The stability of the demand for money and M1 velocity: Evidence from the sectoral data

The stability of the demand for money and M1 velocity: Evidence from the sectoral data

The Quarterly Review of F.conomics and Wnance, Vol. 35, No. 3, Fall, 1995, pages 233-243 Cop-t 0 1995 Trustees of the Universityof Illinois AU lights ...

735KB Sizes 1 Downloads 37 Views

The Quarterly Review of F.conomics and Wnance, Vol. 35, No. 3, Fall, 1995, pages 233-243 Cop-t 0 1995 Trustees of the Universityof Illinois AU lights of reproduction in any form reserved.

ISSN 0033-5797

The Stability of the Demand for Money and Ml Velocity: Evidence from the Sectoral Data JAMES L. BUTKIEWICZ University

of Delaware

MARGARET MARY MCCONNELL Ohio State University

Sectoral monqr demandfunctions relationships

are found

fw

are estimated usingflw

household

and

offunds

(nonfinancial)

data. Cointegrating

business sectors. Estimated

error-correction moak!s for both sectors, however, exhibit parameter nonconstancy for time @riook which roughly correspond with importantfinancial Thus, sectoral estimates of monq innovations

innovations and deregulations.

demand relationships are consistent with the view that

and deregulation contribute to the nonstationatity of Ml velocity.

The apparent nonconstancy of the parameters of the aggregate Ml demand function for the U.S., and the corresponding deterioration of the relationship between money and aggregate output is one of the most extensively examined macroeconomic phenomena of recent years. Studies examining the demand for money and/or stationarity of velocity include those by Brocato and Smith (1989)) Carlson (1989)) Dickey, Jensen and Thornton (1991)) Friedman (1984), Goldfeld (1976)) Hall and Noble

(1987),

Hendry and Ericsson

(1991),

Mascar and Meltzer

(1983),

Mehra

(1989), Melnick (1990), Miller (1991), and Rasche (1987), while Friedman and Kuttner (1992) have examined the decline in the relationship between money and income. The majority of studies conclude that there is no stable Ml demand function, but the existence of a stable M2 demand function is supported by the data.’ Typical explanations for the breakdown of Ml demand include changes resulting from disinflation, deregulation and financial innovations, and uncertainty due to increased volatility of monetary policy. This paper presents the results of another look at the demand for Ml. The approach used is unique in that the data set is the Federal Reserve’s flow of funds data disaggregated by sectors. The methodology used is the Engle-Granger (1987) approach of estimating error correction models for cointegrated relationships.

234

QUARTERLY REVIEW OF ECONOMICS AND FINANCE

Hendry and Ericsson (1991) use this approach to estimate money demand functions (M2) for the UKand Miller (1991) uses the same approach for U.S. money demand. While Dickey, Jensen and Thornton

(1991) note that the Engle/Granger

is sensitive to the order of normalization, approach finds evidence of cointegration obtained

from the maximum

Juselius (1990).

, the cointegrating

likelihood

approach

of nonfinancial

vector is similar to that

developed

In fact, we find that the Engle/Granger

gration for all three normalizations normalizations

approach

they also find that when the Engle/Granger by Johansen

and

approach supports cointe-

of household Ml demand, and for two of three

business demand for Ml. We also find that the scale

and interest rate variables in the cointegrating regressions differ between the house hold and business sectors as do estimated elasticities. This suggests that aggregation across sectors which respond in different magnitudes to different measures of transactions

and opportunity costs may obscure long-run relationships.

Finally, the

sectoral models indicate that instability of Ml demand in the 1960s and 1970s emanated primarily from the business sector, while in the late 1970s and 1980s the source of instability was the household sector. One explanation

consistent with this

pattern of instability is that it is the result of financial innovation and deregulation. Alternatives to holding checking deposits developed first for businesses in the form of repurchase agreements and Eurodollar

deposits. Money market mutual funds and

other alternatives to checking balances became available to households in the 1970s and the 1980s.’ Part I of this paper reviews the flow of funds ownership data, the Goldfeld study of sectoral Ml demand, and discusses briefly updated results of this approach. The methodology of the Engle/Granger error correction model and the necessary pretests are discussed in Section II. Estimated results for the household and nonfinancial business sectors are presented

in Section III. The implications

of the esti-

mates for parameter constancy, including the widespread use of dummy variables in money demand models, are discussed in Section IV Concluding comments are made in Section V.

I.

FLOW OF FUNDS DATA AND THE GOLDFELD STUDY

This study examines the demand for Ml using the flow of funds estimates of sector-al ownership of Ml. The flow of funds ownership data differs in several ways from aggregate monetary data (Federal Reserve Board, 1971). Ml data are averages of daily data, while flow of funds data are end-of-month point estimates. The flow of funds data include balances held by Edge Act corporations and agencies of foreign banks in the U.S. in financial sector balances. Mail float is not included in any sector, but the Fed (1971, p. 464) suggests that mail float is largely comprised of deposits of the business sector, and Goldfeld (1973) includes mail float in the business sector estimates3

DEMAND FOR MONEY AND Ml VELOCITY

235

Goldfeld ‘s (1973) classic study of money demand includes estimation of models for the household, financial, (nonfinancial)

business (including mail float), and state

and local government sectors. Holdings by the “rest of the world” are not examined. The estimates

are obtained

using the partial-adjustment

specification,

which is

frequently called the Goldfeld model of money demand: In($),

= clo + clrln Y,, - a&r,

+ a&($-r

+ El,

S=1,...,4.

(1)

In = the natural logarithm of a variable, where: (M, /p) = the real Ml balances for each of the sectors except finance where nominal levels are used,4 Y, = a measure of transactions for each sector, r = an opportunity cost measure, and E = a random error. All of his estimated equations used a correction for first-order serial correlation. Goldfeld reported that the household and financial sectors were well explained by the model, while the results were not satisfactory for the other two sectors. As a preliminary to estimating cointegrating regressions and error correction models, we reestimated Goldfeld ‘s equations using data from the third quarter of 1952 through the first quarter of 1990.5P6 Also, we experimented with various scale and interest rate variables. The estimates are tested for coefficient constancy for various subsamples. Using Chow tests, the hypothesis of coefficient constancy is rejected for all sectors when the sample is divided at the fourth quarter of 1979. This evidence of nonconstant parameters is the same problem that plagues the aggregate money demand estimates using the partial adjustment model. As is true for aggregate money demand modeling, alternative approaches may provide greater insight into sectoral money demand.

II.

COINTEGRATION

Recent

AND ERROR CORRFXXION MODELS

advances in econometric

techniques

have established

the importance

of

testing for cointegrated relationships. If economic series must be differenced one or more times to obtain stationarity, they are said to be integrated of order d, where d is the number of differences required to obtain stationarity of the series. If one or more series are integrated of the same order, they may follow a common trend. If the residual from a regression of one of these series on the others is integrated of order d- b, where b > 1, the series are said to be cointegrated, and they follow a common trend. Before testing for cointegration, it must be established that all series are integrated of the same order. Augmented Dickey-Fuller tests are reported in Table 1 for

236

QUARTERLY

REVIEW

Table 1.

AUGMENTED

(Sample

1952.4-1990.1)

OF ECONOMICS

AND FINANCE

DICKEY-FULLER TESTS FOR UNIT ROOT

Variable

1st Difference

2nd Difference

In Household Real Ml (LRH)

-2.11

-7.77**

In Business Real Ml (LRB)

-1.03

-8.95**

In State and Local Government Real Ml (LRSL)

-1.76

-9.79**

0.95

-7.82**

In Real Disposable Personal Income (LRDPI)

-1.84

-6.14*+

In Real GNP (LRGNP)

-2.69

-6.67**

In Real State and Local Government Expenditures (LRSLX)

-1.34

+.68**

0.91

-9.36**

In Smonth Treasury Bill Rate (LTBJ)

-2.84

-6.34**

Smonth Treasury Bill Rate (TB3)

-2.73

+.45**

Financial Ml (FIN)

Deposits (DEP)

first and second differences

of all variables used in the models. In all cases, the tests

fail to reject the null hypothesis of a unit root for first differences, hypothesis of a unit root for second differences. As all variables are integrated of order one, cointegrating

but reject the null

regressions are run to

determine if long-run relationships exist among sets of variables that would be included in a money demand model. Engle and Granger (1987) recommend using the augmented Dickey/Fuller test to test for cointegration. The tests indicate cointegration of real household Ml (LRH), real disposable personal income, and the Treasury bill rate, all in logarithms, for all three normalizations.’ These results are reported in Table 2. Table 2 also reports the results of cointegrating regressions for the logarithm of real business money holding (LRB), the logarithm of real GNP, and the level of the three-month cointegration

Treasury Bill rate. ADF tests support the hypothesis of

for two of the three normalizations.

Tubb 2.

COINTEGRATING

(Sample

1952.1-1990.1) Household

No evidence of cointegration

REGRESSIONS Business Sector

Sector

Cointegrating Vector

Cointegrating Vector 1 LRH LRDPI LTB3

2

3

1

-1.27

2.3

-0.6

1

-2.57

0.103

ADF

-3.5*

NOMS: *Rejects **R+crs Augmented

-0.243

1

-4.8**

A.8**

1 LRB

1

LRGNP TB3

0.044 -3.9*

null hypothesis

of unit root in residual

at 5 percrnt

level ofconfidence.

null hypothesis

of unn root in residual

at 1 prrcen~

level ~Cconfidcnce.

tc~ts for coinleg?arion

use four lagged

Dickq-Fuller

-0.072

difTcrence

w”ns.

2

3

-0.477

12.03

1 -0.103

-4.26

-2.5

-5.o**

1

is

237

DEMAND FOR MONEY AND Ml VELOCITY found for various specifications business sectors account

for other sectors.

for 83 percent

estimation of error correction

However, the household

of all sectoral

holdings

and

in 1990. Thus,

models for these two sectors will provide insight into

the aggregate behavior of Ml.

III.

ERROR CORRECTION

MODELS OF SECTORAL HOLDINGS

Error correction models allow for flexible specification of short-run dynamics while obeying the long-run constraints implied by the existence of cointegration. Specifically, error correction models are vector autoregressions which include the lagged levels of the variables in the model or the lagged residual from the cointegrating regression. Both Engle and Granger (1987) and Hendry and Ericsson (1991) use a general to specific approach for estimation of error correction models. In the first phase, unrestricted VARs are estimated for differences of the variables for an arbitrary number of lags. These VARs include lagged levels of the cointegrated variables. The significance, signs, and magnitude of the lagged levels indicate the appropriateness of the error correction specification.* The model is then reestimated, deleting the VAR terms which are least significant and have the smallest magnitude. The lagged levels of the variables from the cointegrating regression are replaced with the lagged residual from that regression. This is the final error correction specification. The error correction columns

estimates are reported

are the unrestricted

estimates

in Table 3. The first and third

for the household

and business sectors

respectively, while columns 2 and 4 are the final error-correction In both Equations 1 and 3 (as expressed in corresponding

specifications.

columns), the relative

size of the lagged level variables is approximately the same as the cointegrating regressions with real sectoral money balances as the dependent variables. The signs indicate that excess supplies in the previous period lead to reduced real balances in the current period, which is appropriate estimated

coefficients

for an error correction

of the lagged levels are significant

specification.

in the household

The sector

equation, and the lagged level of real balances is significant in the business equation. Elimination of the least significant terms and replacing the lagged level terms with the lagged error from the cointegrating correction

regression9

obtains the final error

model. The lagged residual terms are significant in Equations 2 and 4 (in

columns of same numbers), supporting the use of an error correction specification. For both models 2 and 4, the standard error of the regression is reduced compared to the unrestricted regressions 1 and 3 respectively. This is evidence that the final error correction equation is an improved specification. A further test of the error correction specification is the convergence restriction. This is the restriction that the implied long-run income elasticity in the error correction model equals the coeffkient on the income term in the cointegrating

238

QUARTERLY REVIEW OF ECONOMICS AND FINANCE

Tab/k 3.

ERROR

CORRECTION

MODEL

ESTIMATES

(Sample 1952.3-1990.1) Household

Sector

Dependent 1 Constant ALRH (-1) ALRH (-2) ALRH (-3) ALRH (4) ALRDPI ALRDPI (-1) ALRDPI (-2) ALLRDPI(-3) ALRDPI (-4) aTB3 hLTB3 (-1) MTB3 (-2) ALTBJ (-3) L\LTBS(4) LRH (-1) LRDPI (-1) LTB3 (-1)

0.059 (.652) -0.244 (-2.93) 0.106 (1.25) 0.039 (0.45) 0.355 (4.33) 0.407 (1.89) 0.313 (1.38) 0.076 (0.34) -0.159 (-0.721) -0.244 (-1.14) -0.015 (-0.979) -0.018 (0.992) 0.007 (0.34) a.018 (-1.04) 0.02 (1.24) -0.119 (-3.25) 0.086 (3.03) -0.017 (-1.86)

e(-1) R* 3 S.E. SSR F Log Likelihood

0.39 0.31 0.023 0.066 4.95 360.5

Business Sector

Variable ALBH

Dependent

2 -0.002 (-0.47) -0.215 (-2.75) 0.115 (1.51)

3 Constant ALRB (-1) ALRB (-2) ALRB (-3)

0.347 (4.61) 0.410 (2.01) 0.293 (1.38)

ALRB (-4) ALGNP ALGNP (-1) ALGNP (-2) ALGNP (-3)

-0.232 (-1.16) -0.015 (-1.04) a.015 (-1.06)

ALGNP (-4) ATB3 ATB3 (-1) ATB3 (-2) ATB3 (-3)

0.01 (0.71)

ATB3 (4) LRB (-1) LRGNP (-1) TB3 (-1)

-0.12 (-3.55) 0.37 0.32 0.022 0.069 7.999 357.6

0.415 (2.2) a.153 (-1.88) 0.10 (1.24) -0.06 (-0.79) 0.56 (7.57) 0.40 (1.62) 0.24 (0.92) -0.29 (1.12) -0.07 (-0.27) -0.14 (-0.54) -0.00 (4.10) -0.01 (-3.4) 0.01 (1.36) -0.01 (-2.35) 0.01 (2.55) a.10 (-2.38) 0.014 (0.93) -0.01 (-1.67)

e(-1) R* ii2 SE. SSR F Log Likelihood

0.57 0.51 0.0259 0.087 10.09 340.2

Variable &RB 4 a.002 (-0.54) -0.22 (-3.36)

0.58 (9.0) 0.414 (1.89)

a.15 (-0.69)

-0.013 (4.14)

-0.01 (3.59) 0.01 (2.65)

-0.08 (-2.57) 0.55 0.52 0.0257 0.09 20.86 336.2

DEMAND FOR MONEY AND Ml VELOCITY

regression.

The probability that this restriction

obtains for the household

239

sector is

96 percent, but it is only 44 percent for the business sector.” An F test that the sum of the income coefficients in the business money- demand equation is different from zero is significant at only the 19 percent level of confidence, while the same test for the interest rate variables is significant at the 1 percent level. The existence of strong interest rate effects and weak or no income effects in business holdings is consistent with the hypothesis that the introduction of alternatives to demand deposits had a significant effect on the business sector’s demand for money. To further test the error-correction specification, one-step static forecasts of the sectoral models are compared to a naive random walk forecast of sectoral money holdings of the form:

(2)

yt’ Cl

The Theil coefficient for the household sector is .844 and for the business sector is .674.” These test results provide further support of the error-correction specification. Tests for autocorrelation, heteroskedasticity, and normality of the residualsI fail to reject the null hypotheses of no autocorrelation or heteroskedasticity for both models. However, both models exhibit nonnormality of residuals, skewness, and excess kurtosis. These problems are discussed below.

IV.

STABILITY AND THE USE OF DUMMY VARIABLES

Inclusion

of dummy variables in money demand estimates to account for certain

“shifts” in the demand function is frequently dummy variables are used include:

observed.

Common

periods when

1. 2.

1973.4-1976.1 - the period of the “missing money.” Certain quarters in 1980 through 1983 to adjust for the introduction

3.

and OCD accounts.13 Dummies for 1986 to allow for a shift in demand.

of NOW

When these dummy variables are included in Equation 2, the problems of nonnormality, skewness, and excess kurtosis are eliminated. Similar improvements are obtained when dummies for certain quarters of the 1970s are used in the estimation of the business sector in Equation 4.14 While use of dummy variables in certain cases, such as strikes or natural disasters, is acceptable, inclusion of dummies to eliminate large residuals often fails to provide insight into the economic forces which possibly account for these shifts. As an alternative to the use of dummy variables, the parameter constancy of sectoral money demand is examined using recursive least-squares estimates of Equations 2 and 4 from Table 3. Results of the CUSUM of squares tests for the household

240

QUARTERLY

REVIEW OF ECONOMICS AND FINANCE

1.25 -

CUSUM of squares 5% significance

-----

1.00

*- ._

-I

c*

*-*

.’

.’

(/

0.75

0.50 0.25 0.00 -0.25 Figure 1.

I

I

1960 CUSUM

I

1965

I

I

I

I

1970

1975

1980

1985

1’ 30

of Squares Statistic and Confidence Interval Household

Sector.

and business sectors are displayed in Figures 1 and 2 respectively. Movement of the test statistic outside the 5 percent confidence bands is an indication of parameter nonconstancy. The household

sector begins to display parameter

evidence of nonconstancy

nonconstancy

in 1974, but

is greatest during the years immediately preceding

and

1.25

1.00

0.25 0.00 -0.25

Figure 2.

I

CUSUM

I

1960

I

1965

I

1970

1

1975

I

1980

r

1985

1 90

of Squares Statistic and Confidence Interval Business Sector.

DEMAND FOR MONEY AND Ml VELOCITY following passage of the Monetary Control Act of 1980. Parameter nonconstancy household

241 of

money demand is greatest during the period of financial deregulation

and the development of innovative alternatives to demand deposits for the household sector. The business sector displays initial parameter but evidence of nonconstancy demand parameters

nonconstancy

in the mid-1960s,

is greatest in the late 1960s and 1970s. Business money

appear constant during the 1980s. Financial innovations such

as REPOs and overnight Eurodollar

deposits were developed in the late 1960s as an

alternative to business demand deposits. The timing of parameter nonconstancy

in

business money demand coincides with these innovations.

V.

CONCLUSION

The evidence presented are cointegrated account

here indicates that household

with measures

for 83 percent

of income

and business real balances

and interest

of total money balances.

rates. These

No evidence

two sectors

of cointegrated

relationships

are found for the remaining sectors, although a number of alternative

specifications

were tested. For the household and business sectors, the scale variables

in the cointegrating

regression differ, and one uses the log of the Treasury bill rate

while the other uses the level. All of these results indicate that tests for cointegration of aggregate Ml may be plagued by problems of aggregation bias.15 Estimated error correction obtain acceptable nonconstant

models for household and business money demand

results. However, both estimated equations

display evidence of

parameters during certain periods. The periods of parameter noncon-

stancy generally coincide with the development

of innovations such as negotiable

CDS, REPOs , and money market mutual funds and deregulations

such as interest-

bearing checking accounts. Thus, the results presented here are consistent with the hypothesis that observed money demand instability may be due to financial innovations and deregulations.16 Acknowledgment:

This paper was written when McConnell was a graduate student

at the University of Delaware. The authors acknowledge helpful comments from Phil Rothman and two anonymous referees. Responsibility for any errors is the authors.

NOTES *Direct all correspondence to: James L. Butkiewicz, University of Delaware, Department of Economics, Purnell Hall, Newark, DE 197162701. 1. The above-mentioned studies by Friedman, Hall and Noble, and Dickey,Jensen and Thornton do provide results supportive of a stable Ml demand function.

242

QUARTERLY REVIEW OF ECONOMICS AND FINANCE 2.

The first significant household utilization of money market mutual funds occurred

in the mid-1970s due to a sharp increase in short-term interest rates. As rates fell, household deposits in money mutual funds stopped growing until interest rates again rose rapidly, beginning in 1978. Money market mutual fund balances included in M2 increased from $9.5 billion in 1978 to $235.6 billion in 1982. 3.

The Fed includes all currency in the household

sector, but acknowledges

that a

substantial portion of currency is actually held by the foreign sector (Federal Reserve Board, 1971, p. 464n). 4.

The price index is the GNP deflator in all cases, although alternative indices could

be used. As all sectors are active throughout the economy, use of a general price index seems justified. The model presented coefficient

above accepts the theoretical

on the price index is unity. Goldfeld

(1973)

assumption that the implied

found that this assumption was

supported by the data. 5.

First-order autocorrelation

corrections are used for every sector-al regression except

the financial sector, where there is no evidence of first-order autocorrelation. 6.

Estimates of sectoral holdings ofM1 were discontinued at this time, so no additional

data is available. 7.

A number of potential cointegrating

vectors were examined, including the use of

opportunity cost measures for Ml. These measures are the difference between the Treasury bill rate and the rate on demand deposits and the difference between the return on M2 and on demand deposits. The results of these estimates fail to support the hypothesis of cointegration. 8.

Engle and Granger

(1987, p. 271) indicate that the signs and magnitudes of the

variables should reflect those of the cointegrating

vector, and that these variables should be

statistically significant. 9.

The error from the cointegrating

regression with real balances as the dependent

variable is used in each case. 10.

Respecifications

of the household

error correction

interest rate as the dependent variables result in insignificance

model with the income

and

of the error correction

term

in the income equation but not the interest rate equation. This suggests that income is weakly exogenous. 11.

The sectoral models used in these comparisons exclude dummy variables. Inclusion

of dummy variables reduces the Theil coefficients

to .75 and .658 for the household

and

business sectors respectively. 12.

The tests are the Breush/Godfrey

LM test, the Box/Pierce

statistics, the Arch, White and Ramsey reset tests, and the Jarque/Bera 13.

and Ljung/Box

Q

normality statistic.

Dummies used are for the interval from 1773.1 through 1976.4, the first quarter of

1981 and separate dummies for the first and second quarters of 1983. 14.

The dummies are for years 1970 and 1979.

15.

However, Dickey, Jensen

and Thornton

(1991) find evidence of cointegration

aggregate Ml demand using both the Engle/Granger

and Johansen/Juselius

for

tests for cointe-

gration. 16.

Mehra (1992) draws similar conclusions

from examination

using the same methodology as is used in this paper.

of aggregate Ml data

DEMAND FOR MONEY AND Ml VELOClTY

243

REFERENCES Board of Governors of the Federal Reserve System. 1971. “Survey of Demand Deposit Ownership.” Federal Reserve Bulletin, 57tJune) : 456467. Brocato, Joe and Kenneth L. Smith. 1989. “Velocity and the Variability of Money Growth: Evidence from Granger-Causality Tests-A Comment.” Journal of Monqr, Credit and Banking, 21: 258-261. Carlson, John B. 1989. “The Stability of Money Demand, Its Interest Sensitivity, and Some Implications for Money as a Policy Guide.” Federal Reserve Bank of Cleveland Economic Review, 3: 2-13. Dickey, David A., Dennis W.Jansen and Daniel L. Thornton. 1991. “A Primer on Cointegration with an Application to Money and Income. ” Federal Reserve Bank of St. Louis Ret&w, (March/April): 58-78. Engle, Robert E and C.W.J. Granger. 1987. “Cointegration and Error Correction: Representation, Estimation, and Testing.” Econometrica, 55: 251-276. Federal Reserve Board. 1971. “Survey of Demand Deposit Ownership.” Federal Reserve Bulletin, 57: 456-467. Friedman, Benjamin M. and Kenneth N. Kuttner. 1992. “Money, Income, Prices, and Interest Rates.” American Economic Review, 82: 472492. Friedman, Milton. 1984. “Lessons from the 1979-82 Monetary Policy Experiment.” American Economic Review, 74: 397400. Goldfeld, Stephen M. 1973. “The Demand for Money Revisited.” Brookings Papers on Economic Activity, 3: 577-638. -. 1976. “The Case of the Missing Money.” BmokingsPaperson EconomicActivity, 3: 683-730. Hall, Thomas E. and Nicholas R. Noble. 1987. “Velocity and the Variability of Money Growth: Evidence from Granger-Causality Tests.” Journal of Monqr, Credit and Banking, 21: 258-261. Hendry, David F. and Neil R. Ericsson. 1991. “An Econometric Analysis of U.K. Money Demand” in Monetary Trends in the United States and United Kingdom by Milton Friedman and Anna J. Schwartz. American Economic Review, 81: 8-38. Johansen, Soren and KatarinaJuselius. 1990. “Maximum Likelihood Estimation and Inference on Cointegration-with Application to the Demand for Money.” Oxford Bulletin of Economics and Statistics, 52( 2) : 169-2 10. Mascara, Angelo and Allen H. Meltzer. 1983. “Long- and Short-Term Interest Rates in a Risky World.” Journal of Monetary Economics, 12: 485-518. Mehra,Yash P. 1989. “Some Further Results on the Source of Shift in MlDemand in the 1980s.” Federal Reserve Bank of Richmond Economic Review, 75: 3-13. -. 1992. “In Search of a Stable, Short-Run Ml Demand Function.” Federal Reserve Bank of Richmond Economic Review, 78: 9-23. Melnick, Rat?. 1990. “The Demand for Money in Argentina 1978-1987: Before and After the Austral Program.” Journal of Business and Economic Statistics, 8: 427-434. Miller, Stephen M. 1991. “Monetary Dynamics: An Application of Cointegration and ErrorCorrection Modeling.” Journal of Morq Credit and Banking, 23: 139-154. Rasche, Robert H. 1987. “Ml-Velocity and Money-Demand Functions: Dc; Stable Relationships Exist?” Pp. 9-88 in Empirical Studies of Velocity, Real Exchange Rates, Unemployment and Productivity, edited by Karl Brunner and Allen H. Meltzer. Carnegie-Rochester ConferPnce Series on Public Policy, Vol. 27.