Examining the sustainability of current account deficit in India: Evidence from asymmetric error correction mechanism with threshold cointegration

The export sector has to be encouraged and the import sector has to be restricted by implementing tight import restricted policies (export promotion and import duties) and [r]

Examining the sustainability of current account deficit in India: Evidence from asymmetric error

correction mechanism with threshold cointegration

Lingaraj Mallick * †Abstract The study examines the sustainability of current account deficit in India from 1960 to 2014 with

an application of recent advanced non-liner cointegration method The study applies threshold cointegration test developed by Enders and Sikols (2001) for the purpose of examining the long-run relationship between exports and imports The findings reveal threshold cointegration existence with asymmetric adjustment for export and import in India with a long-run coefficient less than one which represents weak sustainability of current account deficits in India Further, in order to understand the potential asymmetric transmission mechanism between exports and imports, the study has applied asymmetric error correction and the results of asymmetric error correction present unidirectional Granger causality running from export to import This reveals that after certain threshold level of current account deficit i.e 3%, both export and import have different speed of adjustment towards long-run and their adjustment is asymmetric in nature These results reveal that after certain threshold level of current account deficit, there should be systematic policy to adjust short-run behaviour of imports to bring back to the long-run for sustainability of current account deficit in India The policy makers should consider the nonlinear behaviour of current account deficit while formulating any policy prescription towards sustainability of current account deficit India

Key Words: Current account deficits, Unit root, Structural break, threshold cointegration, Asymmetric error correction mechanism

JEL Code: F30, F32, C22

_

†PhD Scholar, School of Economics, University of Hyderabad, India

*Assistant Professor, ASCW, Budgam, Srinagar,

Satellite Campus, Maulana Azad National Urdu University, Hyderabad

lingaraj.pu@gmail.com , Contact: +9697207863

The most significant study can be traced out to the study of Hush and Husted in 1991 conducted for the examination of sustainability of current account deficit in USA Many studies have based their works by incorporating the intertemporal budget constraint model developed by Hasted (1992) Hausted (1992) has extended that US current account deficit is sustainable by considering quarterly data from 1967 to 1989 through the application of cointegration test developed by Engle and Granger (1987) He also, in his argument, has suggested that the short run fluctuation of export and imports could be brought back to their long-run equilibrium level through the effective and appropriate macroeconomic policies In contrast to the study of Hausted (1992), Fountas and Wu (1999) found the rejection of the hypothesis of long-run relationship between exports and imports in USA which argues in favour of unsustainability of trade deficit in USA from the quarters of 1967 to 1994 Since, then many other studies both in developed as well as developing countries (Bahmani and Rhee, 1997; Arize, 2002; Tang, 2003;

Irandoust and Ericsson, 2004; Narayan and Narayan, 2005; Herzer and Nowak-Lehman, 2006; Marial Yol, 2009; Mukhtar and Rasheed, 2010; Tiwary, 2012; Nag and Mukherjee, 2012; and Ndoricimpa and Leah-Achandi, 2014) been carried out by considering the symmetrical long-run relationship between exports and imports of respective countries to arrive at its conclusion if the trade deficit sustainable or unsustainable under the framework of intertemporal budget constraint model developed by Hakko and Rush (1991) and Hausted (1992) Narayan and Narayan (2005) and Mukhtar and Rasheed (2010) support the arguments delivered by Hausted (1992) as , if both imports and exports exhibit symmetrical long-run relationship or cointegrated , then the current account deficits problem is attributed to short-run phenomenon and in long-run the deficits are termed as sustainable which is not in violation of its international budget constraint

This situation of current account deficits also has penetrated the Indian economy and India is not free from the unfavourable impacts on her macroeconomic fundamentals too This phenomenon of existence of current account deficit in India also has led many policy makers and researcher to evaluate its situation of sustainability by taking into consideration of different time periods The current account deficit story is well known and reflects an overall failure to curb imports and boost exports in India This leaves India dependent on foreign capital to fund the current account deficit It also exposes the vulnerability of India’s overall balance of payment (BoP) in a scenario of sudden exit of foreign capital as was seen in 2009 and 2012, the only two instances in the past 12 years when India clocked a fall in the BoP and had to draw down on its foreign exchange reserves As a result, India’s foreign exchange reserves peaked in 2008 and have stagnated ever since Hence a higher portion of the current account deficits leads to greater vulnerability of economy on external front and could again lead to a stronger depreciation of currency CAD touched a record high of 6.7 per cent of the GDP in the third quarter ended December 2012 (RBI)

The main objective of this current study is to re-evaluate the sustainability of current account deficits in India by taking into consideration of nonlinear cointegration framework unlike earlier works, between real total exports and imports, in a framework of Hakkio and Rush (1991) and Husted (1992) The rest of the study is followed as Section 2 deals with theoretical underpinnings, Section 3 stresses on reviews of past literature, Section 4 gives data sources and desired methodology, Section 5 deals with result analysis and finally, Section 6 extends conclusion and policy implication

Where we have is for current income, is for consumption and is for income respectively

is the current borrowing, is initial debt amount and is termed as world interest rate

By solving for in equation (1) we can have the trade balance t expression and is discounting factor:

t For the verification of the hypothesis of sustainability Husted (1992) suggests the following assumption i.e where is the expenditure on imports

From equation (3), after solving for we have

By substituting equation (5) and (6) in equation (4) we will have the following equation:

By letting and , equation (7) can be written as:

The equation (8) can be expressed as follows where and assuming that

of current account deficit in the concerned country (Herzer and Nowak-Lehman, 2005; Serdar, 2008; Rahman, 2008 and Triwary, 2012) If exports ( ) and imports ( ), are cointegrated then current account deficits will be sustainable of a country (Hakkio and Rush, 1991; Husted, 1992) But , the question here erupts as the above liner cointegration between total exports and imports of any country speak sustainability of current account deficit of concerned country as directed by (Hakkio and Rush, 1991; Husted, 1992) would be violated in its methodological respect if we carry forward to testing liner combination between the above said variables Because of macroeconomic instability, impact of policy changes of both internal as well as external, the liner framework of examining the sustainability of current account deficit would lead inadequate and misleading results and conclusions for policy makers Hence, this study has tried to address the nonlinear cointegration and its error correction mechanism to have robust results and policy implications on sustainability of current account deficit in India

3 Reviews of Literature

There is a plethora of empirical literatures that account for examining current account deficit in various countries throughout the world The problem of current account deficits also has encouraged significant attention not only in other countries but also in India, in order to check whether the current account deficits problem is sustainable or unsustainable by examining the celebrated long-run relationship between exports and imports from various time periods Some of the empirical studies can be discussed in this current study to understand various dimensions of both theoretical and methodological aspects of evaluation of sustainability of current account deficits problem from different countries along with India The past literature can

be reviewed into three broad categories of conclusion of sustainability, unsustainability and weak sustainability of current account deficits that has been carried out in different countries below

A study by Zubaidi A B et.al (2005), suggest that the current account deficits is not sustainable in eight Asian countries by considering non-existence of panel cointegration between real exports and real imports before crisis period which amounts large swings in exchange rates

of the concerned countries The study, further, concludes that there are macroeconomic frictions due to the persistent current deficits In a similar manner, Ozer Mustafa, et.al, (2011) in Iran, argue that current account deficit is an indicator of worsening external balance with increasing macroeconomic fragility and can be solved in connection with supportive economic policies towards attracting green filed foreign direct investment with equity portfolio Holmes, Mark J et.al (2011) conclude in favour of unsustainability of current account deficits in India before pre-liberalization and finds that current account deficit is sustainable after liberalization periods by taking into consideration of both parametric and non-parametric cointegration methods to find out the long-run relationship between exports and imports from 1950 to 2011 In contrast to such above studies, there are some studies which argue in favour of strong and weak sustainability of current account deficits

Adedej O S and Jagdish Handa (2008) find sustainability of current account deficits in Nigeria by employing Johansen's cointegration technique on consumption and output from 1960

to 2003 Heidari Hassan, et.al (2007) explain the current account deficits in Iran using bound testing to co-integration for the period 1960-2007 by consideration of multitude structural breaks

in macroeconomic variables and finds that the current account deficit in Iran is sustainable But, against this backdrop, Ozer Mustafa and Oya Innci Coskuun (2011) find weak sustainability of current account deficits after crisis period and suggest current account deficit is as an indicator of worsening external balance with increasing external fragility in Turkey Kumar A Tiwari (2012) examines the current account sustainability of India by taking into consideration of non-oil and oil exporting and importing goods during the period of 1970 to 2007 and suggest strong evidence

of sustainability of current account deficits which arises due to non-oil exports and non-oil imports but evidence for no sustainability of current deficits arises due to oil exports and oil imports Sohrabji Niloufer (2010) discusses the current account position of India after the reform period by using an inter-temporal solvency model of Hakkio Rush (1991) and Husted (1992) to empirically verify sustainability of current account deficits in India Applying dynamic GLS he finds a cointegration between inflows and outflows current account in India, and hence India's current account position is sustainable Nag Biswajit and Jaydeep Mukherjee (2012) find no sustainability of trade deficit in India from 1950 to 2009 by applying cointegration in the presence of endogenous structural breaks in Indian economy

However, all the above empirical literatures suffer from severe methodological laxity in the advent of newly developed and most robust model of threshold cointegration which, in general, assumes asymmetric adjustment unlike symmetric adjustment towards the long-run equilibrium analysis of two time series Most of the earlier studies have used extensively the cointegration test developed by Engle and Granger (1987), Johansen and Juselius (1990), and which are neither correct nor accurate in the presence of transaction costs and asymmetries in price transmission (Balke and Fomby, 1997), that draw conclusions based on asymmetric adjustment Hence, Balke and Fomby (1997) proposed a threshold cointegration analysis which assumes the adjustment towards the long-run equilibrium holds when the deviation from the equilibrium exceeds some threshold level (Stigler, 2012) To avoid such laxity, the current study, has applied Threshold Autoregressive (TAR) and Momentum Threshold Autoregressive (MTAR) models to develop a threshold cointegration test which allows asymmetric adjustment towards the long-run equilibrium (Balke and Fombay, 1997; Enders and Granger, 1998; and Enders and Sikols, 2001) The study have examined the relationship between export and import

by applying the asymmetric error correction with threshold cointegration developed by Enders and Sikols (2001) to evaluate the sustainability of current account deficit in a framework of Hakkio and Rush ( 1991) and Husted (1992) in India

4 Data and methodology

In order to evaluate the sustainability of current account deficit in India, the study has taken into consideration of total exports and imports in terms of US Dollars All data have been collected from Hand Book of Statistics on Indian Economy, Reserve bank of India 2014, for the period

1960 to 2014 The study incorporates both the method of liner cointegration developed by Engle

and Granger (1987), Johansen and Juselius (1990), and Threshold cointegration with error correction mechanism in the presence of threshold cointegration (Balke and Fomby, 1997; Enders and Granger, 1998)

The main intention of this method of asymmetric cointegration and error correction is to check whether exports have asymmetrical influence on imports or vice-versa The study also has applied unit root test in presence of structural breaks The above methodologies are explained below:

As the macroeocnomic data are not free from the impact of various shocks ( oil crisis, gold crisis, 2008 economic crisis , and other external shocks), it is inevitable to check their

possible impacts on the macroeconomic varibales Due the presence of structural breaks , the slope coefficient may change and lead to baised results for conclusions if we apply unit root tests that are solely based on the assumption of linearity of the data In order to check the possible existence of structural breaks , the study has used various unit root tests which take into account the presence of sturutucal breaks for robust inference about intergation propoerties of the varibales The widely applied Perron and Vogelsang (1992) and Clemente, Montañés, Reyes (1988) structural break unit root test are applied to check unit root properties of the export and import in the presence of possible structural breaks

4.1 Unit root test in the presence of structural breaks

Unit root tests in time series data play very prominent role in order to determine the order

of integration of the concerned time series The most celebrated one is ADF test which do not consider the presence structural breaks in the macroeconomic data Hence, the ADF tests are biased towards the non-rejection of the null-hypothesis (Perron, 1989) in the presence of structural break The study has applied the unit root test developed by Perron and Vogelsang (1992) to find out the presence of possible one structural break in the data Additive outlier (AO) models consider for a sudden change in mean (Crash model) This AO model has been estimated with the help of two-step procedures In the first step, it removes the deterministic part of the series by estimating the following regression

Where the obtained residuals from Eq (11), TB is the structural date, if t= TB+1 and is zero, otherwise Eqn (11) and Eqn (12) are estimated by OLS which takes into consideration of each break year TB=k+2,… ,T-1, where T presents the number of observations and k presents the truncation lag parameter If the t-statistic on becomes non-zero (significantly different from zero), the null hypothesis of a unit root will be rejected At this situation, the case

of any variable (say, export) will be a stationary time series around a structural break The break would bring temporary changes in export Against this, if the t-statistic on is not significantly

different from zero, the export would be a non-stationary time series and any sudden break would have permanent effects on the long-run level of the export

4.2 Unit roots in the presence of Double structural breaks

Most variables also don’t consider one break So, it is very essential to check for double structural breaks in the variables In the first test, we discuss the detection of possible one structural break and now, we can go for double breaks that may be existed in the variables For more than one break, Clemente, Montañés, Reyes (CMR), 1988 test is applied to both the variables of the study

Clemente et al (1998) estimated the following regression in order to examine the unit root

in the presence of more than one structural break by considering Perron and Vogelsang (1992) procedure So, the Eq (11) and (12) can be changed to

Where and zero, otherwise becomes equal to one if

and zero, otherwise TB1 and TB2 are the time period where the mean is being modified

For the verification of unit root null hypothesis, Eq (11) has been firstly estimated by OLS to remove the deterministic part of variables and then test is carried out by searching for the minimal Pseudo t-ratio for the hypothesis in Eq (12) for all beaks The null-hypothesis of

a unit root is rejected if the t-statistic on is significantly different from zero In this case, the variable export becomes time series around two structural breaks One shock on break can cause temporary movements of the variable export, but the case of two breaks could cause permanent effects Similarly, if the t-statistic on is not statistically different from zero, the variable export would be a non-stationary time series and a sudden shock could have permanent impacts on the long-run level of the variable export

4.3 Liner cointegration

The liner cointegration by Engle Granger (1987) does not speak about the asymmetric transmission between the variables So the traditional cointegration is of severe criticism by different scholars like (Balkee and Fomby, 1997; Enders and Siklos, 200) for not capturing nonlinearity of the time series Each and every time series exhibits nonlinearity and hence it is very essential to check the nonlinearity and need to apply asymmetric error correction in the presence of the threshold cointegration between the variables in the study Generally the cointegration has been widely applied to investigate the long-run relationship among the variables There are two major cointegration methods based on Johansen and Engle-Granger two-step procedures Both of the tests assume symmetric relationship between variables The threshold cointegration has been also come into force in order to assess the price transmission studies Balke and Fomby (1997) propose a two-step approach developed by Engle and Granger (1987) Enders and Granger (1998) and Enders and Siklos (2001) further developed the standard Dickey-Fuller test by allowing for the possibility of asymmetric transmission movements in time series data This usually gives a path breaking road map for evaluating the asymmetric adjustment in the time series data without maintaining the hypothesis of a symmetric adjustment

to a long-run equilibrium A hand full of studies has been carried out by different authors to assessing asymmetric transmission mechanism widely (Abdulai, 2000; Abdulai, 2002; Sun Changyou, 2011; Chang Tsangyao et.al , 2011; Jou, Rong-Chang et.al , 2011; Chang Tsangyao and Han-Wen Tzeng, 2011; Jou Zare, Roohollah and M Azali , 2014; Chena, Haiqiang & Yau, H.Y and Chien-Chung Nieh , 2009; Koutroumanidis, T et al 2011; Alaabed Alaa and M Masih, 2016; and Matemilola, B.T, et.al 2015 etc ) to find out long-run equilibrium relationship between two or more time series.)

The most widely applied cointegration tests of Johansen and Engle-Granger two-step approach can be discussed below The Johansen approach is an extend version of multivariate generalization of Dickey-Fuller test (Johansen, 1988; Johansen and Juselius, 1990) It is based on the relationship between the rank of a matrix and its characteristics roots in a vector autoregression It starts with a vector autoregressive model and then recomposes it into a vector error correction model as follows:

where is a vector of export volume at year t for India K is number of lags, and is the error term The relationship among the coefficients for the two equations is

where I is an identity matrix The number of cointegration vectors, r, can be detected between the

variables in from the trace statistics and maximum eigenvalue The Engle-Granger two-step procedure is based on time series property of the residuals from the estimated long-run equilibrium relationship (Engle and Granger, 1987) It can be explained as:

be avoided by taking number of lags (P) based Akaike Information Criteria (AIC), Bayesian

Information Criteria (BIC), or Ljung-Box Q test If the null hypothesis of = 0 is rejected, then

the residual series from long-run equilibrium is stationery and hence the variables and will

be cointegrated

4.4 Threshold cointegration Analysis

The above written cointegration is symmetric in nature and does not speak the asymmetric transmission between the variables The asymmetric adjustment in cointegrated variables is estimated by Enders and Siklos (2001) Enders and Siklos (2001) proposed a two-regime threshold cointegration approach to entail asymmetric adjustment in cointegration analysis The threshold model is modified from Eq (17) such as:

Where I is a heavy side indicator, P is the number of lags in the model to be undertaken,

, and are the coefficients and the threshold value to be estimated The lag P can be included in the model to avoid serial correlation based on the selection of AIC, BIC, or Ljung-

Box Q test The heavy side indicator I can be expressed as with two alternative definitions of the

threshold variable with the lagged residuals ( ) or the change of the lagged residuals ( ) The Threshold Auto regression (TAR) model is specified with the Eq (18) and (19a) whereas the Momentum Threshold Autoregression (MTAR) model with Eq (18) and (19b) The TAR model

is specially modelled to capture the potential asymmetric deep movements in the residuals (Enders and Granger, 1998; Enders and Siklos, 2001) The speciality of the MTAR model is, it can capture the more dynamic asymmetries in the residuals when the adjustment exhibits more momentum in one direction than other In the model the negative deepness (i.e., ) of the residuals implies that increases trend to persist, whereas decreases trend to revert quickly towards to equilibrium (Enders and Granger, 1998)

The threshold value has been termed as zero from the regression based on residual series But alternatively, Chen (1993) proposes a search method to estimate the consistent threshold value The super consistent threshold value can be obtained through several procedures First, we have to arrange the threshold variables i.e for TAR model or the for MTAR model Secondly the possible threshold values are determined To be meaningful of the threshold value that must actually cross the threshold value (Enders, 2004) and the threshold value should lie between the maximum and minimum values of the threshold variables So the highest and lowest 15% of the values are excluded from the procedure to ensure an adequate number observation on each side The middle 70% values of the sorted threshold arranged variables are used as potential threshold values In third, the TAR and MTAR model is estimated with each potential threshold value In each procedure the sum of squared error can be estimated and the relationship between the sum of squared errors and the threshold value can be evaluated Finally, the threshold value that minimizes the sum of squared errors is deemed to be consistent estimates of threshold Out of the equations of the TAR and MTAR model the consistent TAR and consistent MTAR model can be estimated with the estimated threshold value of both TAR and MTAR They are TAR in Eq (19a) with = 0; consistent TAR in Eq (19a) with estimated ; MTAR in Eq (19b)

with = 0; and consistent MTAR in Eq (19b) with estimated The appropriate model for adjustment mechanism is selected based on the lowest AIC and BIC (Enders and Siklos, 2001)

The asymmetric adjustment mechanism in the cointegration relationship can be estimated

by F-test of rejecting null hypothesis of no cointegration ( ) against the alternative of cointegration with either TAR or MTAR threshold adjustment The test statistics is presented by where this test does not consider a standard distribution and the critical values in Enders and Siklos (2001) will be used The null hypothesis of symmetric adjustment in the long-run can be tested by a standard F-test ( The rejection of null hypothesis of symmetric adjustment mechanism indicates the existence of asymmetric adjustment process

4.5 Asymmetric error correction with threshold cointegration equations

In the cointegration analysis of the time series models Granger representation theorem (Engle and Granger, 1987) expresses the error correction mechanism when the two time series are cointegrated The cointegration and error correction mechanism of Engle and Granger considers symmetric adjustment process But in case of the asymmetric cointegration with error correction mechanism can be explained by extending on the standard specification of error correction model The standard specification has been extended by Granger and Lee (1989) to address the asymmetric adjustments Both the first differences variables and error correction terms are generally decomposed into positive and negative components This mechanism reveals

an extensive exposition on whether positive and negative impacts of trade differences have asymmetric effects on the dynamic behaviour of trade If the threshold cointegration is present then a threshold error correction terms are modelled further (Balke and Fomby, 1997; Enders and Granger, 1998) The asymmetric error correction model with threshold cointegration is represented as:

) if > and equal to 0, otherwise; is equal to ( ) if < and equal to 0 otherwise Similarly in case of is equal to ( ) if > and equal to 0, otherwise; is equal to ( ) if < and equal to 0

otherwise For this the maximum number of lags J has been selected based on AIC statistic and

Ljung-Box Q test in order to avoid autocorrelation The error correction term E, expressed as

and = (1 ) , are designed from threshold cointegration regression in Eqs (18), (19a) and (19b) This error correction terms take into consideration of possible asymmetric error correction trade balance in response to positive and negative shocks to the deviations from long-run equilibrium and also takes into account of the impact of threshold cointegration through the construction of Heaviside indicator in Eq (19a) and (19b) The estimated coefficients can reveal the presence of asymmetric behaviour and response of individual variables on the disequilibrium in previous periods Here the driving force should be export and the expected sign for the error term of export should be positive ( and negative for imports ( )

The study has made four kinds of hypothesis and F-tests to check the Granger causality, distributed lag asymmetric effect, cumulative asymmetric effect and lastly equilibrium adjustment path asymmetry between exports and imports The first test of Granger causality is tested on weather exports Granger causes its own or imports based on restricting all exports to be zero and followed by the F-test ( for all lags i simultaneously) Similarly we

can test in case of imports for all lags) In case of the second test we have to check the distributed lag asymmetric between export s and imports of India Here the hypothesis

is weather distributed lag asymmetric effect is there in between exports and imports of India The null hypothesis of presence of distributed lag asymmetric effect of exports on its own or on imports is tested by the hypothesis ( ) This can be adopted for each lag and both

the variables (i.e., ) The third hypothesis of cumulative asymmetric effect can be tested by setting the null hypothesis ( ) for exports and for imports ( ).The last hypothesis of equilibrium adjustment path asymmetry can be examined by setting the null hypothesis of for each equation estimated

Fig 1

Graph of log export and log import from 1960-2014

5 Results and Discussion

5.1 Descriptive statistics and Unit root test

The descriptive statistics of export and import in India are reported in Table 1 The trends

of both series are demonstrated in Fig 1 From 1960 to 2014, the average export volume is 0.893 and 0.972 for imports It shows that the average for import volume is higher than the exports This explains that the volume of import is always higher than the export volumes in India During the period of late 90s there were huge imports as compared to the post-reform periods In the post reform periods, the export has also increased tremendously Over all the import volume in terms of dollars has been also increased Form the Fig 1 it is very clear that both the variables have evolved to be closer and more correlated which represents co movement

of the variables in the long-run

We have examined the nonstationarity properties of the two variables by using ADF test The lag length for the ADF test is determined by the AIC statistic and Ljung-Box Q test The

null hypothesis of non-stationarity of the variables cannot be rejected at the appropriate significance level at level form which reveals that the variables are cointegrated The Table 1 shows that variables are integrated of order one and hence require first difference to be stationary Thus it is concluded that the both export and import are cointegrated in India

Table 1

Descriptive statistics and unit root test results for exports and imports of India

Level (X t ) 1st Diff ( ∆X t ) Level (M t ) 1st Diff (∆M t )

Notes: The critical values are and for ADF test with trend , and for ADF test with a drift at the 1%, 5%

and 10% level of significance respectively (MacKinnon ,1996 ) *** Denotes significance at 1% level

However, the ADF test gives biased conclusion in the presence of possible structural break of the concerned variables in the study (Perron, 1989) Hence, it is very essential to check unit root which could account the presence of structural break The study has performed the unit root test developed by Perron and Vogelsang (1992) and Clemente, Montañés, Reyes (CMR, 1988) to test possible structural break presence in the data

Table 2

Unit roots in the presence of a one-time structural break (AO Model)

Variables Structural breaks year t-statistics P-value

The results of possible one time structural break in the variables in a framework of AO model to testing unit root are reported in Table 3 The null hypothesis of unit root is rejected at 5% level of significance The result of structural breaks for real import is negative which reflects