You are on page 1of 18

Int Rev Econ (2014) 61:399–415

DOI 10.1007/s12232-014-0199-3

RESEARCH ARTICLE

Exploring the nexus between remittances and economic


growth: a study of Bangladesh

Ronald R. Kumar • Peter J. Stauvermann

Received: 24 August 2013 / Accepted: 9 March 2014 / Published online: 22 March 2014
Ó Springer-Verlag Berlin Heidelberg 2014

Abstract In this article, we explore the much debated nexus between remittances
and economic growth in Bangladesh. Drawing on an annual data from 1979 to 2012
and using the augmented Solow framework with the autoregressive distributed lag
bounds procedure, we examine the cointegration relationship, the short-run and
long-run effects and the causality nexus between remittances per worker, capital per
worker and the output per worker. The results show that remittances have a mixed
effect in the short-run, however, a momentous positive effect in the long-run
(0.11 %), on the output per worker. From the Granger causality assessment, we find
inter alia, a bidirectional causality between remittances and output (in per worker
terms) and a unidirectional causation from capital to remittances (in per worker
terms). Our results therefore support remittance led growth hypothesis in
Bangladesh.

Keywords Remittances  ARDL approach  Granger causality  Bangladesh

JEL Classifications C22  F24  F43  O53

1 Introduction

Bangladesh, according to the United Nation (UN) classification, is a least developed


country (LDC). The average per capita income for the economy is around US$588

R. R. Kumar (&)
School of Accounting and Finance, The University of the South Pacific, Laucala Campus, Suva, Fiji
e-mail: kumar_RN@usp.ac.fj; ronaldkmr15@gmail.com

P. J. Stauvermann
Department of Economics, Changwon National University, Changwon, South Korea
e-mail: pstauvermann@t-online.de

123
400 R. R. Kumar, P. J. Stauvermann

(constant 2000). Nevertheless, the country receives large amount of remittances.1


The top ten destination countries for Bangladeshis include India, Saudi Arabia, the
United Kingdom, Kuwait, Oman, the United States, Malaysia, the United Arab
Emirates, Italy, and Jordan.
Although, there are lot of studies done to explore the role of remittances on
economic growth in Bangladesh, the contribution of remittances to output, and the
direction of causality remains controversial. Bangladesh has experienced an
exceptional growth in remittances on the back of growing output per worker.
However, it is not clear whether remittances have any significant effect on the
growth rate of output per worker, since the results presented in the earlier studies
(Siddique et al. 2012; Paul et al. 2011; Ahmed 2010) are mixed. Therefore, in this
article, we present an alternative method and model (Rao 2007; Kumar 2013c;
Kumar and Kumar 2013) to examine the nexus between remittances and economic
growth. While our results in few respects coincide with Siddique et al. (2012), Paul
et al. (2011), and Ahmed (2010) who looked at Bangladesh in the recent past, it
significantly deviates from them in many other ways. Henceforth, we would like of
highlight them in the rest of the article.
Briefly, we examine the plausible effects of remittances vis-à-vis per worker
income in Bangladesh over the period 1979–2012. In what follows, we establish the
cointegration relationship between remittances and output per worker, estimate the
short-run and long-run effects of remittances, and explore the direction of causality.
We use the augmented Solow framework (Solow 1956) and apply the ARDL
bounds approach (Pesaran et al. 2001) to examine cointegration and short-run and
long-run effects. In addition, we utilize the Toda and Yamamoto (1995) approach to
examine the causality nexus. The rest of the article is organized as follows. In Sect.
2, we provide a brief literature survey to account for the huge literature on this topic.
Section 3 provides the econometric modeling, estimation techniques, and results. In
Sect. 4, we conclude our results.

2 A brief literature survey

The studies focusing on the impact of remittances on economic activities are


growing. Personal remittances (formerly workers’ remittances) received are defined
as a composition of personal transfers and compensation of employees. The
personal transfers include current transfers in cash or in kind received by resident
households from non-resident households. The compensation of employees includes
income of border, seasonal, and other short-term workers who are employed in a
country where they are not resident, and of residents employed by non-resident
entities (World Bank 2013). It is argued that remittances have a welfare-enhancing
effect, particularly when they support consumption, capital investment, education
and human development, entrepreneurship, and poverty reduction efforts (Ratha
2007; Buch and Kuckulenz 2010; Rao and Hassan 2012). However, high remittance

1
Note that the World Bank has renamed the Workers’ Remittances to Personal Remittances since 2012.
We are therefore using the latter definition.

123
A study of Bangladesh 401

transfer costs discourage the flow of remittances through formal channels thus give
rise to informal money transfers. This is a more pressing issue in developing
countries where besides poor infrastructure, high cost of transfer constraints the ease
of remittances inflow subsequently compelling remitters to send money via informal
channels such as postal mails, visiting migrants or migrant’s relatives and friends,
and the informal money transfer services (IFTs) (Coxhead and Linh 2010). The
formal channels used by remitters often include Western Union money transfers,
bank drafts, and automated teller machines. Notably, the remittances flow is also
influenced by a remitter’s job stability and the remittance-sending country’s
economic performance.
Empirically, it has been shown that remittances have both a growth-enhancing
and a poverty-reducing potential. For instance, Adams and Page (2005) study 71
developing countries with the aim to analyze the effects of migration and
remittances on inequality and poverty. Their results show that both international
migration and remittances significantly reduce the level, depth, and severity of
poverty in developing countries. Nevertheless, the remittances-led growth (RLG)
hypothesis has shown mixed results. For example, Pradhan et al. (2002) examine
the effect of workers’ remittances on economic growth in a sample of 39
developing countries using a panel data from 1980 to 2004 and a standard growth
model. Their results show a positive impact of remittances on growth. In a study,
Chami et al. (2003) consider the role of remittances in development and economic
growth by constructing a framework which connects the motivational (altruistic)
aspects of remittances to the impacts on economic activities, and find that
remittances in fact have a negative effect on economic growth largely due to the
moral hazard problem in remittances. Gupta et al. (2009) assess the effect of
remittances in Sub-Saharan Africa within the context of financial development and
poverty reduction and find remittances have a direct poverty-mitigating effect and
the potential to support financial development. Giuliano and Ruiz-Arranz (2009)
explore the links between remittances and growth within the context of financial
development for 100 countries and find remittances can boost growth in countries
with a less developed financial system by providing an alternative way to finance
investment and help overcome liquidity constraints. Acosta (2008) studies ten
countries in Latin America and the Caribbean (LAC) and finds remittances support
growth and reduce inequality and poverty. Mundaca (2009) analyzes the effect of
workers’ remittances and financial intermediation on economic growth with a
panel data for selected countries in the LAC and states that remittances when used
appropriately with effective financial intermediation could result in growth
possibilities. Nyamongo et al. (2012) investigate the role of remittances and
financial development on economic growth in a panel of 36 countries in Africa
over the period 1980–2009. They find, inter alia, (a) remittances are an important
source of growth for African countries; (b) volatility of remittances has a negative
effect on growth; (c) remittances seem to compliment financial development; and
(d) financial development is a weak contributor to growth. Contrary to Giuliano
and Ruiz-Arranz (2009), Bettin and Zazzaro (2012) show where financial system
functions effectively, the RLG hypothesis is highly plausible relative to where
banking systems are weak. In a recent study, Kumar (2013b, e) considers the

123
402 R. R. Kumar, P. J. Stauvermann

effects of capital inflows including remittances in the LAC region by grouping the
region into two clusters led by Brazil and Mexico and finds remittances have a
positive contribution in the Mexico-led cluster, negative in the Brazil-led cluster,
and overall, there is a net positive effect of remittances in the region which
nevertheless is relatively lower due to the predominant negative effects in the
Brazil-led cluster.
The positive effect of remittances on income is also noted in a number of
country-specific studies. For instances, some studies conclude remittances have a
growth-enhancing capacity in the long-run for countries which are recipients of
large amount of remittances. Among these, countries include Tonga, Samoa, Fiji,
Vanuatu, India, Pakistan, Nepal, and Guyana (Jayaraman et al. 2011a, b, 2012;
Kumar 2011a, b; Kumar et al. 2011; Kumar 2012a, 2013c). Similarly, other studies
also show when remittances are effectively linked with technology (Philippines) and
financial development (Sub-Saharan countries), the effect on growth is positive
(Kumar 2012b; 2013d). On the contrary, some studies find remittances have a
negative effect on growth. For instance, Rao and Takirua (2010) examine the
plausible sources of growth in a small island economy of Kiribati using the general-
to-specific (GETS) technique and find remittances have a long-run negative effect.
In examining the effects of tourism and remittances in Kenya, Kumar (2013a)
concludes that the latter has a positive influence on per worker output in the short-
run only.

2.1 Remittances and growth nexus in Bangladesh

Quite a lot of studies have explored the role of remittances in Bangladesh, with
mixed outcomes (Mahmud and Osmani 1980; Stahl and Habib 1989; Murshid et al.
2002; Siddiqui and Arbar 2003). Hasan (2006) confirms that remittances have a
positive impact on household consumption in Bangladesh and suggests the need to
strengthen the micro-finance institutions for effective and efficient mobilization of
remittances. Ahmed (2010) points out that despite Bangladesh being one of the top
recipients of remittances, remittances have a negative effect on its economic
growth. Ahmed therefore emphasizes that the motivation for remittance differs
significantly to that of investment and exports in Bangladesh, and hence, remittance
does not play a significant role in the capital development. Moreover, although the
evidence of many policy initiatives and activities to increase the inflow of
remittances is clear, there is a lack of effective use of remittances in the country
(Chowdhury et al. 2010). Barai (2012) argues that remittances are mainly used for
consumption as a means to overcome poverty, and hence, remittances contribute
substantially to the marginal propensity of consumption. In a recent study, Siddique
et al. (2012) examine the causal link between remittances and economic growth for
three countries namely—India, Sri Lanka, and Bangladesh using the Granger
causality test in a vector autoregression (VAR) framework and over a 25-year
period (1975–2006). Their results show a unidirectional causality from remittances
to economic growth.

123
A study of Bangladesh 403

3 Econometric modeling and estimation techniques

3.1 Framework

For the purpose of modeling and analysis, we use an approach introduced by Sturm
et al. (1998) and Rao (2010) which are related to the augmented Solow (Solow,
1956) framework.2 The initial equation is defined as:
yt ¼ At kta ; a [ 0 ð1Þ
where A = stock of technology and k = capital per worker, and a is the capital
share. The Solow model assumes that the evolution of technology is given by:
Ut ¼ A0 egt ð2Þ
where A0 is the initial stock of knowledge and t is time.
Next, we introduce remittances per worker, rmt, as shift variable:
Wt ¼ f ðrmt Þ ð3Þ
The effect of rmt on total factor productivity (TFP) can be captured when the
latter is entered as a shift variable into the production function (c.f. Rao 2010).
Subsequently, we identify rmbt [where b [ (0, 1) represents elasticity of remittances]
as part of the stock of technology and redefine At as follows:
At ¼ Ut Wt ¼ Ao egt rmbt ð4Þ
In this case, egt includes catch-all factors. Hence,
 
yt ¼ Ao egt rmbt kta ð5Þ

3.2 Data

We use the perpetual inventory method to build the data for capital stock. We
assume depreciation rate (d) of 0.08 and an initial capital stock (K0) as 1.2 times the
real GDP of 1978 in constant USD (US dollars at 2005 prices). The gross fixed
capital formation in constant 2005 USD is used as a suitable proxy for aggregate
investment (It). Hence, Kt = (1–d) Kt-1 ? It. The data on labor stock are estimated
using average rate of employment (which was 69.1 % over the period 1991–2011).
Remittances per worker data are converted to constant 2005 prices. A total of
34 years of annual data over the period 1979–2012 are used in the analysis. Notably,
the data for gross fixed capital formation reported by the World Bank (2013) are
from 1980 to 2012, and therefore, our sample size is restricted to 1979–2012
although data are available for remittances since 1976. The data on key variables are
sourced from World Development Indicators and Global Development Finance
database (World Bank 2013). All data are duly transformed into natural log form for
analysis. A descriptive statistics and correlation matrix and the graphs of log of

2
Note that if a Cobb–Douglas function is used, the form of neutrality of technical progress (Harrod‘-
neutral, Solow-neutral, and Hicks-neutral) does not play a role since the effect remains the same.

123
404 R. R. Kumar, P. J. Stauvermann

output, capital, and remittances, in per worker terms, are provided in Table 1 and
Fig. 1, respectively.

3.3 ARDL bounds procedure

Next, we specify the ARDL specifications as (6–8) below. Note that each equation
has a dummy (dum) associated that represents the structural break in the respective
level series which is identified by applying the Perron (1997) unit root test with
structural break. Including the dummy therefore provides a relatively more robust
computation of bound statistics.
DLyt ¼ b10 þ b11 Lyt1 þ b12 Lkt1 þ b13 Lrmt1 þ a10 dumy
Xp Xp X
p
ð6Þ
þ a11i DLyti þ a12i DLkti þ a13i DLrmti þ e1t
i¼1 i¼0 i¼0

DLkt ¼ b20 þ b21 Lyt1 þ b22 Lkt1 þ b23 Lrmt1 þ a20 dumk
Xp Xp X
p
ð7Þ
þ a21i DLyti þ a22i DLkti þ a23i DLrmti þ e2t
i¼1 i¼0 i¼0

DLrmt ¼ b30 þ b31 Lyt1 þ b32 Lkt1 þ b33 Lrmt1 þ a30 dumrm
Xp Xp X p
ð8Þ
þ a31i DLyti þ a32i DLkti þ a33i DLrmti þ e3t
i¼1 i¼0 i¼0

The autoregressive distributed lag (ARDL) approach is used because it is


relatively simple and recommended for small sample size (Ghatak and Siddiki 2001;
Pesaran et al. 2001). To examine the cointegration based on the computed
F-statistics, it is recommended to use the critical bounds from Narayan (2005)
which are specifically constructed for a small sample size. The critical bounds of
Pesaran et al. (2001), however, are suitable in cases when sample size exceeds 80.
Although one may not test for unit roots and investigate cointegration thus
overlooking the order of integration, we emphasize the need to conduct the unit root
tests for a couple of reasons. First, to ensure that the series are indeed I(0) and/or
I(1) in order to apply the ARDL bounds procedure instead other approach such as
ordinary least squares (OLS) method which is not recommended for variables in the
presence of unit root; and second, examining the unit root provides information on
the maximum lags which are useful when performing the Toda and Yamamoto
(1995) non-Granger causality procedure. Therefore, we use the ADF, Phillips–
Perron (PP), and KPSS tests to examine the time series properties of the variables
and compute the unit root statistics. Based on these tests, we conclude that all
variables are stationary at most in their first differences (Table 2) duly confirming
the maximum order of integration is one. Furthermore, we use the Perron (1997)
unit root test to determine structural breaks in the series. The presence of structural
breaks may influence the computed bounds F-statistics, and hence, the cointegration
results. We find all variables in the levels series have a unit root with the structural
breaks. Moreover, I(1) series are stationary for all variables except the capital per

123
A study of Bangladesh 405

Table 1 Descriptive statistics


Ly Lk Lrm
and correlation matrix
Mean 6.1695 6.4813 3.4130
Median 6.0959 6.3563 3.2510
Maximum 6.7624 7.3384 4.4355
Minimum 5.8648 6.0397 2.7705
Std. Dev. 0.2734 0.4209 0.48460
Skewness 0.7174 0.6371 0.8885
Kurtosis 2.2912 2.0552 2.5679
Jarque–Bera 3.7349 3.6694 4.8769
Probability 0.1545 0.1597 0.0873
Ly 1.0000 – –
Lk 0.9981 1.0000 –
Source: Authors’ calculations Lrm 0.8922 0.8899 1.0000
using Eviews 8

8.0

6.0

4.0

2.0
Ly Lk Lrm
0.0
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
Fig. 1 Log of output, capital, and remittances (in per worker terms) constant 2005 US$. Source Authors’
calculation and World Bank (2013)

worker. In this case, we rely on the ADF, PP, and KPSS tests to pursue cointegration
tests while controlling for the relevant period breaks in the respective series. As
evident from Table 3, the structural breaks in the level series are noted in 1992,
1987, and 1988 for Ly, Lk, and Lrm, respectively. In case of the first difference
series, breaks in the series are noted in the years 1984, 2006, and 1988 for DLy, DLk,
and DLrm, respectively. We factor this information while computing the bounds F-
statistics by setting the break period to one in the respective dummy variable in Eqs.
(6–8).
Next, the bounds F-statistics are presented in Table 4. The results show evidence
of long-run cointegration when the per worker output (Lyt) is set as the dependent
variable. In this case, the computed F-statistics of 14.6512 exceeds the upper critical
bound of 9.413 at the 1 % level of significance. Notably, when the Lk and Lrm are
set as dependent variable, separately, the computed F-statistics does not satisfy the
critical bounds duly confirming a single cointegrating vector.3
3
Note that the critical bounds are presented for samples of 35 and 30 since the Narayan (2005) bounds
are for samples from 30 to 80 with 5-year intervals. The sample size in this study is 34 and the computed
F-statistics of Ly as depended variable clearly exceeds both samples of 30 and 35.

123
406 R. R. Kumar, P. J. Stauvermann

Table 2 Unit root tests results


Variables in log ADF Phillips and Perron KPSS
form
Level First Level First Level First
difference difference difference

Lyt -0.9058 -7.7782A -0.7024 -7.4921A 0.1557C 0.4825C


A A
Lkt -2.2128 -7.4567 -3.1336 -7.2564 0.2476 0.6767C
A A
LRM -1.0547 -5.5949 -1.2133 -5.6013 0.2220 0.1684A

The ADF critical values are based on Mackinnon (1996). The optimal lag is chosen on the basis of the
Akaike information criterion (AIC). The null hypotheses for ADF and Phillips–Perron tests are that a
series has a unit root (non-stationary) and for KPSS, the series is stationary, respectively. A, B, and C
denote 1, 5 and 10 % level of significance denotes the rejection (and acceptance) of null hypothesis in
case of ADF and Phillips–Perron (and KPSS) tests. Source: Authors’ calculation using Eviews 8

Table 3 Unit root tests with single structural break


Variables Level 1st Diff.

T-stat Break T-stat Break

Ly -1.624435[2] 1992 -8.862123[0]A 1984


Lk -3.660159[1] 1987 -2.779164[0] 2006
LRM -2.853569[2] 1988 -6.590713[0]A 1988

Critical values are obtained from Perron (1997). The null hypothesis is that a series has a unit root with a
structural break in both the intercept and the trend; A denotes rejection of the null hypothesis at the 1 %
level of significance. Source: Authors’ calculation using Eviews 8

Table 4 Results of bound tests based on 1 % critical bounds

Dependent variable Computed F-statistic

Ly 14.6512A
Lk 6.0642
LRM 4.2949

Sample size Lower bound value Upper bound value

35 7.977 9.413
30 7.643 9.063

Critical values are from Narayan (2005)—Critical values for the bounds test: case V: unrestricted
intercept and unrestricted trend, p. 1990; A—indicates significance at 1 % level. Source: Authors’
calculation using Mocrofit 4.1

Once the existence of a long-run relationship is confirmed, the next step is to


estimate the long-run and short-run equations. During this process, the diagnostic
tests are looked at based on the initial ARDL (autoregressive distributed lag) model
(Table 5). These includes: (1) Lagrange multiplier test of residual serial correlation
[v2(1) = 0.2078]; (2) Ramsey’s RESET test using the square of the fitted values for

123
A study of Bangladesh 407

a
15

10

-5

-10

-15
1995 2000 2005 2010

b
1.5

1.0

0.5

0.0

-0.5
1995 2000 2005 2010

Fig. 2 a Cumulative sum of Recursive residuals. b Cumulative sum of squares of Recursive residuals.
The straight lines represent critical bounds at 5 % significance level.

Table 5 Diagnostic tests from the ARDL lag estimates


Test types LM version p value F version p value

(I) Serial correlation v2(1) = 0.2078A 0.649 F(1, 16) = 0.1116A 0.743
2 A
(II) Functional form v (1) = 0.8006 0.371 F(1, 16) = 0.4387A 0.517
(III) Normality v2(2) = 3.9403A 0.139 Not applicable
(IV) Heteroscedasticity v2(1) = 2.5009A 0.114 F(1, 28) = 2.5464A 0.122

A indicates rejection of null hypothesis of the presence of the respective test types, (I), (II), (III), and (IV)
at 1 % level of significance, respectively. Source: Authors’ calculation using Mocrofit 4.1

correct functional form [v2(1) = 0.8006]; (3) normality test based on a test of
skewness and kurtosis of residuals [v2(2) = 3.9403]; and (4) heteroscedasticity test
based on the regression of squared residuals on squared fitted values
[v2(1) = 2.5009]. As noted from the v2 results, the diagnostic tests show the
equation performed well as the disturbance terms are normally distributed and
serially uncorrelated with homoscedasticity of residuals duly confirming the models
have correct functional form. Moreover, the CUSUM and CUSUM of squares plot,

123
408 R. R. Kumar, P. J. Stauvermann

which measures stability of the parameters in the model, show the model is
stable (Fig. 2a, b).

3.4 Short-run and Long-run results

3.4.1 Short-run

The short-run results show mixed outcomes (Table 6: panel a). The capital
productivity coefficient is not statistically significant in the current (t), lag-one (t -
1) and lag-two (t - 2) periods. However, capital per worker coefficient at lag-three
(t - 3) is negative (D0t-3 = -0.9528) and statistically significant at the 10 % level
of significance. Moreover, the coefficient of remittances per worker, although
positive, is not statistically significant in period t. Notably, the coefficients of
remittance per worker are negative and statistically significant in periods t - 1
(DLrmt-1 = -0.0482) and t - 2 (DLrmt-1 = -0.0454), at the 10 and the 5 %
levels of significance, respectively. Therefore, remittances per worker have a net
negative effect in the short-run. The error-correction term (ECTt-1 = -0.6640),
which measures the speed at which prior deviations (errors) from equilibrium are

Table 6 Estimated long-run coefficients and error-correction representation

Panel a: long-run (dependent variable Lyt) Panel b: short-run (dependent variable DLyt)

Regressor Coefficient Standard t-ratio Regressor Coefficient Standard t-ratio


error error

Lk 0.3314 0.1544 2.1466B DLkt 0.2202 0.5308 0.4148N


Lrm 0.1087 0.0527 2.0635C DLkt-1 -0.4218 0.7869 -0.5360N
Constant 3.1424 0.6593 4.7664A DLkt-2 0.3805 0.7661 0.4967N
A
Time 0.0157 0.0045 3.4799 DLkt-3 -0.9528 0.5285 -1.8026C
DLrmt 0.0104 0.0200 0.5179N
DLrmt-1 -0.0482 0.0247 -1.9495C
DLrmt-2 -0.0454 0.0197 -2.3003B
DLrmt-3 0.0230 0.0137 -1.6810N
Constant 2.0865 0.7790 2.6784B
Time 0.0105 0.0043 2.4511B
ECTt-1 -0.6640 0.1752 -3.7897A

Panel c: Short-run dynamics test statistics

R-squared 0.9015 R-bar-squared 0.8319


SE of regression 0.0067 F-stat. F(10, 19) 15.5512
Mean of dependent Variable 0.0297 SD of dependent variable 0.01642
Residual sum of squares 0.0008 Equation log-likelihood 115.9751
Akaike information criterion 102.9751 Schwarz Bayesian criterion 93.8674
DW-statistic 1.8800 ARDL(1,4,4) N = 34

A, B, and C refer to the 1, 5; and 10 % level of significance, respectively; N = not significant within the
1–10 % level. Source: Authors’ calculation using Mocrofit 4.1

123
A study of Bangladesh 409

corrected (in this case, about 66 %), has the correct (negative) sign and is significant
at the 1 % level duly indicating a relatively speedy convergence to the long-run
equilibrium.

3.4.2 Long-run

The long-run results (Table 6: panel b) show that the capital per worker
(Lk = 0.3314) is the dominant driver of the output per worker. The coefficient of
Lk is statistically significant at the 5 % level and is close to the stylized value of
one-third (0.33) (Rao 2007; Ertur and Koch 2007). However, in many instances, the
capital share can deviate from the stylized value due to a number of reasons. These
include: (a) when the capital and labor inputs tend to grow at relatively similar rates;
(b) when an economy is predominantly developing, and hence, a large number of
self-employed persons earn income from both capital and their own labor (Gollin
2002) thus making it difficult to obtain meaningful measures of income shares; and
(c) the quality of data and the sample size which also makes it difficult to compute
the capital stock (Bosworth and Collins 2008) that can ideally exhibit decreasing
returns to scale and thus conform to a desirable steady-state convergence. Moreover,
the elasticity coefficient of remittances per worker is 0.11 % (Lrmt = 0.0187),
which is statistically significant at the 10 % level. We also include the trend (Trend)
variable is in the regression. The results show Time trend is significant and has a
marginal positive effect on the output per worker both in the short-run
(Time = 0.0105) and the long-run (Time = 0.0157), respectively. Hence, the
marginal positive effect over time is an indication of the other factors contributing
positively to the growth in the economy of the Bangladesh that are not controlled
explicitly in the regression.

3.5 The Toda–Yamamoto approach to Granger non-causality test

Next, we examine the Granger non-causality test proposed by Toda and Yamamoto
(1995) (henceforth, T–Y approach). The T–Y approach is suitable when the series
are either integrated of different orders, not cointegrated, or both. In these cases, the
ECM (error-correction method) cannot be applied for Granger causality tests, and
the standard (pair-wise) Granger causality test may not give robust results. Hence,
the T–Y approach provides a method to test for the presence of non-causality,
irrespective of whether the variables are I(0), I(1) or I(2), not cointegrated or
cointegrated of an arbitrary order. In order to carry out the Granger non-causality
test, we present the model in the following VAR system:
X
k dX
max X
k dX
max
lyt ¼ a0 þ a1i lyti þ a2j lytj þ g1i Lkti þ g2j Lktj
i¼1 j¼kþ1 i¼1 j¼kþ1
X
k dX
max
þ /1i Lrmti þ /2j Lrmtj þ k1t ð9Þ
i¼1 j¼kþ1

123
410 R. R. Kumar, P. J. Stauvermann

Table 7 Granger non-causality test based on v2 (p value)


Dependent variable (Y)

Ly Lk Lrm

Excluded variables (X) ?


Ly – 1.5209 (0.9106) 16.1776A (0.0064)
C
Lk 10.1832 (0.0702) – 13.9653B (0.0158)
A
Lrm 21.3215 (0.0007) 0.4032 (0.9952) –
Combined 41.0459A (0.0000) 3.6877 (0.9603) 27.1659A (0.0025)

Based on the maximum order of integration (p = 1) and maximum lag used in the ARDL estimation
(m = 4), the maximum lag in the Toda-Yamamoto non-causality test is (p ? m) 5. A, B, C indicate
causality at the 1, 5, and 10 % levels of significance, respectively. Granger causality is denoted as X
causing Y, that is: X ? Y. Source: Authors’ calculation using Eviews 8

X
k dX
max X
k dX
max
Lkt ¼ b0 þ b1i Lkti þ b2j Lktj þ h1i lyti þ h2j lytj
i¼1 j¼kþ1 i¼1 j¼kþ1
X
k dX
max
þ #1i Lrmti þ #2j Lrmtj þ k2t ð10Þ
i¼1 j¼kþ1

X
k dX
max X
k dX
max
Lrmt ¼ c0 þ c1i Lrmti þ c2j Lrmtj þ u1i lyti þ u2j lytj
i¼1 j¼kþ1 i¼1 j¼kþ1
X
k X
dmax
þ l1i Lkti þ l2j Lktj þ k3t ð11Þ
i¼1 j¼kþ1

The null hypothesis of non-causality is rejected when the p-values fall within the
conventional 1–10 % level of significance. Hence, in (9), Granger causality from lkt
to lyt and lrmt to lyt implies g1i = 0Vi and /1i = 0Vi, respectively. Similarly, in
(10), lyt and lrmt Granger causes lkt if h1i = 0Vi and #1i = 0Vi, respectively; from
(11) lyt and lkt Granger causes lrmt if u1i = 0Vi and l1i = 0Vi, respectively. From
the unit root results, the maximum order of integration is 1 (m = 1), and the optimal
lag length chosen in the ARDL estimates using the Akaike information (AI) and
Schwarz Bayesian (SB) Criteria (Table 6: panel c) is four (p = 4). Hence, the
appropriate lags to carry out the Granger non-causality test (p ? m = 5) is five.
Before proceeding to causality test, we examine the inverse roots of the AR (auto-
regressive) characteristics polynomial which should lie within the positive and
negative unit boundary in order to obtain reliable results. If the inverse roots lie
outside unity, the appropriate lag and/or trend variables are included as the
exogenous variables to correct this. We ensure that the AR inverse roots are within
the unit boundary by including the trend and the appropriate lags as exogenous
variables before proceeding to the causality assessment. The results of the causality
tests are reported in Table 7.
The Granger causality results (Table 7) show a bidirectional causation between
output per worker (Ly) and remittances per worker (Lrm) which are statistically
significant at 1 % level. In other words, Lrm ? Ly (v2 = 21.3215) and Lrm / Ly

123
A study of Bangladesh 411

(v2 = 16.1776) (Ly $ Lrm) duly indicating the mutually reinforcing effect of the
remittances and output. Moreover, a unidirectional causation is noted from capital
per worker to output per worker (Lk ? Ly, v2 = 10.1832) at the 10 % level and from
capital per worker to remittances per worker (Lk ? Lrm, v2 = 13.9653) at the 5 %
level of significance. The latter indicates the plausibility of remittances being
channeled into investment activities and capital accumulation. Finally, the causation
from the combined effects (which we define as the causation due conjoint interaction)
show that the capital per worker and the remittances per worker conjointly cause the
output per worker (Lk 9 Lrm ? Ly, v2 = 41.0459) at the 1 % level of significance;
and the output per worker and the capital per worker conjointly cause the remittances
per worker Ly 9 Lk ? Lrm, v2 = 27.1659) at the 1 % level of significance.
Our findings to some extent are similar to Siddique et al. (2012), who find a
unidirectional (one-way) causation from remittances to economic growth, and Paul et al.
(2011) who find that output alone determined long-run movements of remittances.
However, our study deviates both from Siddique et al. (2012) and Paul et al. (2011) in that
we find a bidirectional causation between economic growth (measured by output per
worker) and the remittances per worker and hence conclude that remittances and output
are have mutually reinforcing effect in Bangladesh.4 Furthermore, we also estimate the
long-run elasticity coefficient of remittances per worker, which is 0.11 %.

4 Conclusion

In this paper, we set out to explore the much controversial topic of whether
remittances influence economic growth in Bangladesh. We used the augmented
Solow framework and the ARDL bounds procedure to examine the short-run and
long-run effects, and further extended the study to examine the causality nexus. The
results clearly show that the remittances per worker have a long-run momentous
effect on the output per worker in the long-run. However, the short-run results
indicate a net negative effect. Furthermore, the causality results show a mutually
reinforcing effect between remittances and output, and a unidirectional causation
from capital per worker to remittances. Subsequently, we contend that remittances
in Bangladesh exhibit a mixed effect on the economic growth in the short-run.
However, there is evidence that remittances have a momentous long-run effect and
therefore support the remittances-led growth hypothesis (RLG). Noting these
outcomes, cost-effective and efficient remittance inflows to Bangladesh need to be
encouraged which duly need to be channeled to ongoing and new productive
activities. Developing pro-remittance infrastructure, which includes financial sector
development, technology, financial literacy, and supporting institutions will
therefore be critical for optimizing benefits from remittances and propelling the
economic growth of Bangladesh.

4
Note that the study also considered the role of financial development and information and
communications technology (ICT). We find that both have no significant effect on per worker output
both in the short-run and long-run, and therefore they are not included in the study.

123
412 R. R. Kumar, P. J. Stauvermann

Acknowledgments The authors thank and appreciate the comments and suggestions from the Editor-in-
Chief, Professor Dr. Pierluigi Porta and the anonymous reviewers. Peter J. Stauvermann thanks the
financial support from the College of Economics and Business of the Changwon National University,
2013. Finally, the views expressed in the article are those of the authors and does not necessarily reflect
the views of their affiliated institutions.

Appendix

See Table 8.

Table 8 Sample used for Bangladesh (1976–2012)


Year GDP (constant Investment GDP Remittances Population Employment
2005 USD) (constant 2005 deflator (constant 2005
USD) USD)

1976 17294707649 n.d. 19.61 132821263 73944664 51092241


1977 17757004877 n.d. 18.98 576919295 75880084 52429524
1978 19013106589 n.d. 23.85 672139549 78011624 53902316
1979 19926046494 n.d. 26.84 885252803 80241773 55443243
1980 20089269083 2154148534 31.56 1490219630 82498440 57002493
1981 20853062701 2402310648 34.88 1517015872 84764142 58567985
1982 21348628816 2596987724 38.26 1910807716 87060582 60154715
1983 22206008546 2728607549 41.51 2148666504 89399666 61770911
1984 23356429079 2989267213 47.35 1468564465 91804318 63432411
1985 24109272332 3169398866 52.63 1325798445 94287722 65148325
1986 25133571752 3380632534 56.84 1407904119 96851505 66919777
1987 26071622934 3656264232 63.02 1647687593 99476987 68733860
1988 26634562018 3878569485 67.81 1563679492 102133217 70569188
1989 27330350482 4150442939 73.57 1430527108 104779345 72397537
1990 28954130547 4413090044 78.24 1382364111 107385847 74198505
1991 29921011909 4475590408 83.40 1281005204 109934590 75959566
1992 31428771530 4674371925 85.88 1474215571 112430968 77684444
1993 32866441571 5119569454 86.12 1624153859 114897543 79388730
1994 34208938270 5598389580 89.37 1788080285 117369492 81096729
1995 35893746330 6108127109 95.94 1739227952 119869585 82824174
1996 37552743789 6755394079 100.00 1867130760 122400896 84573189
1997 39575917287 7504178630 103.09 2056087360 124945315 86331262
1998 41644760785 8409346789 108.53 2054889918 127478524 88081588
1999 43672540057 9241097131 113.58 2208858666 129966823 89800884
2000 46268657157 9913701045 115.69 2361446537 132383265 91470531
2001 48708872666 10485766834 117.53 2486481922 134729503 93091669
2002 50859569550 11341651201 121.28 3272175044 137006279 94664813
2003 53532745612 12233797551 126.77 3495841089 139185986 96170887
2004 56889517979 13354973365 132.15 3765685642 141235035 97586682
2005 60277560976 14784406504 138.86 4314502846 143135180 98899592

123
A study of Bangladesh 413

Table 8 continued

Year GDP (constant Investment GDP Remittances Population Employment


2005 USD) (constant 2005 deflator (constant 2005
USD) USD)

2006 64273563733 16009777113 146.04 5160590417 144868702 100097373


2007 68404967193 17365406452 155.95 5834863216 146457067 101194858
2008 72639529866 17678803431 169.65 7305016919 147969967 102240199
2009 76809880500 18772290123 180.72 8083592904 149503100 103299521
2010 81471733197 20446395151 192.42 7829930636 151125475 104420505
2011 86936936296 22210657422 206.91 8098627870 152862431 105620659
2012 92429549174 23875336365 223.46 9020602471 154695368 106887131

The data are sourced from World Bank (2013). Data on employment are computed from the average
employment rate, and remittances data are converted to 2005 constant prices, n.d. = no data available.
Source: World Bank (2013)

References

Acosta P (2008) What is the impact of international remittances on poverty and inequality in Latin
America? World Dev 36:89–114
Adams RH, Page J (2005) Do international migration and remittances reduce poverty in developing
countries? World Dev 33:1645–1669
Ahmed M (2010) Migrant workers remittance and economic growth: evidence from Bangladesh. ASA
Univ Rev 4:1–13
Barai MK (2012) Development dynamics of remittances in Bangladesh. Sage Open 2:1–13
Bettin G, Zazzaro A (2012) Remittances and financial development: substitutes or complements in
economic growth? Bulletin of Economic Research 64:509–536
Bosworth B, Collins SM (2008) Accounting for growth: comparing China and India. Journal of Economic
Perspectives 22:45–66
Buch MC, Kuckulenz A (2010) Worker remittances and capital flows to developing countries.
International Migration 48:89–117
Chami R, Fullenkamp C, Jahjah S (2003) Are immigrant remittance flows a source of capital for
development? IMF Working Paper 03/189, Washington, DC
Chowdhury R, Hamid F, Chatterjee D (2010) Remittance as a tool of economic development, Bangladesh
Perspective. Bangladesh Res Publ J 4:286–296
Coxhead I, Linh VH (2010) The effects of global shocks on poverty in Vietnam institute of social
economic research. Osaka University, Japan
Ertur C, Koch W (2007) Growth, technological interdependence and spatial externalities: theory and
evidence. J Appl Econom 22:1033–1062
Ghatak S, Siddiki J (2001) The use of ARDL approach in estimating virtual exchange rates in India.
J Appl Stat 28:573–583
Giuliano P, Ruiz-Arranz M (2009) Remittances, financial development, and growth. J Dev Econ
90:144–152
Gollin D (2002) Getting income shares right. J Polit Econ 110:458–474
Gupta S, Pattillo CA, Wagh S (2009) Effect of remittances on poverty and financial development in sub-
saharan Africa. World Dev 37:104–115
Hasan RA (2006) Harnessing remittances for economic development of Bangladesh. Bangladesh
Working Paper 1, INAFI, Philippines. http://www.inafiasia.net/download/new/Microfinance%
20&%20Remittance/1.pdf. Accessed 1 Dec 2012
Jayaraman TK, Choong CK, Kumar R (2011a) Financial sector development and remittances in pacific
island economies: how do they help the world’s two most recipient-dependent countries? Perspect
Glob Dev Technol 10:386–405

123
414 R. R. Kumar, P. J. Stauvermann

Jayaraman TK, Choong CK, Kumar R (2011b) Role of remittances in small Pacific Island economies: an
empirical study of Fiji. Int J Econ Bus Res 3:526–542
Jayaraman TK, Choong CK, Kumar R (2012) Role of remittances in India’s economic growth. Glob Bus
Econ Rev 14(3):159–177
Kumar RR (2011a) Nexus between financial and technology inclusion, remittances and trade openness
vis-à-vis growth: a study of Nepal. Econ J Nepal 34:13–33
Kumar RR (2011b) Exploring the role of trade, aid, remittances and financial development in Pakistan.
J Soc Policy Sci 2:119–158
Kumar RR (2012a) Role of trade openness, remittances, capital inflows, and financial development in
Vanuatu. In: Sirkeci I, Cohen JH, Ratha D (eds) Migration and remittances during the global
financial crisis and beyond. World Bank, Washington, pp 325–333
Kumar RR (2012b) Exploring the interactive effects of remittances, financial development and ICT in
Sub-Saharan Africa: an ARDL bounds approach. Afr J Econ Sustain Dev 1(3):214–242
Kumar RR (2013a) Exploring the nexus between tourism, remittances and growth in Kenya. Qual Quant.
doi:10.1007/s11135-013-9853-1
Kumar RR (2013b) Exploring the nexus between capital inflows and growth in Latin America and the
Caribbean: a study of clusters led by Brazil and Mexico. Qual Quant. doi:10.1007/s11135-013-9906-5
Kumar RR (2013c) Remittances and economic growth: a study of Guyana. Econ Sys 37:462–472
Kumar RR (2013d) Linking remittances with financial development and ICT: a study of the Philippines.
Int J Econ Bus Res 5:379–399
Kumar RR (2013e) Exploring polarization and uniformity in sectoral and capital vis-à-vis growth: a study
of Brazil-led and Mexico-led clusters in the region. Manag Decis 51:1579–1595
Kumar RR, Kumar R (2013) Effects of energy consumption on per worker output: a study of Kenya and
South Africa. Energy Policy 62:1187–1193
Kumar RR, Naidu V, Kumar R (2011) Exploring the nexus between trade, visitor arrivals, remittances
and income in the Pacific: a study of Vanuatu. Economica 7:199–217
Mackinnon JG (1996) Numerical distribution functions for unit root and cointegration tests. J Appl
Econom 11:601–618
Mahmud W, Osmani SR (1980) Impact of emigrant workers’ remittances on the Bangladesh economy.
Bangladesh Dev Stud 8:1–28
Mundaca G (2009) Remittances, financial market development, and economic growth: the case of Latin
America and the Caribbean. Rev Dev Econ 13:288–303
Murshid KAS, Iqbal K, Ahmed M (2002) A study on remittance inflows and utilization. IOM, Regional
Office for South Asia, Dhaka http://www.un-bd.org/pub/unpubs/IOM-A%20Study%20on%
20Remittance%20Inflows%20and%20Utilization.pdf. Accessed 1 Dec 2012
Narayan PK (2005) The saving and investment nexus for China: evidence from cointegration tests. Appl
Econ 37:1979–1990
Nyamongo EM, Misati RN, Kipyegon L, Ndirangu L (2012) Remittances, financial development and
economic growth in Africa. J Econ Bus 64:240–260
Paul BP, Uddin MGS, Noman AM (2011) Remittances and output in Bangladesh: an ARDL bounds
testing approach to cointegration. Int Rev Econ 58:229–242
Perron P (1997) Further evidence on breaking trend functions in macroeconomic variables. J Econom
80:355–385
Pesaran MH, Shin Y, Smith R (2001) Bounds testing approaches to the analysis of level relationships.
J Appl Econom 16:289–326
Pradhan G, Upadhyay M, Upadhyaya K (2002) Remittances and economic growth in developing
countries. Eur J Dev Res 20:497–506
Rao BB (2007) Estimating short and long-run relationships: a guide for the applied economist.Appl Eco
39(13):1613–1625
Rao BB (2010) Estimates of the steady state growth rates for selected Asian countries with an extended
Solow model. Econ Model 27:46–53
Rao BB, Hassan GH (2012) Are the direct and indirect growth effects of remittances significant? World
Economy 35:351–372
Rao BB, Takirua T (2010) The effects of exports, aid and remittances on output: the case of Kiribati. Appl
Econ 42:1387–1396
Ratha D (2007) Leveraging remittances for development, migration policy institute. World Bank,
Washington

123
A study of Bangladesh 415

Siddique A, Selvanathan EA, Selvanathan S (2012) Remittances and Economic growth: empirical
evidence from Bangladesh, India and Sri Lanka. J Dev Stud 48:1045–1062
Siddiqui T, Arbar CR (2003) Migrant worker remittances and micro-finance in Bangladesh. Working
Paper No. 38, Social Finance Programme, International Labour Office. http://www.ilo.org/wcmsp5/
groups/public/@ed_emp/documents/publication/wcms_117970.pdf. Accessed 1 Dec 2012
Solow RM (1956) A contribution to the theory of economic growth. Quart J Econ 70:65–94
Stahl CW, Habib A (1989) The Impact of workers’ remittances on indigenous industries: evidence from
Bangladesh. Dev Econ 27:269–285
Toda HY, Yamamoto T (1995) Statistical inferences in vector autoregressions with possibly integrated
process. J Econom 66:225–250
World Bank (2013) World development indicators and global development Finance. World Bank

123
Reproduced with permission of the copyright owner. Further reproduction prohibited without
permission.

You might also like