Abstract: This paper explains large empirical literature and
demonstrates evidence on the relationship between foreign remittances and distribution
of income. Using time series data covering period 1971-2006 and three advanced
techniques [Johansen Trace-Test, DOLS
Method and ARDL Bounds Testing]
model for co-integration, we found robust evidence of long run relationship
among the said variables. In the case of Pakistan Pakistan
Keywords: Migration, Foreign Remittances, Income Inequality
JEL Classification: O11, O15, D31
1.
Introduction
Remittances inflows are a key and
stable source of foreign capital and revenue in developing economies because
there is no need to depend on external factors like foreign loans and aids etc.[1] In
literature, the relationship between foreign migrants’ remittances and income
inequality is scarce and incongruous. Some empirical evidences showed that
remittances worsens the income distribution (Milanovic, 1987; Stark et
al., 1988, Taylor, 1992; Taylor and Wyatt, 1996; Adams, 1989;
Rodriguez, 1998; Leones and Feldman, 1998; Adger, 1999; Gonzalez-Konig and
Wodon, 2005). In contrast, other argued that foreign
remittances decrease the income inequality (Barham and Stepehn, 1998; Ahlburg,
1996; Handa and King, 1997), but Adams (1992) found no significant impact of
remittances on rural income distribution in case of Pakistan .[2]
In the late 1980s, Lipton in his
pioneer work viewed that migrants’ remittances generate negative externalities
of various types and these externalities are responsible for an increase in
income inequality. He also described that remittances do not reimburse for said
undesirable impact because the migrants’ remittances are either very small or
go disproportionately to those better off. In contrast, Stark (1978) and Stark et al. (1986) in their models,
emphasized on equality-enhancing consideration associated with migration such
as risk-diversification, mitigation of credit constraints and various sharing
and filtering-down mechanisms pertaining to migrants remittances.[3]
They also concluded that income inequality does not mean a social welfare loss,
and may be consistent with situation that is preferred under both general
social welfare and Pereto criteria.[4]
Stark et al. (1988) distinguished
that the distributional impacts of migration are not the same for all types of
migration and concluded that foreign remittances contribute in raising income
inequality while internal remittances are having equalizing favorable impact on
village income distribution.[5]
Stark et al. (1986) found the relationship between foreign remittances
and income inequality is an inverse U-shaped (Kuznets Curve). It means that
initially un-equalizing effects of remittances on income distribution of a
village decreases at latter stages of the village’s migration history. Jones
(1998) distinguished three stages of migration: first is the “innovative stage” in which only the most
adventurous and better off people migrate, in that case international
remittances tend to increase income inequality; second is the “early
adopter stage”, when people from lower segments of population also start to
migrate and comparatively, remittances become more equalizing the income
distribution; and lastly, the “latter
adopter stage”, when due to accumulation over time of remittances in the
families with migrants, those families move farther apart from the families
without migrants and therefore remittances may be worsening the income
distribution. Stark et al.(1986)
found that, in a Mexican village with little migration to United-States,
remittances are contributing in inequality to rise while opposite is true for
village where migration to United-States has long migration history.[6]
But, Konig and Wodon (2005) extended the mythology developed by Yitzhaki (1982)
and Stark et al. (1986) and
accomplished that impact of remittances on inequality depends where those who
migrate are located in the distribution of income.
Lermann and Yitzhaki (1998)
concluded that remittances income distributes more to the Gini-Coefficient than
its share of total income indicates and it tends to increase inequality in the
community utilizing the Gini-Coefficient decomposition method. Carrington et al. (1996) found that international
remittances are likely to have different impact at different levels of a
villages’ migration history while net migration cost becomes endogenous to
migration process.[7]
Similarly, Quibria (1997) argued that foreign remittances do not affect all the
classes in the society symmetrically utilizing the two-factor, two-commodity,
two-class general equilibrium model that makes a distinction between traded and
non-traded goods.[8] The
division of losers and gainers depends on the volume of remittances, the
distribution of factors of endowments and type of emigration.[9]
Koechlin and Leon (2006) provided
the evidence in the support of existing theoretical literature that accounts
for the network effects that, the initial stages of migration history generates
inequality-worsening impact of remittances. Then as opportunity cost of
migration lowers, remittances sent to those households reduce income
inequality: a clear indication of inverted U-shaped relationship between
remittances and income inequality.[11] David
and Rapoport (2004) investigated the overall impact of migration on
distribution of income in Mexican rural communities. This impact is composed of
direct and indirect effects of remittances, and other potential spillover and
general equilibrium effects of migration. They found that migration ability
with their accords raise expected benefits and decease the costs of migration,
and generate a Kuznets-type relationship between remittances and income inequality.[12]
All these studies based on the Gini-Index decomposition with exogenous
distribution of domestic incomes may yield biased estimates of the inequality
impact of international remittances, with the direction of bias being
theoretically uncertain and depending on the initial distribution of wealth. Gini-Index decomposition only provides
information about point that decomposed.
Despite the fact, that a wide
rage of strand of economic research has investigated the effects of remittances
on variety of topics along-with employing variety of techniques with different
aspects. There is no formal study to analyze either remittances improve income
distribution or worsen in the case of Pakistan utilizing large time
series data. One may say that this small effort is advancement over previous
literature in a number of important aspects. First, it uses the data set that
is sufficiently large to enable robust conclusions to be drawn from econometric
results; specifying the sample utilized in this paper consists of annual data
covering the period 1971-2006. Second,
this paper includes other important determents, to investigate their impacts on
income inequality like government spending, human capital formation, financial
development, unemployment rate and local migration (urbanization) which are
ignored in all previous literature.[13] Third, this paper employs three advance
techniques (Johansson Trace-Test, DOLS, Dynamic Ordinary Least Squares and ARDL
Bounds Testing) for co-integration and robustness of long run relationships
among variables while Error Correction Method (ECM) for short run dynamics.[14]
Section 2 explains the model, data collection procedure and methodological
framework; Section 3 investigates and interprets empirical results. Finally,
Section 4 presents the conclusions and also provides some policy implications.
2. Modeling, Data and Methodology
According to assumptions of Barro
(1991) and Koechlin and Leon
(2006), the most important econometric classification to investigate the
significant relationship between income inequality and foreign remittances; we
utilize log-linear modeling specification as given below. Bowers and Pierce
(1975) suggest that Ehrlich’s (1975) findings with a log linear specification
are sensitive to functional form. However, Ehrlich (1977) and Layson (1983)
argue on theoretical and empirical grounds that the log linear form is superior
to the linear form. Both Cameron (1994) and Ehrlich (1996) suggest that a
log-linear form is more likely to find evidence of a deterrent effect than a
linear form. This makes our results more favorable to the deterrence
hypothesis. Following the above discussion empirical equation is being modeled
as:
(1)
(2)
The inequality-narrowing
hypothesis predicts 
<0, in contrast the inequality-widening hypothesis
predicts >0 and, inverted U-shaped hypothesis predicts if 
>0 and 
<0, if <0 and >0 U-shaped hypothesis predicts and CV are some control variables that effect income inequality.
<0, in contrast the inequality-widening hypothesis
predicts >0 and, inverted U-shaped hypothesis predicts if >0 and <0, if <0 and >0 U-shaped hypothesis predicts and CV are some control variables that effect income inequality.
Where Gini is measure of income
inequality proxied by Gini-coefficient while REM is a measure of international
remittances, expecting declining impact on income inequality as described in
literature but may be positive through its detrimental channels.[15]
Real GDP per capita (GDPPC) is assuming inverse impact through its development
process. HDI represents the human capital formation, which may be affecting
income inequality either positively or negatively. CPI indicates the inflationary situation in
the economy and assuming that, monetary instability hurts poor and middle class
relatively more than rich because latter class has better access to financial
instruments that allow them to hedge their exposure to inflation and expecting
positive effect of inflation on inequality[16];
while measure of government general consumption expenditures conjecturing
negative effect on inequality, but there can be opposite effect of government consumption
on inequality if rich households use their political powers to exploit the
poor.[17]
Urbanization is represented by URB which might decline the income inequality
through the creation of employment opportunities.[18]
Unemployment rate worsens distribution of income through its direct and
detrimental impacts, indicated by UMP. FD is measure of financial sector’s
development proxied by Credit to private as share of GDP may equalizes the
income distribution directly and indirectly through growth sound effects.[19]
Table 1: Descriptive Statistics and Correlation Matrix
|
Variables
|
GINI
|
GS
|
HDI
|
REM
|
CPI
|
GDPPC
|
|
Observations
|
36
|
36
|
36
|
36
|
36
|
36
|
|
Mean
|
3.599
|
26.108
|
-0.825
|
1.223
|
3.961
|
9.484
|
|
Median
|
3.612
|
26.336
|
-0.804
|
1.565
|
3.950
|
9.496
|
|
Maximum
|
3.758
|
27.000
|
-0.616
|
2.360
|
5.101
|
10.200
|
|
Minimum
|
3.413
|
25.150
|
-1.108
|
-2.338
|
1.950
|
8.950
|
|
Std. Dev.
|
0.106
|
0.549
|
0.151
|
1.008
|
0.890
|
0.340
|
|
Skew-ness
|
-0.252
|
-0.308
|
-0.271
|
-1.546
|
-0.471
|
0.309
|
|
Kurtosis
|
1.843
|
1.753
|
1.790
|
5.639
|
2.418
|
2.372
|
|
Jarque-Bera
|
2.389
|
2.901
|
2.637
|
24.806
|
1.843
|
1.166
|
|
Probability
|
0.302
|
0.234
|
0.267
|
0.0000
|
0.397
|
0.558
|
|
Sum
|
129.59
|
939.91
|
29.732
|
44.035
|
142.62
|
341.42
|
|
Sum Sq.Dev.
|
0.394
|
10.571
|
0.806
|
35.580
|
27.732
|
4.059
|
|
GINI
|
1
|
|
|
|
|
|
|
GS
|
0.9744
|
1
|
|
|
|
|
|
HDI
|
0.997
|
0.973
|
1
|
|
|
|
|
REM
|
0.348
|
0.340
|
0.328
|
1
|
|
|
|
CPI
|
0.983
|
0.937
|
0.985
|
0.396
|
1
|
|
|
GDPPC
|
0.925
|
0.912
|
0.912
|
0.163
|
0.854
|
1
|
Source:
Author’s own calculations.
Descriptive statistics in Table 1 reveals that government
spending, human capital formation, consumer prices and GDP
per capita are highly but remittances are weakly correlated with income inequality. Human capital formation, consumer prices
and economy size are associated with government spending positively but
correlation of remittances is weak as mentioned in descriptive table. The data
has been collected from IFS (International Financial Statistics, 2007), WDI
(World Development Indicators, 2007) and Economic Survey of Pakistan (Various
Issues) except income inequality.[20]
2.2 KPSS Unit Root
Test
Literature reveals that ADF and P-P test are having low
explaining power especially in small sample data set. Shift has been focused to
KPSS (1992)[21] to
investigate the order of integration for concerned actors in the model. KPSS (1992) test assumes null hypothesis is
stationary and vice versa. KPPS test states that any time series variable
like 
is the combination of
three mechanisms i.e. deterministic trend, random walk plus error term as given
below:
is the combination of
three mechanisms i.e. deterministic trend, random walk plus error term as given
below:
(3)
In
Equation (4), random walk is specified as following:
(4)
To
check out the stationarity of, is regressed on
constant and trend terms, residuals from equation are used finally i.e. 
for statistics of KPSS
test:
, is regressed on
constant and trend terms, residuals from equation are used finally i.e. for statistics of KPSS
test:
(5)
where
and
. In KPSS test 
, main reason is that KPSS test is highly sensitive with
change of truncation lags.
2.3 Johansen
Co-integration Technique
Engle and Granger (1987) discussed that, a set of economic
series is not stationary, there may have to exist some linear combination of
the variables that is stationary. Now, when all the variables are
non-stationary at their level but stationary in their 1st
difference, this allows proceeding further for the implementation of Johansen
co-integration technique (1991; 1995). Economically speaking, two variables
will be co-integrated if they have a long-term relationship between them. Thus,
co-integration of two series suggests that there is a long integration tests
and of course, the system approach developed by Johansen (1991,1995) can also
applied to a set of variables containing possibly a mixture of I(0) and I(1) (Pesaran
and Smith (1998) and Pesaran et al. (1997).
The general form of the vector error correction model is as follows:
This
can also be written in standard form as:
(6)
where
and 
where p is
represents total number of variables considered in the model. The matrix
captures the long run relationship between the p-variables. Now for the Johansson Test
we employ the Trace test, which is based on the evaluation of 
against the null
hypothesis of
, where r indicates
number of co-integrating vectors. The co-integration test provides an
analytical statistical framework for investigating the long run relationship
between economic variables in the model.
captures the long run relationship between the p-variables. Now for the Johansson Test
we employ the Trace test, which is based on the evaluation of against the null
hypothesis of, where r indicates
number of co-integrating vectors. The co-integration test provides an
analytical statistical framework for investigating the long run relationship
between economic variables in the model.
2.3 DOLS Procedure
for Long Run Relationship
Stock and Watson (1993) developed a model for the
investigation of long run relationships among dependent variable and
explanatory variables. This procedure involves regressing the dependent variable
on all explanatory variables in levels, leads and lags of the first difference of all I(1)
explanatory variables (Masih and Masih, 1996). This method is superior to a
number of other estimators as it can be applied to systems of variables with
different orders of Integration (Stock and Watson, 1993). The inclusion of
leads and lags of the differenced explanatory variables corrects for
simultaneity bias and small sample bias among the regressors (Stock and Watson,
1993). The specification of DOLS model is follows given below:
(7)
where Xt is Gini-Coefficient, Zt is
a vector of explanatory variables and D is lag operator.
2.4 ARDL Approach
for Co-Integration
The structure of the study has
suggested to employ autoregressive distributed lag (ARDL) bounds testing
approach developed by Pesaran et al.
(2001) to carry out co-integration analysis among income inequality and
international remittances with battery of other explanatory variables.
Statistical literature reveals that ARDL bounds approach has numerous
advantages than other conventional co-integrating methods. In ARDL technique,
there is no need of having same order of integration to find out co-integration
among macroeconomic variables. It is concluded that ARDL can be applied
irrespectively of whether the variable are integrated of order I(0) or integrated of order I(1)[22] or having mixed order of integration.
Fortunately, ARDL bounds testing overcomes all problems faced by traditional
techniques for co-integration in literature. It has better explaining power and
properties such small sample data. Moreover, a dynamic error correction model
(ECM) can be derived from ARDL through a simple linear transformation
(Banerrjee et al., 1993). The dynamic
ECM integrates the short-run dynamics with the long-run equilibrium without
losing long-run information.
Conditional error correction
version of the ARDL model for co-integration is given below:
(8)
where 
is the drift component
and 
is assumed to be white
noise error process. The ARDL approach estimate 
number of regression in order to obtain optimal lag length
for each variable, where ‘p’ is the
maximum number of lag to be used and “k” is the number of variable in
the equation (1). The optimal lag order of the first difference regression is
based on minimum value of Schwarz-Bayesian criteria (SBC) to ensure that there
is no serial correlation in the estimated residual[23]
of equation. Following Pesaran et al.
(2001), two different statistics are implemented to “bound test” for ensuring
of long-run relationship: an F-test
for the joint significant of the coefficients of the lagged levels in Eq. (8)
(with null hypothesis 
means no evidence of existence of long run relationship in
concerned model and vice versa. Two asymptotic critical value bounds provide a
test for co-integration when the independent variables are I(d) ( Where 0 ≤ d ≤ 1), a lower value assuming
the regressors are I(0), and an upper value assuming purely I(1) regressors. If
the F-statistics exceeds the upper critical value, we can conclude that there
prevails a long run relationship regardless of whether the underlying order of
integration of the variables is I(0)
or I(1). If the F-statistics falls below the lower
critical values one cannot reject the null hypothesis of no co-integration. If
the F-statistics exceeds the
upper bounds, one may reject the hypotheses of no long run relationship.
However, if the F-statistics
falls between these two bounds, inference would be inconclusive.[24]
If any variable is having I(2) integrating order, then ARDL bounds testing will
be collapsed.
is the drift component
and is assumed to be white
noise error process. The ARDL approach estimate number of regression in order to obtain optimal lag length
for each variable, where ‘p’ is the
maximum number of lag to be used and “k” is the number of variable in
the equation (1). The optimal lag order of the first difference regression is
based on minimum value of Schwarz-Bayesian criteria (SBC) to ensure that there
is no serial correlation in the estimated residual[23]
of equation. Following Pesaran et al.
(2001), two different statistics are implemented to “bound test” for ensuring
of long-run relationship: an F-test
for the joint significant of the coefficients of the lagged levels in Eq. (8)
(with null hypothesis means no evidence of existence of long run relationship in
concerned model and vice versa. Two asymptotic critical value bounds provide a
test for co-integration when the independent variables are I(d) ( Where 0 ≤ d ≤ 1), a lower value assuming
the regressors are I(0), and an upper value assuming purely I(1) regressors. If
the F-statistics exceeds the upper critical value, we can conclude that there
prevails a long run relationship regardless of whether the underlying order of
integration of the variables is I(0)
or I(1). If the F-statistics falls below the lower
critical values one cannot reject the null hypothesis of no co-integration. If
the F-statistics exceeds the
upper bounds, one may reject the hypotheses of no long run relationship.
However, if the F-statistics
falls between these two bounds, inference would be inconclusive.[24]
If any variable is having I(2) integrating order, then ARDL bounds testing will
be collapsed.
Then, the long-run connection is investigated through use of
selected ARDL model. If variables are co-integrated, the conditional long run
model can then be produced from the reduced form of Eq. (8), when the first
differenced variables jointly equal to zero, i.e. 
. Thus,
. Thus,
(9)
where 
, and 
are the random
errors. These long run coefficients are empirically estimated by the ARDL,
model in Equation (1) through OLS method. When
there is long relationship between variables, there also exists error
correction representation. Therefore, the error correction model is estimated
generally as in the following given reduced form:
, and are the random
errors. These long run coefficients are empirically estimated by the ARDL,
model in Equation (1) through OLS method. When
there is long relationship between variables, there also exists error
correction representation. Therefore, the error correction model is estimated
generally as in the following given reduced form: 
(10)
To ascertain the goodness of fit of the ECM model, the sensitivity analysis is
conducted. The diagnostic test examines the serial correlation, functional
form, normality and heteroskedasticity associated with the model. The stability
test is conducted by employing the cumulative sum of recursive residuals (CUSUM)
and the cumulative sum of squares of recursive residuals (CUSUMsq). Examining
the prediction error of the model is another way of ascertaining the
reliability of the ARDL model. If the error or the difference between the real
observation and the forecast is infinitesimal, then the model can be regarded
as best fitting.
3. Interpretation
Design
Prior step to investigate the
order of integration of individual series; ADF (Augmented Dickey-Fuller) Phillips-Perron and KPSS tests have been utilized in
the present study. Results of both tests are reported in Table 2; all variables
are stationary at their 1st difference form through employing ADF,
P-P and KPSS tests.[25] Results
of latter test also prove that all the series are trend stationary.
Table 2: Unit-Root
Estimation
|
Variables
|
ADF Test at 1st Difference
|
Phillips-Perron at 1st Difference
|
KPSS at 1st Difference
|
|||||
|
Intercept and trend
|
Prob.
Values
|
Lags
|
Intercept and trend
|
Prob.
Values
|
Bandwidth
|
Intercept and trend
|
Bandwidth
|
|
|
LGINI
|
-3.945
|
0.021
|
2
|
-3.477
|
0.0580
|
10
|
0.1432**
|
9
|
|
LREM
|
-9.048
|
0.000
|
0
|
-7.765
|
0.0000
|
4
|
0.1087***
|
3
|
|
LGS
|
-3.564
|
0.048
|
1
|
-6.626
|
0.000
|
3
|
0.0996***
|
1
|
|
LGDPC
|
-5.420
|
0.000
|
0
|
-5.240
|
0.0009
|
3
|
0.1064***
|
4
|
|
LHDI
|
-3.244
|
0.094
|
3
|
-16.145
|
0.000
|
4
|
0.0852***
|
5
|
|
LCPI
|
-5.104
|
0.001
|
0
|
-5.281
|
0.0007
|
2
|
0.1236**
|
1
|
|
Lag Length Selection of VAR Model
|
||||||||
|
Lags
|
AIC
|
SBC
|
Maximum Likelihood
|
|||||
|
1
|
-22.95081
|
-21.08439
|
443.6392
|
|||||
|
2
|
-26.87495
|
-23.37330
|
534.8742
|
|||||
Note:
** and *** are MacKinnon (1996) one-sided p-values at 5 percent and 10 percent
level of confidence, respectively.
This tends to support for the
employment of the most advanced techniques for the investigation of long run
relationships among the macroeconomic variables. Lag length of VAR model is
selected 2 on the basis of (AIC) and (SBC). After having information about
order of integration of all series and before
going to ARDL, Johansen Co-integration test is applied and results are
shown in Table 3. Starting with the null hypothesis of no Co-integration (
) among the variables, the trace-test statistics is 236.1014,
which is above 1 percent and 5 percent critical values as shown by the
Prob-value. Hence it rejects null hypothesis
in the favor of
general alternative
. Likewise null hypothesis of no Co-integration () among the variables, the trace-test statistics is 142.0879,
which is above 1 percent and 5 percent critical values as shown by the
Prob-value. So, it also rejects null hypothesis
in the favor of
general alternative
. Consequently, one may conclude that there are six
Co-integrating relationships among income inequality, remittances as share of
GDP, human capital formation, GDP per capita, government spending and consumer
prices.
) among the variables, the trace-test statistics is 236.1014,
which is above 1 percent and 5 percent critical values as shown by the
Prob-value. Hence it rejects null hypothesis in the favor of
general alternative. Likewise null hypothesis of no Co-integration () among the variables, the trace-test statistics is 142.0879,
which is above 1 percent and 5 percent critical values as shown by the
Prob-value. So, it also rejects null hypothesis in the favor of
general alternative. Consequently, one may conclude that there are six
Co-integrating relationships among income inequality, remittances as share of
GDP, human capital formation, GDP per capita, government spending and consumer
prices.
Table 3: Johansen’s Multiple Co-integration Test Results
|
Hypotheses
|
Trace-Test
|
0.05 percent
critical value
|
Prob.**
|
|
236.1014
|
117.7082
|
0.000
|
|
|
R £ 1
|
142.0879
|
88.80380
|
0.000
|
|
R £ 2
|
91.02466
|
63.87610
|
0.000
|
|
R £ 3
|
46.72793
|
42.91525
|
0.019
|
|
R £ 4
|
27.35478
|
25.87211
|
0.033
|
|
R £ 5
|
12.66226
|
12.51798
|
0.047
|
Note:
**MacKinnon-Haug-Michelis (1999) p-values
Therefore, over annual data from
1971 to 2006 appears to support the proposition that in Pakistan ; there does exist a stable
long-run relationship among above-mentioned variables in the model. The results
of DOLS (Dynamic Ordinary Least
Square ) are reported in Appendix-A; only
significant regressors are shown in the Table 7. Adjusted-R2 value of 0.999692 is an
indication of good-fit for the dataset, the F-statistics-10714.32 (Prob-value = 0.00) is statistically
significant at 1 percent level of significance, concluding that the explanatory
variables are jointly significant in influencing distribution of income for a
small developing economy like Pakistan. The results of DOLS regression show
that in long run international remittances, human capital formation and
increase in per capita income along with inflation and government spending are
having positive impact on income inequality, however, leads of inflation and
human capital formation. Finally, differenced terms of government spending,
remittances and GDP per capita
improve distribution of income (Appendix-A).
Finally, ARDL for long run relationships is also
applied and results are expressed in Table 4. The main assumption of ARDL is
that the included variables in model are having co-integrated order I(0) or
I(1) or mutually. This tends to support for the implementation of bounds
testing, which is tree step procedure, in the first step we selected lag order
on the basis of SBC because computation of F-statistics for Co-integration is
very much sensitive with lag length, so lag order 2 is selected on lowest value
of SBC[26].
The total number of regressions estimated following the ARDL method in the
Equation 2 is 
729.
729.
Table 4: F-statistic of Co-integration
Relationship
|
Test-statistic
|
Calculated Value
|
Lag-order
|
Significance level
|
Bound Critical Values(restricted intercept and no trend)
|
|
|
F-statistic
|
9.17
|
2
|
1 percent
5 percent
10 percent
|
I(0)
|
I(1)
|
|
5.25
3.79
3.17
|
6.36
4.85
4.14
|
||||
|
Short Run
Diagnostic Tests
Serial Correlation
LM Test =0.195(0.824)
ARCH Test = 0.196(0.664)
White
Heteroskedasticity Test = 1.721(0.138)
Normality J-B Value
= 1.998(0.368)
Ramsey RESET Test =
10.255(0.0038)
|
|||||
Source: Author’s own
calculations.
Given the existence of a long run
relationship, in the next we used the ARDL co-integration method to estimate
the parameters of log-linear Equation (1) with a maximum order of lag set to 2
minimize the loss of degrees of freedom. The results of bounds testing approach
for long run relationship represent that; calculated F-statistic is 9.17, which is higher than the upper bound (6.36)
and lower bound (5.25) at 1 percent level of significance, implying that the
null hypothesis of no co-integration cannot be accepted and, indicating that
there is indeed a co-integration relationship among the variables as
long-established by J-J maximum likelihood approach..
Long-run coefficients of the
variables under investigation are shown in the Table 5. Model 1 reveals that
increase in remittances seems to worsen the income distribution significantly. The
reason for this is that increase in cost of migration pushes the poor segments
of population downward and rich are obtaining the fruits of migration and
getting richer and richer, while rapid increase in per capita income also
pushes the income inequality upward because growth effects did not trickle down
to lowest quintile of population (bottom 20 percent) according upper-echelon
phenomenon in the country. Low down income inequality is coupled with low rate
of human capital formation. Consumer prices rise makes income distribution more
skewed through its detrimental direct and indirect channels[27].
Government spending affects the income inequality positively and significantly
because rich households use their political powers to exploit the poor, which
worsens the income distribution swiftly. There is also evidence of an increase
in income inequality as international migration history continues, means,
relationship between remittances and income inequality is Kuznets inverted U-shaped.
This shows another proof of linear association between foreign remittances and
income distribution, while internal migration (rural-urban) makes income
distribution more equal through obtaining fruits from employment opportunities
in urban areas.[28]
Combined impact of remittances and human capital formation on income inequality
is negative which indicates that improvements in human capital formation
through remittances equate the distribution of income.
Table 5: Long Run OLS (Ordinary Least Squares) Results
|
Dependent Variable: LGINI
|
|||||
|
Variable
|
Model 1
|
Model 2
|
Model 3
|
Model 4
|
Model 5
|
|
Constant
|
2.741***
(7.77)
|
3.164***
(9.17)
|
2.729***
(6.90)
|
2.717***
(6.98)
|
2.698***
(6.81)
|
|
LREM
|
0.003**
(2.41)
|
0.002**
(2.34)
|
-0.0179*
(-1.70)
|
-0.0157
(-1.47)
|
-0.0374
(-0.93)
|
|
LREM2
|
…….
|
…….
|
0.0013*
(1.90)
|
0.0008
(1.03)
|
0.0007
(1.01)
|
|
LGDPC
|
0.052**
(6.87)
|
0.053***
(7.75)
|
0.059***
(6.93)
|
0.055***
(6.05)
|
0.055***
(6.00)
|
|
LHDI
|
0.325***
(3.71)
|
0.327***
(4.19)
|
0.275***
(3.52)
|
0.272***
(3.49)
|
0.274***
(3.47)
|
|
LCPI
|
0.033*** (3.41)
|
0.032***
(3.79)
|
0.038***
(4.38)
|
0.034***
(3.69)
|
0.033***
(3.49)
|
|
LGS
|
0.019** (2.104)
|
0.006
(0.62)
|
0.016*
(1.65)
|
0.018*
(1.81)
|
0.020*
(1.87)
|
|
LURB
|
…….
|
-0.058***
(-2.96)
|
-0.047**
(-2.42)
|
-0.046**
(-2.45)
|
-0.044**
(-2.20)
|
|
LEMP
|
…….
|
…….
|
…….
|
0.009*
(1.71)
|
0.010*
(1.76)
|
|
LFD
|
…….
|
…….
|
…….
|
0.003
(0.26)
|
-0.003
(-0.20)
|
|
LREM*LHDI
|
…….
|
…….
|
-0.019*
(-1.82)
|
-0.017
(-1.57)
|
-0.021
(-1.58)
|
|
LREM*LFD
|
…….
|
…….
|
…….
|
…….
|
0.005
(0.56)
|
 = 0.998294
Durban-Watson = 1.74
F-stat = 3510.42
|
= 0.998690
Durban-Watson = 1.71
F-stat = 3685.73
|
= 0.998889
Durban-Watson = 1.76
F-stat = 3034.93
|
= 0.999005
Durban-Watson = 1.97
F-stat = 2510.68
|
= 0.999018
Durban-Watson = 1.94
F-stat = 2220.263
|
|
Note: ***, ** and * represent 1 percent, 5
percent and 10 percent level of significance, respectively.
Inequality is being positively
affected by an increase in unemployment level, while access to credit
(financial development) seems to worsen the income distribution insignificantly
due to low quality and efficiency of financial institutions.[29]
Combined impact of remittances and financial development is positive but
insignificant describing that remittances are increasing inequality because
these are not properly channelizing through banking sector due to low incentive
to migrants. Net effect of financial sector’s development improves distribution
of income insignificantly.
Having found a long run
relationship, we applied the ARDL method to investigate the long run but for short
run estimation, we followed the Equation 1 and utilized the below given model
for short run dynamics.
(11)
Table 6: Short-Run Dynamic Model (1, 1, 2, 2, 1, 1)
|
Dependent Variable: DLIGNI
|
||
|
Variable
|
Coefficient
|
Prob-value
|
|
Constant
|
0.0041
|
0.1056
|
|
DLREM
|
-0.0008
|
0.6806
|
|
DLGDPC
|
0.0550
|
0.0030
|
|
DLGDPC(-1)
|
-0.0087
|
0.5851
|
|
DLCPI
|
0.0347
|
0.0458
|
|
DLCPI(-1)
|
-0.0023
|
0.8811
|
|
DLHDI
|
0.0602
|
0.4485
|
|
DLGS
|
0.0143
|
0.1040
|
|
CR(-1)
|
-0.6043
|
0.0218
|
|
= 0.542
-Adjusted =0.395
Durban-Watson =
1.74 F-stat =
3.69(0.005)
|
||
Source: Author’s own
calculations.
Following the above ECM equation,
results are reported in Table 6. The results indicate that, remittances improve
the income distribution in short run insignificantly but increase in the volume
of per capita income worsens the income distribution. Lag impact of GDP per
capita declines the income inequality but insignificantly, and this effect
converges to its future affect. Consumer prices rise and enhancement in
government spending increase income inequality. Finally, human capital
improvement pushes income distribution to skew insignificantly.
The error correction term CRt-1, which measures
the speed of adjustment to restore equilibrium in the dynamics model, appear
with negative sign and is statistically significant at 5 percent level,
ensuring that long run equilibrium can be attained. Bannerjee and Newman (1993)
holds that a highly significant error correction term is a further proof of the
existence of stable long run relationship.[30]
The coefficient of CR(-1) is equal to (-0.604) for short run model respectively
and imply that deviation from the long-term income inequality is corrected by
(0.604) percent over the each year. The lag length of short run model is
selected on the basis of AIC and SBC. The short run dynamics of the income
inequality based on ARDL (1, 1, 2, 2, 1, 1) model for Pakistan is reported in Table 4.
The diagnostic statistics indicate that the equation is mis-specified. The
model fulfilled the conditions of non-serial correlation, no autoregressive
conditional heteroskedasticity and normality of disturbance term. There is no
heteroskedasticity in the short model.
Finally, we examine the stability of the long run
parameters together with the short run movements for the equation. To
this end, we rely on cumulative sum (CUSUM) and cumulative sum squares
(CUSUMSQ) tests proposed by Borensztein et al. (1995; 1998). The same
procedure has been used by Pesaran and Pesaran (1997) to test the stability of
the long run coefficients. The tests applied to the residuals of the ECM
model (Table 2) along with the critical bounds are graphed in Figures 1. As
can be seen in the figures, the plot of CUSUM stay within the critical 5
percent bounds for all equations but and CUSUMsq statistics exceeds the
critical boundaries due to misspecification of short run model.
4. Conclusions and Policy Implications
Present paper explains large
empirical literature and explores empirical evidence on the relationship
between international remittances and distribution of income. Using large time
series data covering period 1971-2006 and three advanced techniques Johansen
Trace-test, DOLS and ARDL model for co-integration, we found robust evidence of
long run relationship among the said variables. In the case of Pakistan ,
the relationship between international remittances and income inequality is
U-shaped indicating that inequality initially decreases and then increases as migration
history continues.
Our empirical findings support
the evidence that international remittances increases inequality in the long
run while local migration improves income distribution. Increase in per capita
income pushes the income inequality upwards means inequality is encouraged by
an increase in per capita income. Improvements in human capital formation
worsen the situation of income distribution. Inflation lowers the purchasing
power of poor individuals in the economy and raises inequality through its
detrimental channels like unemployment. Increase in government consumption
worsens the income inequality, which indicates that rich households use their
political powers to exploit the poor, which raises the income inequality
swiftly.
The main policy implication of
this study is that while encouraging migration, it may increase the economic
development in less developed areas or region of the country, which will
improve the income distribution and alleviate poverty. A higher development in
financial institutions and markets will allow an easier and cheaper
transmission of migrants’ remittances; lower transaction cost will also allow
poorer households to receive remittances at earlier stages of migration,
compared to how long they would have to wait if financial markets are less
developed. Therefore, financial sector should regulate its institutions and
markets to launch such policies to provide some particular incentives for
remittance senders through proper formal banking system. Government should adopt
such policies to enhance the volume of skilled labor through technical
education at rural areas. More opportunities could be enhanced through
regulation of recruitment process and safe transport facilities through
supporting working rights for poor class. There will be a need to take
initiatives to promote transport and safe mechanism for migrants to send small
sums of money, and to create a more attractive investment climate in the
country particularly in rural areas to alleviate poverty from its roots. More
migrants’ remittances from abroad mean more national saving which is
pre-requisites for development process and economic growth.
References
Adams, R. 1989. “Workers
remittances and inequality in rural Egypt ” Economic Development and Cultural Change 38 (1): 45-71.
Adams, R. 1992. “The impact of
migration and remittances on inequality in rural Pakistan ”
Pakistan
Adger, W. N. 1999. “Exploring
income inequality in rural, coastal Viet Nam ” Journal of Development Studies 35(5): 96–119.
Ahlburg, D. A. 1996. “Remittances
and the income distribution in Tonga ”
Population Research and Policy Review
15(4): 391–400.
Bahmani-Oskooee,
M. and R. Ng 2002. “Long run demand for
money in Hong Kong : an application of the ARDL model” International
journal of Business and Economics 1(2): 147-155.
Banerjee, A. and A. Newman 1993.
"Occupational choice and the process of development." Journal of Political Economy 101(21):
274-298.
Barham, B. and B. Stephen 1998.
“Migration, remittances, and inequality: estimating the net effects of migration
on income distribution” Journal of
Development Economics 55(2): 307–31.
Barro, R. 1991. “Economic
growth in a cross-section of countries” Quarterly
Journal of Economics 106(2): 407-444.
Borensztein, E., J. Gregorio and J. W. Lee
1995. “How does foreign direct investment affect economic growth?” NBER working paper No. 5057.
Borensztein, E., J. Gregorio and J. W. Lee
1998. “How does foreign investment affect growth?” Journal of International Economics 45(1): 115-135.
Bowers, W. and G. Pierce 1975.
“The illusion of deterrence in isaac ehrlich’s work on the deterrent effect of
capital punishment” Yale Law Journal 85(2): 187-208.
Carrington,
W. J., E. Detragiache and T. Vishwanath 1996.
“Migration
with endogenous moving costs” American Economic Review
86(4): 909-930.
Cameron, S. 1994. “A review of
the econometric evidence on the effects of capital punishment” Journal
of Socio-Economics 23(2):
197-214.
David, M. and H. Rapoport 2004
“Network effects and dynamics of migration and inequality: theory and evidence
from Mexico ”
BREAD Working Paper.
Ehrlich, I.1975. “The deterrent
effect of capital punishment – a question of life and death” American
Economic Review 65(3): 397-417.
Ehrlich, I.
1977. “The deterrent effect of capital punishment reply” American Economic
Review 67(3): 452-58.
Ehrlich, I.
1996. “Crime, punishment and the market for offences” Journal of Economic
Perspective 10(1): 43-67.
Engle, R. F. and C. W. J.
Granger 1987. “Co-integration and error correction representation: estimation
and testing” Econometrica 55(2): 251–276.
Gonzalez-Konig,
G. and Q. Wodon 2005. “Remittances and inequality” Working Paper Series
EC200506, World Bank.
Government
of Pakistan , Economic Survey of Pakistan Islamabad .
Handa, S. and D. King 1997.
“Structural adjustment policies, income distribution and poverty: a review of
the Jamaican experience” World
Development 25(6): 915–30.
Haroon, J. 2005. “Does
inequality matter for poverty reduction” SPDC’s Research Report 58.
International
Monetary Fund, 2008. International Financial Statistics CD-ROM-2008. Washington , D.C.
United States
Johansen, S. 1991. “Estimation
and hypothesis testing of co-integrating vectors in Gaussian vector
autoregressive models” Econometrica 59(6): 1551–1580.
Johansen, S. 1995. Likelihood-Based
Inference in Cointegrated Vector Autoregressive Models. Oxford University Press: Oxford .
Jones, R. C. 1998. “The renewed
role of remittances in the New World Order” Economic
Geography 74(1): 1-7.
Koechlin, V.
and G. Leon 2006. “International
remittances and income inequality: an empirical investigation international
remittances and income inequality” Journal of Policy Reform 10(2): 228-
241.
Knowles, J.C.
and R. Anker 1981. “An analysis of income transfers in developing country” Journal of Development Economics 8(3):
205-226.
Layson, S.
1983. “Homicide and deterrence: another view of the canadian time series
evidence” Canadian Journal of Economics 16(3): 52-73.
Lerman, R. and S. Yitzhaki 1985. “Income inequality effects by
income source: a new approach and application to the U.S. ” Review
of Economics and Statistics 67 (2): 151-156.
Lucas, R. E. B. 1987.
"Emigration to south Africa 's
mines" American Economic Review
77(3): 313-30.
Masih, A. M. M. and R. Masih 1996.
“Energy consumption, real income and temporal causality” Energy Economics 18(3): 165-183.
Milanovic, B. 1987. “Remittances
and income distribution” Journal of
Economic Studies 14(5): 24-37.
Oberai, A. S. and H. K. M. Singh 1980.
“Migration, remittances and rural development” International Labor Review 119(2): 229-241.
Peseran, M. H. and B. Peseran 1997. Working with Microfit 4.0: Interactive
Econometric Analysis. Oxford University Press: Oxford .
Pesaran, M. H., Y. Shin and R. J. Smith
2001. “Bounds testing approaches to the analysis of level relationships” Journal of Applied Econometrics 16(3): 289–326.
Quibria, M. G. 1997.
“International migration, remittances and income distribution in the
source country: a synthesis” Bulletin of
Economic Research 49
(1): 29-46.
Rodriguez, E. R. 1998. “International
migration and income distribution in the Philippines ” Economic Development and Cultural Change 46(2): 329–350.
Stark, O. 1978.
“Economic-demographic interaction in the course of agricultural development:
the case of rural-to-urban migration, Rome ”
Food and Agricultural Organization of the
United Nations.
Stark,
O. and D. Levhari 1982. “On migration and risk in LDCs, economic development
and cultural change” Journal of
Development Economics 31(1): 191–196.
Stark, O., J.
E. Taylor and S. Yitzhak 1986.
“Remittances and inequality” Economic
Journal 96(3): 722-740.
Stark, O., J.
E. Taylor and S. Yitzhak 1988. “Migration, remittances and inequality: a
sensitivity analysis using the extended Gini index” Journal of Development Economics 28(3): 309–22.
Stock, J. and M. Watson 1993.
“A simple estimator of co-integrating vectors in higher order integrated
systems” Econometrica 61(4): 783-820.
Taylor, J. E. and T. J. Wyatt
1996. “The shadow value of migrant remittances, income and inequality in a
household-farm economy” Journal of
Development Studies 32(6): 899-912.
Taylor, J. E. 1992.
"Remittances and inequality reconsidered: direct, indirect and
intertemporal effects" Journal of
Policy Modeling 14(2): 187-208.
World
Bank, 2008. World Development Indicators CD-ROM-2008. Washington , D.C. United States
Yitzhaki,
S. 1982. “On an extension of the Gini inequality index” International
Economic Review 24(3): 617-628.
Appendix-A
Table 7: Estimated
DOLS MODEL for Income Inequality
|
Dependent variable =LGINI
|
|||
|
Variables
|
Coefficient
|
T-Statistic
|
Prob-value
|
|
Constant
|
2.356451
|
11.97780
|
0.0000
|
|
LREM
|
0.004635
|
8.267295
|
0.0000
|
|
LHDI
|
0.192813
|
3.645794
|
0.0013
|
|
LCPI
|
0.052890
|
8.083849
|
0.0000
|
|
LGS
|
0.022064
|
4.766731
|
0.0001
|
|
LGDPC
|
0.064113
|
14.66855
|
0.0000
|
|
DLGDP
|
-0.014576
|
-2.221755
|
0.0364
|
|
DLCPI (+1)
|
0.023289
|
2.256998
|
0.0338
|
|
|
-0.011272
|
-2.594354
|
0.0162
|
|
DLHDI (+1)
|
0.194594
|
5.252829
|
0.0000
|
|
DLREM
|
-0.005885
|
-5.898395
|
0.0000
|
|
R2 = 0.999785 Adjusted R2
= 0.999692
S.E. of regression =
0.0018 AIC = -9.576276
Durbin-Watson = 1.792835 F-statistic = 10714.32
|
|||
[1] Remittance income refers to regular cash payments
received from household members working outside the community for periods of 6
months or more (Leon and Feldman, 1998).
[2] Also Portes and Rumbaut (1990) and Lipton (1980)
[3] However, Stark and Yitzhaki, (1982) proposed a social
welfare measure with desirable properties for assessing the effect of changes
in income inequality and inequality upon social welfare.
[4] Lack of techniques fro properly assessing the
contribution of out-migration to village income distribution, use of analytical
frameworks which preclude unambiguous welfare judgments about changes in income
distribution and lack of systematic empirical studies focusing on the
appropriate income-earning entities are major reasons for opposing views
regarding migration and the distribution of income in rural areas (Stark et al., 1986 and Stark et al., 1988)
[5] Remittances from internal migrants embody large
returns to schooling component, and education is highly associated with household
income in villages (Stark et al.,, 1986).
[6] None of both these authors provide a formal model of
the decision to migrate; they argue that impact of remittances on inequality
depends on the stage of migration in the home country or location.
[7] Munshi (2003) explained that individuals with large
networks are more likely to be employed and to hold higher paying jobs upon
arrival in the U.S.A, which increases income inequality at their home place.
[8] Espinosa and Massey (1997) argued that social
networks play a crucial role in mitigating the hazards of crossing the borders,
with friends and relatives with previous migrant experience often accompanying
new migrants across the border, showing them preferred routs and techniques of
clandestine entry.
[9] The model in this study is static one that ignores
number of dynamic economic consideration, especially those relating to saving
and investment. It does consider the different uses of remittances.
[10] Urban families who have members who migrate are
likely to occupy a lower position in the distribution of income within urban
areas than their rural counterparts. That is, with similar level of income,
migrants’ families will tend to be comparatively richer in rural areas (when
compared to other rural households) and poorer in urban areas (when compared
other urban households (Konig and Wodon, 2005, pp. 3).
[11] They used cross-country data of 78 economies
utilizing ordinary least square, instrumental variables and panel data methods.
[12] Indeed, migration appears to increase inequality at
low levels of society migration prevalence and reduces inequality at higher
levels (David and Rapoport, 2004).
[13] In many developing economies migrants’ who send
remittances to their families have vary little access to financial
intermediaries: a small percentage have bank accounts, saving accounts or
access to credits.
[14] This paper KPSS (Kwiatkowski, Philips, Schmidt, Shin,
1992) have been utilized to find the order of integration of variables in
model.
[15] The estimates of Gini-Coefficient (proxy for Income
Distribution) are available in the Economic Survey up to 1996-1997. After this
a simple interpolation technique is applied to take the decline or growth in
trend between two points in time and fill the data gapes between successive
observations. However, a slightly more sophisticated method is applied to
generate an interpolated series for inequality (Jamal, 2005).
[16] See for example, Easterly and Fisher (2001)
[17] Government current consumption expenditures for the
purchase of goods and services including compensation of employees. It also
includes most expenditure on national defense and security, but excludes
government military expenditures that are part of government capital formation.
[18] Migration of population to cities as share of total
population
[19] Private Credit captures the amount of credit
channeled from savers, through financial intermediaries, to private firms. Private
Credit is a comparatively comprehensive measure of credit issuing
intermediaries since it also includes the credits of financial intermediaries
that are not considered deposit money banks.
[20] see Jamal Haroon (2005)
[21] Theoretical form of KPSS test is based on
Bahmani-Oskooee 2002, pp: 2497.
[22] In this study Augmented Dickey-Fuller (ADF) and KPSS
(Kwiatkowski, Philips, Schmidt, Shin, 1992) tests were applied.
[23] SBC is known as selecting the smallest lag length to
specify a parsimonious model. The mean prediction error of AIC based model is
0.0005 while that of SBC based model is 0.0063 (Min B. Shrestha, 2003).
[24] Moreover, when the order of integration of the
variable is known and if all the variables are I(1), the decision is made on the basis of upper bound.
Similarly, if all the variables are I(0),
then the decision is made on the basis of lower bound.
[25] LCPI is also stationary at level with trend and
constant term
[26] At lower value of SBC, value of AIC is also low as
shown in Table 2.
[27] For details see Easterly (2001)
[28] See Stark, Taylor and Yitzhaki, (1986)
[29] In financial development there is an interaction
between remittances and financial institutions. First, providing financial
intermediaries through remittances increase benefits to senders and recipients
because it improves opportunities to save, borrow, buy other financial services
like insurance, invest, and helps financial institutions to mobiles in local
part of society where money is located. Due to international migration,
increases in households’ assets have national affects on growth and development
in the economy. Thus national savings and investment ratio can improve growth
rate when foreign saving are allocated to productive projects to strengthen
productive base of local economy in the development process.
[30] Indeed, he has argued that testing the significance of
CEt-1, which is supposed to carry a negative coefficient, is
relatively more efficient way of establishing Co-integration.
