perhaps a newer result, though not necessarily a surprising one, is that the 1980's – a period not systematically addressed before in Australia – has also been a period in which conventional equations appear to break down.

It must be conceded that the stability or otherwise of broad money has not been considered. This is a shortcoming since broad money has attracted increasing attention in recent years, not least because of some of the changes to the financial system outlined in the previous section. One attraction of broad money is that shifts of business between banks and non-banks do not, in the first instance at least, affect it. In principle then, while shifts between direct and intermediated financing could still lead to instability in the demand for broad money, at least some of the potential sources of instability in the demand for narrow aggregates should not affecf it. Empirically, this remains to be demonstrated.

What are we to make of these results?

Laidler (1986) gives a practical interpretation of the problem, noting that conventional money demand functions may remain useful as long-run descriptions but suffer periods in which errors are large. Such periods will show formal instability. Laidler argues that, despite this, the demand functions may be useful as a framework for thinking about money if properly interpreted.

Taking the more general view of the tests reported in this paper as being simply another set of diagnostics, it could be argued that what we have taken to be instability is in fact evidence that the equations were wrongly specified in the first place. Going one step further would take us back to a closer examination of the models themselves, both theoretically and empirically. In this context, the work of Milbourne (1985). is sobering. He found that, for Australian data, no one type of model was able to completely dominate all others in empirical performance, suggesting that the whole gamut of monetary models may need to be re-thought.

Alternatively, returning to an earlier theme, it may be inappropriate to model the demand for money as a single equation in isolation from the economic system. This could mean that the results here are ambiguous.^[8]

A third alternative is to accept the results here as simply confirmation of what would be expected based on theory. The insights due to Lucas, which in the monetary sphere find their expression in what has become known as Goodhart's Law, suggest that the very attempt to depend on a stable demand for money for policy purposes may itself upset that stability.

Evaluation of the results should probably give heed to all of the above. To return to the notions of mispecification and simultaneity, Gordon (1984) provides a timely reminder of a key point about identification. In the context of an orthodox, though more completely specified, model of the demand for and supply of money, changes in the money supply process, such as a move away from interest rates and towards financial quantities as proximate policy objectives, introduce ambiguities into the econometric procedure: it is no longer clear that the researcher is actually estimating a demand function at all. Observed changes in estimated co-efficients could simply reflect the changing institutional regime, rather than instability in the demand for money itself.

An analogous point is made by Laidler (1982), as a result of drawing the distinction between adjustment of real balances at an individual level, and adjustment at an aggregate level. It is at the individual level that the microeconomic notion of adjustment costs is invoked to explain lagged adjustment. At this level, real balances adjust by changes in the level of nominal balances. But it is the aggregate level we observe. At that level, under the assumption that the supply of nominal balances is given by the authorities, real balances must adjust by the price level changing. Thus Laidler's point is that the empirical demand for money equation may in fact be some sort of price adjustment equation in disguise. Shifts in those equations can then be interpreted as changes in pricing behaviour in the economy.

Other possible explanations for unsatisfactory econometric performance include measurement error. “Money” may not correspond closely to measured financial aggregates. Similarly, “income” may not be well represented by GDP. It is possible that alternative measures of money may provide different results to those above.

For example, monetary aggregates constructed on a functional basis, where deposits of the same type are combined across a range of institutions (including non-bank financial intermediaries), may have a more stable relationship with income and interest rates, since they would be less affected by re-intermediation. However, the data for such aggregates are not easily available in Australia.

Clearly, these questions have not been satisfactorily resolved in this paper. But even if the results reported here are not taken to be indicative of instability in the demand for money so much as evidence of mis-specification – they do provide evidence that the conventional approach to the relationship between monetary aggregates, interest rates and economic activity has not yet yielded results which are sufficently robust for stability to be claimed.

Footnote

Against this point of view, Pagan and Volker (1980) utilised Sargan's Generalised Instrumental Variables Estimator to try to gauge the extent of any simultaneous equations bias in their single-equation model and [8]