Professional Documents
Culture Documents
Economic Modelling
journal homepage: www.elsevier.com/locate/ecmod
a r t i c l e i n f o a b s t r a c t
Article history: Economists have investigated the relationship between output and export in order to explain economic
Accepted 18 June 2013 growth for long years. Numerous studies have found very close correspondence between the growth of out-
put and export. It is commonly known that Thirlwall's papers indicate very tight relationship between the
JEL classification:
growth of output and the ratio of the growth of exports to the income elasticity of demand for imports. This
C32
paper aims to apply Thirlwall's balance-of-payments-constrained (BPC) model for the Turkish economy for
F43
O40
1968–2011 period. This research also evaluates the procedures of testing Thirlwall's principle by estimation
of the income elasticity of demand for imports using the test of stationarity and cointegration methods. The
Keywords: findings are in accordance with the Harrod–Thirlwall growth model. The test results of Johansen cointegration
Thirlwall model procedure and the comments on these results are presented as well.
Balance-of-payments constrained growth © 2013 Elsevier B.V. All rights reserved.
model
Johansen cointegration methodology
0264-9993/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.econmod.2013.06.019
A. Gökçe, E. Çankal / Economic Modelling 35 (2013) 140–144 141
determined by Thirlwall's model. The capital flows will not be longer arranging Eq. (1) for Z and replace it in Eq. (4) gives
than the duration determined by Thirlwall either. Later, Elliot and
Rhodd (1999) and Thirlwall and Hussain (1982) showed that more Y t ¼ ½ð1=πÞð1−α Þ P dt −P ft þ ð1=πÞX t : ð5Þ
advanced estimations can be obtained including the effect of borrow-
ing using the same countries and time period.
Moreno-Brid (1998–99) contributed to the extended BPC growth When relative prices are measured in the same monetary units,
model developed by Thirlwall and Hussain in 1982 by including the they will be constant and their difference will be equal to zero
restriction that the current account deficit is constant in the long according to the Marshall–Lerner condition. Hence, the Eq. (5) can
run. If the capital flows are permitted but the current deficit relative be rewritten as:
to domestic income remains constant, then Thirlwall's basic princi-
ples will still be valid. Similarly, McCombie and Roberts (2002), Y t ¼ ð1=πÞX t : ð6Þ
argued that under reasonable assumptions related to sustainable net
foreign capital inflows as the rate of national income, the capital
Eq. (6) shows Harrod foreign-trade multiplier relation, where (1/π)
inflows in the BPC growth model do not make a great contribution
is a foreign-trade multiplier. The economic growth model with the
in loosening the balance of payments restriction. However, according
balance of payments constrained is called Thirlwall model. The model
to Barbosa-Filho (2001), the contribution of Moreno-Brid (1998–99)
is a different version of Harrod's foreign trade multiplier. The Thirlwall
has two limitations; namely, the model with the BPC is not necessary
model shown in Eq. (6) indicates that the growth rate with the
to be stable and it does not distinguish the interest payments from
constrained balanced of payments in the long-run is obtained by divid-
nonfactor services and good imports. So, Barbosa-Filho (2001) ex-
ing the export (or the export growth rate) by the income elasticity of
tended the model to allow a sustainable foreign borrowing by taking
demand for imports.
into account the effects of interest rates. Moreno-Brid (2003) argued
that Thirlwall's principle does not take the interest payments to
4. Data, model, and methodology
borrowed countries into account without referring to the studies
done by Elliot and Rhodd (1999) and Barbosa-Filho (2001). They de-
Real exports and real income have been moving together, which re-
veloped a BPC model that contained interest payments. The model
veals that these series may be cointegrated. An empirical confirmation
was implemented for the Mexican economy and the results were
of the Harrod–Thirlwall model requires the account of balance to be
found to be supportive of the model.
consistent with data. In the long-run output should be cointegrated
The Thirlwall model has still been investigated for various coun-
with export, where (1/π) is the cointegration coefficient. In this
tries. There is an increasing number of empirical studies on this topic.
study, the objective is to test the presence of long term relationship
between the variables in Eq. (6) using Johansen (1991, 1995)
3. Theoretical framework
cointegration method. Here, (1/π) is a cointegration parameter as men-
tioned previously. Johansen's methodology ascribes to the fundamen-
Thirlwall model (1979) expresses the long-term economic growth
tals of VAR methodology and uses maximum likelihood ratio test. In
via the dynamic Harrod foreign trade multiplier. According to the
the first step, the order of cointegration of the series is determined
model, demand side factors are primary actors on economic growth
with the augmented Dickey–Fuller (ADF) test. When the long-term
and the dominant constraint on demand is the balance of payments.
equilibrium relationship is revealed by the Johansen cointegration
The primary aim of the Thirlwall model (1979) is to bring a light on
method, the hypothesis of the Thirlwall model will not be rejected
how the balance of payments may affect the countries' performances
and the validity of the model for Turkish Economy will be proved.
on economic growth. Thirlwall model is expressed by the following
The Thirlwall model is estimated for the Turkish economy using
three equations:
the annual data for 1968–2011 period. The Johansen cointegration
analysis is performed between GDP and export for this period. The
X t ¼ φ P dt −P ft þ ρZ t ð1Þ Johansen cointegration methodology is employed in order to show
the long-run relationship between GDP and export. GDP and export
series are used at their current values in million dollars. The series
Mt ¼ α P dt −P ft þ πY t ð2Þ are in their natural logarithmic forms.
The data taken from the World Bank shows the movements of
X t þ P dt ¼ Mt þ pft : ð3Þ lnGDP and lnexport in time on Graph 1.
Y t ¼ ½ð1 þ φ−α Þ=π P dt −P ft þ ðρ=πÞZ t ð4Þ
Graph 1. lnGDP and lnexport for the Turkish Economy (1968–2011).
142 A. Gökçe, E. Çankal / Economic Modelling 35 (2013) 140–144
on the Turkish export after 2009. The path of export seems to grow Lags logL LR FPE AIC SCI HQ
more steadily compared to that of GDP.
0 −65.45654 NA 0.145418 3.747585 3.835559 3.778290
Two break tests of Lee and Strazicich (2003) are applied to lnGDP 1 39.68680 192.7628a 0.000528a −1.871489a −1.607569a −1.779374a
and lnexport series. However, the breaks that are statistically significant 2 40.78217 1.886472 0.000622 −1.710121 −1.270254 −1.556595
could not be found. Eviews 7.0 and Gauss 6.0 programs are used in order 3 41.66481 1.422030 0.000745 −1.536934 −0.921121 −1.321999
4 42.43430 1.154232 0.000901 −1.357461 −0.565702 −1.081116
to determine the breaks and to obtain the Johansen cointegration esti-
5 47.75517 7.390099 0.000853 −1.430843 −0.463137 −1.093087
mations, and to perform the other time series analyses. 6 54.92190 9.157492 0.000736 −1.606772 −0.463120 −1.207607
7 55.61844 0.812632 0.000920 −1.423247 −0.103648 −0.962672
4.1. Johansen methodology 8 60.09057 4.720581 0.000947 −1.449476 0.046069 −0.927491
logL: logarithmic likelihood, FPE: final prediction error, AIC: Akaike information
Johansen's methodology takes its starting point in the vector criteria, SIC: Schwarz information criteria, and HQ: Hannan–Quinn criteria.
a
autoregression (VAR) of order p given by Optimum lag lengths are shown in bold.
yt ¼ μ þ A1 yt−1 þ ⋯ þ Ap yt−p þ εt It is known that most of the time series are not stationary at their
own levels and that the cointegration can occur at different orders. Em-
where yt is an n × 1 vector of variables that is integrated of order one, pirical studies have shown that economic series are usually integrated
which is denoted as I(1) and εt is an n × 1 vector of innovations. This at order one. Before starting the Johansen cointegration method, the
VAR can be rewritten as order of cointegration of the series has to be known. However, two
series are integrated of different orders, they cannot be cointegrated.
X
p−1
Therefore, the first step in testing a time series model is to determine
Δyt ¼ μ þ Πyt−1 þ Γt Δyt−1 þ εt
i¼1
the order of integration by means of testing for unit roots. Hence,
ADF unit root test is employed and its results are presented in Table 1.
where Π = ∑ pi = 1Ai − I and Γi = − ∑ pj = i + 1Ai.
If the coefficient matrix Π has reduced rank r b n, then there exist 5. Findings
n × r matrices α and β each with rank r such that Π = αβ′ and β′yt is
stationary. r is the number of cointegrating relationships, the According to ADF test, while H0 is accepted at the log levels of GDP
elements of α are known as the adjustment parameters in the vector and export series, H0 is rejected at first differences at 1% significance
error correction model and each column of β is a cointegrating vector. level. Based on this result, the first differences of the series are
It can be shown that for a given r, the maximum likelihood estimator stationary and integrated of order one. In this situation, the long-run
of β defines the combination of yt − 1 that yields the largest canonical relationship of the series can be investigated. The structure of I(1) of
correlations of Δyt with yt − 1 after correcting for lagged differences the series represents the growth rates of the variables and consistent
and deterministic variables when they present. Johansen proposes with the Thirlwall rule. A crucial determination in using the Johansen
two different likelihood ratio tests of the significance of these canon- procedure is the lag length. The first step of the Johansen cointegration
ical correlations and thereby the reduced rank of the Π matrix: the analysis is to estimate unconstrained VAR (p) model and determine the
trace test and maximum eigenvalue tests are shown in Eqs. (7) and lag length. Hence, the VAR model is estimated and the lag lengths are
(8) respectively. summarized in Table 2.
All the criteria have shown that the appropriate lag-length is 1,
X
n namely, the model has VAR (1) structure. The goodness of fit tests for
λtrace ¼ −T ^
ln 1−λ ð7Þ
i residuals are performed following VAR (1) estimation. Breusch–
i¼rþ1
Godfrey Lagrange Multiplier (LM) is used to test for autocorrelation in
residuals and Jarque–Bera tests for normality. Based on the test results,
^
λmax ¼ −T ln 1−λ ð8Þ
rþ1 autocorrelation does not exist for 12 lags in residuals (Appendix 1). The
normality test shows that the variables have normal distributions both
Here, T is the sample size and λ^ i is the ith largest canonical correla- individually and jointly (Appendix 2). Whether the model is stable or
tion. The trace test tests the null hypothesis of r cointegrating vectors not have also been investigated in addition to the tests. For the VAR
against the alternative hypothesis of n cointegrating vectors. The model to have a stable structure, inverse roots of characteristic polynom
maximum eigenvalue test, on the other hand, tests the null hypothesis of AR should be within unit circle. That all AR roots are in the unit circle
of r cointegrating vectors against the alternative hypothesis of r + 1 shows that the estimated model is stable (Appendix 3 and Appendix 4).
cointegrating vectors. Neither of these test statistics follows a chi
square distribution in general; asymptotic critical values can be found
Table 3
in Johansen and Juselius (1990).
Johansen cointegration test.
λtrace statistics
Table 1
Augmented Dickey–Fuller test results. Hypothesis Eigenvalues λtrace 0.05 p
a
H0 : r = 0, H1 : r ≥ 1 0.542104 37.41915 20.26184 0.0001
Variables k ADF test statistics (τ) k ADF test statistics (τ)
H0 : r ≤ 1, H1 : r ≥ 2 0.085246 3.831316 9.164546 0.4378
(intercept) (intercept with trend)
Appendix 4. Inverse roots of characteristic polynomial Bairam, E.I., 1990. The Harrod trade multiplier revisited. Applied Economics 22,
711–718 (June).
Bairam, E.I., 1993. Static versus dynamic specification and the Harrod foreign trade
Inverse Roots of AR Characteristic Polynomial multiplier. Applied Economics 25, 739–742.
1.5 Barbosa-Filho, N., 2001. The balance-of-payments constraint: from balanced
trade to sustainable debt. Banca Nazionale del Lavoro Quarterly Review 219,
381–400 (December).
Elliot, D., Rhodd, R., 1999. Explaining growth rate differences in highly indebted countries:
1.0 an extension to Thirlwall and Hussain. Applied Economics 31.
Holland, M., Vieira, F., Canuto, O., 2004. Economic growth and the balance-of-payments
constraint in Latin America. Investigacion Economica LXIII, 45–74.
0.5 Hussain, M.N., 1999. The balance-of-payment constraint and growth rate differences
among African and East Asian economies. African Development Review vol. 11,
103–137 (African Development Bank).
Johansen, S., 1991. Estimation and hypothesis testing of cointegration vectors in Gaussian
0.0 vector autoregressive models. Econometrica 55, 1551–1580.
Johansen, S., 1995. Likelihood-based Inference in Cointegrated Vector Autoregressive
Models. Oxford University Press, Oxford.
-0.5 Johansen, S., Juselius, K., 1990. Maximum likelihood estimation and inferences on
cointegration—with applications to the demand for money. Oxford Bulletin of
Economics and Statistics 52, 169–210.
Lee, J., Strazicich, M., 2003. Minimum Lagrange multiplier unit root test with two struc-
-1.0 tural breaks. The Review of Economics and Statistics 85 (4), 1082–1089.
López, J., Cruz, A., 2000. Thirlwall's law and beyond: the Latin American experience.
Journal of Post Keynesian Economics 22 (3), 477–495.
-1.5 MacKinnon, J.G., Haug, A., Michelis, L., 1999. Numerical distribution functions of likelihood
ratio tests for cointegration. Journal of Applied Econometrics 14 (5), 563–577.
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
McCombie, J., Roberts, M., 2002. The role of balance of payments in economic growth.
The Economics of Demand-Led Growth.Edward Elgar, MA.
McCombie, J., Thirlwall, A., 1997. The dynamic Harrod foreign trade multiplier and the
demand oriented approach to economic growth—an evaluation. International
References Review of Applied Economics 11, 1.
Moreno-Brid, J.C., 1998–99. On capital flows and the balance-of-payments constrained
Ansari, H., Xi, Y., 2000. The chronicle of economic growth in Southeast Asian countries: growth model. Journal of Post Keynesian Economics 21.
does Thirlwall's law provide an adequate explanation. Journal of Post Keynesian Moreno-Brid, J.C., 2003. Capital flows, interest payments and the balance-of-payments
Economics 22 (4), 573–588 (summer). constrained growth model: a theoretical and empirical analysis. Metroeconomica
Atesoglu, H.S., 1993a. Balance of payments constrained growth: evidence from the 54 (2–3), 346–365.
United States. Journal of Post Keynesian Economics 15, 507–514 (Summer). Razmi, A., 2005. Balance of payments constrained growth model: the case of India. De-
Atesoglu, H.S., 1993–4b. Exports, capital flows, relative prices and economic growth in partment of Economics, University of Massachesetts Amherst, Working Paper
Canada. Journal of Post Keynesian Economics 16 (2), 289–297 (Winter). 2005-05.
Atesoglu, H.S., 1994. An application of a Kaldorian export-led model of growth to the Thirlwall, A.P., 1979. The balance of payments constraint as an explanation of international
United States. Applied Economics 26, 479–483 (January). growth rate differences. Banca Nazionale del Lavoro Quarterly Review 45–53.
Bairam, E.I., 1988. Balance of payments, the Harrod foreign trade multiplier and eco- Thirlwall, A.P., Hussain, M.N., 1982. The balance of payments constraint, capital flows,
nomic growth: the European and North American experience, 1970–85. Applied and growth rate differences between developing countries. Oxford Economic Pa-
Economics 20, 1635–1642. pers 34 (10), 498–509.