Does Harberger–Laursen–Metzler (HLM) Exist in Pakistan? Cointegration,Causality and Forecast Error Variance Decomposition Tests

Abstract

This article investigates the validation of Harberger–Laursen–Metzler hypothesis using Pakistani data over the period of 1978Q1–2012Q4. We have used Gregory–Hansen cointegration approach accommodating structural break in the series. The results confirm the presence of cointegration between the variables. The impact of terms of trade is positive on Pakistan’s trade balance. Domestic income improves local trade balance. Foreign income has a positive impact on Pakistan’s trade balance. The causality analysis indicates the presence of feedback effect between terms of trade and trade balance. Domestic income Granger causes trade balance. The unidirectional causality is running from foreign income to Pakistan’s trade balance. This study opens up new directions for policymakers for improvements in terms of trade to sustain trade balance in Pakistan.

Keywords

Terms of trade trade balance Pakistan

Introduction

Since the advent of industrialization, the economies have opened themselves to rest of the world, global trade emerged and led the small and developing economies to grow faster and developed enormously. The trade liberalization enabled global economies to gain prosperity through trade effect but simultaneously they are facing the major macroeconomic issues of exchange rate and external trade imbalances. In a small open economy, the common policy tool used in response of external trade deficit is to devalue the currency but it causes inflation and high depreciation cost. The changes in terms of trade have a direct effect on trade balance and an indirect effect on savings, income and inflation but unfortunately, this effect varies from country to country. Therefore, researchers have tried to find out the exact relationship between terms of trade and its impact on balance of trade from different economies. The study of Harberger (1950) and Laursen and Metzler (1950) described this phenomenon as terms of trade deterioration will result in a decline in savings due to decrease in real income, and therefore real depreciation will increase real expenditures. This effect is known as the Harberger–Laursen–Metzler (HLM) effect. Recently, the relationship between terms of trade and exchange rate has been the recurrent research topic in the macroeconomic literature. Further, HLM states that the temporary increase in terms of trade may have a positive impact but if shock persists over long run, it would deteriorate the current account and savings. Regrettably, most of the literature on this topic has focused on the short-run analysis and implicitly ignored the long-run effect.

A number of studies on developing countries have examined the relationship between terms of trade and trade balance using different methods, that is, Bahmani-Oskooee (1985) investigated the four developing countries for short-run impact and results support the pattern of movement and satisfied the Marshall–Lerner¹ (ML) condition. Fry (1986) studied the 14 Asian countries to test the HLM effect using three equation model for savings, investment, economic growth and analyzed the terms of trade effect on current account. Grilli and Yang (1988) tested the terms of trade impact over commodity prices rather than on income and concluded that commodity prices effect is different in oil-exporting and non-oil-exporting developing economies. Correia, Neves and Rebelo (1995) checked the same effect on small open economy for terms of trade and real exchange rate. The later studies included Otto (2003), study of Broda (2004) on Malaysian data confirming the long-run relationship, studies by Kaplinsky (2006), Hung-Ju and Hsu (2009), Santos-Paulino (2010), Jawaid and Waheed (2011), Hirose and Ikeda (2012), Idrees and Tufail (2012) and Jääskelä and Smith (2013) and the most recent studies of Bahmani-Oskooee and Xu (2013) and Islam, Shahbaz and Tahir (2013) on the HLM effect on Korea–US and Bangladesh, respectively.

During the last three decades, Pakistan’s economy has also been struggling with the consistent external imbalance shocks and currency devaluation is the sole policy measures taken by the government to offset such shocks. Resultantly, country has been facing double digit inflation and current account has been largely fragile to changes in global and national economic policies. As terms of trade is considered as one of the key operators of change in real income, therefore, it directly impacts national savings and increases dependency on foreign debts. Country’s current account balance has been under stress since decades and expanding on a year-to-year basis. During the last two fiscal years (FY), deficit has increased from US$2.072 billion (FY2012 est.) to US$2.36 billion (FY2013 est.). Again, $1.37 billion is recorded for the first two months of FY2014. Therefore, it is imperative to investigate the effect of increased terms of trade on Pakistan’s trade balance. Following Harberger (1950) and Laursen and Metzler (1950), HLM effect suggests that an exogenous increase in terms of trade in a small but open economy results in an improvement in that country’s trade balance which is appropriate for a developing economy (Berument & Dincer, 2005; Bouakez & Kano, 2008; Chen & Hsu, 2006; Huang & Meng, 2007; Jawaid & Raza, 2013; Sobrino, 2011; Tsen, 2006). Similarly, Pakistan is also one of the developing countries that largely depends on foreign debt liabilities and consecutively facing current account deficit problem. This notion proposes to conduct this study in order to find out the exact relationship between terms of trade and trade balance in Pakistan.

This study tests the HLM effect for Pakistan using autoregressive distributed lag (ARDL) bound testing approach of cointegration and checks both long-run and short-run behaviours of variables in vector error correction model (VECM). However, Granger causality test is also applied in order to explore the exact direction of causality. The results of this study would potentially provide policy guidelines which would lead in achieving steady growth and immutable balance of trade. Subsequently, it may contribute to resist the shocks and stabilize saving and inflation rates in our economy. The remaining article is divided into following sections: the second section presents a brief literature review, model construction and data is given in the third section, the fourth section presents the methodological framework, results and discussion are given in the fifth section and the sixth section presents the conclusion and policy recommendations.

Literature Review

Significant amount of literature is produced which examined the relationship between terms of trade and trade balance but the evidence so far remains inconclusive. However, the scholars generally agree that the impact of change in terms of trade on trade balance may not be similar in every economy. In the theoretical perspective, this notion is known as the HLM effect, based on the seminal works of Harberger (1950) and Laursen and Metzler (1950). The HLM effect predicts that an increase in terms of trade originating from an exogenous shock to a small open economy leads to an improvement in that country’s trade balance. Otto (2003) examined the HLM effect for a number of developing and small Organisation for Economic Co-operation and Development (OECD) economies.² The study found that a positive shock to terms of trade improves trade balance but deteriorates later as the shocks become persistent. Dibooglu (2000) pointed out that positive shock to terms of trade is important in the short run but may not sustain in the long run once other factors tend to offset the initial effect. Hoque (1995) found the long-run relationship among current account deficit, terms of trade, domestic income and foreign income using fixed exchange rate regime in the case of Australia. Kouassi, Decaluwe and Colyer (1998) investigated the relationship between terms of trade and current account deficits using the VECM. For Cote d’Ivoire, they find a long-run relationship but point out that current account deficit in Cote d’Ivoire cannot be explained by terms of trade. A strong unidirectional causality runs from current account deficit to terms of trade. The dynamic simulations indicate that a significant portion of fluctuations in terms of trade is explained by current account deficit. The study of Santos-Paulino (2010) provides a similar evidence from small island developing states. Tsen (2009) investigates the impact of terms of trade and oil process on trade balance of Asian economies. The empirical results found that the impact of income terms of trade is different across economies.

A large proportion of available literature exploring the direct relationship between trade balance and terms of trade focus on the correlation between two variables. Only a few have examined the long-run relationship between two series. Haynes and Stone (1982) found a positive relationship between US terms of trade and trade balance. Warner and Kreinin (1983), Gylfason and Risager (1984), Bahmani-Oskooee (1986) and Marquez (1990) found that a deteriorating terms of trade improves trade balance. Bahmani-Oskooee and Jonardhanan (1995) in a 24-country study reported no long-run relationships between two series. Wong (2006) shows that there is a long-run relationship between trade balance and commodity terms of trade but none between income terms of trade and trade balance for Malaysia. Later, these results were validated by Tsen (2006) for Malaysia. Nwachukwu and Egwaikhide (2007) examined the determinant of private saving in Nigeria using HLM hypothesis and found that external terms of trade have a positive impact on private savings. Berument and Dincer (2005) and Zortuk (2008) studied the Turkish economy to examine the long-run impact between terms of trade (income terms of trade and commodity terms of trade) and trade balance. Their study utilized Johansen’s cointegration method, the VECM and Granger causality approaches to test the relationship between income terms of trade and trade balance which showed that there is no causality between the series. Bahmani-Oskooee and Tankui (2008) summarized the relationship between terms of trade and trade balance for 20 African countries. They adopted Pedroni’s (2004) approach using cointegration and confirmed the long-run relationship between both the variables. The results validated the ML requirements where income terms of trade showed a positive impact on trade balance. Another study of Wong (2009) examined the HLM effect for Korea, Hong Kong and Singapore. The empirical results for the long-run relationship between variables indicate that an increase in the terms of trade causes a decline in trade balance. The feedback hypothesis is found between income terms of trade and trade in Hong Kong and Singapore while trade balance Granger causes income terms of trade in Korea. For other Asian countries, Wong (2009) tested the same relationship and found a long-run relationship between the variables but affect of terms of trade on trade balance varies across countries. The empirical evidence showed that domestic demand, foreign income, terms of trade and oil prices are major determinants of trade balance in short run as well as in long run. Besides the work of Wong (2006), very few studies from the above-mentioned literature examined the long-run relationship and most of them used commodity terms of trade³ which is a form of barter terms of trade. Appleyard and Field (2001) approximate income terms of trade⁴ by ratio of exports value to import price. Commodity terms of trade focuses on the relationship between export price and import price, while income terms of trade quantifies the trend of export-based capacity of a country to imports of goods. The high value of commodity terms of trade implies that price of exports is high relative to import prices. However, the direction between commodity terms of trade and income terms of trade may not imply the same thing. An increase in price of exports relative to import could lead to higher commodity terms of trade but income terms of trade could worsen if offset by a decline in quantity of exports. The income terms of trade is a better measure of terms of trade compared to commodity terms of trade because the former can rise faster relative to the latter (see Appleyard & Field, 2001).

Therefore, the present study aims to focus on the income terms of trade for validating that hypothesis in order to draw meaningful insights for Pakistan and provide a certain policy guidelines. Moreover, the methodology and empirical techniques utilized to investigate the relationship between terms of trade and trade balance also play a crucial role in deriving inconclusive and challenging results, hence misleading the policymakers. Therefore, it is imperative to adopt relevant and current techniques which could ensure maximum accuracy in order to derive conclusive evidence which could provide guidelines to policymakers. For example, Idrees and Tufain (2012) test HLM effect using VAR test in a recursive form, and this type of technique only captures the linear interdependence of multiple time series; however, the study of Islam et al. (2013) and Jawaid and Raza (2013) employed ARDL, the VECM Granger test to check HLM effect for Bangladesh and India, respectively. The later study uses the latest technique of estimation and also finds the long-run relationship among the series and hence gives more reliable results for policy design. Therefore, this study uses ARDL model approach of cointegration to empirically examine the small open developing economy of Pakistan. Being a part of a globalized village and to ensure sustainable development goals, it is now imperative to know the relation between terms of trade and balance of trade for Pakistan. The contribution of this article seems significant and the policy implications of this study are highly valuable for both government and international institutions.

The Model Construction and the Data

The aim of the present study is to examine the relationship between income terms of trade and trade balance in Pakistan. The equation of trade balance can be derived by the exports and imports equations, respectively, indicated as follows:

X_{t} = f (P x_{t})

(1)

M_{t} = f (P m_{t})

(2)

T B_{t} = X_{t} - M_{t}

(3)

T B_{t} = f (P x_{t} / P m_{t})

(4)

\ln T B_{t} = α_{11} + α_{22} \ln T T + μ_{t}

(5)

\ln T B_{t} = β_{11} + β_{22} \ln T T + β_{33} \ln Y_{t} + β_{44} \ln Y F_{t} + μ_{t}

(6)

where X_t (Px_t) is real exports (unit price of exports) per capita, M_t (Pm_t) is real imports (unit price of imports) per capita, TB_t is trade balance which is the difference between exports and imports. The ratio (Px_t/Pm_t) between unit value of exports to unit value of imports is termed as terms of trade. Real domestic income per capita is indicated by Y_t proxies real gross domestic product (GDP) per capita. Real foreign income (YF_t) per capita is proxied by real GDP of trading partners of Pakistan. If terms of trade is positively linked with trade balance, then we expect ^TT_t/^TB_t > 0 otherwise ^TT_t/^TB_t < 0. The rise in domestic income has a burden on import bill due to hike in imports demand and we expect ^Y_t/^TB_t < 0, otherwise ^Y_t/^TB_t < 0. If an increase in foreign income will increase the demand for Pakistan’s exports, then foreign income has a positive impact on trade balance and we expect ^YF_t/^TB_t > 0, otherwise ^YF_t/^TB_t < 0. We have converted all the series into logarithm for results reliability and efficiency. The study covers the period of 1978Q1–2012Q4. We have collected data for real effective exchange rate, unit value of export and unit value of imports from International Financial Statistics (CD-ROM, 2013). The real GDP per capita of Pakistan and her trading partners is collected World Development Indicators (CD-ROM, 2013).⁵

Methodological Framework

This study employs ARDL bounds testing approach to cointegration proposed by Pesaran, Shin and Smith (2001) to check the long-run relationship among terms of trade, domestic income, foreign income and trade balance in the case of Pakistan over the period of 1978Q1–2012Q4. The ARDL cointegration framework using bound testing approach is a more suitable approach because it analyzes the relationship between dependent variable and regressors even when it is uncertain whether the regressors under observation are trend or stationary. In this process, two separate critical values are associated with all regressors that generate critical value bounds and classified as purely I(1) or purely I(0) or mutually I(1)/I(0) cointegrated. Following this process, a conclusion can be drawn on the basis of F-statistics⁶ without knowing the order of integration of variables and this method is also suitable for small sample analysis, that is, Pakistan. Assuming the possibility of structural break in the underlying series, in addition to the ARDL approach to cointegration, we also utilized Gregory–Hansen structural break cointegration test developed by Gregory and Hansen (1996). This test has a uniqueness to point out the structural break in the series and these dates are endogenously determined without affecting the rest of the series. Moreover, a dynamic unrestricted error correction model (UECM) is also generated from the ARDL bounds testing through a simple linear transformation. The UECM combines the short-run dynamics with the long-run equilibrium without losing any long-run information. The UECM is expressed as follows:

\begin{array}{l} \begin{matrix} Δ \ln T B_{t} = ϑ_{1} + ϑ_{T} T + ϑ_{T B} \ln T B_{t - 1} + ϑ_{T T} \ln T T_{t - 1} + ϑ_{Y} \ln Y_{t - 1} + ϑ_{Y F} \ln Y F_{t - 1} \end{matrix} \\ \begin{matrix} + \sum_{i = 1}^{p} ϑ_{i} Δ \ln T B_{t - i} + \sum_{j = 0}^{q} ϑ_{j} Δ \ln T T_{t - j} + \sum_{k = 0}^{r} ϑ_{k} Δ \ln Y_{t - k} + \sum_{l = 0}^{s} ϑ_{l} Δ \ln Y F_{t - l} + μ_{t} \end{matrix} \end{array}

(7)

\begin{array}{l} \begin{matrix} Δ \ln T T_{t} = α_{1} + α_{T} T + α_{T B} \ln T B_{t - 1} + α_{T T} \ln T T_{t - 1} + α_{Y} \ln Y_{t - 1} + α_{Y F} \ln Y F_{t - 1} \end{matrix} \\ \begin{matrix} + \sum_{i = 1}^{p} α_{i} Δ \ln T T_{t - i} + \sum_{j = 0}^{q} α_{j} Δ \ln T B_{t - j} + \sum_{k = 0}^{r} α_{k} Δ \ln Y_{t - k} + \sum_{l = 0}^{s} α_{l} Δ \ln Y F_{t - l} + μ_{t} \end{matrix} \end{array}

(8)

\begin{array}{l} \begin{matrix} Δ \ln Y_{t} = β_{1} + β_{T} T + β_{T B} \ln T B_{t - 1} + β_{T T} \ln T T_{t - 1} + β_{Y} \ln Y_{t - 1} + β_{Y F} \ln Y F_{t - 1} + \sum_{i = 1}^{p} β_{i} Δ \ln Y_{t - i} \end{matrix} \\ \begin{matrix} + \sum_{j = 0}^{q} β_{j} Δ \ln T B_{t - j} + \sum_{k = 0}^{r} β_{k} Δ \ln T T_{t - k} + \sum_{l = 0}^{s} β_{l} Δ \ln Y F_{t - l} + μ_{t} \end{matrix} \end{array}

(9)

\begin{array}{l} \begin{matrix} Δ \ln Y F_{t} = ρ_{1} + ρ_{T} T + ρ_{T B} \ln T B_{t - 1} + ρ_{T T} \ln T T_{t - 1} + ρ_{Y} \ln Y_{t - 1} + ρ_{Y F} \ln Y F_{t - 1} + \sum_{i = 1}^{p} ρ_{i} Δ \ln Y F_{t - i} \end{matrix} \\ \begin{matrix} + \sum_{j = 0}^{q} ρ_{j} Δ \ln T B_{t - j} + \sum_{k = 0}^{r} ρ_{k} Δ \ln T T_{t - k} + \sum_{l = 0}^{s} ρ_{l} Δ \ln Y_{t - l} + μ_{t} \end{matrix} \end{array}

(10)

In the above equations, the first^t-difference operators are denoted by Δ and μ_t represents residual terms. Keeping in view of the sensitivity of F-statistic with lag order selection, the suitable lag length of the first-differenced regression is selected on the basis of the minimum value of akaike information criterion (AIC) since the inappropriate lag length selection may lead to misleading results. Pesaran et al. (2001) suggested an F-test to evaluate the joint significance of the coefficients of lagged level variables. For example, the null hypothesis for cointegration between the variables is H₀: α_TT = α_TB = α_Y = α_YF = 0, while the rejection of null hypothesis for cointegration is H_a: α_TT ≠ α_TB ≠ α_Y ≠ α_YF ≠ 0. Pesaran et al. (2001) generated two asymptotic critical values, that is, upper critical bound (UCB) and lower critical bound (LCB), that are used to determine the existence of cointegration between the series without knowing the order of integration. The LCB is used to test cointegration if all the series are integrated at I(0); otherwise, we use UCB. Our computed F-statistics are F_TB(TB/TT, Y, YF), F_TT(TT/TB, Y, YF), F_Y(Y/TB, TT, YF) and F_YF(YF/Y, TB, TT) for equations (7)–(10), respectively. In the ARDL bound testing approach, the existence of a long-run relationship between the variables is calculated in the basis of F-statistic. If the F-statistic falls outside the UCB, then there is cointegration between the series or if our calculated F-statistic falls within LCB, then there is no cointegration. Similarly, the decision regarding cointegration is inconclusive if calculated F-statistic falls between LCB and UCB. In such scenario, an error correction method is adopted which is an easy and suitable way to examine cointegration between the variables. Here, we have used critical bounds generated by Narayan (2005) to test cointegration rather than those generated by Pesaran et al. (2001) and Turner (2006).

The VECM Granger Causality

After determining the long-run relationship among the variables, we use the Granger causality test to assess the causality between the variables. In case the cointegration exists between the series, the VECM can be developed as follows:

\begin{array}{l} \begin{matrix} [\begin{matrix} Δ \ln T T_{t} \\ Δ \ln T B ​_{t} \\ Δ \ln Y_{t} \\ Δ \ln Y F_{t} \end{matrix}] = [\begin{matrix} b_{1} \\ b_{2} \\ b_{3} \\ b_{4} \end{matrix}] \end{matrix} + [\begin{matrix} B_{11, 1} B_{12, 1} B_{13, 1} B_{14, 1} \\ B_{21, 1} B_{22, 1} B_{23, 1} B_{24, 1} \\ B_{31, 1} B_{32, 1} B_{33, 1} B_{34, 1} \\ B_{41, 1} B_{42, 1} B_{43, 1} B_{44, 1} \end{matrix}] \times [\begin{matrix} Δ \ln T T_{t - 1} \\ Δ \ln T B_{t - 1} \\ Δ \ln Y_{t - 1} \\ Δ \ln Y F_{t - 1} \end{matrix}] + ... + [\begin{matrix} B_{11, m} B_{12, m} B_{13, m} B_{14, m} \\ B_{21, m} B_{22, m} B_{23, m} B_{24, m} \\ \begin{array}{l} B_{31, m} B_{32, m} B_{33, m} B_{34, m} \\ \begin{matrix} B_{41, m} B_{42, m} B_{43, m} B_{44, m} \end{matrix} \end{array} \end{matrix}] \\ \begin{matrix} \times [\begin{matrix} Δ \ln T T_{t - 1} \\ Δ \ln T B_{t - 1} \\ Δ \ln Y_{t - 1} \\ Δ \ln Y F_{t} \end{matrix}] \end{matrix} + [\begin{matrix} ζ_{1} \\ ζ_{2} \\ ζ_{3} \\ ζ_{4} \end{matrix}] \times (E C M_{t - 1}) + [\begin{matrix} μ_{1 t} \\ μ_{2 t} \\ μ_{3 t} \\ μ_{4 t} \end{matrix}] \end{array}

(11)

It is mentioned in the VECM model that the difference operator is represented by (1 – L) and ECM_t–1 denotes the lagged error correction term, generated from the long-run association. The statistical significance of estimate of the lagged error term, that is, the negative sign of ECT_t–1 confirms the existence of the long-run causal relation using the t-statistic. However, the short-run causality is identified by the joint χ² statistical significance of the estimates of first-difference lagged independent variables. For example, the significance of α_22,i ≠ 06_i suggests that the terms of trade Granger causes trade balance and causality runs from trade balance to terms of trade can be indicated by the significance of β_22,i ≠ 06_i. The same inference can be drawn for rest of causality hypotheses. Finally, we use Wald or F-test to test the joint significance of estimates of lagged terms of independent variables and error correction term.

Results and Discussions

The descriptive statistics and correlation matrix are reported in Table 1. The results showed that all the series are normally distributed which is indicated by the statistics of Jarque–Bera normality test. This implies that series have zero mean with constant variance. The correlation analysis reported indicates that terms of trade and domestic income are positively correlated with trade balance. A negative correlation is also found between foreign income and trade balance. There exists a positive correlation from domestic and foreign incomes to terms of trade. Finally, foreign income is positively correlated with domestic income.

Table 1.

Descriptive Statistics and Correlation Matrix

Variables	ln TB_t	ln TT_t	ln Y_t	ln YF_t
Mean	–0.3762	9.2811	10.0588	6.1125
Median	–0.2854	9.5861	10.1545	6.2083
Maximum	0.2875	11.1736	10.4692	6.5212
Minimum	–1.2147	7.3246	9.5491	5.6029
Std. Dev.	0.4259	0.8507	0.2881	0.2880
Skewness	–0.4366	–0.3713	–0.3562	–0.3572
Kurtosis	1.9444	2.9323	1.9040	1.9035
Jarque–Bera	3.2061	0.9502	2.9191	2.9260
Probability	0.2012	0.6218	0.2323	0.2315
ln TB_t	1.0000
ln TT_t	0.5468	1.0000
ln Y_t	0.1905	0.2799	1.0000
ln YF_t	–0.0924	0.1869	0.1630	1.0000

Source: Authors’ own.

The long-run relationship among the variables has been examined using the ARDL cointegration framework with bound testing approach. However, the condition for utilizing this approach is that if the variables are integrated either at I(0) or I(1) or I(0)/I(1). Therefore, to ensure this condition that none of the variables is stationary at I(2) or beyond this level, we have applied augmented Dickey–Fuller (ADF) unit root test by Dickey and Fuller (1979), Dickey and Fuller generalized least-squares (DF-GLS) unit root test by Elliot, Rothenberg and Stock (1996) and Ng–Perron unit root test by Ng and Perron (2001). These unit root tests determine that all the variables have unit root problem at their level form but were found to be integrated at I(1).⁷ However, Baum (2004) pointed out that unit root analysis by ADF, DF-GLS and Phillips–Perron (P–P) unit root tests may provide biased results due to structural break that might occur in the series.

In order to resolve this issue, we have also obtained two structural break unit root tests such as Zivot and Andrews (1992) unit root test which provides information about one structural break and Clemente–Montanes–Reyes (1998) de-trended structural break unit root test that contains information about two structural breakpoints in the series. Moreover, Clemente–Montanes–Reyes unit root test administers information about two possible structural breakpoints in the series through (i) an additive outliers (AO) model that seeks out an abrupt change in the mean of the series and (ii) an innovational outliers (IO) model that directs gradual shifts in the mean of the series. As a result, the AO model is more consistent for series having sudden structural changes when compared with gradual shifts.

The results reported in Table 2 show that trade balance, terms of trade, domestic income and foreign income are found to be non-stationary at their level. The variables are stationary at first-differenced form. This implies that the series are integrated at I(1). The unique level of integration of the variables leads us to apply the ARDL bounds testing approach to test cointegration among trade balance, terms of trade, domestic income and foreign income in the case of Pakistan over the period of 1978Q1–2012Q4. After confirming that all the series are integrated at I(1), the model is now ready for the analysis of cointegration relationship between the series using the ARDL bounds testing approach. The selection of an appropriate lag length of the variables using AIC and Schwarz’s Bayesian criterion criteria is a necessary step before the ARDL bounds test is applied (Tiwari & Shahbaz, 2013). It is described by Lütkepohl (2006) that AIC lag length criteria provide efficient and consistent results to capture dynamic relation between the variables. Therefore, using the AIC criteria, the optimal lag length of the variables is mentioned in the second column of Table 3 along with the results of the cointegration test. In order to determine whether cointegration between the variables exists or not, we have to compare our calculated F-statistic by following null hypothesis, that is, no cointegration with critical bounds such as LCB and UCB. The results reveal that there are two cointegrating vectors. This represents the cointegration relationship at 1 and 5 per cent significance levels when terms of trade and trade balance are treated as response variables. The results reported in Table 3 show that a long-run relationship among trade balance, terms of trade, domestic income and foreign income exists in the case of Pakistan.

Table 2.

Zivot–Andrews Structural Break Trended Unit Root Test

Variable	At Level		At First Difference
Variable	t-statistic	Time Break	t-statistic	Time Break
ln TB_t	–4.099 (2)	1981	–7.973 (0)*	1980
ln TT_t	–4.367 (0)	2003	–7.868 (2)*	1977
ln Y_t	–3.582 (1)	1993	–5.663 (0)**	1982
ln YF_t	–3.586 (1)	1993	–5.659 (0)*	1982

Source: Authors’ own.

Note: * and ** represent significance at 1% and 10% levels, respectively. The lag order is shown in parenthesis.

Table 3.

The Results of ARDL Cointegration Test

Bounds Testing to Cointegration			Diagnostic Tests
Estimated Models	Optimal Lag Length	F-statistics	$χ_{N O R M A L}^{2}$	$χ_{A R C H}^{2}$	$χ_{R E S E T}^{2}$
F_TB(TB/TT, Y, YF)	2, 2, 2, 2	7.308*	2.7823	[1]: 0.2974	[1]: 2.1921
F_TT(TT/TB, Y, YF)	2, 2, 1, 2	7.3836**	0.4410	[1]: 0.2411	[1]: 0.0416
F_Y(Y/TB, TT, YF)	2, 2, 2, 1	1.2917	0.9897	[1]: 0.0965	[1]: 0.2098
F_YF(YF/TB, TT, Y)	2, 2, 2, 1	0.9466	0.4973	[1]: 2.4996	[3]: 1.8647

Significant Level	Critical Values (T = 40)^#
Significant Level	Lower Bounds I(0)	Upper Bounds I(1)
1% level	6.053	7.458
5% level	4.450	5.560
10% level	3.740	4.780

Source: Authors’ own.

Note: The asterisks * and ** denote significance at 1%, 5% and 10% levels. The optimal lag length is determined by AIC. [] is the order of diagnostic tests. # denote that critical values are collected from Narayan (2005).

It is also noticed that the presence of structural break in the time series makes long-run relations biased, weak and unreliable. To avoid this issue, we have applied Gregory–Hansen (1996) structural break cointegration test that does not only remove this deficiency of the ARDL bounds testing approach to cointegration but also ensure the robustness of long-run relationship among trade balance, terms of trade, domestic income and foreign income. The Gregory–Hansen cointegration test is powerful over residual-based cointegration tests and adjusts the presence of one structural break in the series. The results are reported in Table 4. The results show that cointegration relationship exists among trade balance, terms of trade, domestic income and foreign income after allowing structural breaks in 1981 and 2003 which was investigated by applying fully modified ordinary least-square (FMOLS) approach. This approach indicated the statistical significance of the dummy variable for structural break in trade sectors of Pakistan. The structural breaks in trade reforms include competitiveness, concentration of exports and openness for agricultural in 1980s and implementation of national export policy in 2002. The empirical evidence indicated that there is a cointegration relationship between the variables as trade balance and terms of trade are used as forcing variables including dummy variable. This implies that a long-run relationship exists between the variables and long-run results are robust.

Table 4.

Gregory–Hansen Structural Break Cointegration Test

Model	T_TB(TB/TT, Y, YF)	T_TT(TT/TB, Y, YF)	T_Y(Y/TB, TT, YF)	T_YF(YF/TB, TT, Y)
Break period	1981	2003	1993	1993
ADF test	–4.7119**	–4.8279**	–3.3579	–3.7297
Prob. values	0.0000	0.0033	0.0014	0.0005

Source: Authors’ own.

Note: * shows significance at 1% level. The ADF statistics show the Gregory–Hansen tests of cointegration with an endogenous break in the intercept. Critical values for the ADF test at 1%, 5% and 10% are –5.13, –4.61 and –4.34, respectively. The results of other alternative tests such as Z_t and Z_a are available upon request from authors.

Table 5 reports the results of long- and short-run analyses. The results reveal that commodity terms of trade have a positive impact on trade balance in Pakistan. All else same, a 1 per cent increase in terms of trade leads trade balance by 0.2795 per cent.⁸ It is statically significant at 1 per cent level of significance. This validates the HLM effect which reveals that an increase in terms of trade improves trade balance in the case of Pakistan. These findings are consistent with Wong (2006) for Malaysia, Zortuk (2008) for Turkey but contradictory with Wong (2009) for Hong Kong. The positive impact of domestic income on trade balance shows the presence of absorption theory. Theory of absorption discloses that an increase in domestic income raises the demand for money and that would subsequently increase exports that will improve in trade balance. Keeping other things constant, a 1 per cent increase in per capita income adds in trade balance by 0.1229 per cent. Shahbaz, Awan and Ahmad (2011) also indicated a positive effect of income on trade balance but it is statistically insignificant. This may be due to the use of different proxies for terms of trade and data span as well as data frequency.⁹ A positive and statistically significant relationship is found from foreign income to trade balance in the case of Pakistan. This shows that an increase in income of world will raise demand for Pakistani products which in result improves trade balance of the country. A 1 per cent increase in world income is linked with 0.1094 per cent improvement in trade balance, all else same. This finding is same as that of Shahbaz et al. (2011) but coefficients are different due to differences in data frequencies and time spans.

Table 5.

Long- and Short-run Results

Dependent Variable = ln TB_t
Long-run Analysis
Variables	Coefficient	Std. Error	t-statistic	Prob. Values
Constant	–21.7047	4.7792	–4.5414*	0.0001
ln TT_t	0.2795	0.0758	3.6832*	0.0008
ln Y_t	0.1229	0.0421	2.9213*	0.0061
ln YF_t	0.1094	0.0618	1.7685***	0.0857
Short-run Analysis
Variables	Coefficient	t-statistic	Coefficient	t-statistic
Constant	0.0773	0.0309	2.4948**	0.0178
Δln TT_t	0.4751	0.0839	5.6617*	0.0000
Δln Y_t	0.4926	0.7427	0.6632	0.5118
Δln YF_t	–0.1169	0.0748	–1.5621	0.1278
ECM_t–1	–0.4575	0.1047	–4.3674*	0.0001
R ²	0.5092
F-statistic	8.5615*
Durbin–Watson	1.7245
Short-run Diagnostic Tests
Test	F-statistic	Prob. Values
$χ_{N O R M A L}^{2}$	0.4073	0.8157
$χ_{S E R I A L}^{2}$	0.6846	0.5117
$χ_{A R C H}^{2}$	0.1583	0.6930
$χ_{W H I T E}^{2}$	0.4056	0.9080
$χ_{R E M S A Y}^{2}$	0.0619	0.8050

Source: Authors’ own.

Note: * and ** show significance at 1% and 5% levels, respectively.

The short-run results are shown in Table 5. Our results indicate that terms of trade improve trade balance and it is statistically significant at 1 per cent level. The impact of national income (foreign income) has a positive (negative) impact on trade balance but insignificant. The results show the significance of lagged error term, that is, ECM_t–1 while estimate has a negative sign. The statiscally significant lagged error terms verify our established cointegration relationship among terms of trade, domestic income, foreign income and trade balance in the case of Pakistan. The coefficient of ECM_t–1 is –0.4575 and it is significant at 1 per cent level. We note that changes in terms of trade are modified by 45.75 per cent in each quarter. This implies that convergence would take more than 2 years to range the stable path of equilibrium which is an indication of very fast and significant adjustment process for Pakistan in any shock to trade balance equation.

The model is also tested for sensitivity analysis by using stability test of cumulative sum of recursive residuals (CUSUM) and the CUSUM of square (CUSUM_SQ) suggested by Brown, Durbin and Evans (1975). Hansen argued that misspecification of model may provide biased results that further lead to misinterpretation of results. The CUSUM and CUSUMsq tests are applied as diagnostic tests in order to ensure the adherence of parameters. Further, Brown et al. (1975) suggested that these tests provide help in analyzing the possible changes in parameters. The expected value of recursive residual is zero which leads us to accept that null hypothesis of parameter constancy is correct, otherwise not.

The plots of both CUSUM and CUSUMsq are shown in Figures 1 and 2 at 5 per cent level of significance. The results indicate that plots of both tests are within critical bounds at 5 per cent level of significance. Table 5 shows the results of diagnostic tests. The results signify that the short-run model successfully passes all the tests of normality, serial correlation, autoregressive conditional hetero-skedasticity, white heteroskedasticity and functioning form of the model. It is also evident that normality of residual term is proved by Jarque–Bera estimates, and no serial correlation is observed during short span of time. Similarly, no evidence of autoregressive conditional heteroskedasticity is found and same inference can be seen for white heteroskedasticity. The functional form of the short-run model is well justified and is confirmed by the estimates of Ramsey regression equation specification error (RESET) test. The stability and sensitivity analysis show that the ARDL and short-run results are stable and reliable for policy purpose regarding trade balance in the case of Pakistan.

Figure 1.

Source: Authors’ own.

Figure 2.

Source: Authors’ own.

The long-run and short-run analyses just show the impact of independent variables on dependent variables and ignore the cause and effect of the variables (direction of causal relationship between the variables). This is solved by applying the VECM Granger causality approach. Table 6 reports the empirical findings of the VECM Granger causality framework. In the long run, we find that the feedback effect exists between trade balance and terms of trade. Domestic and foreign incomes Granger cause trade balance. Terms of trade is Granger caused by domestic and foreign incomes. In shortrun causality analysis, trade balance Granger causes terms of trade and in result, terms of trade Granger cause trade balance. The unidirectional causality is found to be running from foreign income to trade balance. Foreign income is Granger caused by terms of trade and trade balance in Pakistan.

In an economic literature, it is argued that the Granger causality approaches, such as VECM Granger causality test, possess some limitations, that is, the causality test does not acquire the relative strength of causal relation between the variable beyond the specified time limit. This weakens the reliability of causality results by the VECM Granger approach. However, the innovative accounting approach (IAA) provides solution to this problem. The IAA is a combination of variance decomposition method (VDM) and impulse response function (IRF). The variance decomposition approach determines the response of the dependent actor to shocks stemming from independent actors. Therefore, we applied the IRF and VDM. The IRF is an alternative to VDM. Table 7 shows the results of VDM, while the IRF graph is shown in Figure 3. The variance decomposition approach indicates the magnitude of the predicted error variance for a series accounted for by innovations from each of the independent variable over different time horizons beyond the selected time period.

Table 6.

The VECM Granger Causality Analysis

Dependent Variable	Direction of Causality
	Short Run				Long Run	Joint Long- and Short-run Causality
	Δln TB_t–1	Δln TT_t–1	Δln Y_t_–1	Δln YF_t–1	ECT_t–1	Δln TB_t–1, ECT_t–1	Δln TT_t–1, ECT_t–1	Δln Y_t_–1, ECT_t–1	Δln YF_t–1, ECT_t–1
Δln TB_t	—	161786*[0.0000]	1.0827[0.3524]	4.8410**[0.0156]	–0.6522*[–4.7747]	—	17.1158*[0.0000]	19.0913*[0.0000]	14.5403*[0.0000]
Δln TT_t	8.4948*[0.0013]	—	0.8074[0.4561]	0.4382[0.6495]	–0.3649*[–3.3012]	6.8840*[0.0013]	—	5.4644*[0.0044]	4.4847**[0.0108]
Δln Y_t	0.1160[0.8908]	0.6694[0.5192]	—	0.0838[0.9197]	—	—	—	—	—
Δln YF_t	3.5974**[0.0394]	6.2084*[0.0053]	0.1025[0.9028]	—	—	—	—	—	—

Source: Authors’ own.

Note: *, ** and *** show significance at 1%, 5% and 10% levels, respectively.

Table 7.

Variance Decomposition Method

Period	S.E.	ln TB_t	ln TT_t	ln Y_t	ln YF_t
Variance Decomposition of ln TB_t
1	0.1275	100.0000	0.0000	0.0000	0.0000
2	0.1562	86.2587	0.3969	1.2284	12.1158
3	0.1601	85.1947	1.2152	1.2512	12.3387
4	0.1686	77.9464	8.1303	1.8452	12.0779
5	0.1849	66.2377	19.7662	2.2046	11.7913
6	0.2010	58.5778	28.3183	2.1639	10.9399
7	0.2127	55.6391	32.3480	1.9725	10.0402
8	0.2203	55.0003	33.7218	1.8441	9.4336
9	0.2252	54.9118	34.1771	1.8066	9.1043
10	0.2288	54.6859	34.5006	1.8031	9.0102
11	0.2321	54.2032	34.8863	1.7956	9.1148
12	0.2352	53.5989	35.3037	1.7804	9.3167
13	0.2381	53.0654	35.6542	1.7673	9.5129
14	0.2406	52.6955	35.8732	1.7632	9.6679
15	0.2425	52.4641	35.9705	1.7677	9.7976
Variance Decomposition of ln TT_t
1	0.1783	41.9810	58.0189	0.0000	0.0000
2	0.2702	29.7024	65.2860	0.2465	4.7648
3	0.3235	28.1993	65.6518	0.2811	5.8677
4	0.3612	28.7108	64.1752	0.6121	6.5017
5	0.3881	29.1961	62.6072	1.2806	6.9159
6	0.4077	29.4297	61.4196	2.1380	7.0125
7	0.4228	29.4519	60.5425	3.0846	6.9208
8	0.4351	29.3658	59.7933	4.0662	6.7745
9	0.4451	29.2535	59.0387	5.0600	6.6476
10	0.4533	29.1398	58.2397	6.0532	6.5671
11	0.4598	29.0137	57.4286	7.0251	6.5324
12	0.4650	28.8621	56.6581	7.9488	6.5308
13	0.4691	28.6857	55.9637	8.8008	6.5496
14	0.4724	28.4949	55.3545	9.5680	6.5824
15	0.4751	28.3009	54.8229	10.2482	6.6278
Variance Decomposition of ln Y_t
1	0.0191	2.3693	4.3456	93.2850	0.0000
2	0.0297	3.1383	8.5308	87.1981	1.1327
3	0.0372	3.1168	11.3966	84.3643	1.1221
4	0.0431	3.2345	14.6770	81.2431	0.8452
5	0.0479	3.6161	18.5024	76.9010	0.9802
6	0.0522	4.3280	22.1656	72.0632	1.4431
7	0.0557	5.3500	25.1300	67.4920	2.0278
8	0.0587	6.5435	27.2892	63.5186	2.6485
9	0.0612	7.7374	28.8031	60.1407	3.3187
10	0.0634	8.8163	29.8859	57.2269	4.0707
11	0.0653	9.7434	30.6955	54.6512	4.9097
12	0.0671	10.5343	31.3173	52.3426	5.8056
13	0.0687	11.2206	31.7886	50.2770	6.7135
14	0.0702	11.8268	32.1275	48.4482	7.5973
15	0.2667	22.3959	68.3167	0.9108	2.9001
Variance Decomposition of ln YF_t
1	0.0195	0.1174	6.3734	0.5540	92.9550
2	0.0308	1.1424	21.2148	0.2233	77.4193
3	0.0379	3.7166	27.2338	0.1550	68.8945
4	0.0417	7.69536	27.7798	0.1909	64.3338
5	0.0438	11.1668	26.4743	0.2991	62.0596
6	0.0451	13.0787	25.1818	0.3588	61.3805
7	0.0462	13.7400	24.1542	0.3470	61.7586
8	0.0474	13.7794	23.3561	0.3552	62.5092
9	0.0486	13.6448	22.8615	0.4432	63.0502
10	0.0498	13.5888	22.5890	0.5930	63.2290
11	0.0508	13.6736	22.3720	0.7672	63.1870
12	0.0517	13.8279	22.1331	0.9502	63.0887
13	0.0524	13.9615	21.8806	1.1448	63.0129
14	0.0531	14.0342	21.6435	1.3579	62.9643
15	0.0538	14.0535	21.4430	1.5905	62.9128

Source: Authors’ own.

Table 7 reports the empirical evidence regarding VDM and results reveal that innovative shocks of terms of trade, domestic income and foreign income contribute to trade balance by 35.97, 1.76 and 9.79 per cent and rest is explained by innovative shock of trade balance itself. Trade balance is explained by 52.46 per cent of its own shocks. The response of terms of trade due to shocks in trade balance is 28.30 per cent. The contribution of domestic income and foreign income to explain trade balance is 10.24 and 6.62 per cent, respectively.

Trade balance contributes to economic growth by 22.39 per cent. Domestic income is 68.31 per cent contributed by terms of trade. Foreign income contributes to domestic income by 2.90 per cent. Trade balance, terms of trade and domestic income contribute to foreign income by 14.05, 21.44 and 1.59 per cent, respectively. Overall results show that the feedback effect exists between trade balance and terms of trade. Trade balance and terms of trade cause domestic income. Foreign income is caused by terms of trade.

The IMF traces the time path of the impacts of shocks of independent variables on the dependent variables in a VAR system. The IMF is an alternative to VDM showing how long independent variable reacts to shock stemming in the dependent variables. We can see the magnitude of the response of trade balance to its own shock, those of terms of trade, domestic income and foreign income. Trade balance responds positively due to forecast error stemming in terms of trade. The response in domestic income is inverted U-shaped and foreign income responds positively after third time horizon due to forecast error stemming in terms of trade. The response in terms of trade, domestic income and foreign income is fluctuating due to forecast error stemming in trade balance. The contribution of domestic income to trade balance, terms of trade and foreign income is positive. Trade balance and terms of trade respond positively but response of foreign income in domestic income is minimal and fluctuating.

Concluding Remarks and Policy Recommendations

This article explored the validation of HLM effect by incorporating domestic income and foreign income in trade balance function in case of Pakistan. We have employed the ARDL bounds testing to investigate the long-run relationship between the variables. We find that trade balance, terms of trade, domestic income and foreign income are cointegrated for long-run relationship. Moreover, improvement in terms of trade improves trade balance which validates the HLM effect in Pakistan. Domestic income improves trade balance. Foreign income is positively linked with trade balance. The causality analysis by IAA reveals that terms of trade causes trade balance and trade balance leads terms of trade. Domestic income is cause of trade balance, whereas terms of trade causes foreign income.

Figure 3.

Source: Authors’ own.

The results of this study suggest that Pakistan has the capability to benefit from the improved terms of trade if foreign investment is attracted. However, in the past, the conventional way of dealing with deteriorating terms of trade has been the currency devaluation that has not been successful and balance of trade remained fragile to external shocks due to weak price elasticity of demand for Pakistani exports. This orthodox policy must be replaced by paying attention towards the new policy to increase foreign investment. It does not only improves balance of payment and growth but also improves terms of trade as a rebound in the long run. Due to increase in the foreign investment, the exports flourish both in terms of volume and composition. The analysis also suggests that if the investment is diverted more towards the production of such goods and services that cover larger part of country’s imports, it would guarantee higher outcomes of improved terms of trade.

We are aware about the limitations of our study. For instance, this article indicates the partial impact of income terms of trade on trade balance in Pakistan. To capture the complete impact of terms of trade on trade balance, commodity terms of trade should also be considered for empirical analysis (we have used income terms of trade). Few studies have investigated the impact of currency devaluation on trade balance (Shahbaz, Awan & Ahmad, 2011, Shahbaz, Jalil & Islam, 2012a) and domestic output (Shahbaz, Islam & Aamir, 2012b). The Marshal Learner Condition was also empirically investigated (Aftab & Aurangzeb, 2002; Aftab & Khan, 2008). Such studies could not provide any help to policymakers in designing a comprehensive trade policy for the removal of trade deficit. This shows the importance of further comprehensive research investigating the impact of currency devaluation and terms of trade (income terms of trade and commodity terms of trade) on trade balance, exports, imports and domestic output using time-varying cointegration, time-varying Granger causality and time-varying long-run estimates. The time-varying cointegration, causality and long-run estimates provide efficient empirical evidence due to their high explanatory powers.

Footnotes

Acknowledgements

The authors/author is/are grateful to the anonymous referees of the journal for their extremely useful suggestions to improve the quality of the article. Usual disclaimers apply.

Notes

References

Aftab

, & Aurangzeb

(2002). The long run and short run impact of exchange rate devaluation on Pakistan’s trade performance. Paper presented at South and Southeast Asia Econometrics Society Conference, held by Lahore University of Management Sciences (LUMS), Lahore, Pakistan.

Aftab

, & Khan

(2008). Bilateral J-curves between Pakistan and her trading partners (PIDE Working Paper No. 45). Pakistan Institute of Development Economics.

Appleyard

, & Field

A.J., Jr.

2001. International economics (4th ed.). Singapore: McGraw-Hill.

Bahmani-Oskooee

(1985). Devaluation and the J-curve: Some evidence from LDCs. The Review of Economics and Statistics, 67(3), 500–504.

Bahmani-Oskooee

(1986). Determinants of international trade flows: The case of developing countries. Journal of Development Economics, 20(1), 107–123.

Bahmani-Oskooee

(1995). Real and nominal effective exchange rates for 22 LDCs: 1971: 1–1990. Applied Economics, 27(7), 591–604.

Bahmani-Oskooee

, & Tankui

(2008). The black market exchange rate vs. the official rate in testing PPP: Which rate fosters the adjustment process? Economics Letters, 99(1), 40–43.

Bahmani-Oskooee

, Hegerty

S. W.

, & Xu

(2013). Exchange-rate volatility and US–Hong Kong industry trade: is there evidence of a ‘third country’effect?. Applied Economics, 45(18), 2629–2651.

Baum

C. F.

(2004, July). Topics in time series regression modeling. InUnited Kingdom Stata Users’ Group Meetings 2004 (No. 7). Stata Users Group.

10.

Berument

, & Dincer

(2005). Denomination composition of trade and trade balance: Evidence from Turkey. Applied Economics, 37(10), 1177–1191.

11.

Bouakez

, & Kano

(2008). Terms of trade and current account fluctuations: The Harberger–Laursen–Metzler effect revisited. Journal of Macroeconomics, 30(1), 260–281.

12.

Broda

(2004). Terms of trade and exchange rate regimes in developing countries. Journal of International Economics, 63(1), 31–58.

13.

Brown

, Durbin

L.J.

, & Evans

J.M.

(1975). Techniques for testing the constancy of regression relationships over time. Journal of the Royal Statistical Society, 37(2), 149–192.

14.

Chen

H. J.

, & Hsu

C. M.

(2006). Current account, capital formation and terms of trade shocks: a revisit of the Harberger-Laursen-Metzler effect. Journal of Economics, 88(2), 179–201.

15.

Correia

, Neves

J.C.

, & Rebelo

(1995). Business cycles in a small open economy. European Economic Review, 39(6), 1089–1113.

16.

Dibooglu

(2000). International monetary regimes and incidence and transmission of macroeconomic shocks: Evidence from the Bretton Woods and modern floating periods. Southern Economic Journal, 66(3), 590–608.

17.

Dickey

, & Fuller

W.A.

(1979). Distribution of the estimates for autoregressive time series with unit root. Journal of the American Statistical Association, 74(366a), 427–431.

18.

Elliot

, Rothenberg

T.J.

, & Stock

J.H.

(1996). Efficient tests for an autoregressive unit root. Econometrica, 64(4), 813–836.

19.

Fry

M.J.

(1986). Terms-of-trade dynamics in Asia: An analysis of national saving and domestic investment responses to terms-of-trade changes in 14 Asian LDCs. Journal of International Money and Finance, 5(1), 57–73.

20.

Gregory

A. W.

, & Hansen

B. E.

(1996). Practitioners corner: tests for cointegration in models with regime and trend shifts. Oxford bulletin of Economics and Statistics, 58(3), 555–560.

21.

Grilli

E.R.

, & Yang

M.C.

(1988). Primary commodity prices, manufactured goods prices, and the terms of trade of developing countries: What the long run shows. The World Bank Economic Review, 2(1), 1–47.

22.

Gylfason

, & Risager

(1984). Does devaluation improve the current account? European Economic Review, 25(1), 37–64.

23.

Harberger

A.C.

(1950). Currency depreciation, income, and the balance of trade. The Journal of Political Economy, 58(1), 47–60. Retrieved from http://www.journals.uchicago.edu/doi/10.1086/256897

24.

Haynes

S.E.

, & Stone

J.A.

(1982). Impact of the terms of trade on the US trade balance: A reexamination. The Review of Economics and Statistics, 64(4), 702–706.

25.

Hirose

K.I.

, & Ikeda

(2012). Decreasing and increasing marginal impatience and the terms of trade in an interdependent world economy. Journal of Economic Dynamics and Control, 36(10), 1551–1565.

26.

Hoque

(1995). Co-integrating relationship between terms of trade and current account deficit: The Australian evidence. Applied Economics Letters, 2(4), 130–133.

27.

Huang

K.X.

, & Meng

(2007). The Harberger–Laursen–Metzler effect under capital market imperfections. Journal of International Money and Finance, 26(6), 1001–1015.

28.

Hung-Ju

, & Hsu

C.-H.

(2009). Demand changes and real exchange rate dynamics in a finite-horizon model with sectoral adjustment costs. Southern Economic Journal, 75(4), 1191–1211.

29.

Idrees

, & Tufail

(2012). The Harberger–Laursen–Metzler effect: Evidence from Pakistan. The Lahore Journal of Economics, 17(2), 87–110.

30.

Islam

, Shahbaz

, & Tahir

M.I.

(2013). Income terms of trade and trade balance: The long run evidence from Bangladesh. Bangladesh Development Studies, 36(2), 109–122.

31.

Jääskelä

J.P.

, & Smith

(2013). Terms of trade shocks: What are they and what do they do? Economic Record, 89(285), 145–159.

32.

Jawaid

S.T.

, & Raza

S.A.

(2013). Effects of terms of trade on growth performance of India. Economic Modelling, 33, 940–946.

33.

Jawaid

S.T.

, & Waheed

(2011). Effects of terms of trade and its volatility on economic growth: A cross country empirical investigation. Transition Studies Review, 18(2), 217–229.

34.

Kaplinsky

(2006). Revisiting the revisited terms of trade: Will China make a difference? World Development, 34(6), 981–995.

35.

Kouassi

, Decaluwe

, & Colyer

(1998). Is it real? The long-run relation between terms of trade and current account deficits: The Ivory evidence. Applied Economics Letters, 5(7), 437–440.

36.

Laursen

, & Metzler

L.A.

(1950). Flexible exchange rates and the theory of employment. The Review of Economics and Statistics, 32(4), 281–299.

37.

Lütkepohl

(2006). Structural vector autoregressive analysis for cointegrated variables. AStA Advances in Statistical Analysis, 90, 75–88.

38.

Marquez

(1990). Bilateral trade elasticities. The Review of Economics and Statistics, 72(1), 70–77.

39.

Mendoza

E.G.

(1995). The terms of trade, the real exchange rate and economic fluctuations. International Economic Review, 36(1), 101–137.

40.

Narayan

P.K.

(2005). The saving and investment nexus for China: Evidence from cointegration tests. Applied Economics, 37(17), 1979–1990.

41.

, & Perron

(2001). Lag length selection and the construction of unit root test with good size and power. Econometrica, 69(6), 1519–1554.

42.

Nwachukwu

T.E.

, & Egwaikhide

F.O.

(2007, July). An error-correction model of the determinants of private saving in Nigeria. Paper presented at the African Economic Society (AES) Conference, Cape Town, South Africa.

43.

Otto

(2003). Terms of trade shocks and the trade balance of trade: There is a Harberger–Laursen–Metzler effect? Journal of International Money and Finance, 22(2), 155–184.

44.

Pedroni

(2001). Purchasing power parity tests in cointegrated panels. Review of Economics and Statistics, 83(4), 727–731.

45.

Pedroni

(2004). Panel cointegration: Asymptotic and finite sample properties of pooled time series tests with an application to the PPP hypothesis. Econometric Theory, 20(3), 597–625.

46.

Pesaran

M.H.

, Shin

, & Smith

(2001). Bounds testing approaches to the analysis of level relationships. Journal of Applied Econometrics, 16(3), 289–326.

47.

Santos-Paulino

A.U.

(2010). Terms of trade shocks and the current account in small island developing states. The Journal of Development Studies, 46(5), 855–876.

48.

Shahbaz

, Awan

, & Ahmad

(2011). The exchange value of the Pak-Rupee and Pak-Trade balance: An ARDL bounds testing approach. Journal of Developing Areas, 44(2), 69–93.

49.

Shahbaz

, Islam

, & Aamir

(2012b). Is devaluation contractionary? Empirical evidence for Pakistan. Economic Change and Restructuring, 45(4), 299–316.

50.

Shahbaz

, Jalil

, & Islam

(2012a). Real exchange rate changes and the trade balance: The evidence from Pakistan. The International Trade Journal, 26(2), 139–153.

51.

Sobrino

C.R.

(2011). Current account, productivity and terms of trade shocks in Norway. Applied Economics Letters, 18(18), 1745–1750.

52.

The World Bank. (2013). World Development Indicators CD-ROM. World Bank.

53.

Tiwari

A.K.

, & Shahbaz

(2013). Modelling the relationship between whole sale price and consumer price indices: Cointegration and causality analysis for India. Global Business Review, 14(3), 397–411.

54.

Tsen

W. H.

(2006). Is there a long-run relationship between trade balance and terms of trade? The case of Malaysia. Applied Economics Letters, 13(5), 307–311.

55.

Tsen

W. H.

(2009). Terms-of-trade and trade balance: Some empirical evidence of Asian economies. The International Trade Journal, 23(4), 422–457.

56.

Turner

(2006). Response surfaces for an F-test for cointegration. Applied Economics Letters, 13(8), 479–482.

57.

Warner

, & Kreinin

M.E.

(1983). Determinants of international trade flows. The Review of Economics and Statistics, 65(1), 96–104.

58.

Wong

H.T.

(2006). Is there a long-run relationship between trade balance and terms of trade? The case of Malaysia. Applied Economics Letters, 13, 307–311.

59.

Wong

H.T.

(2009). Terms of trade and trade balance in Korea, Hong Kong and Singapore. Labuan Bulletin of International Business & Finance, 7, 53–76.

60.

Wong

H.T.

(2010). Terms of trade and economic growth in Japan and Korea: An empirical analysis. Empirical Economics, 38(1), 139–158.

61.

Zivot

, & Andrews

(1992). Further evidence of great crash, the oil price shock and unit root hypothesis. Journal of Business and Economic Statistics, 10(3), 251–270.

62.

Zortuk

(2008). Testing the relationship between trade balance and terms of trade: The case of Turkey. Problems and Perspectives in Management, 6(2), 39–43.