THE ROLE REMITTANCES IN THE STABILITY OF MONEY DEMAND

IN PAKISTAN: A COINTERGRATION ANALYSIS

SUBMITTED BY ABDUL UR REHMAN

SUBMITTED TO SIR MUHAMMAD UMAR

CLASS BBA 5TH A (MORNING)

ROLL NO 2181022
 SECTION 1: INTRODUCTION

Summary:
Remittances are the sum of compensation and personal transfers in the balance of
payment. In open economy countries it impacts on the development and welfare along with other
factors. Remittances are one source of financial inflow and substitute for foreign aid in
developing countries such as Pakistan and may affect money demand. Developing economies
depends upon remittances to cover scarce domestic resources, inadequate private capital inflow,
and scarce official development assistance.

Remittance transfers bring households nearer to a country’s available financial services.

Levine and Zervos (1996) confirm the satisfactory and active role of financial institutions in
enhancing a country’s growth, which may affect monetary transactions.

A rise in remittances should encourage households to increase their money holdings for
transaction motives. Remittance encourages households in Pakistan to use USD as a medium of
exchange which to substitution of PKR. More use of USD tends to decrease demand of domestic
currency. On the other hand, foreign income tends to increase foreign reserves and money supply.

Adenutsi and Ahortor, Vergas-Silva studies are focused on remittances in estimation of

money demand for Ghana and Mexico. There is no study in Pakistan which includes remittances
as independent variable. Pakistan is a labor-intensive country like Ghana and Mexico. It is very
dependent on remittances from emigrants working abroad. So, this study adds remittances as
exogenous variable in money demand. Pakistan differs from Ghana and Mexico in many aspects
such as population, GDP and inflation rate. So, the role of remittances in Pakistan’s money
demand is not similar as Ghana and Mexico.

Thus, this research paper contributes to the literature by including remittances as an

independent variable along with other variables in estimation of money demand in Pakistan.
Section two gives brief history of remittances. Section three summarizes the literature on money
demand. Section four explains methodology and data. Section five discusses results and Section
six provides conclusion.
 SECTION 3: LITERATURE REVIEW

Summary:
Literature review consists of following studies:

AUTHORS YEAR ESTIMATION TECHNIQUE RESULTS

Muco, Sanfey, & Taci 2004 - Remittance inflows help to bring stability between
inflation and exchange rate.
Muco, Marta, 1999 - Empirical evidences that in developing countries
Papapanagos, & Sanfey remittances explain more variation in inflation than
Haderi, Papapanagos, 1999 - money supply.
Sanfey, & Talka
Lewis 1954 - Labor surplus, innovative technologies, foreign aid,
Solow 1956 - Foreign Direct Investment (FDI) and the
Denison 1967 - opportunity cost of investment are the source of
Romer 1986 - economic growth.
Barro 1991 -
Ratha 2007 - Remittance play and important role in boosting
income and stable investment.
Buch, Kukulenz & 2002 - Remittances are more valuable than foreign direct
Manchec investment and private capital flows.
Ball, Lopez & Reyes 2013 - Under fixed exchange rate regime, they temporarily
do not effects money supply, reduce inflation and
increase real exchange rate.
Reinhart, & Rogoff 2004 - Increase in remittances cause increase in prices by
shifting resources from trade sector to non-trade
sector.
Verga-Silva & Haug 2006 - Remittances improve macroeconomic conditions in
host country as compared to home country.
Ratha 2005 - Remittances are source of foreign exchange, a tool
to alleviate poverty, risk sharing mechanism,
investment source and increase future consumption.
Shahbaz & Amir (Pakistan) 2009 -
Amjad et al (Pakistan) 2013 -
Mughal (Pakistan) 2013 -
Abiad et al (91 Countries) 2013 - Studies on the remittances include in literature. But
Bang et al (India) 2013 - none of these studies investigates the impact of
Keshri & Bhagat (India) 2013 - remittances on money demand.
Akkoyunlu & Siliverstovs 2013 -
(Turkey)
Imai et al (Asia) 2013 -
Adentusi & Ahortor 2008 VAR Remittances have no impact on money demand for
Ghana.
Vergas-Silva 2009 Variance decomposition, impulse Remittances negatively impact US dollars money
response and vector error correction demand with statistical significance. Remittances
model Granger-cause money demand for Mexico.
 SECTION 4: METHODLOGY AND DATA

Summary:
Money and demand model for Pakistan’s economy is as follow:

LnMt = b0 + b1LnYt + b2Lnit + b3nt + b4LnRERt + b5LnRMt + et

Where Mt is the money demand a dependent variable, consisting of M1 and M2 the broad money
demand. Y is the real income and its coefficient is expected to have positive sign, i is the
discount rate and the coefficient has a negative sign and π is the inflation rate and its coefficient
is negative. Explanatory variable RER (real exchange rate) and RM (remittances), both of them
have unclear sign. Nominal monetary aggregate serves as a econometric warning as compared to
real aggregate. Money demand function deals with real balances. M1 consist of currency in
circulation and demand deposits. It is transformed into real terms as LnM1 by deflating with
GDP deflator for the year 2006. Similarly, M2 consisting of M1 and quasi-money is converted
into real terms LnM2.

Real income LnY, expressed in millions PKR at price base year 2006. As there is a
positive relation between real income and money demand, the expected coefficient of real
income is b1>0, interpreting an increase in money demand as real income increases. The bank
discount rate Lni is used as a proxy for domestic interest rate. As there is a negative relation
between discount rate and money demand the sign coefficient discount rate is expected to be
negative. b2<0 interpret as increase in opportunity cost of holding money as an increase in
discount rate. Inflation rate is also an opportunity cost of holding money, which continuously
changes in the price level in the economy. Inflation rate uses the proxy of a GDP deflator and is
defined as follow:

Where P(t) and P(t-1) are the current and previous year’s GDP deflators. The coefficient of
inflation rate is negative as the relationship between inflation rate and money demand is
negative. The exchange rate calculates the number of units of domestic currency in terms of
foreign currency. The nominal exchange rate can be converted into real exchange rate by
following formula:

Where EX is the exchange rate of PKR, Pd and P*are the price level in Pakistan and USA. The
sign of coefficient is undecided. A negative sign results in an appreciation, results in decrease of
money demand whereas positive sign favors wealth effect hypothesis. Remittances (LnRM)
expressed in million PKR, are converted into real terms by deflating GDP of base year 2006. The
presences of dollarization in economy facilitates household of holding both currencies.
Therefore, sign of coefficient is unstill unclear and its impact may be positive or negative on
money demand.

The ARDL Bound Testing Approach:

Autoregressive Distributed Lag (ARDL) Bound Testing approach, A modern cointegration

technique for examining long and short-term relationships between independent and dependent
variables is used to exam effect of remittances on money demand. This technique is appropriate
for small sample size unlike other cointegration approaches it do not need all independent
variables in same order. ARDL is possible for the same number of optimal lags, which is not
possible with traditional techniques. ARDL model suggested by Pesaran AND Shin (1999) and
Pesaran et al. (2001) is used in this paper:

The parameter ℽj j=1, 2, 3, 4, 5. portrays long term corresponding variable normalized by ℽ0.
Meanwhile bi, ci, di, ei, fi, and gi indicators of money demand in Pakistan depict short-run
effect.

The null hypothesis is stated as (ℽ1 = ℽ2 = ℽ3 = ℽ4 = ℽ5). There is no cointegration among the
variables. The bound testing approach by Pesaran et al. (2001) is used comparing F-valued and
tabulated value. If the calculated F-value is above the bound there is long-run relationship; if
below no cointegration is confirmed. The result becomes inconclusive if the calculated value of
the F-statistics falls between the lower bound and upper bound. An error correction can be
included in the ARDL model to analyze the long-run relationship and is another way to confirm
cointegration:

ARDL model provides better results for small dataset and is more dynamic than traditional
techniques.

Data:

This paper analyzes as annual dataset covering the period 1972-2014. The data is collected
domestically from annual reports of State Bank of Pakistan, the Pakistan Statistical Bureau, and
the Pakistan Economic Survey. Internationally collected from International Financial Statistics
(IFS) issued by IMF and World Data Indicators (WDI) (CD ROM 2014), owned by World Bank.
The units of data are local currency (PKR), international currency (USD), and the percentages
that are unit-less. Logarithmic transformation is used expect for inflation rate. The dependent
variables in the model are LnM1 and LnM2. Explanatory variables are real income, discount
rate, real exchange rate and remittances.
 SECTION 5: EMPERICAL RESULTS
Summary:
In 1969 the Augmented Dickey fuller stationary test was conducted to confirm non availability of
stationary at second order I(2) for all the variables. The results reveal that LnM1, Ln1, π, and
LnRM are stationary at zero order with constant term in the model, while LnM2, LnY, and
LnRER are non-stationary at the level but at first difference with constant term. All the variables
are I(0) or I(1) and none is I(2). Results are reported in Table 2 below:

Table 2: Augmented Dickey Fuller (ADF) Test Statistics

Level First Difference
C C+T C C+T
LnM1 0.15 -3.57** -5.42*** -5.38***
LnM2 -0.35 -2.25 -5.05*** -4.99***
LnY -0.37 -1.17 -6.56*** -6.81***
LnI -3.12** -3.24* -4.54*** -4.44***
π -4.62*** -4.58*** -7.49*** -7.38***
LnRER -1.05 -0.45 -4.81*** -4.79***
LnRM -3.08** -3.79** -6.50*** -5.96***
Note: 1. *, **, and *** show the results’ significance at the level of 10%, 5%, and 1% respectively.

First step in ARDL is to select a lag order based on concurrent criterion in literature, as F-
statistics is sensitive to lag length. The Hannan-Quinn Criterion (HQC) is preferred for model
parsimonious specification clearly. So, this paper uses HQC, select optimal three lags, and reports
the results of models for M1 and M2 in Pakistan in Table 3 and 4. The computed values of F for
M1 and M2 are 27.348 and 31.891 respectively; they fall above bound 5% and 1% level
respectively. Test statistics are found significant for M1 and M2 at 1% and 5% respectively. The
result rejects the null hypothesis of no cointegration. Therefore, there is valid long-run
relationship between money aggregates and explanatory variables for Pakistan.

The results show that coefficient of remittances (LnRM) is positively related to M1 and M2 in
both short-run and long-run. M1 is significant at 1% in long-run and insignificant in short-run.
Whereas, it is positively related to M2 and remains statistically insignificant even at 5% level in
short-run and long-run. This suggests remittance in Pakistan is for consumption purpose. The
coefficient of real income (LnYt) is positive and statistically significant in long-run at 1% but
insignificant in short-run. Whereas, the coefficient of real income (LnYt) is also positive with M2
and significant at 1%, which explains that 1% is the real income results in an increase of five
basis points in real broad money demand. Income does not follow quantity theory of money
demand in long-run and is also insignificant in short-run.

The coefficient of discount rate (LnI) has a negative relation with both M1 and M2 and is not
statistically significant in both long and short run. The coefficient of inflation rate (π) has the
expected sign with both monetary aggregate and is statistically significant with M1 and M2 even
at 10%. The exchange rate (LnRER) is positively related to the M1 and M2 in long-run and short-
run. However, it is significant with both money aggregates at 1% in long-run, supporting wealth
hypothesis and is not significant in short-run.

Table 3: ARDL (1, 2, 0, 0, 3, 1) Model for Narrow Money Demand (M1) in Pakistan

Panel A: Short-Run Effects

Lag Order
0 1 2 3
1.082
ΔLnM1 (7.571)
0.009 -0.108
ΔlnY (1.345) (2.506)
-0.017
ΔlnI (0.262)
-0.010
Δπ (2.437)
-0.155 0.303 -0.759
ΔLnRER (0.655) (0.962) (3.614)
0.033
ΔLnRER (1.006)

Panel B: Long-Run Effects

Constant LnY LnI π LnRER LnRM ECM(-1)
10.260 0.038 -0.063 -0.036 2.292 0.443 -0.262
(6.899) (2.227) (0.253) (1.656) (10.335) (4.048) (2.975)

Panel C: Battery of Diagnostic Tests

Bounds Test RESET ARCH-Test LM
F Prob χ2 Prob χ2 Prob χ2 Prob
27.348(0.000) 0.197(0.846) 0.114(0.736) 4.540(0.103

Panel D: CUSUM and CUSUMQ Plots for M1

Note:

a. The values in parentheses are absolute t-values in Panel A and B and p-values in Panel C.

b. The upper bound critical value is 3.5 for the F-statistics in Pesaran et al. (2001, Table CI-Case III, p. 300) at 5%.
c. RESET is Ramsey's specification test. It follows χ2 distribution with one degree of freedom. 3.84 is the critical value at 5%.
d. ARCH test is for heteroskedasticity of residuals. It follows χ2 distribution with one degree of freedom. 3.84 is the
critical value at 5%. .
e. LM is the Lagrange multiplier test for serial correlation. It follows χ2 distribution with four degrees of freedom. 9.48 is
the critical value at the level of significance 5%.

Table 4: ARDL (2, 2, 3, 0, 3, 0) Model for Broad Money Demand (M2) in Pakistan
Panel A: Short-Run Effects
Lag Order

0 1 2 3
0.295
ΔLnM2 (1.999)
0.008 -0.014
ΔLnY (1.248) (2.518)
-0.082 -0.156 0.238
ΔLnI (0.743) (1.114) (2.286)
-0.007
Δπ (2.086)
0.194 0.265 -0.407
ΔLnRER (0.956) (1.073) (2.448)
0.026
ΔLnRM (0.864)

Panel-B: Long-Run Effects

Constant LnY LnI Π LnRER LnRM ECM(-1)

14.608 0.049 -0.299 -0.033 1.839 0.124 -0.213
(6.604) (1.872) (0.692) (1.449) (7.463) (1.004) (2.548)

Panel C: Battery of Diagnostic Tests

Bounds Test RESET ARCH-Test LM
F Prob χ2 Prob χ2 Prob χ2 Prob
31.891(0.000) 1.251(0.275) 0.023(0.879) 1.095(0.578)
Panel D: CUSUM and CUSUMQ Plots for M2
Note:
a. The values in parentheses are absolute t-values in Exhibit A and B and p-values in Exhibit C.

b. The upper bound critical value is 3.5 for the F-statistics in Pesaran et al. (2001, Table CI, Case III, p. 300) at 5%.
c. RESET is Ramsey's specification test. It follows χ2 distribution with one degree of freedom. 3.84 is the critical value at
5%.
d. ARCH test is for heteroskedasticity of residuals. It follows χ2 distribution with one degree of freedom. 3.84 is the
critical value at 5%.
e. LM is the Lagrange multiplier test for serial correlation. It follows χ2 distribution with four degrees of freedom. 9.48 is
the critical value at 5% level of significance.

Economic policies are concerned with long-run than short-run while estimating coefficients.
Cointegration through lagged explanatory variable in ARDL makes is meaningful for long-run
(Ericsson and McKinnon 2002). The Error Correction Model (ECM) coefficient shows how
quickly variables return from short-run to long-run equilibrium and it should be highly significant
with negative sign, which proves stable long-run relationship in model (Banerjee et al. 1998). The
value of ECM for M1 and M2 are reported in panel D of Table 3 and 4. The coefficient for ECM
(-1) for real narrow money demand (M1) is -0.262 and statistically significant. Which describes
every year long-run equilibrium is corrected at the speed of 26.2% of short-run equilibrium. Real
broad money demand (M2) is -0.213 and significant at 1%, so equilibrium is corrected at 21.3%
for M2 through explanatory variables.

Panel C and D of Table 3 and 4 shows all the diagnostic tests including Ramsey RESET test,
ARCH test, and Breusch-Godfrey LM test. Panel D of Table 3 and 4 provides evidences that M1
and M2 are stable money demand functions in Pakistan as they fall within 5% level of
significance in CUSUM and CUSUMQ plots.

SECTION 6: CONCLUSION

Summary:
This paper examines the dynamic relationship between series of monetary aggregates for the
period 1972-2014. M1 and M2 are dependent variables and real income, discount rate, real
exchange rate and remittances are explanatory variables. It uses ARDL approach to cointegration
to find out existence of long and short-run effects of remittances. M1 and M2 money demand
functions are stable in Pakistan. Remittance have long-run effects, with M1 quicker to adjust to
equilibrium (26.2%) than M2 (21.3%). The authorities should use narrow money supply rather
than broad money supply to control monetary policy in Pakistan as remittances impacts only M1
in long-run. The authorities should take steps to avoid illegal money transfers to improve and
enhance economic growth by providing easy and proper channel for money transfer.