You are on page 1of 10

World Journal of Economic and Finance

Vol. 4(1), pp. 075-084, January, 2018. © www.premierpublishers.org. ISSN: XXXX-XXXX WJEF

Research Article

Asymmetric Co-integration between Exchange Rate and


Trade Balance in Thailand
Saad Buba1, Salim Mohammed Ibrahim Al-Jadi2 Garba Mohammed Guza3
1
Adamawa State Polytechnic, Yola, Nigeria.
2
Faculty of Economics, Sabha University, Libya.
3
Mai Idriss Alooma Polytechnic, Geidam, Nigeria.

This paper empirically examines the long-run exchange rate pass-through into trade balance in
Thailand. The study incorporates political stability in the short run model to ascertain its effect
on the trade balance. Asymmetric co-integrating adjustment method proposed by Enders and
Siklos (2001) is employed for the study. The empirical findings revealed that there exists an
asymmetric cointegration relationship between exchange rate and trade balance as well as
exchange rate and imports & exports volumes after conducting a momentum-threshold
autoregressive (M-TAR) and threshold autoregressive (TAR) tests respectively. The results of the
short run effects showed that political stability has no meaningful effect on trade balance of
Thailand. The findings have further shown that changes in real exchange rate have contributed
to the presence of trade balance deficit in Thailand during the period under study; which is most
likely to be as a result of massive imports of crude oil between the late 1990’s and through to
2010.

Keywords: Asymmetric adjustment; Exchange rate; Trade balance; Thailand; Import and Export

INTRODUCTION

In Thailand, just like in other developing countries, the In 1999, after the economic and financial crisis in 1997,
trade deficit is one of its major hiccups. Import Prices relate effort has been made by the Royal Thai Government to
to the rate of variation in the prices of goods and services stabilize the economy of Thailand. In 2003, the economy
bought by people of the country from foreign sellers and was set back on track; the exchange rate stability has been
delivered by same foreign suppliers. A question may be achieved, (Trade Policy Review report, 2003). Between
asked, do import prices greatly affected by exchange 2000 and 2009, the average exchange rate of Baht to US
rates. In January 2015 import prices in Thailand were dollar depreciated and the trade deficit emerge due to the
91.65 index points, one month later, it declined to 91.41 absence of effective fiscal policy. The economist report
Index Points. Import prices in Thailand, from 2000 to 2015, 2014 revealed that Thailand would remain the slowest
averaged at 88.46 Index Points, getting an all-time, high to growing economy in South-East Asia, probably till 2017.
103.35 Index Points in January 2014 and the lowest at Melitz (2003), in his model of heterogeneous firms,
67.79 Index Points in January 2000 (Bank of Thailand, predicts that a change in trade conditions causes a change
2015 report). The response of international prices to in the variety of goods that are traded.
changes in exchange rate remains a central issue, the
exchange rates pass-through to import prices dictates the *Corresponding author: Saad Buba, Adamawa State
degree of change to the current account and international Polytechnic, Yola, Nigeria. E-mail:
transmission of inflation, (Cook, 2014). saadbubadamare12@gmail.com

Asymmetric Co-integration between Exchange Rate and Trade Balance in Thailand


Buba et al. 076

This can be explained by the fact that when trade costs This study is theoretically, based on the standard theory of
decreased, firms that were unable to export before the international trade by Hecksher-Ohlin (1990’s), which
decrease will now do so. But the new firms’ productivity will states that changes in real exchange rate yield impacts on
be lower or will have higher marginal cost than the firms both the value and volume of trade; the higher the real
already exporting before the decrease. In addition, the new exchange rate for the home country, the more the trade
firms with lower productivity also have a pricing behavior surplus the country obtains and vise-versa. The major
that will determine how aggregate pass-through to a different between this study and Duasa (2009) besides the
country’s import prices changes resulting from a decrease choice of country of study is the ability of the current study
in trade costs. to incorporate political stability in the short-run model and
the idea employed by the study to conduct robust checks
In recent years, Thailand has recorded a fall in its trade using fully modified ordinary least squares (FMOLS) and
balance from USD 36.6 billion in 2009 to USD 6.8 billion in the dynamic ordinary least squares (DOLS).
2013, making a short of USD 29.8 billion (Focus
Economics, 2015). A question may be asked; is it the
depreciation of Baht to US dollar that contributed to this RELATED LITERATURE
trade balance deficit faced by Thailand? This study will
examine the exchange rate pass-through into to trade There is various literature concerning trade balance,
balance in Thailand. To the best of our knowledge, there Kayhan et al. (2013) have examined the causality between
is no study on Thailand that examined exchange rate pass- the trade deficit and government expenditure in the
through into trade balance using asymmetric cointegration Turkish economy; they employed Toda-Yamamoto
with threshold autoregressive (TAR) and momentum causality (bootstrap process based) and frequency domain
threshold autoregressive (M-TAR). It is imperative, analysis method, their findings revealed a bi-directional
therefore, to study this pass-through of exchange rates causality between the trade deficit and government
into trade balance in Thailand using asymmetric expenditure. This is evident that each method applied in
cointegration approach proposed by Enders and Siklos their study is robust; government expenditure affects trade
(2001). deficit in the shorter period, whereas trade deficit affects
government expenditure even in the long run. They
Previous studies have used cointegration analysis to conclude based on theoretical and empirical interpretation
examine the relationship between exchange rates and that reduction in government expenditure will help reduce
trade balance such as the Engle-Granger (1987), two-step the trade deficit. A more recent study was conducted by
cointegration analysis and Johansen Juselius (1999), to Prakash and Maiti (2016) who investigated the devaluation
examine the long run relationship between exchange rate effectiveness in improving trade balance in Fiji. They used
and trade balance, see (Hsing 2009), Wong and Tang econometric models to empirically examine devaluation
(2007), Bagchi et.al (2004), Khan (2005). The Engle- impact; they found that real exchange rate has a strong
Granger (1987) and Johansen Juselius (1999), assumed relationship with trade balance in the long run.
symmetric cointegration; they show the changing Appreciation of foreign currency is responsible for the
coefficient values related either negatively or positively in rising trade deficit. This is due to Fiji’s over-dependent on
the symmetry error. Studies dealing with symmetry imports and its failure to cure import bills by raising
relationship do not consider the account of asymmetric domestic export. Another recent study by Chiu and Sun
adjustment that occurs in the endogenous variables, (2016) examines the relationship between trade balance,
(Duasa, 2009). Precisely, Cook (2006) argued that Engel savings and exchange rates in a panel of 76 countries. The
and Granger (1987) and Johansen (1988) were unable to study uses panel smooth transition regression model with
identify the cointegration between stock prices and instrumental variables to address the potential non-linear
industrial production for a long span of U.S data. However, effects of the savings rate. They found that countries with
Narayan (2007) has employed asymmetric approach using savings rate above 14.8 percent threshold can induce their
TAR and M-TAR for nominal exchange rates and price trade balance by increasing the savings rate.
levels for Canada, the UK, Japan, Germany, Italy and
France (G6) vis-à-vis the US dollar. Shen et.al (2007) has Aliyu and Tijjani (2015) examined the long-run exchange
employed the asymmetric cointegration test proposed by rates pass-through into trade balance in Nigeria by means
Enders and Siklos (2001) to examine the long run of threshold cointegration and asymmetric error correction
asymmetric equilibrium relationships between the Chinese modelling. Niemenien (2015) found that the introduction of
Shanghai and Shenzhen stock markets. The asymmetric a common currency in the EU area did not increase the
relationship entails differentiating positive and negative elasticity of net capital flows to per capita income; hence
effects of the error obtained from the cointegration the understanding of imbalance in the Euro area
regression, (Enders and Siklos, 2001). Asymmetry has increased. Interestingly, countries with the most
been a significant matter in macroeconomics analysis with sophisticated financial markets have had positive intra
findings giving proof of the existence of asymmetric balances and negative extra balances. Whalley and Wang
changes of most of the macroeconomic indicators. (2011) found that the effect of Renminbi revaluation on
Asymmetric Co-integration between Exchange Rate and Trade Balance in Thailand
World J. Econ. Fin. 077

surplus is proportionally larger than on trade flows; elasticity for imports. We rewrite equations (1) and (2) by
changes in trade flows can be substantial. China’s taking their natural log:
processing trade was incorporated into their model, they
found that the impact of revaluation on the surpluses are 𝑙𝑛𝑋𝑡 = 𝜂[ 𝑙𝑛𝑃𝑡 − 𝑙𝑛𝑃𝑡∗ − 𝑙𝑛𝐸𝑡 ] + 𝜀𝑙𝑛𝑌𝑡∗ (3)
elasticity dependent; huge substitution elasticities between
domestic and foreign production in demand in China yield 𝑙𝑛𝑀𝑡 = 𝛾[𝑙𝑛𝑃𝑡∗ − 𝑙𝑛𝐸𝑡 − 𝑙𝑛𝑃𝑡 ] + 𝑧𝑙𝑛𝑌𝑡 (4)
larger effects on trade, its flow and surpluses. Devaluation
of US dollar deteriorates her bilateral trade balance with Where 𝑙𝑛𝑒𝑡 = [𝑙𝑛𝑃𝑡∗ + 𝑙𝑛𝐸𝑡 − 𝑙𝑛𝑃𝑡 ] designates natural log
13 trading partners, but improves it with 37 trading of real exchange rate. Under normal condition, balance of
partners, especially for China, and the U.S. trade trade (TB) is the ratio of exports to imports or otherwise
imbalance can be eliminated by way of dollar devaluation and its equation could be written as in Duasa (2009).
in the long run, (Chiu et al. 2010). Contrary to many
empirical studies but consistent with the theoretical 𝑙𝑛𝑇𝐵𝑡 = 𝑧𝑙𝑛𝑌𝑡 + 𝜀𝑙𝑛𝑌𝑡∗ + 𝜃𝑙𝑛𝑒𝑡 (5)
expectation regarding the effect of the contractionary
policy, Ivrendi and Guloglu (2010) found that monetary Where 𝜃 = - (𝜂 + 𝛾). The amount of 𝑙𝑛𝑒𝑡 shows the
policy affects the level of price, domestic demand and real condition of Marshal-Lerner (M-L), whether achieved or
exchange rate in the short run. This is consistent with both not. The M-L condition occur when exchange rate
Keynesian and Monetarist approaches. Other studies on devaluation will improve balance of trade as the sum of
this matter also include Chang (2008), Ghoshray (2008), long-term export and import demand elasticities less than
Ewing et.al (2006), Yau and Nieh (2008), and Heimonen one, (Davidson, 2009). In order to achieve the M-L
(2006), Duasa (2009) among others. condition, 𝜂 and 𝛾 are expected to be non-positive then
𝜀 and 𝑧 are expected to be non-negative hence, M-L holds
The inability to compel into consideration, the currently at any time 𝜃 is non-negative showing that rate of
acknowledged evidence of the existence of asymmetric exchange is higher.
changes of macroeconomic variables, may lead to
inappropriate interpretations (Duasa 2009). This is Empirical Approach
because according to Balke and Fomby, (1997), the
movement towards the long-run steady state will not be Enders and Siklos (2001), have expanded Engle-Granger
essentially constant, Hence the need for this study. The two-step cointegration tests, to integrate an asymmetric
rest of the work of this paper is structured as follows: the error correction term, assumed two series{𝑦𝑡, 𝑥𝑡 }, The
theoretical framework is presented in the next section, that ordinary least squares (OLS) is employed in the first step
is section 3, empirical methods are presented in section 4; to examine the symmetric correlation among the 𝑦𝑡 and 𝑥𝑡
the highlights of the empirical findings and analysis are in the long-run; thus:
presented in section 5. Section 6, the final section 𝑦𝑡 = ģ0 + ģ1 𝑥𝑡 + 𝜀𝑡 (6)
concludes and draws policy recommendation from the
major findings. Where ģ0 and ģ1 designate the parameters to be estimated
and 𝜀𝑡 is an error term. The conceivable cointegration
Theoretical Framework among 𝑦𝑡 and 𝑥𝑡 is then estimated through the order of
incorporation of the residuals 𝜀𝑡 in equation (6) by means
In this section, we provide an overview of the theoretical
of a Dickey-Fuller test as follows:
relationship between exchange rate and trade volumes
(import & export volumes). Rose and Yellen (1989) and
∆𝜀̂𝑡 = 𝜌𝜀̂𝑡−1 + 𝑣𝑡 (7)
Rose (1990) presented a model which specified in a
reduced form, the performance of trade balance in a given
economy as equation (1) and (2) below. The model is By way of suitable degree of augmentation put in place
presented based on the real exchange rate and real through the insertion of lagged values of the endogenous
incomes generated within and outside the economy. variable, (𝐻0 : 𝜌 = 0) as null hypothesis of no cointegration
𝜂 can be properly examined through the assessment of the
𝑋𝑡 = (𝑃𝑃∗𝐸) . (𝑌𝑡∗ )𝜀 (1)
𝑡. t-ratio of the changing parameter 𝜌 and precisely produced
𝑃∗ 𝐸 𝛾 unusual critical values. Though, according to Enders and
𝑀𝑡 = ( ) . (𝑌𝑡 ) 𝑧 (2) Siklos (2001), the misspecification of Engle-Granger
𝑃 𝑡
method occurs when the estimated time series have a
According to the models, X denotes exports, M is imports, fundamental asymmetric correlation. Hence, stationary
E is the minimum exchange rate (as in for example Duasa testing of the error term is done by integrating the
(2009)) and P, 𝑃 ∗ and Y, 𝑌 ∗ indicate the domestic and asymmetric changes. A model was proposed by Enders
foreign levels of prices and incomes; 𝜂 and 𝛾 denote the and Siklos (2001) in the second step as follows:
elasticities of the rate of real exchange for exports and
imports, respectively, while ́ 𝜌́ ∆𝜀̂ + 𝜑
∆ε𝑡 = 𝐼𝑡 𝜌1 𝜀̂𝑡−1 + (1 − 𝐼𝑡 )𝜌2 𝜀̂𝑡−1 + ∑𝑘𝑖=1 (8)
𝑡−𝑖 𝑡
𝜀 stands for export ′ s elasticity of income, z is the income
Asymmetric Co-integration between Exchange Rate and Trade Balance in Thailand
Buba et al. 078

Where 𝜌1 𝜌2 and 𝜌́ 𝑖 are coefficients; 𝜑𝑡 is a white-noise if 𝑦𝑡−1 is above its long run equilibrium value(= 𝛾0 +
disturbance; k is the number of lags; and 𝐼𝑡 is an indicator 𝛾1 𝑥𝑡−1 ), but 𝜌2 if 𝑦𝑡−1 is below the equilibrium.
function such that: The error-correction model of the asymmetric
1𝑖𝑓𝜀𝑡−1 ≥ 0 correspondingly exists for 𝑦𝑡 and 𝑥𝑡 once they are made in
𝐼𝑡 = { (9) an asymmetric cointegration relationship. Thus:
0𝑖𝑓𝜀𝑡−1 < 0
∆𝑦𝑡 = 𝛼0 + 𝜃11 𝑀𝑡 𝜀𝑡−1 + 𝜃12 (1 − 𝑀𝑡 )𝜀𝑡−1 +
Tong (1983, 1990) revealed that the OLS estimates of 𝜌1
∑𝑘𝑖=1 𝛼1𝑖 ∆𝑦𝑡−𝑖 + ∑𝑘𝑖=1 𝛼2𝑖 ∆𝑥𝑡−𝑖 + 𝜗1𝑡 (11)
and 𝜌2 contained an asymptotic multivariate normal
distribution. Equations 6, 8 and 9 are cointegration model And
and are called the Enders and Siklos (2001)’s threshold ∆𝑥𝑡 = 𝛽0 + 𝜃21 𝑀𝑡 𝜀𝑡−1 + 𝜃22 (1 − 𝑀𝑡 )𝜀𝑡−1 +
autoregressive (TAR) model. It remains imperative to
∑𝑘𝑖=1 𝛽1𝑖 ∆𝑦𝑡−𝑖 + ∑𝑘𝑖=1 𝛽2𝑖 ∆𝑥𝑡−𝑖 + 𝜗2𝑡 (12)
know that the indicator function 𝐼𝑡 is influenced by the rate
of 𝜀𝑡−1 in equation (9).
Where 𝜃11 and 𝜃12 signify the adjustment coefficients’
speed of ∆𝑦𝑡 if 𝑦𝑡−1 is above and below its long run steady
An alternative threshold has been suggested by Enders
state, respectively. Likewise, 𝜃21 and 𝜃22 denote the
and Siklos (2001) and Enders and Granger (1998). This
adjustment coefficients’ speed of ∆𝑥𝑡 of the two systems,
suggested threshold depends on the adjustment in 𝜀𝑡−1 in
respectively, 𝛼0 and 𝛽0 are the constant terms, 𝛼1𝑖, 𝛼2𝑖, 𝛽1𝑖
the preceding period and consequently, the new indicator
𝑀𝑡 is emerged as: and 𝛽2𝑖 are coefficients of lagged change terms, and 𝜗1𝑡
and 𝜗2𝑡 are white-noise disturbances.
1𝑖𝑓∆𝜀𝑡−1 ≥ 0
𝑀𝑡 = { (10)
0𝑖𝑓∆𝜀𝑡−1 < 0 In addition, we apply the Granger causality test in order to
observe the lead-lag correlation among 𝑦𝑡 and 𝑥𝑡 . The null
The model that contains equations 6, 8 and 10, is devoted hypothesis; 𝑥𝑡 does not lead 𝑦𝑡 is H0: 𝛼2𝑖 = 0, i = 1,……..,k
to the momentum threshold autoregressive (M-TAR) and the null hypothesis that 𝑦𝑡 does not lead 𝑥𝑡 is H0: 𝛽1𝑖 =
model. The threshold (τ) value is set to 0 in equations 9 0, i = 1,…….,k.
and 10. When the threshold (τ) is not known; the use of
grid process is suggested by Enders and Siklos (2001), in
order to come up with a consistent and dependable EMPIRICAL RESULTS AND ANALYSIS
estimate of the threshold. To be precise, we bear in mind
that TAR model, the residual series {𝜀̂𝑡 } is orderly and Data description and unit root test
arranged as {𝜀̂10 < 𝜀̂20 <. … . < 𝜀̂𝑇0 }. After thrown away 15
percent of the largest and smallest {𝜀̂𝑡 }, the dominant The observed series of trade variables in this study are all
observations which is 70 percent of the array are formally together, collected from the Bank of Thailand (BOT)’s data
reflected as thresholds in equations (8) and (9) as every and statistics database. A monthly data of fifteen years,
single of the observation could be the conceivable from M1: 2000 to M12: 2014; see Chen, et al. (2005) and
threshold. The estimated threshold considered yielding the Ibrahim (2011) on the use of the data sample size. The
lowermost residual sum of squares stood to be believed as measure of the exchange rate is the exchange rate of
the right estimate of the threshold. A related method can commercial banks in Bangkok, with the base year of 2000.
be applied for M-TAR model. With 70 percent of The imports and exports volumes are both in indices with
observations of the sequence {𝜀̂10 < 𝜀̂20 <. … . < 𝜀̂𝑇0 } 2000 as the base year, and finally, the balance of trade is
measured as values of the threshold for equation (10), basically the difference between export volume and import
possibly, the value may offer least sum of residual squares volume, (Luo, Lv & Zhang 2010). Imports and exports
emanating from the estimation of equations (8) and (10), values are in log form, except for the trade balance.
is then described as the consistent threshold. Political stability data were taken from World Bank’s world
governance indicator (WGI); an annual data resampled to
To check for asymmetric cointegration; firstly, we monthly using Eviews.
determine whether or not 𝑦𝑡 and 𝑥𝑡 are cointegrated in the
TAR and M-TAR models. We carry out this test by means The first test conducted in the unit root test; breakpoint unit
of F test meant to investigate the null hypothesis of no root, in particular, using a modified Dickey-Fuller test with
cointegration, H0: 𝜌1 = 𝜌2 = 0. Enders and Siklos (2001) Eviews 9 software. It carries out a test which allows for a
denoted Ф as the F statistics and it has a non-standard structural break in the process. The latest Eviews 9
distribution. Secondly, the null hypothesis H0: 𝜌1 = 𝜌2 can supports the computation of modified Dickey-Fuller tests
be tested using the standard F-statistics, however this can which allow for levels and trends that differ across single
be done in the existence of asymmetric cointegration. The break date. The framework follows the work of Perron
evidence supporting the asymmetric adjustment of the (1989), Perron and Vogelsang (1992), Vogelsang and
error correction term is specified when both H0: 𝜌1 = 𝜌2 = Perron (1998), Banerjee, et al. (1992).The results of the
0 and H0: 𝜌1 = 𝜌2 are rejected. Possible adjustment is 𝜌1 unit root tests are displayed in Tables 1 and 2 below.

Asymmetric Co-integration between Exchange Rate and Trade Balance in Thailand


World J. Econ. Fin. 079

Table 1: Breakpoint Unit Root Tests Model 1: 𝑡𝑏𝑡 = 𝛼0 + 𝛼1 𝑟𝑒𝑟𝑡 + 𝜀𝑡 (13)


Minimize Dickey-Fuller test statistics Model 2: 𝑥𝑡 = 𝛽0 + 𝛽1 𝑟𝑒𝑟𝑡 + 𝜀𝑡 (14)
Variable Level First Difference Model 3: 𝑚𝑡 = 𝜔0 + 𝜔1 𝑟𝑒𝑟𝑡 + 𝜀𝑡 (15)
Tb -3.9798 -6.6659***
X -3.6053 -5.2154** Model 1 forms the relationship between trade balance and
M -3.1415 -5.9743*** real exchange rate; Model 2 institutes the link between real
rer -3.9745 -10.128*** exchange rate and volume of exports while Model 3
establishes the link between real exchange rate and the
Note: *, **, and *** indicate level of significance at 10%, 5% volume of imports. The results of these long-run
and 1% level, respectively. relationships are presented in Table 3.
Table 1 above showed that all the variables are stationary Table 3: Long-run model
at first difference and at 1 percent significant level. This
means that the data are not spurious and no presence of Equation/Model Dependent Variables
unit root. Variable Ln tb Ln x Ln m
Constant -0.79*** 21.21*** 22.00***
Table 2 Breakpoint Unit Root Test (-4.25) (64.63) (56.65)
Ln rer 0.22*** -2.31*** -2.54***
Minimize Intercept break test statistics
(4.32) (-25.36) (-23.50)
Variable Level First Difference
Observation 180 180 180
Tb -4.6409* -16.600***
Adjusted R-square 0.08 0.78 0.75
X -2.9157 -4.7754**
F-statistic 18.69*** 6.43*** 5.52***
m -4.0771 -22.1629***
rer -4.1997 -9.5069***
Before testing for the existence of asymmetric
Note: *, **, and *** indicate level of significance at 10%, 5% cointegration, we conduct the Engle-Granger (1987)’s two-
and 1% level, respectively step cointegration test; based on equation (7), the results
of these tests are presented in Table 4.
Our test results for two structural breaks (minimize Dickey-
Fuller t-statistics and minimize intercept break t-statistics) Table 4: Engle-Granger ADF Cointegration tests
showed statistics coefficient with p-values less than 0.01
all at first difference except for trade balance under Model t-statistic Critical values
minimize intercept break t-statistics. This has led us to 1% 5% 10%
reject the null of a unit root. The break dates and the effect (tb) 1 -17.069 -2.578 -1.942 -1.615
sign of break dummy variable on each of our variables and (x) 2 -4.218 -2.578 -1.942 -1.615
their significance levels were also provided. (m) 3 -19.759 -2.578 -1.942 -1.615

For real exchange rate, the break date is 2001, the fifth The Engle-Granger cointegration test results (Table 4)
month is shown by both Dickey-Fuller t-statistics and above reveal the existence of long-run relationships
autoregressive coefficient graphs; the effect of break between real exchange rate and trade balance, exports
dummy variable on the exchange rate is negative but not volume & imports volume in models 1, 2 and 3
significant. The import volume has its break date in 2012, respectively. With this, therefore, the null hypothesis of no
third month; the effect of its break dummy on it was positive cointegration is hereby rejected at 1% significance levels.
(0.331) and significant at one percent. The break date for Thus there might be a chance to discover an asymmetric
export volume is 2008, seventh month; the effect of its cointegration in the model without further prove from
break dummy variable is positive (0.1748) and significant Johansen Jusellius (1990) cointegration test. We now
at five percent. The trade balance has a break date in explore further analysis called threshold cointegration
2009, second month and the effect of the break dummy using TAR and M-TAR for the three models. The results
variable on it is also positive but insignificant. Lastly, are presented in Table 5.
political stability has a break date in 2007, sixth month; the
effect of its break dummy variable is positive (1.466) and Table 5: Enders-Siklos asymmetric cointegration tests
significant at ten percent. See Appendix 1 and 2 for break H0: 𝜌1 = 𝜌2 = 0 H0: 𝜌1 = 𝜌2
dates and break dummy variables effects. TAR: Ф k M-TAR: k TAR: M-TAR:
Ф* F test F test
The long run relations Model 7.48** 1 9.52** 1 0.218 3.98
1(tb)
We present the following functions which are considered
Model 2(x) 14.43*** 1 11.38*** 1 6.54** 1.12
suitable for long run correlation between exchange rate
Model 12.03*** 1 11.01*** 1 5.02 3.18
and trade balance, export volume & import volume, as in 3(m)
the work of Duasa (2009):
Asymmetric Co-integration between Exchange Rate and Trade Balance in Thailand
Buba et al. 080

Note: k Notation is the default lag periods of lagged After the estimate of M-TAR error-correction model of
difference term, ** indicates significance at 5% level, the trade balance, we found that long-run relationship exists
critical values of Ф and Ф* statistics are given in Enders and between real exchange rate and trade balance and with a
Siklos (2001). F indicates F statistic for the null hypothesis causal adjustment manner being asymmetric. Indeed, the
of symmetric adjustment, 𝜌1 = 𝜌2 . rate of the adjustment parameters show that as trade
balance and real exchange rate set off from their original
In Table 5 above, we have found, using model 1, steady-state temporarily, the movement or adjustment
asymmetric cointegration occurs between trade balance back to the steady state, is faster when there is relative
and real exchange rate in both TAR and M-TAR under the increase in trade balance (above long-run value),
joint null hypothesis. Exhausting model 2 and 3, we also compared to relative decrease in trade balance (below
found asymmetric cointegration between exports demand, long-run value). We explain this scenario in the following
imports demand and real exchange rate respectively, models.
using TAR model as Ф and M-TAR model as Ф* statistic, ∆𝑡𝑏𝑡 = 𝐾 − 0.442[𝑡𝑏𝑡−1 + 0.79 − 0.22𝑟𝑒𝑟],
both significance at 1% and TAR F-statistic of Wald for 𝜀̂𝑡−1 > 0 (16)
coefficients tests is significant at 5% level. All results are ∆𝑡𝑏𝑡 = 𝐾 − 0.269[𝑡𝑏𝑡−1 + 0.79 − 0.22𝑟𝑒𝑟],
significant at 5% and 1 % levels. for 𝜀̂𝑡−1 < 0 (17)
Threshold error-correction models
In the above models, K represents constant and lagged
changes in equations 11 and 12. The specification of M-
Having confirmed the existence of asymmetric
TAR disclosed a return to an underlying equilibrium
cointegrations between real exchange rate and all the
relationship is faster, 44% adjustment speed, as a short-
trade variables used in this study, we now go further to
term movement from equilibrium relationship, are
estimate TAR and M-TAR error correction models so as to
instigated by relative increases in the trade balance, or
evaluate the short run dynamically. We set the maximum
equally, decreases in the real exchange rate. However, the
lag order of first differenced variables to 12. The lag order
return to the original equilibrium is only 27% as the
is to be trimmed down if and only if the last lag is not found
temporary moves from equilibrium point are triggered by
significant at least 10% level. Table 5, Table 6 and Table
comparative decreases in the trade balance, or equally,
7 present the results; M-TAR for trade balance model and
increases in the real exchange rate.
TAR for export and import models respectively. All the
three models are diagnosed with robust evidence.
From the investigations made, an inference is drawn that
there is an evidence of deficit trade balance in Thailand.
Table 6: Momentum Threshold Autoregressive (M-
This can be explained by the fact that the adjustment of
TAR) Error Correction Model for trade balance
trade balance back to the long run value is slower when it
Independent Variable Dependent Variable: (M- faces deficit than when it faces surplus. The apprehension
Equation TAR) ∆𝒕𝒃 is that a trade deficit that lasts long can lead to more
Constant -0.003 foreign debt as the country pays more than its earnings,
𝑀𝑡 𝜀𝑡−1 -0.442*** and eventually devaluation of the currency. Thailand
(1 − 𝑀𝑡 )𝜀𝑡−1 -0.269*** historically has been having a negative trade balance with
∆𝑡𝑏𝑡−1 -0.308*** the rest of the world for most of the times. If a trade balance
∆𝑡𝑏𝑡−2 -0.281*** surplus is present, it remains relatively small especially the
∆𝑡𝑏𝑡−9 0.244*** period between 2001 and 2010. Though if not for crude oil
∆𝑟𝑒𝑟𝑡−1 0.023* imports, the trade balance would have been positive all the
∆𝑟𝑒𝑟𝑡−10 -0.026** time. There was also a noticeable striking trade deficit in
Adjusted R-square 0.420 the 1995 – 2000 periods, as a result of Asian economic
F-statistics 18.128*** crisis. Prior to the said crisis, Thailand was importing a lot
Diagnostic tests: of capital goods; which was a good thing to reckon,
Serial correlation 0.867 however, it is obvious that too much of a good thing turned
Heteroskedasticity 0.379 out to be bad at the time. These trade deficits occurred
Wald test: despite various trade and investment policies adopted by
F-statistics 8.690*** the government of Thailand, such as export promotion
Chi-square 43.45*** policies that have to do with refinancing facilities,
Note: ***, ** and * denote 1%, 5% and 10% significance investment promotion for exports and tax refunds. These
levels respectively policies date back from the 1960s through to 1990s.

Asymmetric Co-integration between Exchange Rate and Trade Balance in Thailand


World J. Econ. Fin. 081

Table 7: Threshold Autoregressive (TAR) Error


Correction Model for export demand Table 8: Threshold Autoregressive (TAR) Error
Correction Model for import demand
Independent Variable Dependent Variable
Equation (TAR) ∆𝒙 Independent Variable Dependent Variable:
Constant 0.003 Equation (TAR) ∆𝒎
𝐼𝑡 𝜀𝑡−1 -0.389*** Constant 0.013**
(1 − 𝐼𝑡 )𝜀𝑡−1 -0.095** 𝐼𝑡 𝜀𝑡−1 -0.095**
∆𝑥𝑡−1 -0.215*** (1 − 𝐼𝑡 )𝜀𝑡−1 -0.385***
∆𝑥𝑡−12 0.425*** ∆𝑚𝑡−1 -0.420***
∆𝑟𝑒𝑟𝑡−1 1.117*** ∆𝑚𝑡−7 -0.118*
Adjusted R-square 0.475 ∆𝑚𝑡−9 0.119*
F-statistics 31.024*** ∆𝑚𝑡−11 -0.154**
Diagnostic tests: ∆𝑟𝑒𝑟𝑡−3 -0.724*
Serial correlation 0.4870 Adjusted R-square 0.332
Heteroskedasticity 0.4077 F-statistics 12.873***
Wald test: Diagnostic tests:
F-statistics 30.056*** Serial correlation 0.992
Chi-square 90.167*** Heteroskedasticity 0.167
Wald test:
Note: *** and ** denote 1% and 5% significance levels
F-statistics 9.548***
respectively.
Chi-square 47.74***
From Table 7, we noticed the existence of long-run Note: ***, ** and * denote 1%, 5% and 10% significance
cointegration between real exchange rate and export levels respectively.
volume. We also observed a causal adjustment manner
being asymmetric. When export volume and real In Table 8, we have import demand model for TAR error-
exchange rate depart temporarily from the state of correction, the model also advocates speedy adjustment
equilibrium relationship, adjusting back to steady of import demand, and however, the adjustment is faster
equilibrium state is faster resulting from a comparative when below long-run value. On the other hand, the model
increase in export volume (above long-run value) as provides very little evidence concerning adjustment in
compared to a relative decrease in export demand which import demand when it is above long-run value. This can
has very low percentage point of 9%. This means that the be demonstrated as follows:
error-correction export demand model provides a
∆𝑚𝑡 = 𝐾 − 0.095[𝑚𝑡−1 − 22.00 + 2.54𝑟𝑒𝑟𝑡−1 ],
somewhat meaningless evidence for the adjustment of
export demand when it is below long-run value. We have for 𝜀̂𝑡−1 > 0 (20)
the following as an illustration for this non-meaningfulness. ∆𝑚𝑡 = 𝐾 − 0.385[𝑚𝑡−1 − 22.00 + 2.54𝑟𝑒𝑟𝑡−1 ],
for 𝜀̂𝑡−1 < 0 (21)
∆𝑥𝑡 = 𝐾 − 0.389[𝑥𝑡−1 − 21.21 + 2.31𝑟𝑒𝑟𝑡−1 ],
for 𝜀̂𝑡−1 > 0 (18) Where K, expresses other terms, which is lagged changes
∆𝑥𝑡 = 𝐾 − 0.095[𝑥𝑡−1 − 21.21 + 2.31𝑟𝑒𝑟𝑡−1 ], and constant in imports and real exchange rates. The
for 𝜀̂𝑡−1 < 0 (19) assessed coefficient of error correction recommends that
38% deviation of imports in last-period from its long-run
Where K, signifies other terms, which is constant and value will be corrected by imports change. Specifically, the
lagged changes in export and real exchange rates. Error- values of the conformity parameter showed that when real
correction coefficients estimate suggest that return to the exchange rate and imports coincidentally move from their
underlying equilibrium correlation is faster, with 39% underlying equilibrium relationship, alteration back to
adjustment speed as temporary departures from the balance is faster after a relative reduction in imports or
original equilibrium is caused by a relative increase in similarly, an increment in the real exchange rate. On the
export demand. On the other hand, the return to the other hand, imports adjustment is found to be less
underlying equilibrium is just as low as 9% speed of meaningful when it is above its long-run value. This result
adjustment as temporary departures from it are triggered is in opposite with the one in Table 6, and it shows that a
by relative decreases in export volume or consistently, shock of exchange rate on import volume is possible to be
increases in the real exchange rate. This result confirmed temporary in nature. To be precise, the increment in
persistent evidence of trade balance deficit observed in exchange rate has the same impact on both exports and
Table 5. imports demand in Thailand.

Asymmetric Co-integration between Exchange Rate and Trade Balance in Thailand


Buba et al. 082

We have also included a control variable of political analysis to examine the interdependence among
stability in our short-run model, using autoregressive variables. One of such methods is implemented from
distributed lag (ARDL), in order to capture its effects on the Engle-Granger (1987), which assumes symmetric
trade balance in Thailand. We found that, although, the adjustment of the error correction term. For symmetric
variable has a negative effect the effect is not significant. cointegration, adjustment coefficients are alike,
irrespective of their positivity or negativity in the equilibrium
Table 9: Short-Run Model error. This study tries to investigate the long run
equilibrium relationship between real exchange rate and
Regressors Coefficient Standard Error T-Ratio
trade balance, export demand as well as import demand
DLEXR -0.0259** 0.01109 -2.3417
using asymmetric cointegration framework advanced by
DLXVOL 0.3770*** 0.07061 5.3353
Enders and Siklos (2001). The study also incorporates
DLMVOL -0.3973*** 0.07124 -5.5779
political stability in its short run model as a control variable,
DPSTAB -0.4410 0.64620 -0.6823
to see whether the variable has any effect on Thailand’s
Dependent Variable DTB, R-Squared 0.99578, DW-
trade balance. This method is used because; it has been,
statistic 1.9979
nowadays, an important process in macroeconomic
analysis, with quite a number of studies providing clear
Robustness Checks
evidence of asymmetric adjustment of macroeconomic
variables.
It is imperative to have an additional support to our
findings. In doing this, we conduct two types of estimation;
In this study, we found that there exists a long run
the dynamic ordinary least square (OLS) and the fully
asymmetric cointegration between trade balance and real
modified ordinary least squares (FMOLS), to serve as
exchange rate when we conducted the M-TAR model.
robust checks to our findings. We presented the results of
Using TAR model, we also found the existence of
the robustness checks in Table 10 with trade balance as
asymmetric cointegration between real exchange rate and
the dependent variable. The coefficients of exchange rate
exports and imports volumes with the value of threshold
using DOLS is negative and significant at 10 percent which
not equal to zero (𝜏 ≠ 0). Therefore, based on the
is in line with our main findings. The coefficient is also
expectations and objective of this study, we report that
negative and significant at 5 percent using FMOLS. This
changes in the real exchange rate have contributed to the
confirms our findings that exchange rate causes negative
recent trade balance deficit came across by Thailand.
trade balance in Thailand, as a result of excess import over
export. The other two independent variables; import and
From the M-TAR model of the error correction for trade
export volumes showed a negative and a positive
balance, an inference can be drawn that long-run
relationship with trade balance respectively and are
relationship exists between trade balance and real
significant at 1 percent level. Political stability showed a
exchange rate with the underlying adjustment manner
non-significant negative effect on trade balance using both
being asymmetric. Basically, the value of the adjustment
FMOLS and DOLS.
parameters specifies that when trade balance and real
exchange rate depart from their core equilibrium
Table 10: Results for DOLS and FMOLS Estimates
relationship, then adjustment back to equilibrium is more
DOLS FMOLS speedy resulting from an increase in trade balance (above
Dep. Var. : Trade Lag=1, long-run value) as related to decreasing in trade balance
Balance lead=1 (below long-run value). The export model for TAR error
Exchange Rate -0.12* -0.022* correction suggests that when export volume and real
(0.06) (0.01) exchange rate momentarily depart from their equilibrium
Import Volume -1.00*** -1.00*** correlation, adjustment back to original equilibrium is more
(0.01) (0.01) rapid resulting from a relative increase in export volume
Export Volume 0.99*** 0.99*** (above long-run value) compared to a relative decrease in
(0.01) (0.01) export volume (below long-run value).
Political Stability -0.001 -0.004
(0.00) (0.00) As per import demand model for TAR error correction, the
model recommends rapid adjustment of import volume
Note: values in parenthesis are standard errors when it is below long-run value. The model also asserts
that returns or adjustment back to equilibrium point is quite
CONCLUSION slow when it is above long-run value and faster when it is
below long-run value. The results indicate the evidence
In examining the impacts of exchange rate shock on trade that there is the presence of fluctuations of trade balance
variables, most studies currently employed asymmetric deficit and surplus in Thailand, with much effect on trade
threshold cointegration framework. However, prior to this, balance deficit. This is because adjustment of trade
several studies commonly used symmetric cointegration balance back to its long-run value is quite slow when it
Asymmetric Co-integration between Exchange Rate and Trade Balance in Thailand
World J. Econ. Fin. 083

faces deficit than when it faces surplus. The study also Enders, W. & Granger, C.W. (1998). Unit-root Tests and
uses political stability in its short run model as part of the Asymmetric Adjustment with an Example Using the
explanatory variables to capture its effects on trade Term Structure of Interest Rates. Journal of Business
balance; however, its effect on trade balance was not a and Economic Statistics 16: 304-311
meaningful one. For the trade balance deficit, it is most Engle R. F; Granger C. W. J; (1987), Co-integration and
likely due to a policy of massive crude oil imports between Error Correction: Representation, Estimation and
late 1990,s through to 2010 which in turn led to the Testing; Econometricca, Vol. 55, No. 2 (Mar., 1987),
devaluation of Thailand’s currency. The shock of the Pp. 251-276
exchange rate, import demand might be more or less a Ewing, B.T., Payne, J.E., Thompson, M.A., Al-Zoubi, O.M.
temporary one. Therefore, to avoid a negative effect on (2006), Gevernment Expenditures and Revenues:
Thailand’s currency, it is suggested that policy adjustment Evidence from Asymmetric Modeling. Southern
is implemented to boost export and in addition reduce Economic Journal.
over-reliance on crude oil imports. Ghoshray, A. (2008). Asymmetric Adjustment of Rice
Export Prices: The Case of Thailand and Vietnam.
Journal of Applied Economics 5(2): 80-91
REFERENCES Heimonen, K. (2006). Nonlinear Adjustment in PPP-
Evidence from Threshold Cointegration. Empirical
Aliyu A.J., Tijjani S.M., (2015), Asymmetric Cointegration Economics. Vol 31, no. 2, 479-495
between Exchange Rate and Trade Balance in Nigeria: Hsing, Y. (2009). Test of the J-curve for six selected new
Cogent Economics and Finance (2015) 3: 1045213. EU countries.International Journal of Economic Policy
Bagchi, D., Chortareas, G.E. & Miller, S.M.(2004). The in Emerging Economies, 2(1), 76-85.
Real Exchange Rate in Small, Open, Developed Ibrahim, M.H. (2011) Inflation hedging effectiveness of an
Economies: Evidence from Cointegration Analysis. emerging Asian market: the case of Malaysia’, Int. J.
Economic Record. Vol. 80: 76-88 Economics and Business Research, Vol. 3, No. 5,
Balke, N.S. & Fomby, T.B. (1997). Threshold Ivrendi, M., & Guloglu, B. (2010). Monetary shocks,
Cointegration. International Economic Review 38: 627- exchange rates and trade balances: Evidence from
645 inflation targeting countries. Economic
Chang, S.C. (2008). Asymmetric Cointegration Modelling, 27(5), 1144-1155.
Relationship among Asian Exchange Rates. Johansen S; Jusselius K; (1999), Maximum Likelihood
Chen, L. H., Finney, M., & Lai, K. S. (2005). A Threshold Estimation and Inference on Co-integration with
Cointegration Analysis of Asymmetric Price Applications to the Demand for Money: Oxford bulletin
Transmission from Crude Oil to Gasoline Prices. of Economics and Statistics 1990; 52, 169-210.
Economics Letter 89: 233-239. Kayhan, S., Bayat, T., & Yüzbaşı, B. (2013). Government
Chiu, Y. B., & Sun, C. H. D. (2016). The role of savings expenditures and trade deficits in Turkey: Time domain
rate in exchange rate and trade imbalance nexus: and frequency domain analyses.Economic
Cross-countries evidence. Economic Modelling,52, Modelling, 35, 153-158.
1017-1025. Khan, H. (2005). Tourism and trade: Cointegration and
Chiu, Y. B., Lee, C. C., & Sun, C. H. (2010). The US trade Granger Causality Tests. Journal of Travel Research.
imbalance and real exchange rate: an application of the Vol. 4, no. 2: 171-176.
heterogeneous panel cointegration method. Economic Luo, L., Lv, W., & Zhang, R. (2010, August). An empirical
Modelling, 27(3), 705-716. study on balance of trade in China based on random
Cook, J.A. (2014). "The Effect of Firm-Level Productivity forest regression model. In Computer Science and
on Exchange Rate Pass-Through". Economics Letters Education (ICCSE), 2010 5th International Conference
122 (1): 27–30 on (pp. 561-564). IEEE.
Cook, S. (2006). The power of single equation tests for Melitz M.J., (2003), The Impact of Trade on Intra-Industry
cointegration. Applied Economics Letters, 13(5), 265- Reallocations and Aggregate Industry Productivity;
267. Econometrica, Vol. 71, No. 6. (Nov, 2003), pp. 1695-
Davidson, P. (2009). The Keynes solution: the path to 1725.
global economic prosperity. Macmillan. Narayan, P.K. (2007). Are Nominal Exchange Rates and
Duasa J., (2009), Asymmetric Cointegration between Real Price Levels Co-Integrated? New Evidence from
Exchange Rate and Trade Variables: The Case of Threshold Autoregressive and Momentum-Threshold
Malaysia; MPRA Paper, No. 14535, poated 9, April Autoregressive Models: Economic Record. Vol 83,
2009. Economic Change and Restructuring. Vol. 41, No.260: 74-85.
No. 2: 125-141. Nieminen, M. (2015). Trade imbalances within the euro
Enders W; Siklos P. L; (2001), Co-integration and area and with respect to the rest of the world. Economic
Threshold Adjustment: Journal of Business and Modelling, 48, 306-314.
Economics Statistics, Septmber, 2000.

Asymmetric Co-integration between Exchange Rate and Trade Balance in Thailand


Buba et al. 084

Prakash, K., & Maiti, D. (2016). Does devaluation improve Yau, H.Y. & Nieh, C.C. (2008). Testing for Cointegration
trade balance in small island economies? The case of with Threshold Effect between Stock Prices and
Fiji. Economic Modelling, 55, 382-393. Exchange Rates In Japan And Taiwan: Japan and
Rose, A.K. & Yellen, J. L. (1989). In There a J-curve?. World Economy. Article in Press. Corrected Proof.
Journal of Monetary Economics, 24: 53-68.
Rose, A.K. (1990). Exchnage Rates and the Trade Accepted 19 January 2018
Balance: Some Evidence from Developing Countries.
Economic Letters, 34: 271-275. Citation: Buba S, Al-Jadi SMI, Guza GM (2018).
Shen, C.H., Chen, C.F. & Chen, L.H.(2007). An Empirical Asymmetric Co-integration between Exchange Rate and
Study of the Asymmetric Cointegration Relationships Trade Balance in Thailand. World Journal of Economic
among the Chinese Stock Markets: Applied and Finance 3(2): 075-084.
Economics. Vol. 39, No. 11: 1433-1445.
The Bank of Thailand, Report Published 2015.
Tong, H. (1983). Threshold Models in Nonlinear Time-
Series Analysis: Springer-Verlag, New York.
Copyright: © 2018 Buba et al. This is an open-access
Tong, H. (1990). Nonlinear time-series: A Dynamical
article distributed under the terms of the Creative
approach: Oxford University Press, Oxford
Commons Attribution License, which permits unrestricted
Whalley, J., & Wang, L. (2011). The impacts of Renminbi
use, distribution, and reproduction in any medium,
appreciation on trade flows and reserve accumulation
provided the original author and source are cited.
in a monetary trade model. Economic Modelling, 28(1),
614-621.

Asymmetric Co-integration between Exchange Rate and Trade Balance in Thailand

You might also like