# MONEY DEMAND IN ROMANIAN ECONOMY, USING MULTIPLE REGRESSION METHOD AND UNRESTRICTED VAR MODEL

### Text-only Preview

**Quantitative Methods Inquires**

**MONEY DEMAND IN ROMANIAN ECONOMY, USING**

**MULTIPLE REGRESSION METHOD AND**

**UNRESTRICTED VAR MODEL**

**Mariana KAZNOVSKY**

PhD Candidate, University of Economics, Bucharest, Romania

Monetary and Financial Statistics Division, National Bank of Romania, Bucharest, Romania

**Mariana.Kaznovsky@bnro.ro**

E-mail:

E-mail:

**The paper describes the money demand in Romanian economy using two**

Abstract:

Abstract:

econometrics models. The first model consist in a multiple regression between demand

money, monthly inflation rate, Industrial production Index and the foreign exchange rate

RON/Euro. The second model (Unrestricted Vector AutoRegressive model) is applied for the

same variables used in the first model. Identifying a statistically strong model, capable of

stable estimations for the money demand function in Romania’s economy constitutes a

prerequisite to the application of an efficient monetary policy.

**money demand; unrestricted VAR model; Romania**

Key words:

Key words:

Multiple regression estimation of Romanian money demand function

Multiple regression estimation of Romanian money demand function

The theory underlying money demand function is based on the classical

macroeconomic model of Hicks & Hansen IS-LM, specifically LM curve. The theoretical

hypothesis (assumptions) of the dual equilibrium model for the money market in an open

economy are: the perfect mobility of capital, uncovered interest rate parity principle,

monetary policy conducted by the central bank are using the short term interest rate

variable as the operational one without affecting the stability of the exchange rate of the

national currency.

The LM curve is defined by the all possible combinations of interest rate and income

levels for which the demand of money is equal with money supply (Figure 1.).

**Figure 1.**LM curve

**187**

**Quantitative Methods Inquires**

The money demand function is a synthetic way of measuring the dependence

between, on the one side, the monetary aggregates - as the money issued by the monetary

financial institutions: credit institutions and money market funds, and used as financial

resource by the non-banking entities: non-issuing money institutions, and, on the other side,

the money consumers in the economy.

The classic model [3;4]1 estimates this correlation by the degree of explanation of

the endogenous variable “monetary aggregate” by the following exogenous variables:

monthly price growth rate, value of the economic output (GDP, industrial production value),

average passive interest rate practiced by the credit institutions as an expression of the “price

of money” and other variables expressing the cost of opportunity for possessing the currency

- like exchange rate, the dynamics of the domestic capital index or a foreign capital index

related to the analyzed economy. Taking into account changes in the international oil

markets as an indicator of foreign restrictions could be useful in explaining the money

demand pattern.

The specific choice of variables used to estimate the demand of money depends on

the working hypotheses, on the availability of data with adequate frequency, as well as on

conclusions of previous studies and research works regarding the significance of correlations

that point to one indicator being more reliable than others in approximating the variable.

In order to express the monetary aggregate in the Romanian economy, the choice

has been made for the broad money indicator M2 (known as broad money up to 2007, after

which M3 was introduced, M2 becoming the intermediary monetary aggregate). The

explanation of the use of M2 resides in the higher degree of coverness of the financial

instruments by this indicator. Narrow money M1 is almost designed to be a proxy measure of

the exchange transaction incentives of money only, while broad money M2 is designed to

quantify also the accumulation of value purpose of holding money.

Although, the exogenous variables have to be restricted to the most significant

ones, thus avoiding multicollinearity. Out of purely practical reasons, the industrial

production index has been selected to measure the economic output, whereas for the cost of

money we considered significant the average interest rate for one month as a liability of

monetary financial institutions. For medium and long term maturity we used the interest rate

of the one year government bonds. Longer maturities have been left out because of the

discontinuity in issuance, in relation to the investor lack of preference for medium and long

term maturities.

As an indicator of price increases, we used the monthly Consumer Price growth

rate, as the GDP deflator is available, at best, quarterly, starting with 1998.

Our study has been compensating for the inflationary component by studying the

dependence between the deflated monetary aggregate and the real money demand factors.

For the following regressors the ‘t’ statistic significance of the coefficients of the

money demand function has been confirmed: industrial production index, real money

balances as the log level recorded three months ago, monthly inflation rate and the foreign

exchange rate (leu/Euro). The money demand elasticity in respect of interest rate (as the

average cost of monetary financial institutions for the borrowed resources and, implicitely, as

the rate of return of deposits made by non-banking entities with the banks) was not being

confirmed at the 10% level of significance. Thus, the conclusions of some previous work

papers that the interest rate channel is not efficiently working in the romanian economy are

confirmed by the statistical data [1]. The weak sensitivity of the real variables block could be

**188**

**Quantitative Methods Inquires**

explained by the rigidity of the economy to the monetary impulses due to the specific

structural changes in our emerging economy. On the other side, the National Bank of

Romania’monetary policy was focused on the monetary aggregates (base money) as the

operational target, the exogenity of interest rate being a practical issue in the nominal

variables block.

Thus, we have estimated the equation of the money demand for the Romanian

economy using the following specific version:

*m*=

*a*+

*a m*

+

*a prodind*

?

*a r*inf

*l*+

*a curs*_

*eur*+ ?

*t*

0

1

*t*?1

2

*t*?3

3

*t*

4

*t*

*t*

where:

*m*is the real monetary aggregate, deflated using the CPI (real M2) and seasonally adjusted,

decimal logarithm values being considered; seasonally adjustment has been performed

using the Census X11 method, considering the multiplicative method. The need to isolate

and detach the seasonal component from the series of the monetary aggregate was imposed

by the known peak effect during summer and holidays. Introducing unadjusted series would

have led to rejection of the coefficients, the statistical significance being infirmed with a

probability of 90%.

*rinfl*is the monthly CPI rate, logarithm values;

*prodind*is the Industrial Production Index, logarithm vales

*curs_eur*is the foreign exchange rate RON/Euro, logarithm values.

Table 1 presents the estimators of the regression coefficients, obtained using Eviews

4 software, for the time horizon for January 1992-December 2005, statistical series having

monthly frequency.

**Table 1.**Estimators of the regression coefficients

Dependent Variable: LOG(M2_SA/p)

Method: Least Squares

Date: 02/27/08 Time: 15:22

Sample(adjusted): 1992:04 2005:12

Included observations: 165 after adjusting endpoints

Variable

Coefficient

Std. Error

t-Statistic

Prob.

C

0.192377

0.031722

6.064535

0.0000

LOG(m2_SA(-1)/p(-1))

0.951809

0.008594

110.7478

0.0000

LOG(rinfl)

-0.012963

0.002819

-4.599288

0.0000

LOG(PROD_IND(-3))

0.054574

0.027316

1.997866

0.0474

LOG(CURS_EUR)

0.048924

0.011085

4.413621

0.0000

R-squared

0.999853

Mean dependent var

4.075405

Adjusted R-squared

0.999850

S.D. dependent var

1.911885

S.E. of regression

0.023439

Akaike info criterion

-4.639032

Sum squared resid

0.087899

Schwarz criterion

-4.544912

Log likelihood

387.7201

F-statistic

272760.0

Durbin-Watson stat

1.789483

Prob(F-statistic)

0.000000

A powerful influence of the autoregressive component upon the deflated broad

money has been detected (the estimated coefficient being 0.95); the value of the industrial

production as an approximate value of the real aggregate supply, is positively correlated

with monetary aggregate, but transmission of this influence is produced with a time lag of 3

months. Thus, the changes in real variables is reflected in values of the nominal variables

**189**

**Quantitative Methods Inquires**

after 3 months, but the influence is not strong (estimated elasticity is 0.05: at a change in the

industrial production index with 1%, the reaction of the broad money over 3 months is of

size 0.05%). Estimation of simultaneous correlation of this link has been infirmed by the “t”

test, at a probability level of 90%.

The opportunity cost of holding the money has been approximated by means of

introducing the leu/Euro exchange rate: the equation confirms the positive correlation

between the exchange rate and the real broad money. The national currency depreciation

influences the growth of money demand in real terms, as a consequence of the considerable

weight of the foreign currency denominated part of the monetary aggregate. The shifting

from local currencies to USD/Euro, as a process of substitution of the national currency, is

characteristic for emerging markets, marked by significant changes in economic structure,

and for which the tax of holding the money (inflation rate) and inflationary expectations are

high.

The inverse correlation between the inflation rate and the real broad money is

statistically confirmed by the “t” test. Interesting to observe is the small influence of prices

upon the real broad money.

Stationarity of the data has been verified with the ADF (Augumented Dicky-Fuller)

test, for the case of a liniar trend, a constant and eight lags, corresponding to the timespan

of january 1992-december 2005 (results are presented in table 2).

The degree of explanation brought by the exogenous variables in their entirety,

contributes in proportion of 99% (adjusted R2 coefficient) to the obtaining of values for the

adjusted series of money demand, as is visible from figure 2.

The errors terms resulted from the regression, represented as a blue line has been

tested for autocorrelation: the Durbin-Watson statistic confirms the rejection of the

autocorrelation in the residual series; as a consequence, the regression parameters are

relevant and statistically significant.

8

6

4

0.15

2

0.10

0.05

0

0.00

-0.05

-0.10

93 94 95 96 97 98 99 00 01 02 03 04 05

**Figure 2.**Adjusted versus real money demand

**190**

**Quantitative Methods Inquires**

**Tabel 2.**Augumented Dicky-Fuller

Augmented Dickey-Fuller Test for real broad money M2/p

Sample(adjusted): 1992:06 2005:12

ADF Test Statistic

7.864602 1% Critical Value*

-3.4715

5% Critical Value

-2.8792

10% Critical Value

-2.5761

*MacKinnon critical values for rejection of hypothesis of a unit root.

Augmented Dickey-Fuller Test for industrial production index prod_ind

Sample(adjusted): 1992:06 2005:12

ADF Test Statistic

-7.769099 1% Critical Value*

-3.4715

5% Critical Value

-2.8792

10% Critical Value

-2.5761

*MacKinnon critical values for rejection of hypothesis of a unit root.

Augmented Dickey-Fuller Test for exchange rate leu/euro

Sample(adjusted): 1992:06 2005:12

ADF Test Statistic -3.205847 1% Critical Value* -3.4715

5% Critical Value

-2.8792

10% Critical Value

-2.5761

*MacKinnon critical values for rejection of hypothesis of a unit root.

Augmented Dickey-Fuller Test for inflation rate

Sample(adjusted): 1992:06 2005:12

ADF Test Statistic -3.205847 1% Critical Value* -3.4715

5% Critical Value

-2.8792

10% Critical Value

-2.5761

* MacKinnon critical values for rejection of hypothesis of a unit root.

**Estimating the reaction of the broad money to shocks in real economy**

**variables (VAR model for the money demand in the economy)**

An estimation of the correlation between the real exogenous variables and money

aggregates based on the UVAR (Unrestricted Vector AutoRegressive) with three lags has

been applied for the same variable used in the multiple regression. The system of

simultaneous equations comprises thus, the following variables: real broad money M2

(seasonally adjusted levels), IPI (Industrial production index), inflation rate, leu/Euro foreign

exchange rate. Series are monthly and covers the years 1992-2005; ADF stationarity tests

have shown the stationarity of the series of broad money, foreign exchange rate and

inflation rate with a probability of 95%.

The model specification is as follows:

**191**

**Quantitative Methods Inquires**

3

log(

*M*_

*SA*(

*t*) /

*p*(

*t*)) =

*a*+ ?

*a*log(

*M SA*(

*t*?

*i*) /

*p*(

*t*?

*i*)) +

2

0

1

*i*

2 _

*i*1

=

3

3

3

+?

*a*log(

*prod*_

*ind*(

*t*?

*i*)) + ?

*a*log(

*curs*(

*t*?

*i*)) + ?

*a*log(

*r*inf

*l*(

*t*?

*i*)) +

*u*(

*t*)

2

*i*

3

*i*

4

*i*

1

*i*1

=

*i*1

=

*i*1

=

3

3

log(

*prod*_

*ind*(

*t*)) =

*b*+ ?

*b*log(

*M SA*(

*t*?

*i*) /

*p*(

*t*?

*i*)) + ?

*b*log(

*prod*_

*ind*(

*t*?

*i*)) +

0

1

*i*

2 _

2

*i*

*i*1

=

*i*1

=

3

3

+?

*b*log(

*curs*(

*t*?

*i*)) + ?

*b*log(

*r*inf

*l*(

*t*?

*i*)) +

*u*(

*t*)

3

*i*

4

*i*

2

*i*1

=

*i*1

=

3

3

log(

*curs*(

*t*)) =

*c*+ ?

*c*log(

*M SA*(

*t*?

*i*) /

*p*(

*t*?

*i*)) + ?

*c*log(

*prod*_

*ind*(

*t*?

*i*)) +

0

1

*i*

2 _

2

*i*

*i*1

=

*i*1

=

3

3

+?

*c*log(

*curs*(

*t*?

*i*)) + ?

*c*log(

*r*inf

*l*(

*t*?

*i*)) +

*u*(

*t*)

3

*i*

4

*i*

3

*i*1

=

*i*1

=

3

3

log(

*r*inf

*l*(

*t*)) =

*d*+ ?

*d*log(

*M SA*(

*t*?

*i*) /

*p*(

*t*?

*i*)) + ?

*d*log(

*prod*_

*ind*(

*t*?

*i*)) +

0

1

*i*

2 _

2

*i*

*i*1

=

*i*1

=

3

3

+?

*d*log(

*curs*(

*t*?

*i*)) + ?

*d*log(

*r*inf

*l*(

*t*?

*i*)) +

*u*(

*t*)

3

*i*

4

*i*

4

*i*1

=

*i*1

=

where i=1,3.

uj (j=1,4) are the regression residuals called innovations or shocks. The corresponding

innovation is, thus, that part of the evolution of the variable that neither be explained by its

past values (own history), nor by other variables of the model.

The VAR method concentrates mainly on studying the impact of every shock upon

every variable of the system of equations; this analysis is being performed by impulse

response functions, by factorial decomposition of variance.

Table 3. comprises estimated coefficient of the VAR model, obtained with Eviews 4

software.

The impulsion response functions graphically represent the evolution of these

shocks in time across 10 months, identifying the maximum impact upon variables taken into

account by the model; the sizing of these dependencies between innovations and model

variables is expressed in relative terms, that is, standard deviations of the shocks.

**Tabel 3.**The estimated coefficients of VAR model

Sample(adjusted): 1992:04 2005:12

Included observations: 165 after adjusting endpoints

Standard errors and t-statistic in brackets

LOG(M2_SA/p)

LOG(PROD_IND)

LOG(CURS_EUR)

LOG(RINFL)

LOG(M2_SA(-1)/p(-1))

1.107870

-0.151211

0.376171

-1.267625

(0.09407)

(0.24728)

(0.15506)

(2.45033)

(11.7771)

(-0.61150)

(2.42604)

(-0.51733)

LOG(M2_SA(-2)/p(-2))

0.107563

0.322834

-0.960039

-1.195572

(0.13958)

(0.36690)

(0.23006)

(3.63570)

(0.77064)

(0.87989)

(-4.17291)

(-0.32884)

LOG(M2_SA(-3)/p(-3))

-0.215949

-0.246611

0.616877

2.005835

(0.09335)

(0.24538)

(0.15387)

(2.43154)

(-2.31338)

(-1.00501)

(4.00918)

(0.82492)

LOG(PROD_IND(-1))

0.034269

-0.261478 -0.040848

1.036583

(0.02889)

(0.07593)

(0.04761)

(0.75244)

(1.18632)

(-3.44352) (-0.85791)

(1.37764)

LOG(PROD_IND(-2)) -0.017588

-0.315671 -0.105047

0.614014

(0.02858)

(0.07512)

(0.04710)

(0.74434)

**192**

**Quantitative Methods Inquires**

(-0.61547)

(-4.20240) (-2.23021)

(0.82491)

LOG(PROD_IND(-3))

0.047984

-0.313506

0.018482

0.014463

(0.02894)

(0.07608)

(0.04771)

(0.75393)

(1.65781)

(-4.12050)

(0.38740)

(0.01918)

LOG(CURS_EUR(-1))

-0.073281

-0.128218

1.496446

3.231345

(0.04646)

(0.12213)

(0.07658)

(1.21018)

(-1.57730)

(-1.04987)

(19.5411)

(2.67013)

LOG(CURS_EUR(-2))

0.133757

0.263857

-0.706407

-3.294675

(0.07548)

(0.19840)

(0.12441)

(1.96600)

(1.77219)

(1.32991)

(-5.67820)

(-1.67583)

LOG(CURS_EUR(-3))

-0.061791

-0.050815

0.157700

0.493682

(0.04544)

(0.11945)

(0.07490)

(1.18363)

(-1.35983)

(-0.42542)

(2.10550)

(0.41709)

LOG(RINFL(-1))

0.016839

-0.010827

0.011044

0.295220

(0.00376)

(0.00990)

(0.00621)

(0.09807)

(4.47250)

(-1.09394)

(1.77960)

(3.01028)

LOG(RINFL(-2))

-0.001959

-0.006298 -0.011551

0.148286

(0.00452)

(0.01189)

(0.00746)

(0.11784)

(-0.43299)

(-0.52963) (-1.54912)

(1.25840)

LOG(RINFL(-3))

-0.003161

-0.020569

0.011172

0.153754

(0.00389)

(0.01023)

(0.00641)

(0.10132)

(-0.81272)

(-2.01157)

(1.74250)

(1.51746)

C

0.073499

0.173660

-0.071412

0.423674

(0.03838)

(0.10088)

(0.06326)

(0.99964)

(1.91519)

(1.72144)

(-1.12892)

(0.42382)

R-squared

0.999864

0.237022

0.999321

0.668904

Adj. R-squared

0.999854

0.176787

0.999268

0.642765

Sum sq. resids

0.081358

0.562183

0.221043

55.20174

S.E. equation

0.023136

0.060816

0.038134

0.602635

Log likelihood

394.0993

234.6297

311.6410

-143.7914

Akaike AIC

394.2569

234.7872

311.7986

-143.6338

Schwarz SC

394.5016

235.0320

312.0433

-143.3891

Mean dependent

4.075405

-0.000154

-0.123119

-3.762357

S.D. dependent

1.911885

0.067029

1.409360

1.008271

Determinant Residual Covariance

4.92E-10

Log Likelihood

831.6832

Akaike Information Criteria

832.3135

Schwarz Criteria

833.2923

The responses of the variables studied to a standard deviation of innovations

(variation interval ± 2 standard deviations) are graphically represented, for a timespan of 10

monhs.

**193**

**Quantitative Methods Inquires**

**Conclusions**

Identifying a statistically strong model, capable of stable estimations for the money

demand function in Romania’s economy constitutes a prerequisite to the application of an

efficient monetary policy.

Obtaining by econometric means, the series of adjusted money demand, for which

the statistical stability tests are confirmed, allows for the formalization of the link between

the real-sector and monetary block, as well as the impact assessment of the levels of

monetary variables upon the economy.

**194**

**Quantitative Methods Inquires**

**Bibliography**

1. Antohi, D., Udrea, I. and Braun, H.

**Mecanismul de transmisie a politicii monetare in**

**Romania,**Caiete studii BNR nr.13/2003

2. Boughton, J. M.

**Long run money demand in large industrial countries,**International

Monetary Fund, 1991

3. Friedman, M.

**The optimum quantity of money and other essays,**Aldine, Chicago, 1969

4. Sriram, S. S.

**A survey of recent empirical money demand studies,**International Monetary

Fund, 2001

5. Treichel, V.

**Broad money demand and monetary policy in Tunisia,**International Monetary

Fund, 1997

1 Codification of references:

[1]

Antohi, D., Udrea, I. and Braun, H.

**Mecanismul de transmisie a politicii monetare in Romania,**Caiete

studii BNR nr.13/2003

[2] Boughton,

J.

M.

**Long run money demand in large industrial countries,**International Monetary Fund,

1991

[3] Friedman,

M.

**The optimum quantity of money and other essays,**Aldine, Chicago, 1969

[4]

Sriram, S. S.

**A survey of recent empirical money demand studies,**International Monetary Fund, 2001

[5]

Treichel, V.

**Broad money demand and monetary policy in Tunisia,**International Monetary Fund, 1997

**195**