Financial Deregulation, the Demand for Money, and Monetary Policy in Australia

Factors contributing to the deregulation of the Australian financial system are reviewed and the implications of deregulation are discussed for the transmission mechanism of monetary policy, the interest elasticity of money balances, and the stability of money demand. Several models of money demand, using three definitions of money, are estimated by both fixed- and random-coefficient techniques. Empirical results provide evidence that financial deregulation has led to a breakdown in the well-behaved money demand relationships that held in the regulated financial environment.


Factors contributing to the deregulation of the Australian financial system are reviewed and the implications of deregulation are discussed for the transmission mechanism of monetary policy, the interest elasticity of money balances, and the stability of money demand. Several models of money demand, using three definitions of money, are estimated by both fixed- and random-coefficient techniques. Empirical results provide evidence that financial deregulation has led to a breakdown in the well-behaved money demand relationships that held in the regulated financial environment.

IN RECENT YEARS deregulation and innovation in financial markets have had widespread implications for the conduct of monetary policy in several countries. Simple and well-behaved relationships between money and nominal income that held in the regulated framework have apparently broken down in the changed financial environment. Indeed, several writers—including Fama (1980), Hall (1982), and Jonson and Rankin (1986)—have argued that the more or less stable and predictable demand for money relationships estimated in earlier studies were themselves by-products of the system of financial regulations in force over the period examined. As Jonson and Rankin observed, controls on interest rates limited the scope for interest rates to adjust; observed interest rates did not necessarily reflect market-clearing values, thus leading to situations of excess demand or supply in money markets. Quantity rationing in contractionary periods and the unwinding of rationing in expansionary periods established a positive statistical correlation between the supply of money and nominal income. Consequently, the lifting of controls in financial markets resulted in a deterioration of the old correlations between money and nominal income.

In the aftermath of financial deregulation and innovation and the apparent breakdown of stable and predictable money demand relationships, there has been a move away from strict monetary targeting as the primary objective of monetary policy. In some countries, increased reliance is now placed on other indicators of monetary policy, sometimes in association with published growth objectives for the monetary aggregates. For example, in Australia the practice of announcing conditional projections for the broad monetary aggregate, M3, was suspended in 1985.1 Although monetary growth rates continue to be closely monitored in Australia, this is done in concurrence with developments in a wide range of financial and real economic indicators.

This paper examines the factors underlying financial deregulation and innovation in Australia and investigates the implications of such deregulation for money demand stability and monetary policy. The remainder of the paper is divided into five sections. Section I provides a background to Australia’s regulated financial system as of the 1970s and gives an overview of the conduct of monetary policy in that environment. Section II then analyzes the factors that contributed to financial deregulation and discusses both the deregulatory measures that have been implemented so far and the changed nature of monetary policy, especially with regard to a less interest-sensitive demand for money, in such a deregulated financial system.

One consequence of financial deregulation in Australia was a surge in most of the monetary aggregates, particularly over the period 1984:4 through 1985:4. Recent studies of money demand in Australia by Veale, Boulton. and Tease (1985) and Stevens, Thorp, and Anderson (1987) have shown that various specifications of the demand for money, incorporating several definitions of money, exhibit instability over this episode—specifically, all severely underpredict real money balances. These studies, in accord with much recent work on money demand estimation, are based on the traditional econometric assumption of constant parameter values throughout the estimation period. Yet, as Goodhart (1986) has observed, in periods of financial deregulation there is a problem of distinguishing between responses to money supply shocks and underlying money demand changes. Goodhart further stressed that traditional econometrics is not able to deal with this problem. In addition, Hendry (1986), who has been extremely critical of existing methods of money demand estimation,2 has argued that one of the important drawbacks of current estimation methods is that they assume nonchanging parameter values even in periods of structural change.

To address these criticisms of money demand estimation and to investigate their implications, in what follows we estimate several models of money demand incorporating three monetary aggregates—the money base. Ml, and M3.3 Each of these models is estimated by using both a traditional nonchanging-parameter technique (ordinary least squares) and a random-coefficient estimation procedure developed by Swamy and Tinsley (1980). Accordingly, Section III of the paper discusses the money demand models and the data used to estimate them. Section IV presents the empirical results. First, each model, with each of the aggregates used in turn, is estimated with data for both the regulated and deregulated financial environments in order to assess the implications of deregulation for the elasticities of the determinants of money demand and for the stability properties of the models. Next, postsample forecasts for each model, estimated on the basis of both constant-parameter and random-parameter techniques, are presented and compared. The implications of the results, along with concluding observations, are provided in Section V.

I. Financial Deregulation

The financial system in Australia that had emerged by the 1970s was so highly regulated that it led to a strict segmentation of the financial marketplace. The system was centered on a small number of trading hanks that controlled important segments of the market and were closely regulated by the authorities. The trading banks operated the clearing system and had the sole right to issue checking accounts. They were the only institutions eligible to deal in foreign exchange and acted as agents for the Reserve Bank in implementing a comprehensive set of exchange controls. The savings banks, mainly affiliated with the trading banks, were essentially limited to the function of supplying housing finance. Finance companies and merchant banks4 were considerably less restricted 5 and consequently tended to meet the higher-risk credit demands that would not usually be considered by the trading banks. Finance companies, in many cases affiliated with trading banks, were the main suppliers of consumer credit; merchant banks, frequently representing the local presence of foreign banking groups, tended to specialize in trade finance and instruments tailored to the needs of corporate customers. The thrift institutions—building societies and credit unions—were subject to various state government controls; the former dealt predominantly in mortgage financing, the latter in consumer credit. Money market dealers were set up to make a market in short-term government securities and to provide safe investment opportunities for ultra-short-term cash; these institutions had access to Reserve Bank lender-of-last-resort facilities and were required to hold the bulk of their assets in government securities.6 Alternatively known as “authorized dealers,” they continue to constitute the “official” short-term money market. Other intermediaries that constituted a significant portion of the “captive market” (see below) for government securities were life insurance companies and superannuation funds.

Regulation of the financial sector involved several elements. First, a number of restraints applied to bank deposit rates and lending rates. For example, trading banks were prohibited from offering interest on large deposits ($A 50,000 and over) for periods of less than thirty days or on small deposits (less than $A 50,000) for periods of less than three months. Implicit in these short-term maturity controls on trading bank deposits was a prohibition on the payment of interest on current account balances.7 Likewise, savings banks were subject to a range of interest rate and maturity controls on deposits; they were also subject to interest rate ceilings on housing loans, small overdrafts, and personal installment loans. Second, both trading and savings banks were subject to specified asset ratios. For the trading banks, the rationale for these ratios involved both monetary policy and prudential objectives. For example, in an understanding reached with the Reserve Bank in 1956, trading banks agreed to maintain a minimum ratio (usually set at 18 percent) of liquid assets and government securities against their deposits. This so-called LGS convention typically referred to currency, deposits with the Reserve Bank, treasury notes, and other Commonwealth Government Securities (CGS). The deposits lodged with the Reserve Bank under the LGS convention were in addition to a designated proportion (usually set at 7 percent) of all deposits used as statutory reserves (the SRD ratio).8 Similarly, savings banks had been required to hold a long list of prescribed assets, particularly loans on the security of land, and public sector securities. The ratios to which the savings banks were subjected were partly for prudential purposes and partly to meet sectoral assistance objectives. Other financial institutions were also subject to portfolio restrictions. For example, life insurance companies and superannuation funds operated under the 30/20 Regulation; this referred to a tax concession given to these institutions in return for their holding no less than 30 percent of their assets in public securities, with at least 20 percent of total assets in CGS. Hence, this regulation contributed to the captive market—it forced or enticed loan funds into public securities by tax concessions.

Third, a number of regulations restricted entry to the banking sector over the postwar period, including the Banks’ Act, which required that applicants for a banking authority had to find a minimum of eleven unrelated shareholders judged “suitable” by the authorities, requirements for new banks to provide a full range of services, a long-standing prohibition on foreign interests taking a “significant” interest in a new bank, and the perceived difficulty of competing with established banks, which had a substantial base of non-interest-bearing deposits. Consequently, between 1945 (when the Banking Act to regulate banking was passed) and 1981 the only new banks that were established were savings bank subsidiaries of existing banks.9 The net result of these controls on entry was that, through a series of mergers, by the early 1980s the Australian banking system developed into an oligopolistic sector; four rather homogeneous trading banks dominated the banking system, holding more than 85 percent of all trading bank assets and offering a restricted range of services (in comparison with some banks overseas). It also resulted, however, in a situation in which banks often were not able to compete effectively with other financial institutions (OECD (1985)).

Monetary Policy in an Environment of Regulation

In March 1976 the authorities began announcing annual projections for the growth of M3.10 One consideration underlying this decision was that during the 1970s the velocities of the monetary base, Ml, and M3 all exhibited fairly steady upward trends, with M3 showing the highest stability; in the ten years to 1979/80. base velocity with respect to nominal gross domestic product (GDP) rose at a 4.8 percent annual rate, Ml velocity rose 2.9 percent, and M3 velocity rose 1.5 percent. However, because of what was considered to be the insufficient short-term controllability of monetary aggregates and the inadequate short-term stability of the relationship between the aggregates and ultimate economic objectives, the authorities stressed that the projections were not “rigid targets” but “conditional projections.”11 In particular, under the managed exchange rate regime then in operation, any tightening—for example, in monetary policy—could be offset by capital inflows, thus weakening the link between monetary policy action and the monetary aggregates.

For some time the main operational objective of monetary policy had been on the control of the lending capacity (that is, “free liquidity”) of the banks as a means of controlling liquidity conditions within the financial system as a whole. Monetary policy relied on: (1) changes in the reserve (that is, SRD) ratio. (2) quantitative guidance, and (3) interest rate controls on bank deposits and bank lending. In this contest the LGS assets of banks were viewed as the main determinant of bank lending. For example, if the Reserve Bank considered the level of free liquidity to be excessive, it would—in addition to informing the banks of this view by way of consultations and (if desired) increasing the interest rate on CGS—raise the SRD ratio and thus force the banks to reduce their LGS assets.12 Because of the existence of controls, changes in the reserve ratio had a direct impact on banks’ lending. This approach depended crucially on the CGS being readily cashable at the holder’s discretion, without incurring significant capital losses. In fact, under the “tap system” (in effect during the 1970s) of selling government securities whereby the price (and not the quantity) of these securities was fixed, there was very little risk of capital losses. Although open market operations were used to affect liquidity conditions, they were not the primary instrument of monetary policy. Indeed, one consequence of the tap system was that the availability of securities on the primary market discouraged the development of a secondary market, impairing both portfolio adjustments by holders of government debt as well as the government’s ability to conduct open market operations (see Davis (1985)).

Factors Contributing to Deregulation

The system of regulatory measures in force at the end of the 1970s was constructed to protect investors and to maintain confidence in the stability of financial markets and institutions. Also, direct regulation was used to promote certain social and sectoral goals. For example, interest rate ceilings on housing loans by savings banks were instituted to make these loans affordable to low-income earners. Finally, many of the interest rate controls and portfolio restrictions on both the banks and nonbank financial intermediaries were seen as necessary to assist in the implementation of monetary policy and the sale of government securities (OECD (1985)). In particular, an important justification for the special treatment of banks in the financial system was their role in money creation. Banks alone were empowered to offer transactions accounts, and monetary policy was conducted through controls on these and other deposits (Australia (1984, p. 91); hereafter cited as the Martin Report).

The regulations had allowed the banks to emerge as highly profitable institutions, but with a declining market share and at a high cost to depositors. Several developments during the 1970s, however, altered the impact of regulations on financial institutions and had important ramifications for the conduct of monetary policy.13 One was the upsurge of and increased variability in inflation. High and variable rates of inflation increased the opportunity cost of holding money balances, thereby inducing a move away from lower-yielding money balances. This increased sensitivity of savers placed pressure on the financial system to offer new financial instruments that made it possible to economize on lower-yielding financial assets. The increased variability of inflation led investors to prefer short-dated over longer-dated claims; maturity controls imposed on the banks restricted their ability to meet this demand. The limited flexibility of banks in the face of high and variable inflation rates afforded an opportunity for nonbank financial intermediaries to expand. Consequently, over the decade of the 1970s, money market corporations, building societies, and credit unions experienced rapid growth (Harper (1986)).

A second factor during the 1970s was the progressive increase in the size of government budget deficits (Waterhouse (1985)). The net Public Sector Borrowing Requirement (PSBR), which averaged less than 2 percent of GDP in the first half of the decade, averaged 4.4 percent of GDP during the five years to 1979/80. The effect of this rapid growth in the PSBR was to increase the government’s role as a competitor for household savings, placing considerable pressure on existing methods for the sale of public securities.

The third factor to emerge during the 1970s was the rapid advance in financial communications and data-processing technology; technological improvements greatly increased the opportunities for financial innovation and lowered the transaction costs involved in shifting funds among assets. This development, in turn, made it easier for unregulated financial institutions to grow and for regulated institutions to innovate in order to avoid the regulations. An example of the former phenomenon was the growth in cash-management trusts as a means of providing retail investors (by pooling their funds) access to higher interest rates available in the relatively unregulated wholesale finance markets. An example of the latter was the development of the “sweep” account by banks as a device for the avoidance of the official requirement not to pay interest on bank current deposits.14

A fourth factor operating in the 1970s was the strengthening of links between domestic and international markets, in part reflecting an increased presence of foreign-owned institutions (principally representative offices of foreign banks) in domestic financial markets. This increased presence raised the potential for international capital transactions, both borrowing and lending. The stronger linkage of domestic and international markets increased the access to international capital flows as a source of funds to offset fluctuations in domestic liquidity or to achieve a higher rate of return on investment. Consequently, the Australian financial system increasingly reflected the behavior of external influences—such as interest rate portfolio-substitution pressures. The volatility and unpredictability of international capital flows posed an intensifying problem for the conduct of monetary policy under the managed exchange rate system (Harper (1986, pp. 41-42)).

Deregulatory Measures

One result of the foregoing developments was a weakening in the competitive position of regulated financial institutions relative to that of unregulated financial institutions. In this regard, there was a sharp fall in the market share of banks between the 1960s and the early 1980s. In 1962, for example, the assets of trading and savings banks together accounted for about 53 percent of all assets of financial institutions; by 1983 this share had fallen to 41 percent. Meanwhile, the growth of non-bank financial intermediaries operating beyond the direct influence of the Reserve Bank, the development of new financing techniques by banks, and the increasing integration of domestic and overseas markets contributed to a steady erosion in the ability of the monetary authorities to control monetary conditions. In response, a reform process was initiated with the setting up of the Committee of Inquiry into the Australian Financial System (the Campbell Committee) in January 1979 (Australia (1981); hereafter cited as the Campbell Report). The Committee adopted the view that the efficiency of the system would be maximized in an open competitive environment and, accordingly, made recommendations aimed primarily to break down the mass of regulations that had long maintained the system in a segmented and highly structured form. A subsequent review of the financial system (the Martin Report) in 1984 reaffirmed the broad recommendations of the Campbell Committee to deregulate the system.15

The Martin and Campbell Reports propelled a major reform of the Australian financial system, including opening the system to new (including foreign) banks. Among the most notable changes have been (1) introduction of a tender system for selling treasury notes (in December 1979): (2) removal of ceilings on interest rates on some deposits payable by trading and savings banks (in December 1980); (3) in late 1983. lifting of virtually all exchange controls and the float of the Australian dollar; (4) in August 1984, abolition of remaining controls on bank deposit interest rates; and (5) in April 1985, removal of interest rate ceilings on bank loans under $A 100,000 (other than those for owner-occupied housing). The major changes that were announced between December 1979 and September 1988 are summarized in Appendix I (Table 4).

II. Monetary Policy and Money Demand in the Deregulated Financial System

Financial deregulation has set in motion changes in both the manner in which monetary policy is transmitted to the real economy and the stability and interest elasticity of the demand for money. With regard to the transmission mechanism, the introduction of a tender system of selling government securities and the move to a floating exchange rate regime increased the monetary authorities’ potential control over injections of liquidity into the domestic monetary system, thus enhancing their ability to use open market operations to influence domestic monetary conditions. Financial deregulation also contributed to a weakening of nonprice credit rationing as a channel of monetary policy and strengthened the role of market forces in determining financial and credit flows. Consequently, the effects of monetary policy are increasingly transmitted through open market operations to the real economy through changes in interest rates.16 With greater competition in the financial sector, changes in interest rates tend to spread quickly through the whole range of financial assets and liabilities. Nevertheless, the transmission of monetary actions to the real economy has probably lengthened compared with the previous regulated system, which relied on quantity rationing. Specifically, in the deregulated financial environment the volume of deposits is determined by both demand and supply. Consequently, any tightening of monetary policy by the Reserve Bank will induce a rise in deposits’ rates, resulting in an increase in the supply of deposits and offsetting to some extent the Reserve Bank’s effort to reduce the growth of money. Thus, financial institutions (particularly banks) are now better able to protect their deposit base and to sustain their lending than they had been in the regulated framework, in which the volume of deposits was primarily demand-determined.17

The demand for credit may also have become less sensitive to interest rates in the deregulated system. For example, increased use of floating interest rates and more innovative and flexible loan packages may have resulted in less discouragement to marginal borrowers as rates rise. Borrowers and lenders may also have become more accustomed to interest rate variability and, as a result, may not materially alter their behavior until interest rates are perceived to have shifted in a sustained manner.18

The Demand for Money

A precondition for the successful implementation of monetary targeting is the existence of a stable and predictable relationship between the targeted monetary aggregate and economic activity. The recent financial changes, however, have altered the interest elasticity of the demand for money and have made it more difficult to differentiate among the various aggregates, thus contributing to money demand instability.

With regard to the interest elasticity of the demand for money, the move to market-related yields on bank deposits has resulted in a decline in the interest elasticity of the monetary aggregates (such as M3). Financial deregulation has resulted in an increased share of financial instruments within M3 that offer market-determined rates. As Judd (1983) has noted, the yields on those instruments are closely related to the rates on assets held by depository institutions, leaving the differential unchanged. Because the opportunity cost of holding money balances is the spread between the own rate of return on money and the market yield on substitute financial assets, that deposit rates move closely up or down with market rates in the deregulated system signifies that the opportunity cost of money varies less than do market rates. Consequently, the demand for deposits, however sensitive to the differential between market rates and deposit rates, has become less sensitive to the general level of interest rates; a larger change in market yields is required to achieve a particular change in relative yields than was necessary when deposit rates were controlled. Accordingly, if interest rates are used to effect changes in the money supply, larger changes in market rates are now required to attain given changes in the money supply than was the case in the regulated financial system.

In an open economy framework, the reduced interest elasticity of money demand also has implications for exchange rate behavior, and therefore, for the transmission mechanism of monetary policy. In this context, Dornbusch’s well-known extension (1976) of the Mundell-Fleming model is worth noting. Dornbusch assumed that asset markets adjust fast relative to prices in goods markets. Under the usual Mundell-Fleming assumptions (for example, a small open economy, uncovered interest rate parity, regressive exchange rate expectations), in the short run the exchange rate overshoots its equilibrium value. In the Dornbusch model (with output assumed to be at the full-employment level), the degree of overshooting after an expansion in the money supply can be shown to be ds/dm = 1 + 1/a1a2, where ds/dm is the derivative of the nominal exchange rate relative to the money supply, a1 is the interest response of the demand for money, and a2 is the regressive exchange rate coefficient. Overshooting will thus be greater the less is the interest response of money demand.19 Consequently, the impact of monetary policy by way of the external account will be greater in the short run the less interest-elastic is the demand for money.

Financial deregulation has also contributed to shifts in the demand for money. Some deregulatory measures, such as the entry of new banks and the decontrols on deposit interest rates, boosted the growth of banking aggregates relative to other intermediaries and direct financing. Increased competition following deregulation also encouraged nonbank financial intermediaries to seek business from areas previously financed directly, which added to the process of intermediation and thus to the growth of broad money. In addition, changes in the financial environment took place during a period (beginning in early 1983) when a disinflationary process had set in. Consequently, determinants of money demand—notably interest rates and inflationary expectations—changed sharply and perhaps unpredictably. These factors contributed to a surge in M3 growth. In the year to September 1985, M3 growth reached 19.6 percent compared with a rate of 11.7 percent in the year to September 1984. As a result of the surge in M3 growth, the income velocity of M3 fell by over 5 percent in 1985/86.20 The acceleration in the growth of the monetary base was less pronounced, from 13.3 percent to 14.6 percent. Consequently, the income velocity of the base fell less sharply (by 3,5 percent) in 1985/86. Ml growth accelerated from 9.2 percent to 11.1 percent. Ml velocity fell by 2.8 percent in 1985/86. The lower rate of Ml growth reflected the fact that interest rates on demand deposits, a large component of Ml, did not increase in line with rates on other financial instruments included in M3. resulting in a shift into M3 deposits.21

The foregoing shifts in money demand reflect adjustments from one financial regime to another and may, therefore, be transitory. However, financial deregulation may also involve a more sustained degree of instability of money balances. For example, since M3 now includes a larger proportion of financial instruments yielding market-related rates, deregulation probably has induced an increased flow of investment balances into M3, resulting in an amalgamation of funds held for transactions balances with those held for investment balances. Because investment balances tend to be more responsive than transactions balances to small changes in the broad range of interest rate spreads (including yields on long-term bonds and common stocks), the potential exists for M3 to be dominated by shifts in the composition of the public’s portfolio rather than by changes in income and prices, resulting in a greater degree of instability in the demand for money (Judd (1983)).

To summarize, financial deregulation has lowered the interest elasticity of the demand for money and has furnished the potential for both transitory and permanent money demand instability. In terms of the Hicksian IS-LM apparatus, this implies that the LM schedule—which represents various combinations of real expenditures and nominal interest rates that equate the demand for and supply of real money balances—has become steeper; other things remaining equal, larger changes in interest rates are associated with given changes in the supply of money. That money balances are now held for investment purposes, as well as to satisfy transactions needs, however, means that the variance associated with given monetary policy changes has also increased; a wider band of uncertainty is associated with any given change in the quantity of money.

III. Models and Data

How has financial deregulation altered the relationship between the demand for money and its determinants? Which of the monetary aggregates has exhibited less structural change in the deregulated financial environment? Given the large forecasting errors exhibited by traditional econometric techniques, as applied in conventional money demand models, in predicting the growth in real money balances (underpredic-tions for the base, Ml, and M3) from the end of 1984 through the end of 1985, are there available models or estimation methods that can do better in predicting outcomes so as to provide—even in the event of changed relationships—the monetary authorities with reliable forecasts?

To address these issues, we proceeded as follows. First, a conventional money demand model was estimated over quarterly intervals for 1967:1 through 1987:2. Next, the model was estimated over the two subintervals, 1967:1-1979:4 and 1980:1-1987:2, which correspond, respectively, to the regulated and the deregulated financial environments. (Clearly, it is impossible to have a clean break between subperiods, since the dereg-ulatory measures were phased in during the second subperiod.) With regard to the issue of predictability, several other money demand models are also introduced and are estimated over the period 1967:1-1984:3 on the basis of both fixed-coefficient and random-coefficient methods. In testing for predictability, the reason we chose to end the estimation period at 1984:3 is that the demand for money exhibited the greatest instability in the interval 1984:4-1985:4. Accordingly, predictions for each of the aggregates using the alternative models and estimation techniques are made beginning in 1984:4. Measures of forecast accuracy are presented for the subintervals 1984:4-1985:4 and 1986:1-1987:2 and for the entire interval.

The Partial Adjustment Model (Model 1)

As noted, our point of departure was the conventional, partial adjustment model which treats the demand for real money balances in the following form (all variables are in logarithms):


where Md is a measure of real money balances (that is, nominal money balances divided by the price level); Y is a scale variable such as income or wealth, r is an opportunity cost variable, r0 is the own rate of return on money, and u is an error term.22 In estimating equation (1), real money balances were defined as the base. Ml, and M3, respectively, each divided by the GDP price deflator. For the other regressors in equation (1), we used real GDP as the scale variable, the interest rate on ten-year government bonds as the nominal opportunity cost variable, and the interest rate on trading bank fixed deposits (as compiled by the Reserve Bank of Australia) as the own rate of return on money. A more detailed description of the data used is provided in Appendix II.

Price Expectations (Model 2)

Although a good deal of recent empirical work on Australian money demand has involved estimating the partial adjustment model by using various definitions of money as well as various definitions and specifications of the regressor variables, none of the work—at least to our knowledge—has included a variable representing price expectations. Yet such a variable is particularly relevant if the historical data are generated for the most part in the context of a regulated financial environment in which nominal interest rates do not necessarily reflect market-clearing values because of ceilings and maturity controls on deposits. That being the case for most of the estimation period in Australia, we also estimated another specification of the partial adjustment model as follows:


where j=0nwjp˙tjis a measure of inflationary expectations, ṗ is the annualized consumer price inflation rate, and wj is the weight attached to inflation in period r—j in formulating price expectations.

There are several possible lag procedures to estimate the weights attached to previous inflation rates in determining inflationary expectations, but we experimented with the following four: (1) least-squares estimates, (2) posterior mean for Shiller’s smoothness prior on lag coefficients, (3) the Almon polynomial distributed lag, and (4) Ridge regression. Each of these procedures was implemented using several lag lengths and both with and without a correction for first-order serial correlation. After dividing the sample period. 1967:1-1987:2, into the estimation period 1967:1-1984:3 and the prediction period 1984:4-1987:2, the choice of lag length was predicated on which length provided the best forecast of the dependent variable, in terms of yielding the smallest root mean-square forecast error, for the prediction period under each procedure. The addition of a price expectation variable was found to improve the forecasts based on model 1. A more detailed description of the lag procedures used, and the coefficients assigned to the weights, is presented in Appendix III.

The Buffer Stock Model (Model 3)

A substantial and influential body of money demand literature has been devoted to estimation of the buffer stock model. The buffer stock model itself was pioneered by economists at the Reserve Bank of Australia (for example, Jonson (1976) and Jonson and Taylor (1978)) in the context of the Reserve Bank’s macroeconometric model, but more recently it has been used to estimate money demand relationships in a variety of countries.23

The buffer stock model has been used more frequently in part because of several troublesome attributes associated with the partial adjustment money demand model. In particular, empirical estimates of model 1 have often displayed parameter instability, indicating that the demand for money is not stable. Second, an implication of model 1, which does not seem to be borne out in the real world, is that changes in the money supply are accompanied by interest rate overshooting in the short run (see Judd (1983) for discussion).

In response to these difficulties associated with the conventional money demand specification, the buffer stock approach formulates the demand for money in the following form:






and Z is a set of variables that agents assume have a systematic influence on the money supply, M* is the anticipated money supply, g is a vector of coefficients to be estimated, and v is an error term. Unanticipated changes in the money supply. Mu = (Mt-ĝZt) are the residuals from equation (5). As Laidler (1987) demonstrated, the buffer stock model can be interpreted as a reduced-form equation of a complete macro model in which prices adjust sluggishly.24

Incorporation of the term (M - M*) in equation (2) is supposed to be able to account empirically for why short-term overshooting does not occur, at least to the extent implied in model 1 (see Judd (1983)). In addition, if the buffer stock model—which consists of equations (3)-(5)—is the “true” model of money demand, it means that model 1 is misspecified. Omission of (Mt - Mt*) as a regressor would result in biased coefficients on the included explanatory variables to the extent that these included variables are not orthogonal to the omitted variable. This may be one reason that estimated equations based on model 1 have exhibited parameter instability. In particular, incorporation of (M—M*) in the money demand specification where the buffer stock model is the true model means that short-run variations in the observed stock of money would not have to be induced by shifts in people’s underlying demand for money; such variations could also result from independent, exogenous changes in the quantity of money.

The buffer stock model contains at least one specification problem, since the M, component of the unanticipated money supply variable may be correlated with the numerator of the dependent variable, real money balances (see MacKinnon and Milbourne (1984)). Accordingly, in an attempt to deal with this problem, the monetary base was used instead of the money supply to generate a series on the unanticipated monetary base. The motivation in adopting this procedure was that changes in base money are likely to precede and influence changes in money (Ml and M3). Therefore, the unanticipated monetary base was entered into equations for Ml and M3 to test the buffer stock model.25 A second degree, 12-period (beginning in period t - 1) polynomial distributed lag of the base on its past values was used to generate the anticipated component of Mu. Also, since the addition of a price expectations variable in model 2 was found to improve the forecasting performance of the partial adjustment model, the buffer stock model was estimated on the basis of the same approach used to generate price expectations in model 2. Specifically, the same four lag procedures were employed with and without first-order serial correlation corrections; the price expectations series that was retained in each model was that which yielded the best forecast.

Random-Coefficient Estimation

What is random-coefficient estimation, and why use the procedure during periods of structural change? Essentially, under random-coefficient estimation, each coefficient to be estimated comprises two components: a nonstochastic component that corresponds to the usual fixed coefficient and a stochastic component that is represented by a first-order autoregressive process (for details see Swamy and Barth (1987)). The motivation for using it is that in periods involving prolonged structural changes (as opposed to once-and-for-all shifts), it is unreasonable to expect all of the ‘‘noise’1 to affect only the disturbance term in an equation. The noise should also affect all coefficients in the equation, and the more noise there is, the greater would be the expected impact on all coefficients. As mentioned, financial deregulation in Australia was instituted over a period of years and should have been expected to change the coefficients on all the determinants of the demand for money (for example, to have reduced the interest elasticity). Accordingly, it would seem appropriate to test the forecasting properties of random-coefficient estimation in the Australian context. To the extent that money demand relationships were not affected greatly by structural changes, the advantage of using random-coefficient estimation for purposes of forecasting would be less (as will be demonstrated).

The stochastic coefficient representation of the models estimated here is presented in equations (6) through (8):


where xt, βt, β¯, ɛt, vt are all k × 1 vectors, γ and Δ are k × k matrices, xt represents the vector of the explanatory variables, and βt is a vector of coefficients. Note that equations (6)-(8) represent a special case of a more general variable-coefficient specification that allows one to describe variations in coefficients with explanatory variables, allows for’simultaneous equations” complications, and allows for more general specifications of the error processes.26

In equation (7), each coefficient in each period, βit, has two components: a time-independent fixed component,β¯i, and a time-dependent stochastic component, ɛitt is a vector stationary first-order autoregres-sive process, which can also be represented as a vector moving-average process). Combining equations (6), (7), and (8) reveals that the stochastic coefficient representation can be viewed as a fixed-coefficient model with errors that are both serially correlated and heteroscedastic, where the form of serial correlation and heteroseedasticity is quite general:


Estimation of β¯ can be viewed as an application of generalized least-squares or Aitken estimation if it is assumed that γ and Δ (the covariance matrix) are known. Swamy and Tinsley (1980) have developed a minimum average, risk-linear estimator that, for given a priori moments of β¯, γ, and Δ, can be shown to be more efficient than the Aitken estimator. Because β¯, γ and Δ are not known and must be estimated. Swamy and Tinsley developed an iterative estimation procedure in which γ and Δ are initially arbitrarily chosen but through iteration are consistently and efficiently estimated after initial consistent estimates of β¯, γ, and Δ are obtained.27

IV. Empirical Results

To summarize, three money-demand models were estimated for the monetary base. Ml, and M3: the partial adjustment specification (model 1), partial adjustment with price expectations (model 2), and the buffer stock model with price expectations, where the monetary base was used to estimate the monetary shock term (model 3). The partial adjustment model was first estimated over the period 1967:1-1987:2, and over the subintervals 1967:1-1979:4 and 1980:1-1987:2. This was done to assess the effects of deregulation on the determinants of the demand for each of the aggregates. Next, all three models were estimated over the period 1967:1-1984:3 and then used to generate postsample forecasts of the dependent variables. Four lag procedures were used to formulate price expectations, and they were used with and without first-order serial correlation corrections. The models were estimated on the basis of fixed-coefficient and random-coefficient estimation procedures with the aim of comparing prediction results both across models and with respect to the estimation techniques. The forecast interval was 1984:4-1987:2, along with the two subintervals 1984:4-1985:4 and 1986:1-1987:2.

Table 1 reports the regression results for the partial adjustment model estimated over 1967:1-1987:2 and over 1967:1-1979:4 and 1980:1-1987:2. Equations denoted with the suffix “a” pertain to the entire period, whereas equations with the suffixes “b” and “c” refer to the earlier and later subperiods. respectively. Equation set (1) presents results for the monetary base, equation set (2) for Ml. equation set (3) for M3 without the own rate as an explanatory variable, and equation set (4) for M3 with the own-rate series. All the coefficients have a priori plausible values and the right algebraic signs. In all four sets of regressions there is a marked decline in the absolute value of the interest elasticity of real money balances over the two subintervals: from—0.17 to -0.02 for the base, from -0.13 to -0.09 for Ml. from -0.07 to -0.03 for M3 without the own rate as a regressor. and from -0.11 to—0.07 for M3 with the own rate. Indeed, with the exception of Ml (equation set (2)), in all the other equation sets the opportunity cost variable becomes insignificant in the second subinterval. (The fact that Ml remains sensitive to the opportunity cost variable helps explain, as noted, the shift from Ml balances to M3 balances during 1984 and 1985.) The coefficients on all four equation sets indicate substantial structural change between the regulated and deregulated financial environments, and formal stability tests in general support this result.28

Table 1.

Regression Results: Demand for Monetary Aggregates

(Quarterly data, 1967:1–1987:2)

article image
article image
Sources: Reserve Bank of Australia, Bulletin (Sydney), various issues, and data provided by the Research Department of the Reserve Bank; and Department of the Treasury, The Round-Up (Canberra), various issues.Note: PGDP is the implicit GDP price deflator: figures in parentheses are absolute values of the t-ratios; R¯2 is the adjusted coefficient of determination; and DW is the Durbin-Watson statistic.

The regression results support the hypotheses put forth in Section II that, first, financial deregulation has altered the relationships between the monetary aggregates and their determinants, making monetary targeting, which relies on a stable demand for money, more difficult to implement; and, second, that money demand is less interest-elastic in the deregulated financial environment than in the regulated system. To the extent that these stochastic results reflect the operation of a non-stochastic system, larger changes in interest rates are now required to achieve given changes in the quantity of money compared with the regulated system.

The regression results using alternative models and estimation techniques are reported in Table 2.29 Models estimated using the fixed-coefficient procedure are denoted as “FC” and those with the random-coefficient procedure as “RC.” Models with the base as the dependent variable are denoted with the suffix “a,” those with Ml with the suffix “b,” and those with M3 with the suffix “c.” All the coefficients have a priori plausible values and right algebraic signs. In model la.FC, for example—the fixed-coefficient partial adjustment model for the base—all the slope coefficients are significant. Model la, FC has a short-run income elasticity of 0.24 and implies a long-run income elasticity of 1.09. Random-coefficient estimation does not significantly change the values of the coefficients on the determinants of the base and Ml in model set (1), but does change the coefficients on the determinants of M3. This is a reflection of the fact that M3 was more unstable over the estimation period than either the base or Ml; as noted, random-coefficient estimation has a more pronounced effect on equations that exhibit larger structural changes.

Table 2.

Alternative Money Demand Models: Fixed-Coefficient and Random-Coefficient Results

(Quarterly data, 1967:1–1984:3)

article image
article image
Sources: See Table 1.Note: Figures in parentheses are the absolute values of t-ratios; ρ is the coefficient of serial correlation.

Estimated on the basis of the Almon lag. Sum of 22 (t – j, j = 0 through 21) coefficients.

Estimated on the basis of the posterior mean. Sum of 23 (t – j, j =0 through 22) coefficients.

Model set (2) includes the effects of price expectations. Although all four of the lag procedures for estimating price expectations improved the forecasting performances of the models, the Almon procedure for determining expectations resulted in lower root mean-square errors for the base, whereas the Shiller lag worked best for Ml and M3. (The root mean-square errors for each of the alternative lag procedures in models (2) and (3) are reported in Appendix III.) The coefficient on price expectations is more pronounced in the equation for M3, where it equals—1.49 (model 2c, FC), and is significant, indicating that nominal interest rates have not fully captured the Fisher effect over the estimation period; however, the incorporation of the price expectations term does not significantly alter the coefficients of the other nonconstant explanatory variables. Models 2c, FC and 3c.FC were the only instances in which an autocorrelation correction was found to improve the forecasting accuracy of models 2 and 3.

Model set (3, FC) shows the effect of the monetary shock term on the demand for Ml and M3 balances. The monetary shock variable has a positive impact—equal to 0.22 and 0.13—on Ml and M3 balances, respectively, and is significant (in accordance with the buffer stock hypothesis). The coefficients of the other regressors are near their corresponding fixed-coefficient values as obtained in model sets (2b) and (2c).

With the exception of equations in which M3 is the dependent variable, the means of the coefficients of the models estimated on the basis of the random-coefficient procedure are close to the values obtained with the fixed-coefficient technique. The coefficients on the price expectations variables are highest in equations for M3:—1.57 in model 2c.RC and -1.25 in model ic.RC.30 In interpreting the results of random-coefficient estimation, however, care should be taken not to interpret the (-statistics of the estimates in the conventional manner. In some instances the mean estimates of random coefficients are not significantly different from zero in the usual way that a t-ratio is less than 2.0 in absolute value (for example, in model lc.RC), whereas in some other instances they are far more significant than the corresponding fixed-coefficient estimates (for example, in model la.RC). Even if the power of the r-test is very high, the interpretation of this result under random-coefficient estimation differs from that of the result applicable to fixed-coefficient estimation. The reason that this is so is that a nondegenerate distribution of a random coefficient is not the same as a degenerate distribution of a fixed coefficient at zero, even if the mean of the former distribution is zero.31 A nondegenerate distribution with mean equal to zero may help to improve the accuracy of a forecast, whereas a degenerate distribution at zero does not.

The root mean-square errors for the three monetary aggregates over the postsample period, 1984:4 through 1987:2, and the two subperiods, 1984:4-1985:4 and 1986:1-1987:2 are presented in Table 3.32 From the forecasts based on the conventional partial adjustment model, several generalizations can be drawn. First, over the forecast period as a whole, the forecast of the base yielded a smaller root mean-square error(s) (1.166 percent) than the forecasts for Ml (1.86 percent) or M3 (3.627 percent), although the root mean-square error(s) of different variables are not strictly comparable. Second, despite the fact that M3 growth surged unpredictably over 1984:4-1985:4, the root mean-square error(s) for M3 using a fixed-coefficient estimation was higher in the second subinterval (at 3.851 percent) than in the first (3.339 percent). The same applies to the base. This indicates that money demand instability continued to remain a problem throughout the forecast period. Third, random-coefficient estimation improves the forecasting performances of all three aggregates over the forecast interval as a whole and for each subinterval. with the exception of 1986:1-1987:2 for Ml. Random-coefficient estimation is relatively more successful in reducing the root mean-square error(s) for M3 (to 1.902 percent over the period as a whole), which is to be expected given that the M3 equations exhibited greater instability than did the equations for the base and Ml.

Table 3.

Root Mean-Square Forecast Errors: Real Money Balances, 1984:4–1987:2

(In percent)

article image
Sources: See Table 1.

For the partial adjustment model with price expectations, under a fixed-coefficient estimation this model outperforms the partial adjustment model without expectations over the whole period. For the base, the root mean-square error(s) falls to 0.795 percent (from 1.166 percent) and for Ml it falls to 1.660 percent from 1.863 percent; the reduction in the root mean-square error(s) is the most pronounced for M3, falling to 1.394 percent (from 3.627 percent). In fact, the partial adjustment model with price expectations (model 2), estimated with the fixed-coefficient procedure, outperforms for each aggregate the results of the random-coefficient procedure in the partial adjustment model without price expectations (model 1). However, estimation using random coefficients enhances the forecasting results of model 2 compared with fixed-coefficient estimation of model 2 for each of the aggregates over the 1984:4-1987:2 period, and for each subperiod except the first for the base. Among all the aggregates, the lowest root mean-square error(s) over the entire forecast interval is for M3 (0.724 percent) using model 2 and a random-coefficient estimation.

The buffer stock model with price expectations, as applied to Ml and M3, produces a marginal reduction in root mean-square error(s) under a fixed-coefficient estimation over the entire forecast period—from 1.660 percent to 1.594 percent for Ml, and from 1.394 percent to 1.338 percent for M3. However, although the application of a random-coefficient estimation to the buffer stock model generally reduces the root mean-square error(s) relative to a fixed-coefficient estimation, it does not produce root mean-square error(s) as low as those derived with regard to a random-coefficient estimation of model 2.

Thus, in most cases the estimation of money demand models with random-coefficient estimation significantly decreased the root mean-square error(s) compared with conventional fixed-coefficient estimation. The incorporation of price expectations also helped to reduce significantly the root mean-square error(s) of the forecasts for each of the aggregates. The buffer stock model further improved the forecasting results when fixed-coefficient estimation was used; however, the buffer stock models did not further reduce the forecast errors when estimated with the random-coefficient procedure.33 The procedure that yielded the lowest root mean-square error(s) was random-coefficient estimation of the expectations-augmented partial adjustment model.

V. Conclusions

The conduct of monetary policy in Australia has changed substantially in recent years in conjunction with the deregulation of the Australian financial system. In the regulated financial system, monetary policy consisted of changes in the Statutory Deposit Ratio, interest rate and maturity controls on deposits and lending, and quantitative lending guidance. Because of the existence of controls, changes in the reserve ratio had a visible and direct effect on banks’ lending. A number of factors emerged during the 1970s that gave impetus to the deregulation of the financial system, including high and more variable inflation, larger government budget deficits, the rapid advance in communications and data-processing technology, and the emergence of stronger links between domestic and international markets. Monetary policy in the deregulated financial environment is implemented mainly through open market operations that initially affect short-term interest rates and, with a lag, other rates in the spectrum. In contrast to the earlier financial environment in which the volume of deposits was primarily demand-determined, in the deregulated system the volume of deposits is determined by both demand and supply. With banks better able to protect their deposit base and their lending, and with the demand for credit less sensitive to interest rates (because, for example, of floating interest rates), the transmission mechanism of monetary policy has lengthened.

Financial deregulation has also distorted the relationship between the monetary aggregates, on the one hand, and real economic activity and nominal interest rates, on the other. As a result, conventional money demand models severely underpredicted the surge in the targeted aggregate, M3, which occurred in late 1984 and in 1985, culminating in early 1985 in the abandonment of the practice of announcing conditional projections of M3.

The results of this paper indicate that incorporation of price expectations, in addition to nominal interest rates, in money demand models substantially reduces forecast errors (under both fixed-coefficient and random-coefficient estimation) of real money balances; this finding supports the view that price expectations are particularly relevant as a proxy for the opportunity cost to holding money in models estimated on the basis of data taken largely from a sample period in which nominal interest rates may not have reflected market-clearing values.

The paper discussed both transitory and permanent factors that may have contributed to money demand instability in Australia. In either case, there are reasons to caution against forming predictions of real money balances (particularly of M3. which was less stable than the base or Ml) on the basis of fixed-coefficient models that do not incorporate price expectations. If financial deregulation involves a change in financial regimes in which a stable money demand function once again reappears, it is unlikely that the determinants of money demand will bear the same relationship to real money balances in the new regime as they did under the regulated framework—the new money demand relationship may not be in the same guise as the earlier relationship. Indeed, our discussion of why the interest rate elasticity of money demand has been reduced in the new financial environment is just one manifestation of this argument. Consequently, fixed-coefficient money demand functions without price expectations, estimated on the basis of sample data that consist for the most part of observations drawn from the regulated framework, will be contaminated by that data; they will not yield consistent (in the probability sense) parameter estimates of the new stable money demand relationship. Hence forecasts of the demand for money will not be reliable.

Yet if the demand for money exhibits a more sustained degree of instability in the deregulated financial environment (perhaps because of the increased proportion of investment balances held within M3), the rationale for fixed-coefficient estimation is also highly suspect. In either case, the results of this paper support the hypothesis that a greater degree of predictability would be generated by an estimation technique that incorporates changing relationships, rather than assuming them away. More reliable forecasts were obtained in our analysis by estimating the price expectations model with the random-coefficient technique.

APPENDIX I: A Brief History of Recent Australian Financial Deregulation

For ease of reference, the measures taken to deregulate Australia’s financial sector during December 1979-September 1988 are displayed in Table 4.

Table 4.

Financial Deregulation in Australia, December 1979–September 1988

article image
article image
article image
article image

APPENDIX II: Data Definitions

The Australian monetary base consists of holdings of notes and coin by the private sector plus deposits of banks with the Reserve Bank, and Reserve Bank liabilities to the private nonbank sector. Ml is currency plus trading bank current deposits. M3 is currency plus bank deposits (including certificates of deposit of the private nonbank sector). The M3 series excluded deposits with new banks since March 1985. The series for the base. Ml, and M3 were seasonally adjusted.

Real money balances is the base. Ml, and M3, respectively, divided by PGDP. the implicit GDP deflator (1979/80= 1.00).

Y is real GDP at average 1979/80 prices (fiscal year ending in June).

The symbol r0 denotes the interest rate on trading bank fixed deposits. This comprises the rate of 18- to 24-month maturity from 1972, first quarter, to 1974, second quarter, and the midpoint of the range of 24- to 48-month deposits thereafter. This series was provided by the Australian Department of the Treasury.

The symbol r denotes the yield on ten-year commonwealth government securities.

APPENDIX III: Lag Procedures

Some of the demand for money models considered in this paper may be expressed as


where ut follows an autoregressive process of the first order, Pte represents an expected price variable, and zt represents the other variables included on the right-hand side of money demand equations. Four procedures were used to estimate the distributed lags for price expectations: (1) ordinary least squares; (2) Almon polynomial lag distribution (in this second technique, the n coefficients of the lagged explanatory variables were assumed to be on a polynomial in lag of order r; this assumption allowed for a flexible lag structure with a reduction in the number of parameters that require estimation if r is less than n, and it can be viewed as imposing a specific set of linear constraints on ordinary least-squares estimation); (3) Shiller’s distributed lag, a variant of the Almon procedure in which the restrictions are stochastic (the coefficients of the lagged explanatory variable lie close to. rather than on, a polynomial); (4) the Ridge regression method, which confines the coefficient vector to an ellipsoid about the origin covering all smooth coefficient vectors of certain length (the position of the ellipsoid in the parameter space is determined in such a way that the mean-square error of the corresponding ridge estimator is the smallest).

Specifically, we modeled the unobservable price expectations variable. Pte as a distributed-lag model in actual prices: algebraically.


where pt is the actual price in period t. Combining equations (10) and (11) gives


We further assumed that, for j =0.1,2.....n.


where the vector (λ01,....,λp-1)’ is a priori distributed with mean vector zero and covariance matrix A. implied by Shiller’s smoothness restrictions; for an explicit derivation of A and the estimators of the coefficients of equation (12). see Thurman. Swamy, and Mehta (1986) and Kashyap and others (1988).

Equation (13) reduces to Almon’s restrictions if the variance of λj is zero for every j. Thurman. Swamy. and Mehta (1986) and Kashyap and others (1988) derived the Ridge estimators of the coefficients of equation (12) when the Wj lie within an ellipsoid implied by Shiller’s smoothness restrictions. Equation (13) is not binding if the variance of λj is infinity and the posterior mean for the prior (13) reduces to the generalized least-squares estimator in this case.

The root mean-square errors of the alternative lag procedures obtained in models 2.FC and 3.FC (see Table 3) were as shown in Table 5.

Table 5.

Root Mean-Square Errors of Alternative Lag Procedure. Models 2.FC and 3.FC

article image


  • Adams, C, and Porter, M.G. “The Stability of the Demand for Money,” in Reserve Bank of Australia. Conference in Applied Economic Research: Papers and Proceedings (Sydney, September 1976).

    • Search Google Scholar
    • Export Citation
  • Australia, Australian Financial System: Final Report of the Committee of Inquiry, J. Campbell, Chairman (Canberra: Australian Government Publishing Service, September 1981).

    • Search Google Scholar
    • Export Citation
  • Australia, Australian Financial System: Report of the Review Croup, V. Martin, Chairman (Canberra: Australian Government Publishing Service. 1984).

    • Search Google Scholar
    • Export Citation
  • Bailey, Martin J., and George S. Tavlas. “Trade and Investment Performance Under Floating Exchange Rates: The U.S. Experience,” CATO Journal (Washington). Vol. 8 (forthcoming, Fall 1988).

    • Search Google Scholar
    • Export Citation
  • Cuthbertson, Keith, and Mark P. Taylor, “Buffer Stock Money: An Appraisal.” in The Operation and Regulation of Financial Markets, ed. by Charles Good-hart, David Currie, and David Llewellvn (London: Macmillan, 1987), pp. 12549.

    • Search Google Scholar
    • Export Citation
  • Davis, Kevin, “Australian Monetary Policy: Recent Experience and Some Current Issues,” Australian Economic Review (Melbourne), Vol. 72 (Summer 1985), pp. 3750.

    • Search Google Scholar
    • Export Citation
  • Dornbusch, Rudiger, “Expectations and Exchange Rate Dynamics.” Journal of Political Economy (Chicago), Vol. 84 (December 1976), pp. 116176.

    • Search Google Scholar
    • Export Citation
  • Fama, Eugene, “Banking in the Theory of Finance,” Journal of Monetary Economics (Amsterdam), Vol. 6 (January 1980), pp. 3957.

  • Fleming, Marcus, “Domestic Financial Policies Under Fixed and Under Floating Exchange Rates,” Staff Papers. International Monetarv Fund (Washington), Vol. 9 (November 1962), pp. 36980.

    • Search Google Scholar
    • Export Citation
  • Goldstein, Morris, The Exchange Rate System: Lessons of the Past and Options for the Future, Occasional Paper 30 (Washington: International Monetarv Fund, July 1984).

    • Search Google Scholar
    • Export Citation
  • Goodhart, Charles, “Comments on Buffer Stock Money,” Bulletin, Centre for Economic Policy Research (London), August 1986, pp. 68.

    • Search Google Scholar
    • Export Citation
  • Hall, Robert E., “Monetary Trends in the United States and the United Kingdom: A Review from the Perspective of New Developments in Monetary Economics,” Journal of Economic Literature (Nashville, Tennessee), Vol. 20 (December 1982), pp. 155256.

    • Search Google Scholar
    • Export Citation
  • Harper, Ian, “Whv Financial Deregulation?” Australian Economic Review (Melbourne), Vol. 73 (Fall 1986). pp. 3749.

  • Hendry, David, “Monetary Economic Myth and Econometric Reality.” Oxford Review of Economic Policy (Oxford, England), Vol. 1 (No. 1, 1986), pp. 7284.

    • Search Google Scholar
    • Export Citation
  • Hendry, David, and Neil R. Ericsson. “Assertion Without Empirical Basis: An Econometric Appraisal of Monetary Trends in the United States and the United Kingdom,” International Finance Discussion Paper 270 (Washington: Board of Governors of the Federal Reserve System, December 1986).

    • Search Google Scholar
    • Export Citation
  • Horne, Jocelyn, and Mehdi Monadjemi, “Debt, Credit, and Monetary Targeting in Australia,” Economic Record (Melbourne), Vol. 61 (June 1985). pp. 52234.

    • Search Google Scholar
    • Export Citation
  • Jonson, Peter, “Money, Prices, and Output: An Integrated Essav,” Kredit and Kapiial (Berlin), Vol. 9 (No. 4, 1976). pp. 499518.

  • Jonson, Peter, and Robert W, Rankin, “On Some Recent Developments in Monetary Economics,” Economic Record (Melbourne), Vol. 62 (September 1986). pp. 25767.

    • Search Google Scholar
    • Export Citation
  • Jonson, Peter, and Robert W, Rankin, and J.C. Taylor, “Modelling Monetary Disequilibrium.” in The Australian Monetary System in the 1970s, ed. by Michael Porter (Clayton. Australia: Monash University, 1978), pp. 11123.

    • Search Google Scholar
    • Export Citation
  • Judd, John P., “Deregulated Deposit Rates and Monetary Policy.” Economic Review. Federal Reserve Bank of San Francisco, Vol. 20 (Fall 1983), pp. 2744.

    • Search Google Scholar
    • Export Citation
  • Kahn, Mohsin S., and Malcolm D, Knight, “Unanticipated Monetary Growth and Inflationary Finance,” Journal of Money, Credit and Banking (Columbus, Ohio), Vol. 14 (August 1982), pp. 34764.

    • Search Google Scholar
    • Export Citation
  • Kashyap, A.K., P.A.V.B. Swamy, J.S. Mehta, and R.D. Porter, “Further Results on Estimating Linear Regression Models with Partial Prior Information.” Economic Modelling (Guildford, England), Vol. 5 (January 1988). pp. 4957.

    • Search Google Scholar
    • Export Citation
  • Laidler, David, Monetarist Perspectives (Deddington, England: P. Allan, 1982).

  • Laidler, David, “The ‘Buffer Stock’ Notion in Monetarv Economics,” Economic Journal (London), Vol. 94, Supplement (1984), pp. 1734.

    • Search Google Scholar
    • Export Citation
  • Laidler, David, “Buffer Stock Money and the Transmission Mechanism,” Economic Review, Federal Reserve Bank of Atlanta, March-April 1987, pp. 1123.

    • Search Google Scholar
    • Export Citation
  • Laidler, David, “Some Macroeconomic Implications of Price Stickiness,” The Manchester School of Economic and Social Studies (Manchester, England), Vol. 61 (March 1988). pp. 3754.

    • Search Google Scholar
    • Export Citation
  • MacDonald, R., Floating Exchange Rates: Theories and Evidence (London: Unwin Hyman, 1988).

  • MacKinnon, James G., and Ross D. Milbourne, “Monetary Anticipations and the Demand for Monev.” Journal of Monetarv Economics (Amsterdam), Vol. 13 (March 1984). pp. 26374.

    • Search Google Scholar
    • Export Citation
  • Milbourne, Ross D., “Re-Examining the Buffer-Stock Model of Money,” Economic Journal (London), Vol. 97, Supplement (1987). pp. 13042.

    • Search Google Scholar
    • Export Citation
  • Mundell, Robert A., “The Monetary Dynamics of International Adjustment Under Fixed and Flexible Exchange Rates,” Quarterly Journal of Economics (Cambridge, Massachusetts), Vol. 74 (May 1960), pp. 22757.

    • Search Google Scholar
    • Export Citation
  • Narasimham, G.V.L., P.A.V.B. Swamy, and R.C. Reed, “Productivity Analysis of U.S. Manufacturing Using a Stochastic Coefficients Production Function.” Journal of Business and Economics Statistics (Washington), Vol. 6 (July 1988), pp. 33949.

    • Search Google Scholar
    • Export Citation
  • Organization for Economic Cooperation and Development, Economic Surveys 1984/85: Australia (Paris: OECD, 1985).

  • Pagan, A.R., and P.A. Volker, “The Short-Run Demand for Transactions Balances in Australia.” Economica (London), Vol. 48 (November 1981). pp. 38195.

    • Search Google Scholar
    • Export Citation
  • Stevens, Glenn, Susan Thorp, and John Anderson, “The Australian Demand for Money: Another Look at Stability,” Research Discussion Paper 8701, Sydney: Reserve Bank of Australia. January 1987.

    • Search Google Scholar
    • Export Citation
  • Swamy, P.A.V.B., and J. Barth, “Random Coefficients,” in The New Palgrave: A Dictionary of Economics, Vol. 4, ed. by J. Eatwell, M. Milgate, and P. Newman (New York: Stockton, 1987), pp. 5153.

    • Search Google Scholar
    • Export Citation
  • Swamy, P.A.V.B., and George S. Tavlas, “Modelling Buffer Stock Money—An Appraisal,” IMF Working Paper 88/41 (unpublished; Washington: International Monetary Fund. May 1988).

    • Search Google Scholar
    • Export Citation
  • Swamy, P.A.V.B., and Peter Tinsley, “Linear Prediction and Estimation Methods for Regression Models with Stationary Stochastic Coefficients,” Journal of Econometrics (Amsterdam), Vol. 12 (February 1980), pp. 10342.

    • Search Google Scholar
    • Export Citation
  • Thurman, S.S., P.A.V.B. Swamy, and J.S. Mehta, “An Examination of Distributed Lag Model Coefficients Estimated with Smoothness Priors.” Communications in Statistics, Part A: Methods and Techniques (New York), Vol. 15 (1986), pp. 172350.

    • Search Google Scholar
    • Export Citation
  • Veale, J.M., L.F. Boulton, and W.J. Tease, “The Demand for Money in Australia: A Selective Survey and Some Updated Results,” in Meeting on Monetary Issues, Research Discussion Paper 8502 (Sydney: Reserve Bank of Australia, November 1985).

    • Search Google Scholar
    • Export Citation
  • Waterhouse, M.H., “The Australian Financial Services Industry—Evolution or Revolution?” Working Paper 8502A (North Ryde, Australia: Macquarie University, Centre for Studies in Money, Banking, and Finance, June 1985).

    • Search Google Scholar
    • Export Citation

Mr. Swamy is Senior Economist in the Division of Research and Statistics of the Board of Governors of the Federal Reserve System and Adjunct Professor of Economics at George Washington University. He holds a doctorate from the University of Wisconsin, Madison. The views expressed in this paper are not to be interpreted as those of the Board of Governors or staff of the Federal Reserve System. Mr. Tavlas, an economist in the European Department of the Fund, holds a doctorate from New York University. The authors acknowledge helpful comments from Martin Bailey, Michael Dakolias, William Dewald, William Gavin. David Laidler, and their colleagues in the Fund.


M3 is defined as currency plus bank deposits (including certificates of deposit).


In response, Hendry has pioneered the error-correction estimation procedure. See, for example, Hendry and Ericsson (1986).


The monetary base is defined as the stock of reserve money, consisting of currency outside the Reserve Bank, trading banks’ deposits with the Reserve Bank, and Reserve Bank liabilities to the nonbank private sector. Ml is currency plus trading-bank current deposits. M2, which consists of Ml plus savings banks’ deposits, was unstable throughout the 1970s. Accordingly, it has not been considered useful as a target and thus is not tested in this paper. Broad money—M3 plus borrowings from the private sector by nonbank financial intermediaries less the latter’s holdings of currency and bank deposits—is also not considered because it is subject to long information lags and, hence, is not very useful for targeting.


Formally known as Money Market Corporations.


Under the Financial Corporations Act of 1974, the Reserve Bank acquired extensive powers over nonbank financial institutions, empowering it to prescribe asset ratios and determine their lending policies. However, these powers have not been used.


The trading banks also had access to lender-of-last-resort facilities from the Reserve Bank.


A limited range of long-standing exceptions included accounts of governments, other banks, nonprofit organizations, and bank staffs.


The SRD ratio was used for monetary policy purposes. Banks received a below-market rate of interest on the funds held in these accounts. The interest rate paid on SRDs was increased from 2.5 percent to 5.0 percent in May 1982.


he fact that no new Australian trading bank licenses were issued for a period of some thirty-five years may have been due to the enhanced competitiveness of nonbank financial intermediaries, which were in a better position to compete for funds.


he projections were made on the basis of the fiscal year, which runs from July through June.


everal empirical studies of the demand for the aggregates during the 1970s found a structural break in the middle of the decade. See Adams and Porter (1976) and Pagan and Volker (1981). Subsequently, Home and Monadjemi (1985) examined the case for debt targeting over quarterly intervals for 1961:1—1981:1. They found no support for the adoption of total debt as an intermediate target, although bank credit performed reasonably well on all empirical tests.


he SRD ratio was changed six times in both 1976 and 1977, three times in each of 1975 and 1979, and once in 1981. It ranged from a low of 3.5 percent to a high of 10.0 percent. It has not been changed since 1981.


detailed discussion of these developments is provided in Harper (1986), on which this section is based in part.


hese examples were provided by Harper (1986). The “sweep” account involves the automatic shifting by a bank of surplus current account balances into an interest-bearing deposit or money market fund.


he Martin Group was convened by the then new Labor Government.


here remains some residual rationing in the market for housing finance, where interest rates on loans made by savings banks before April 1, 1986, remain regulated.


Everything else remaining the same, the supply of deposits is inversely related to the deposit rate (since an increase in the deposit rate reduces the profit margin) and the demand for deposits is positively related to the deposit rate. As long as the deposit rate is maintained below its equilibrium (that is, market) rate, the quantity of deposits will be determined by the demand schedule for deposits.


However, in the spirit of the Mundell-Fleming model (see Mundell (1960) and Fleming (1962)), after the floating of the Australian dollar the impact of monetary policy on aggregate demand has tended to be reinforced by changes in the exchange rate. Goldstein (1984, pp. 22-23) has provided a critical discussion of the monetary policy consequences associated with the Mundell-Fleming model.


Overshooting will also be greater the lower is the expectations coefficient associated with divergences between spot and long-run exchange rates. For an exposition, see MacDonald (1988, pp. 59-63); for an overview of exchange rate theories, see Bailey and Tavlas (1988).


The cumulative fall in M3 velocity in 1984/85 and 1985/86 was 7 percent.


The acceleration in the growth of broad money (that is, M3 plus borrowing from the private sector by nonbank financial intermediaries) was less pronounced than M3—from 13.1 percent to 16.3 percent. This circumstance primarily reflected the fact that some nonbank financial intermediaries became banks during this period.


Empirical work performed at both the Australian Treasury and the Reserve Bank of Australia supports the hypothesis that variables representing the opportunity cost of holding money balances and the own rate of return on money affect money balances with a one-period lag. Accordingly, these variables were entered into all equations with a one-quarter lag.


Laidler (1984, 1987, 1988) has provided excellent discussions of the buffer stock concept. See also Cuthbertson and Taylor (1987). For a discussion of recent empirical studies of buffer stock money, see Swamy and Tavlas (1988) and Milbourne (1987).


Laidler also argued that the coefficients of the reduced-form equation are likely to be variable because they are not structural parameters.


An alternative procedure would have been to use instrumental-variables estimation to generate the unanticipated series. Such a procedure was tried by the authors for M3. The results did not differ significantly from the results obtained by using the base.


ldquo;See the general model developed by Swamy and Tinsley (1980).


Comparisons of the Swamy and Tinsley procedure with the Kalman and Bayesian procedures are given in Narasimham. Swamy, and Reed (1988). Note that, as emphasized by Narasimham, Swamy, and Reed, the statistical notions of consistency and efficiency require the existence of the true values of parameters and hence do not apply to the estimators of β¯, γ, and Δ (or to any other fixed-parameter estimators) if these parameters do not represent “real” physical quantities or if equation (9) does not have a natural interpretation in terms of Mtd


he Chow test for stability was performed on the equations reported in Table 1. Only the equations for the base exhibited parameter stability, with a Chow statistic of 2.02 that was insignificant at the 5 percent level. All the other equation sets were unstable, with the Chow statistic for Ml equal to 3.79, forM3 (without the own rate) equal to 7.13, and for M3 (with the own rate) equal to 6.23. The Chow test is subject to several criticisms, including its reliance on the assumptions of a normal disturbance term, no autocorrelation, and no hetero-scedasticity


Note that usual summary statistics such as the R¯2 and the Durbin-Watson statistic are not reported in Table 2, since these statistics do not provide an adequate basis for comparing different models and the estimators that we considered. In this study of alternative models and estimators, we used predictive methods—extrapolation to data outside the sample—that permit sharper discrimination.


Note that the coefficients on price expectations in the random-coefficient equations are determined by multiplying the sum of their corresponding fixed-coefficient weights (Σwj) by a5. The magnitudes of the coefficients found in this study are in line with previous work, which finds coefficients in the range from -0.5 to -3.0. See. for example, Kahn and Knight (1982).


A degenerate distribution is a distribution concentrated entirely at one point.


Note that the predictions for the fixed-coefficient models la, lb, and lc are based on equations for which no serial correlation corrections were performed, since p was not significant in any equation. In an alternative set of regressions (not reported), such corrections were performed on the fixed-coefficient equations and produced a slight decrease in root mean-square error(s) compared with the equations that were not corrected. But the corrections did not change the qualitative nature of the results. The alternative regression results are available from the authors on request.


However, as Laidler (1982) has shown, the buffer stock idea is quite consistent with price expectation effects of the type described in model 2. Only when combined with a particular story about money supply expectations do we get equations (3)-(5). Also, as noted, Laidler argues that the coefficients of the buffer stock model are reduced forms, and thus likely to be variable. For these reasons, the success of the random-coefficient technique is quite consistent with the basic buffer stock message.

IMF Staff papers: Volume 36 No. 1
Author: International Monetary Fund. Research Dept.