RDP 9314: The Demand for Money in Australia: New Tests on an old Topic 1. Introduction

One of the most enduring analytical devices in macroeconomics has been the aggregate money demand function. As the theoretical and empirical appeal of monetarism grew through the 1970s, most countries devoted more attention to movements in the monetary aggregates. In an attempt to wind back relatively high inflation, several countries, Australia included, adopted the practice of announcing targets for monetary growth in order to tie down inflationary expectations.

Despite the initial theoretical appeal of this idea, experience with the use of monetary aggregates as targets, or even as indicators, proved disappointing. Financial innovation and regulatory changes often severely distorted the monetary outcome. Several countries either abandoned targeting altogether, as in the case of Canada, or downgraded its importance, as in the US and UK. Australia shared in this experience too. Monetary targets for M3 were discontinued in early 1985, the authorities noting that shifts in the relationship between M3 and economic activity which were occurring and likely to occur in future, rendered M3 unsuitable as a measure of money, and unreliable as a principal indicator of monetary policy (see Battellino and McMillan (1989) and Grenville (1990)).

While this decision was a matter of judgment at the time, it was well supported by subsequent studies demonstrating that conventionally estimated short-run demand functions for M3 were unstable, in the sense that the estimating equations failed standard prediction tests, both post-sample and within sample (Stevens, Thorp and Anderson (1987) and Blundell-Wignall and Thorp (1987)). Figure 1, which plots both the ratio of M3 to GDP and its trend based on its behaviour over the 1960s and 1970s, provides one illustration of the sometimes sharp and sustained shifts in velocity that have affected all the monetary aggregates at one time or another. In this case, attempts to target growth in M3 in the face of substantial and unpredictable changes in the demand for money, would have led the authorities to impose an unnecessarily tight policy regime. More recent research suggests that the use of monetary rules over the 1980s and early 1990s would have been sub-optimal in terms of the inflation and employment objectives of monetary policy (see Coelli and Fahrer (1992) and Blundell-Wignall et al (1992)).

There have been recent developments in time series analysis, however, which seek to uncover long-run relationships between variables by testing for cointegration between them. Accordingly, while there may be short-run deviations in the presumed relationship between money and income, there may be some underlying economic relationship to which the two series revert over time. If money and income do in fact move together over time, and if reversion to the joint long-run path is stable, then economic analysts are more likely to regard monetary aggregates as containing useful economic information. Figure 1 indicates that this is unlikely to be the case for M3, at least.

Analysis of the long-run relationship between money and income has been undertaken in many countries (for Japan, the UK and the US, for example, see Yoshida and Rasche (1990), Hall, Henry and Wilcox (1989), Hendry and Ericsson (1991) and Mehra (1991)), and also in Australia (see below).

This paper seeks to contribute to this literature within the standard analytical framework in which the long-run demand for real money balances is a function of real income (or other scale variable) and the opportunity cost of holding money. A range of measures of money, economic activity and opportunity cost are tested, with the emphasis not on mining the data for positive results, but on assessing the robustness of empirical relationships by testing sensitivity to sensible changes in specification and estimating technique. This seems especially important in view of the fact that the cointegration literature is relatively young, such that there is not universal agreement as to the ‘correct’ approach or interpretation of results.

Section 2 of the paper briefly revisits basic money demand theory, and summarises recent empirical work using Australian data. Section 3 gives details of the data used and presents preliminary analysis on the statistical properties of the various series under consideration. Section 4 presents the results from applying the Engle-Granger (Engle and Granger (1987)) and the Johansen (Johansen (1988), Johansen and Juselius (1990)) procedures of cointegration, and offers some interpretation and reflection on the results. Section 5 summarises the main points raised in the paper.