Belarus Inflation Model 2005

7/30/2019 Belarus Inflation Model 2005

1/15

Research in International Business and Finance 20 (2006) 200214

Money demand and inflation in Belarus:Evidence from cointegrated VAR

Igor Pelipas

Institute for Privatization and Management, Research Center, 76 Zakharov St.,

Minsk 220088, Belarus

Received 31 January 2005; received in revised form 5 April 2005; accepted 1 September 2005

Available online 19 October 2005

Abstract

The paper analyses the demand for nominal and real money balances (M1) in Belarus on the basis of the

quarterly data for 19922003. Using cointegration analysis and dynamic equilibrium correction models, well

specified and stable long- and short-run money demand functions are derived. On the basis of long-run real

money demand functions the gap between money demand and supply (monetary overhang) is determined.

Within the framework of short-run dynamic models for real money balances, the equilibrium correctionmechanism and speed of adjustment of the system toward steady-state trajectory are investigated. Using the

model of inflation with the equilibrium corrections mechanism taken from real money demand functions,

the determinants of inflation in Belarus are examined.

2005 Published by Elsevier B.V.

JEL classification: C32; E31; E41

Keywords: Money demand; Inflation; Monetary disequilibrium; Cointegration; Equilibrium correction model

1. Introduction

Stable money demand function is an important prerequisite of conducting an effective monetary

policy. Therefore, it is not surprising that a lot of theoretical and empirical studies are devoted

to the problems of money demand.1 Many of the empirical studies provide evidence that stable

money demand function exists both in advanced economies and in some developing countries.

Tel.: +375 17 2361147; fax: +375 17 2361147.

E-mail address: [email protected] See Sriram (1999a) for comprehensive survey of the literature on theoretical and empirical aspects of money demand.

0275-5319/$ see front matter 2005 Published by Elsevier B.V.

doi:10.1016/j.ribaf.2005.09.002


2/15

I. Pelipas / Research in International Business and Finance 20 (2006) 200214 201

In such studies modern techniques of econometric analysis are used. They enable to analyze

long- and short-run aspects of economic dynamics. Wide usage of cointegration analysis and

equilibrium correction models is distinctive for this research of money demand.2

Until recently, implementation of comprehensive econometric tools in analysis of money

demand in transition economies was practically impossible due to absence of necessary sta-tistical data and/or too short time series. However, in recent publications such tools, including

cointegration analysis are used.3 Obviously, as the length of available time series expands, number

of such research will increase too.

We believe that the topicality of the research on money demand in Belarus is determined in

our view by the following. Although there exist the elements of market economy in the country

appeared in the first half of 1990s, subsequent economic policy has turned Belarus into one of the

outsiders amongst transition economies. Intensive government interventions in economic activity

substantially blocks market mechanisms and hampers private sector development. Macroeco-

nomic stability and high inflation remain a problem for Belarusian economy. In such conditions,

the analysis of money demand function allows, on the one hand, to clarify how a demand onmonetary balances is formed in the economy with high degree of state regulation, and how such

economic policy influences on inflation. On the other hand, such analysis provides useful empirical

information for effective monetary policy and anti-inflation measures.

The main aim of this paper is to get answers on the following three questions:

1. Does money demand function exist in Belarus over the period 19922003 and what are its

main determinants?

2. Is money demand function stable in the long-run and short-run?

3. Is there empirical evidence of monetary nature of inflation in Belarus?

Money demand in this paper is investigated by means of cointegration analysis and equilibrium

correction model. The results of the research should be considered definitely as preliminary and

it will be a subject of further analysis, as a longer time series will be available. Nevertheless, we

suppose that this research will stimulate further studies of money demand in Belarus.

The structure of the paper is the following. In the second section, data used in analysis are

described and their order of integration is determined. In the third section, long-run money demand

functions are investigated both for nominal and real balances. The fourth section presents con-

stancy analysis of cointegration relations reflecting money demand. In the fifth section, short-run

equilibrium correction money demand function is estimated. Monetary factors of inflation areanalyzed in sixth section. The seventh section summarizes the results of the research and provides

conclusions.

2. Data and unit root test

Analyzing money demand and inflation in Belarus we have used the following data:

2

Among numerous publications on the subject, one can note special issue of Empirical Economics (1998) devoted toempirical analysis of money demand in the EU and recent research to the IMF, especially Jonsson (1999), Sriram (1999b),

Egoume-Bossogo (2000), Adedeji and Lui (2000), and Nachega (2001).3 See, for instance, Choudhry (1998), Korhonen (1998), Kalra (1998), Babic (2000), Bahmani-Oskooee and Barry

(2000), and Rother (2000), Yang (2001).


3/15

202 I. Pelipas / Research in International Business and Finance 20 (2006) 200214

- consumer price index (CPI);

- monetary aggregate M1;

- real industrial production (RIP = NIP/PPI, where NIP, nominal industrial production; PPI,

producer price index), as a proxy of real income;

- market exchange rate (Belarusian rubles per US$EX M);- refinance rate (REF), as a proxy for interest rate;

- real money balances (M1/CPI).

We used seasonally unadjusted data and included seasonal dummies in appropriate regres-

sions. All data, excluding refinance rate, were transformed into logs: cpi =lnCPI, m1=lnM1,

m1p =lnM1 ln CPI, rip = ln RIP, ex m = ln EX M. The first differences are approximations

of growth rates of the variables. Quarterly data for the period 1992:12003:4 have been used

(48 observations). Since our sample is rather short for cointegration analysis it is necessary

to note the following. First, such a problem is typical of all transition economies. Neverthe-

less, there are a lot of studies, where comprehensive econometric tools, including cointegration

analysis have been successfully implemented. For example, see Choudhry (1998), Korhonen

(1998), Kalra (1998), Bahmani-Oskooee and Barry (2000), Babic (2000), Rother (2000), and

Yang (2001). Secondly, Campos and Ericsson (1999) have shown that time series with small

amount of observations may be rather informative.4 Information content of such data highly

depends on per-observation data variance. Such situation can be observed in developing coun-

tries and transition economies, where time series are short but information content of each

observation is relatively substantial. This consideration is necessary for implementing modern

econometric tools while analyzing transitions economies with rather short time series. Thirdly,

in transition economies duration of the long-run can be much shorter than in advancedeconomies. One can consider long-run here as a time necessary to restore equilibrium after

shocks. If the time of this adjustment is much smaller in comparison with entire sample,

probably we can consider to some extent, this process in terms of long- and short-run. All

above-mentioned argue in favor of carrying out long-run analysis even with relatively short time

series.

Time series used in analysis are presented on Fig. 1. As one can see, some time series

have structural breaks. It is necessary to take them into account while testing for unit root. We

started with ADFGLS test, which is a more powerful modification of standard DickeyFuller

unit root test (see Elliot et al., 1996). The results of the tests are reported in Table 1. All

series in levels are nonstationary variables. The first differences are stationary for all variables,except cpi and m1 where ADFGLS test does not allow to reject the null hypotheses of a unit

root.

As structural breaks are present in time series more careful analysis is needed. We used modified

unit root test allowing for changing mean in time series (Perron, 1992) that is the case in cpi

and m1. Table 2 presents the results and the null hypotheses of a unit root in cpi and m1 is

strongly rejected. Therefore, one can conclude that cpi and m1 are stationary variables with

a changing means. Therefore, all variables in levels are I(1) and can be a subject of cointegration

analysis.

4 Campos and Ericsson (1999) showed that 16 yearly observations for Venezuela, because of high variation in analyzed

variables, are more informative than the same quarterly data for the USA over 40 years.


4/15


Fig. 1. Time series used, 1992:012003:04 (log scale, dis difference operator ).

3. The empirical cointegrated VAR: nominal and real money balances

Our initial empirical model consists of five variables: monetary aggregate, m1; consumer price

index, cpi; real industrial production, rip; market exchange rate, ex m; and refinance rate, ref.5 All

series are quarterly data expressed in natural logs. To correct for seasonality, centered dummies

are included in the models. In contrast to many studies, the restriction of price homogeneity is

5 Data of the Ministry of Statistics and Analysis of the Republic of Belarus and the National Bank of the Republic of

Belarus have been used. Appropriate raw time series are presented in the database of the Research Center of the Institute

for Privatization and Management, http://research.by.
http://research.by/http://research.by/


5/15


Table 1

DickeyFuller GLS unit root test

Variables t-ADFGLS

Constant and trend Constant

cpi 1.705 (1) 0.456 (1)

m1 2.070 (3) 0.598 (3)

m1 cpi 1.945 (3) 1.498 (3)

rip 1.522 (0) 1.048 (0)

ex m 1.860 (2) 0.562 (2)

ref 2.506 (2) 1.677 (2)

cpi 2.509 (0) 1.601 (0)

m1 1.795 (2) 1.165 (2)

(m1 cpi) 3.278 (2)* 2.020 (2)*

rip 6.873 (0)** 5.965 (0)**

ex m 4.265 (0)**

3.364 (0)**

ref 6.227 (1)** 6.144 (1)**

Notes: Numbers in parentheses are optimal lag length chosen by Swartz information criteria. Maximal lag length is four

quarters. ADFGLS tests and appropriate critical values are calculated using Eviews 5.1.* Denotes rejection of the null hypothesis of a unit root at the 5% level.

** Denotes rejection of the null hypothesis of a unit root at the 1% significance level.

not imposed. First, model with nominal money balances is used. Then the hypothesis of price

homogeneity is tested. This approach allows us to model correctly, money demandfor real balances

if the hypothesis of price homogeneity is not rejected.

The appropriate vector model with equilibrium correction mechanism can be written as follows:

Xt = Dt+

k1

i=1

iXti + Xt1 + t, t= 1, . . . , T, (1)

where Xt is a vector of endogenous variables; Dt is a deterministic vector (constant, trend, sea-

sonable dummies, etc.); is a matrix of coefficients Dt; is a difference operator; i is a matrix

of coefficient characterizing the long-run dynamics of variables; t is a vector of serially uncor-

related stochastic errors. The number of cointegrating vectors is equal to the rank of matrix

where is the matrix of cointegrating vectors characterizing the long-term relationship between

Table 2

Unit root test with a changing mean

Variables Tb t-ADF AR 14 (p-value)

cpi 1994:4 6.376 (1)** 0.1940

m1 1995:2 5.672 (0)** 0.1445

Notes: (1) Critical values are from Perron (1992) and equal 5.51 and 4.76 at 5 and 1% significance level, respec-

tively; (2) Tb is the point of structural break. Numbers in parentheses are optimal lag length, which was chosen to yield

uncorrelated residuals in appropriate unit root tests. AR 14 is F-test for serial correlation of residuals of 1 n-order,

H0: serial correlation is not present (Hendry and Doornik, 2001); (3) the following regression is used in testing for a unit

root with a changing mean (Perron, 1992): yt = + DUt+ D(TB)t+ yt1 +

ki=1

ciyti + t, where

yt = ytyt 1; yt =ytyt1; , , , , ci are the parameters of regression; DUt = 1 (t> Tb) and D(TB)t = 1

(t= Tb + 1) are the dummies; kis the number of lags in regression; Tb is the point of structural break; t is an error term.** Denotes rejection of the null hypothesis of a unit root at the 1% significance level.


6/15


variables, is the matrix of the feedback coefficients characterizing the speed of the equilibrium

adjustment of the system. The rank of the matrix and, respectively, the number of cointegration

vectors is determined by using trace statistics LR(trace) = Tk

i=r+1 ln(1 i), where i are

the eigenvalues (1 . . . k), T is the number of observations. The null hypothesis H0 is that

there are rcointegration vectors against the alternative H1: r+ 1 that there are r+ 1 cointegrationvectors. If LR(trace) is statistically significant, then the null hypothesis is rejected.

The lag length of the VAR was chosen according to the LM-type test (it equal two). Trend is

restricted in cointegration space. The model is also checked for serial correlation, normality and

ARCH effect. Multivariate tests of serially uncorrelated residuals indicate that the null cannot be

rejected against the alternative hypothesis of first-order correlation and correlation at the fourth lag,

respectively. Thus, two lags seem to be sufficient for describing the dynamics of the system. We

also have not found any signs of conditional heteroskedasticity, while normality of the residuals

is rejected. Following Bruggeman et al. (2003), we use Bartlett correction for the trace test and

bootstrap p-values since Johansen test for cointegration can to be oversized in small samples (our

sample includes only 48 quarters). The results of cointegration analysis for the system of fivevariables with nominal money balances are presented in Table 3.

Based on Bartlett corrected trace test and appropriate bootstrap p-values, it is reasonable to

conclude that there exists one cointegrating vector in the system of five variables. One can consider

such a vector as a money demand function for nominal balances. A set of appropriate tests has been

carried out (both related to long-run parameter and loading coefficients). As a result, we found

out that hypothesis of long-run unit price homogeneity is not rejected. Moreover, the hypothesis

of long-run unit real income homogeneity is also not rejected.

Weak exogeneity tests show that real industrial production, exchange rate and refinance rate are

weakly exogenous variables. Nominal money and prices are endogenous variables and interact

with each other. The adjustment mechanism is as follows: when nominal money balances are

above their equilibrium, then they should be decreased and at the same time prices will increase

(and vice versa). It means that at least in the long-run nominal money (monetary gap) influences

prices.

Long-run price homogeneity permits to model correctly money demand function for real

balances. As in the first case, we used VAR with two lags and trend restricted in cointegration

space. Model includes centered seasonal dummies. The set of the variables is the same and the

nominal to real transformation leads to m1p. Results of cointegration analysis for real money

balances are given in Table 4. We definitely see one cointegrating vector, taken into consideration

small sample Bartlett correction and bootstrap p-values. According to weak exogeneity test, allvariables in the system are weakly exogenous, except real money balances, m1p. This permits

to model long- and short-run aspects of money demand function within single regression with

equilibrium correction mechanism.

4. Constancy analysis

Following Bruggeman et al. (2003), we carried out formal tests to investigate the parameter

constancy of the cointegrated VAR. Usually, recursive estimates over a limited time period and

ocular inspection of recursive Chow forecast, breakpoint, or predictive failure tests have been

used to examine this problem. Definitely, such diagnostics are useful for preliminary analyses butany inferences drawn from these exercises neglect a large fraction of the sample period and do

not take into account that such tests are formal tests only for a single point in time. In this section,

the parameter constancy of three sets of parameters in Eq. (1) is examined. First, we analyze


7/15


Table 3

Results of cointegration analysis: nominal money balances

1. Cointegration test

Eigenvalue 0.7629 0.6601 0.3362 0.2818 0.1583

Null hypothesis, H0 r= 0 r 1 r 2 r 3 r 4

LRtrace 157.85 91.64 42.01 23.16 7.93

Asymptotic p-value 0.0000 0.0000 0.0615 0.1050 0.2577

Bootstrap p-value 0.0005 0.0295 0.4722 0.3887 0.5303

LRtrace (Bartlett correction) 95.86 61.52 29.30 14.19 6.32

Asymptotic p-value 0.0142 0.0778 0.5437 0.6415 0.4203

Bootstrap p-value 0.0170 0.0980 0.5798 0.5243 0.4312

2. Standardized cointegrating vector () and -coefficients

Variables m1 cpi rip ex m ref Trend

Cointegrating vector, 1.0000 0.8253 0.5758 0.07467 0.0642 0.0535

-Coefficients 0.4668 0.2254 0.0697 0.0322 0.0998

3. Test for significance of a given variable in and weak exogeneity test


Significance of a given

variable in , 2(1)

16.559

[0.000]

11.846

[0.001]

4.999

[0.025]

0.414

[0.520]

0.424

[0.515]

15.477

[0.000]

Weak exogeneity, 2(1) 15.415

[0.000]

3.248

[0.072]

0.735

[0.391]

0.013

[0.910]

0.017

[0.897]

4. Structural hypotheses

(1)

= (1 1 ****) 2(1) = 1.4244 [0.2327](2) = (1 1 1 ***) 2(2) = 2.2962 [0.3172]

(3) rip = 0erm = 0ref= 0 2(3) = 0.8671 [0.8334]

(4) = (1 1 1 ***)rip = 0, erm = 0, ref= 0 2(5) = 4.4521 [0.4863]

5. Estimated cointegrating vector according to hypothesis 4


Cointegrating vector, 1.000 1.000 1.000 0.224 0.094 0.049

Standard errors of 0.016 0.030 0.004

-Coefficients 0.421 0.264

Standard errors of 0.048 0.073

Notes: Bootstrap p-values and trace test with Bartlett correction have been performed in Structural VAR, version 0.32

(http://texlips.hypermart.net/warne/code.html); all other computations have been carried out in PcGive 10.4 ( Hendry and

Doornik, 2001). p-values are in brackets.

non-zero eigenvalues used in the cointegration rank analysis. The main tool here is the fluctuation

test suggested by Hansen and Johansen (1999). Second, we examine the constancy of using

the Nyblom (1989) tests studied by Hansen and Johansen (1999). Third, we take a look at the

constancy of the , 1, and parameters using the fluctuation test ofPloberger et al. (1989). All

the formal tests do not require trimming of the sample but for computational reasons, however,

we will use 40 percent of the sample as a base period and examine constancy over the remainder.The results of constancy analysis are presented in Table 5. In addition to asymptotic p-values,

bootstrap p-values were used. It helps to make more reliable inference in our relatively small

sample. As one can see from the different constancy tests, both cointegrated VAR with nominal
http://texlips.hypermart.net/warne/code.htmlhttp://texlips.hypermart.net/warne/code.html


8/15


Table 4

Results of cointegration analysis: real money balances

1. Cointegration test

Eigenvalue 0.6609 0.3504 0.3075 0.3075

Null hypothesis, H0 r= 0 r 1 r 2 r 3

LRtrace 90.91 41.17 21.32 4.42

Asymptotic p-value 0.0000 0.0740 0.1661 0.6806

Bootstrap p-value 0.0045 0.3267 0.4347 0.8709

LRtrace (Bartlett correction) 70.73 32.40 13.74 3.11

Asymptotic p-value 0.0121 0.3673 0.6789 0.8637

Bootstrap p-value 0.0135 0.3642 0.5568 0.8579

2. Standardized cointegrating vector () and -coefficients

Variables m1p rip ex m ref Trend

Cointegrating vector, 1.0000 1.1148 0.2265 0.0212 0.0470

-coefficients 0.6743 0.1377 0.0209 0.3661

3. Test for significance of a given variable in and weak exogeneity test


Significance of a given

variable in , 2(1)

22.036

[0.000]

28.786

[0.000]

23.369

[0.000]

0.3033

[0.582]

20.181

[0.000]

Weak exogeneity, 2(1) 27.814

[0.000]

2.509

[0.113]

0.006

[0.940]

0.425

[0.515]

4. Structural hypotheses

(1)

= (1 1 ***) 2(1) = 0.5544 [0.4565](2) rip = 0erm = 0ref= 0

2(3) = 2.7803 [0.4268]

(3) = (1 1 ***)rip = 0, erm = 0, ref= 0 2(4) = 2.7912 [0.5933]

5. Estimated cointegrating vector according to hypothesis 3


Cointegrating vector, 1.0000 1.0000 0.2223 0.0523 0.0464

Standard errors of 0.0173 0.0251 0.0047

-Coefficients 0.7034

Standard errors of 0.0850

Notes: Bootstrap p-values and trace test with Bartlett correction have been performed in Structural VAR, version 0.32

(http://texlips.hypermart.net/warne/code.html); all other computations have been carried out in PcGive 10.4 ( Hendry and

Doornik, 2001). p-values are in brackets.

money balances and with real money balances do not show any non-constancy in terms of fluc-

tuation eigenvalue test, different types of tests concerning long-run parameters and test relevant

to short-run part of the system and adjustment coefficients as well. Thus, we can conclude that

results of cointegration analysis are constant over the whole sample.

5. Conditional equilibrium correction model for real money balances

Table 6 presents results of modeling dynamics of real money balances. In line with cointe-

grated VAR, this model initially includes the same variables with lag one, seasonal dummies and

equilibrium correction mechanism, taken from cointegration analysis (EqCM1r). The diagnostics
http://texlips.hypermart.net/warne/code.htmlhttp://texlips.hypermart.net/warne/code.html


9/15


Table 5

Constancy analysis of the cointegrated VAR

1. HansenJohansen fluctuation test of the constancy of the non-zero eigenvalues

Model (eigenvalue, ) Conditional on (T) and 1(T) Updating of(t) and 1

(t)

suptT t/T(i) Asymptotic

p-value

Bootstrap

p-value

suptT t/T(i) Asymptotic

p-value

Bootstrap

p-value

Nominal money (1) 0.3265 1.0000 0.3702 0.5004 0.9638 0.3787

Real money (1) 0.3739 0.9992 0.4532 0.6712 0.7585 0.2216

2a. Nublom supremum test of the constancy of parameters of long-run relationship ()

Model Conditional on (T) and 1(T) Updating of(t) and 1

(t)

suptTQtT(i) Asymptotic

p-value

Bootstrap

p-value

suptTQtT(i) Asymptotic

p-value

Bootstrap

p-value

Nominal money 1.6247 0.5277 0.2401 2.9400 0.0752 0.0675

Real money 1.6522 0.3909 0.1556 2.4664 0.1089 0.0810

2b. Nublom mean test of the constancy of parameters of long-run relationship ()

Model meantTQtT(i) Asymptotic

p-value

Bootstrap

p-value

meantTQtT(i) Asymptotic

p-value

Bootstrap

p-value

Nominal money 0.6486 0.3624 0.2716 0.6949 0.3135 0.4272

Real money 0.7249 0.1955 0.1241 0.9619 0.0845 0.1026

3. PlobergerKramerKontrus fluctuation test of the constancy of parameters , 1,

Model Equation S(10), S(9) Asymptotic p-value Bootstrap p-value

Nominal money m1 1.3246 0.4604 0.7529

cpi 1.5437 0.1578 0.5788

rip 1.2065 0.6840 0.8399

ex m 1.5134 0.1870 0.6028

ref 0.8717 0.9966 0.9035

Real money m1p 1.1961 0.6648 0.7269

rip 1.2183 0.6231 0.6728

ex m 1.1628 0.7255 0.7419

ref 0.9063 0.9876 0.9490

Notes: Computations have been carried out in Structural VAR, version 0.32.

of this initial regression, showed in the first part of the Table 6, is fairly good. No serious problems

were detected with residuals and structural breaks. Then this model was reduced using general to

specific approach (Hendry and Krolzig, 2001). The final specific model is presented in second

part of the table. This model also looks good except for some problems with heteroskedasticity

at the 5% level.

Conditional equilibrium correction model for real money balances has three explanatory vari-

ables and seasonal dummies. Models demonstrate strong equilibrium correction mechanism; ittakes about 1.3 quarters to restore equilibrium after shock. All coefficients have theoretically

expected signs. As Fig. 2 shows, the model fits reasonably well and out-of-sample forecast is

also good enough and within 95% confidence interval. Thus, obtained results witness that it is


10/15


Table 6

Conditional equilibrium correction model for real money balances (dependent variable m1 p)

A. General model

Variables Coefficient Standard error t-value t-prob

Constant 1.64260 0.25922 6.337 0.0000

m1pt1 0.36082 0.09699 3.720 0.0007

ript 0.02738 0.16881 0.162 0.8721

ript 1 0.03973 0.16644 0.239 0.8128

ex mt 0.00091 0.05758 0.016 0.9875

ex mt 1 0.00454 0.06630 0.069 0.9458

reft 0.06868 0.02822 2.434 0.0205

reft 1 0.06479 0.02764 2.344 0.0253

Seasonal 0.00752 0.03573 0.211 0.8345

Seasonal 1 0.12444 0.03908 3.184 0.0032

Seasonal 2 0.12160 0.03080 3.947 0.0004

EqCM1rt1 0.71679 0.11704 6.124 0.0000

Value Probability

Chow (1998:2) 0.3908 0.9708

Chow (2002:4) 0.3583 0.8362

Normality test 2.3370 0.3108

AR 14 test 0.7717 0.5525

ARCH 14 test 1.0032 0.4245

Hetero test 2.4412 0.0522

B. Specific model


Constant 1.64819 0.18144 9.084 0.0000

m1pt1 0.36729 0.07481 4.909 0.0000

reft 0.06551 0.02212 2.961 0.0053

reft 1 0.06606 0.02381 2.775 0.0086

Seasonal 0.01071 0.03053 0.351 0.7278

Seasonal 1 0.12651 0.03615 3.499 0.0012

Seasonal 2 0.12185 0.02752 4.428 0.0001

EqCM1rt1 0.71917 0.08541 8.420 0.0000

Value Probability

Chow (1998:2) 0.4951 0.9348

Chow (2002:4) 0.4201 0.7930


AR 14 test 0.7574 0.5604

ARCH 14 test 1.0641 0.3920

Hetero test 2.9034 0.0132

Notes: computations presented in the table and Table 7 have been carried out in PcGets 1.0 (Hendry and Krolzig, 2001).

The liberal strategy (minimize non-selection probability) was used in this research.

possible to build money demand function for Belarusian economy on the basis of quarterly dataover the period 1992:012003:4. This function is consistent with theoretical considerations in the

long run and fits well in the short run. Moreover, parameters of long- and short-run relationships

are rather stable over the sample.


11/15


Fig. 2. Conditional equilibrium correction model for real money balances: actual and fitted values, residuals and 1-step

forecast with 95% confidence bars.

6. Money and inflation

Using results from the analysis of money demand function, it is interesting to examine the role

of monetary factors in inflation process in Belarus over 19922003. One can consider two aspects

of monetary impact on inflation. On the one hand, short-run dynamics of monetary aggregate,

say m1, on the other hand, long-run influence through monetary gap. Since monetary gap is the

difference between money supply and money demand, equilibrium correction mechanism from

long-run analysis of real money demand can be used to test the appropriate hypotheses.

Building general model of inflation, we include all potentially relevant variables: inflation

with lag, reflecting inflation inertia, growth of monetary aggregate m1, real industrial produc-

tion, exchange rate and refinancing rate. Disequilibrium on monetary market is reflected throughequilibrium correction mechanism taken with one lag (EqCM1rm is the mean-zero equilibrium

correction mechanism). Model also includes seasonal dummies and impulse dummy, considering

financial crisis in Russia in 1998. Table 7 shows regression results. General model have no serious

problems with residuals and structural changes. Thus, it is a good basis for further simplification

by exclusion of insignificant variables. As it was stated earlier, using general to specific approach

we have got final model with acceptable characteristics.

The model shows that all monetary variables (m1, ex m, ref) influence inflation in the short run.

Almost all these variables have an expected sign and are significant. One exception is refinance

rate that has wrong positive sign in inflation regression. Such a result is not unusual for transition

economies. For example, Rother (2002) observed such effect, analyzing inflation in Albania. Inour case this can be explained as following: refinance rate is determined by monetary authorities

in accordance with dynamics of inflation and perhaps some kind of simultaneity one can observe

here. If the longer lag length than we use in our analysis will be taken, one can see right


12/15


Table 7

Impact of monetary factors on inflation (dependent variable cpi)

A. General model


cpit 1 0.50355 0.10048 5.012 0.0000

m1t 0.09158 0.27256 0.336 0.7392

m1t 1 0.14009 0.19452 0.720 0.4770

rip 0.03649 0.17027 0.214 0.8318

ript 1 0.03332 0.17608 0.189 0.8512

ex mt 0.21507 0.07877 2.730 0.0105

ex mt 1 0.06984 0.07268 0.961 0.3443

reft 0.05534 0.02502 2.212 0.0347

reft 1 0.06244 0.02550 2.448 0.0204

Constant 0.00613 0.04116 0.149 0.8826

Seasonal 0.02759 0.03437 0.803 0.4286

Seasonal 1 0.03354 0.04401 0.762 0.4520Seasonal 2 0.02194 0.04066 0.540 0.5935

EqCM1rmt1 0.40710 0.13791 2.952 0.0061

D984 0.03238 0.02218 1.460 0.1547

Value Probability

Chow (1998:2) 0.6522 0.7974

Chow (2002:4) 0.2009 0.9356


AR 14 test 1.5732 0.2111

ARCH 14 test 0.4668 0.7594

Hetero test 36.0840 0.0539

B. Specific model


cpit 1 0.51482 0.05563 9.254 0.0000

m1t 1 0.20844 0.07156 2.913 0.0060

ex mt 0.24087 0.04126 5.838 0.0000

reft 0.05862 0.02047 2.864 0.0068

reft 1 0.07068 0.02101 3.364 0.0018

Seasonal 0.05288 0.01828 2.893 0.0063

EqCM1rmt1 0.43895 0.07046 6.230 0.0000

Value Probability

Chow (1998:2) 0.6135 0.8578

Chow (2002:4) 0.1125 0.9773


AR 14 test 1.3635 0.2671

ARCH 14 test 0.6098 0.6587

Hetero test 21.6695 0.0607

positive effect of refinance rate as a proxy of the interest rate in inflation equation. Monetary gaphas a significant impact on inflation in the long-run. It is important to note that money matters

both in short- and long-run. Fig. 3 shows prognostic performance of the model. As one can see,

equilibrium correction model of inflation fits data reasonably well and performs quite good out-


13/15


Fig. 3. Equilibrium correction model for inflation: actual and fitted values, residuals and 1-step forecast with 95% confi-

dence bars.

of-sample forecast. Thus, money contains a useful information concerning dynamics of prices inBelarus during the period of the research.

7. Concluding remarks

Using cointegrated VAR and equilibrium correction model in studying money demand and

inflation in Belarus the following results have been obtained:

1. All data used in analysis (nominal money M1, real money, consumer prices, real industrial

production, exchange rate and refinancing rate) areI(1) variables. The first differences of thesevariables are stationary and have the order of integration I(0). Thus, cointegration technique

is an appropriate tool for analysis of long-run relationships between mentioned variables. It is

important to note that while determining the order of integration of several variables (namely

the first differences of M1, the first differences of consumer prices) the structural breaks have

to be taken into account.

2. As cointegration analysis shows, there exists the long-run function for nominal money bal-

ances. This long-run relationship is consistent with theoretical expectations and stable within

investigated sample. Demand for nominal money in the long run is determined by consumer

prices, real industrial production (as a proxy for real income), nominal exchange rate and

refinancing rate.3. According to our analysis, the hypotheses of price homogeneity cannot be rejected for long-run

money demand function (there is no monetary illusion). It enables us to model demand for

real money balances correctly.


14/15


4. Cointegration analysis provides us with the evidence that long-run function for real money

balances exists. Demand for real money balances in the long run is determined by real industrial

production, nominal exchange rate and refinancing rate. Inflation, as a stationary variable, is

not included in the long-run relationship.

5. In the framework of the model for nominal money balances, equilibrium correction occursthrough two variables, namely M1 and prices, which are endogenous in cointegrated VAR.

Within the model for real money balances equilibrium in money market restores through

endogenous variables such as real money and nominal exchange rate. Speed of adjustment is

approximately 2.4 quarters.

6. Tests for weak exogeneity have shown that modeling short-run money demand functions for

nominal monetary balances has to be done within the system of equations; for real money

balances single equation approach is appropriate. According to our results, for real money

balances there exists well specified and recursively stable short-run money demand function

with clear-cut economic interpretation.

7. Analysis in the framework of dynamic model of inflation with equilibrium correction mecha-nism proves the hypotheses about the monetary nature of inflation in Belarus. Money supply

growth influences inflation both in the long- and short-run.

8. This research shows that in Belarusian economy the adjustment process occurs rather fast. The

period of equilibrium correction on money market is considerably shorter than entire sample

period. This evidence is in favour of acceptability of using comprehensive econometric tools,

including cointegration analysis, for rather short but relatively informative time series.

Acknowledgements

We thank Lucio Vinhas de Souza, Yulia Vymyatnina and all participants of the Second Meeting

of the UACES Study Group on Monetary Policy in Selected CIS Countries (Helsinki, February

1011, 2005) for useful comments and discussions. The usual disclaimers apply.

References

Adedeji, O.S., Lui, O., 2000. Determinants of inflation in the Islamic Republic of Irana macroeconomic analysis. IMF

Working Paper, WP/00/127. International Monetary Fund, Washington.

Babic, A., 2000. The monthly transaction money demand in Croatia. Working Paper W-5, Croatian National Bank.

Bahmani-Oskooee, M., Barry, M.P., 2000. Stability of the demand for money in an unstable country: Russia. J. Post

Keynesian Econ. 22, 619629.

Bruggeman, A., Donati, P., Warne, A., 2003. Is the demand for euro area M3 stable? ECB Working Paper, 255.

Campos, J., Ericsson, N.R., 1999. Constructive data mining: modeling consumersexpenditure in Venezuela. Econometrics

J. 2, 226240.

Choudhry, T., 1998. Another visit to the Cagan model of money demand: the latest Russian experience. J. Int. Money

Finance 17, 355376.

Egoume-Bossogo, P., 2000. Money demand in Guyana. IMF Working Paper, WP/00/119. International Monetary Fund,

Washington.

Elliot, G., Rothenberg, T.J., Stock, J.H., 1996. Efficient tests for an autoregressive unit root. Econometrica 64, 813836.

Empirical Economics, 1998. 23, 263524.

Hansen, H., Johansen, S., 1999. Some tests for parameter constancy in cointegrated VAR-models. Econometrics J. 2,

306333.Hendry, D.F., Doornik, J.A., 2001. Empirical Econometric Modelling Using PcGive 10, vol. I. Timberlake Consultants

Ltd., London.

Hendry, D.F., Krolzig, H.-M., 2001. Automatic Econometric Model Selection Using PcGets 1. 0. Timberlake Consultants

Ltd., London.


15/15


Jonsson, G., 1999. Inflation, money demand, and purchasing power parityin SouthAfrica, IMFWorking Paper,WP/99/122.

International Monetary Fund, Washington.

Kalra, S., 1998. Inflation and money demand in Albania. IMF Working Paper, WP/98/101. International Monetary Fund,

Washington.

Korhonen, I., 1998. A vector error correction model for price, money, output and interest rate in Russia. Review of

Economics in Transition. Bank of Finland 5, 3344.Nachega, J.-C., 2001. A cointegration analysis of broad money demand in Cameron. IMF Working Paper, WP/01/26.

International Monetary Fund, Washington.

Nyblom, J., 1989. Testing for the constancy of parameters over time. J. Am. Stat. Assoc. 84, 223230.

Perron, P., 1992. Nonstationarity and level shifts with an application to purchasing power parity. J. Business Econ. Stat.

10, 301320.

Ploberger, W., Kramer, W., Kontrus, K., 1989. A new test for structural stability in the linear regression model. J.

Econometrics 40, 307318.

Rother, P.C., 2002. Inflation in Albania. Post-Communist Economies 14, 85107.

Rother, P.C., 2000. Inflation in Albania. IMF Working Paper, WP/00/207. International Monetary Fund, Washington.

Sriram, S.S., 1999a. Survey of literature on demand for money: theoretical and empirical work with special reference to

error-correction models. IMF Working Paper, WP/99/64. International Monetary Fund, Washington.

Sriram, S.S., 1999b. Demand for M2 in an emerging-market economy: an error-correction model for Malaysia. IMF

Working Paper, WP/99/173. International Monetary Fund, Washington.

Yang, S., 2001. Estimation of Russian money demand using a general to specific modeling methodology and Johansens

cointegration analysis. Unpublished working paper.

Documents

Belarus Inflation Model 2005