Pakistan's Trade with China: What is the Potential Effect of CPEC?
Neighbouring countries engage in considerable trade with each other using land transportation. Pakistan shares a common border with China, but there is essentially no overland trade between these countries because of unavailability of feasible outes for economical transportation of goods by land. There are well developed land routes between Pakistan and India, but high tariff and nontariff barriers arising from strained relations have significantly impeded Pakistan-India trade. Trade with other neighbouring countries, Afghanistan and Iran, has also been hampered by conflict in Afghanistan and strife in tribal areas and Balochistan.
In earlier times, historically important trade route connecting India with Central Asia passed through what is now Pakistan. These routes are no longer used and Pakistan has relied largely on international trade by sea through the port of Karachi. As Nabi (2013) has pointed out, east-west trade expansion represents an important growth potential for Pakistan. Although significant progress has not been made in liberalising trade with India, CPEC provides an opportunity for Pakistan to expand its trade with China by land.
Will CPEC have a significant effect on Pakistan's overall trade with China? To address this question, I will present empirical evidence drawn from a research project undertaken with Antonio Marasco and Ijaz Nabi. (1) Our approach is to use a large data set to estimate a model that explains bilateral trade flows for most trading pairs in the world. If Pakistan-China trade is subject to additional costs that are not accounted for in the model, then actual trade flows between these countries should be less than the prediction of the model. An important finding of our study is that both Pakistan's imports from and exports to China are significantly less than the values predicted by the model. Higher transportation costs by land appear to be a plausible explanation of this result.
We use an up-to-date version of the Gravity model which has had considerable empirical success in explaining bilateral trade flows and has strong theoretical foundations. The traditional version of the Gravity model explains bilateral flows using only three variables: size of importing and of exporting countries measured by GDP, and bilateral transportation costs proxied by the distance between the trading pair. All three variables are significant determinants of bilateral trade flows. The significance of the distance index points to the importance of transportation costs in determining the volume of trade between country pairs. It is recognised that distance may not be an adequate measure of transportation costs, and bilateral trade costs also depend on tariff and nontariff barriers. Additional variables are thus added to proxy omitted sources of trade costs. Commonly used (binary) variables are: (1) whether the trading pair shares a common border (which would facilitate trade by land), (2) whether the pair uses a common language or shares a common colonial history (common language and colonial history would reduce informational barriers), and (3) whether the pair has a regional trade agreement or an RTA (which would reduce both tariff and non-tariff barriers).
Although the traditional Gravity model was not motivated by international trade theory, recent developments have shown that widely-used new-style international trade models imply a regression equation that has a similar form as the traditional gravity equation [Anderson and van Wincoop (2003); Helpman, et al. (2008)]. Key features of the new trade models are that countries trade differentiated goods produced under monopolistic competition and firms incur additional variable and possibly fixed trade costs to sell in the foreign markets. These features provide an explanation of intra-industry trade and account for the fact that only some of the firms in an industry engage in exports. The new trade models can be used to derive a general form of the Gravity model, in which bilateral trade flows depend on three multiplicative components: (1) a component that includes factors specific to the exporting country, (2) a component that includes factors specific to the importing country, and (3) a component that includes trade cost indexes specific to the trading pair.
Formally, the generalised Gravity model for a panel data set can be stated as
[X.sub.it,t] = [G.sub.t][S.sub.i,t] [U.sub.j,t] [[phi].sub.ij,t,] ... (1)
where [X.sub.ij,t] is the value of exports of country i to country j, [S.sub.i,t], [U.sub.j,t] and [[phi].sub.ij,t] represent, respectively, the effect of factors specific to the exporting country i, importing country j, and the pair i, j in period t; and [G.sub.t] is a term that can vary over time but does not depend on country characteristics. As exports of country i to country j equal the imports of country j from country i ([M.sub.jt,t]), (1) can also be used to explain bilateral imports. Note, however, that [X.sub.ij,t] can differ from [M.sub.ji,t] in data reported by i and j because of differences in reporting procedures and measurement errors. The revised equation differs from the traditional relation in that the importing- and exporting-country specific components, [S.sub.i,t] and [U.sub.j,t] include factors additional to a country's GDP. Ignoring these additional factors can lead to a significant bias in estimation.
Data Set and Pakistan's International Trade
To estimate (1), we assembled a panel data set that includes bilateral trade flows of 183 reporting countries with 253 partner countries from 2004 to 2013. For each year and reporting country, the bilateral data set includes data on the US dollar value of all exports to and all imports from each partner. The data set includes 515120 observations. The source of the data is the U.N. Comtrade Database. A striking feature of the data is that no trade is reported for a large number of country pairs. The proportion of country pairs with zero trade is around 60 percent in 2004 and higher in later years.
Pakistan's imports have grown from less than 20 billion in 2004 to to nearly 45 billion US dollars in 2013 while exports have increased at a slower pace from less than 15 billion to about 25 billion US dollars over the same period (Figure 1). The share of imports in GDP has fluctuated, but has not changed much between 2004 and 2013. The share of exports in GDP has declined from 2004 to 2013 (Figure 2).
Pakistan trades with over 175 countries. However, the bulk of Pakistan's trade takes place with a much smaller number of countries. China is an important trading partner of Pakistan, and was the second largest supplier of Pakistan's imports as well as the second largest consumer of Pakistan's exports in 2013 (Figures 3 and 4). In this year, imports from China acounted for 15 percent of Pakistan's total imports, and exports to China made up 11 percent of Pakistan's total exports. Large values of Pakistan's bilateral trade with China could simply be due to China's large size. To control for the size of the trade partner, we can look at Pakistan's bilateral imports and exports as a percentage of partner's GDP. Imports from China adjusted for size this way are 0.0007 percent, and size-adjusted exports are 0.0003. Indeed, in terms of adjusted values, among Pakistan's trading partners, China is comparable to Senegal and South Korea for imports and to Ghana and Grenada for exports.
Estimation and Results
We estimate the general form of the Gravity model represented by (1). A simple way to estimate this equation is to express it in its log-linear form. In this form, the components that include factors specific to exporting and importing countries (log [S.sub.i,t] log [U.sub.j,t]) are captured by time-variant dummy variables for exporting and importing countries. These country specific dummy variables control for not only long term national characteristics emphasised by trade models, but also short term influences such as underor over-valued exchange rate. For the component representing bilateral trade costs (log [Q.sub.ij,t]), we use the standard indexes represented by the log of distance (between major cities of the pair) and by dummy variables for common border, common official language, shared colonial history and membership in RTA's. We add a bilateral dummy variable for Pakistan and China (equal to 1 for observations representing Pakistan-China trade flows and equal to 0 for all other observations) to measure the effect of trade costs between Pakistan and China, which are not accounted for by the standard indexes of trade costs. A negative coefficient of Pakistan-China dummy variable would indicate a negative difference between the actual bilateral trade and the value predicted by the standard model (averaged over 2004-2013).
One problem with the log-linear form is that the dependent variable canot be expressed in logs if there is zero trade. One simple solution to this problem drops zero trade observations and fits the model to observations with non-zero trade using OLS regressions. This approach, however, inroduces an unknown selection bias. An alternative approach is suggested by a method proposed by Eaton and Kortum (2001) based on the Tobit model. According to this method (EK Tobit), there is a critical value of exports for each country (in a given period) such that if "ideal" trade falls below this level, zero exports are observed, otherwise observed exports equal ideal exports. A theoretical motivation of this approach is that ideal trade represent trade that would occur in the absence of fixed costs. In the presence of fixed costs, however, exports would be profitable only if they reach a critical level. In empirical implementation, critical level can be identified with the lowest value of exports to all destinations and the model can be estimated by a left-censored interval regression. Although this approach has some theoretical justification, it has an element of arbitrariness in its treatment of zero trade.
Another problem associated with the log linear form is highlighted by Santos Silva and Tenreyro (2006). They point out that the expected value of the log of the error term depends on its variance. Thus, the OLS estimator would be biased and inconsistent if the error is heteroskedastic and its variance depends on one or more of the explanatory variables. They show that under weak assumptions, the Poisson Pseudo Maximum Likelihood (PPML) estimator provides consistent estimates of the original nonlinear model. As PPML procedure does not require logarithmic transformation of the dependent variable, it can accommodate zero trade flows although the procedure is not motivated by a model that can explain zero trade. One limitation of this procedure is that a number of observations may need to be dropped in order to achieve convergence to a solution.
We use all three procedures discussed above to estimate the Gravity model with either bilateral imports or exports as the dependent variable. We use a subset of our data which includes only reporter countries--that is, we exclude some very small countries and regions which do not report export and import data, but are listed as trading partners of some reporting countries. In our empirical model, we include bilateral dummy variables for all neighbours of Pakistan; India, Afghanistan and Iran as well as China.
We focus on the effect of the Pak-China dummy variable, and show the estimatesof the coefficient of this variable in Table 1. (2) The estimates for both the import and export regressions are sensitive to the estimation method. The effect of the Pak-China dummy variable is the strongest in the EK Tobit regression and the weakest in the PPML regression. In all cases, the coefficient of the dummy varaible is significantly negative. An important implication of this finding is that trade costs between Pakistan and China are significantly higher than the level determined by the standard indexes of bilateral trade costs. Trade costs consist of transportation costs, tariffs and a variety of nontariff barriers. Pakistan-Chinal trade is not subject to any special tariff or nontariff barriers (informational barriers are already controlled for in the regression). Indeed, Pakistan has already negotiated a Free Trade Agreement with China and the first phase of this agreement was initiated in 2007. Thus the additional trade costs between Pakistan and China captured by the Pakistan-China dummy variable can only be attributed to higher tarnsportation costs (beyond those accounted for by the distance index) arising from barriers to a trade route by land.
How large are barriers to land transportation between Pakistan and China? One way to measure trade costs of these barriers is to express them in terms of an "equivalent tariff'. A tariff eqivalent of a given trade cost is the tariff level that would have the same effect on trade as the trade cost. As shown in the Appendix, the tariff equivalent of the additional Pakistan-China trade costs (asociated with transportation barriers) equals--([[beta].sub.PC]/[epsilon])100 in percentage points, where [[beta].sub.PC] is the coefficient of the Pak-China bilateral dummy and [epsilon] representsthe elasticity of trade with respect to trade costs. A review of estimates of trade elasticities by Head and Mayer (2014) indicates that the median value equals 5.03. Assuming this value, and using estimates of the Pak-China dummy variable form Table 1, we can readily calculate the tariff that would be equivalent to the additional Pak-China trade costs. For example if we use the OLS estimates for the import regression (with non-zero observations), a tariff as high as 37 percent would be equivalent to the additional costs. PPML estimates would imply lower and EK Tobit estimates higher values of the equivalent tariff.
Potential for Trade Expansion
By how much would CPEC reduce transportaion costs between Pakistan and China? The reduction would depend on how well the main corridor is connected with large markets and production centers in different provinces in Pakistan. It would also depend on the improvement of transportation links between Xinjiang region and the industrialised Eastern region in China. Although it is difficult to accurately measure the decrease in transportation costs due to CPEC, our estimates suggest that even a modest reduction in transportation costs would lead to a substantial expansion in trade between Pakistan and China. For example, even if there is only a 10 percent reduction in the Pak-China bilateral component in the general gravity model (1) and if the Pakistan- and China-specific components remain the same, our mid-level OLS estimates of the Pak-China dummy variable indicate a 19 percent increase in bilateral trade according to the import and a 17 percent increase according to the export regression.
These effects, however, represent only partial trade effects and ignore indirect effects arising from the effect of lower trade costs on country-specific factors. A general equilibrium model is needed to estimate the indirect effects. Trade models suggest that indirect effects are likely to further expand bilateral trade. For example, in the Anderson-van Wincoop model, the country-specific components of the exporting and importing countries can be expressed, respectively, as the ratios of real GDP or real aggregate expenditure to "multilateral-resistance" terms (see the Appendix). In this model, the decrease in Pak-China trade cost is likely to lead to further expansion in bilateral trade via the income-expenditure channel.
Recently, there has been much concern about the poor performance of Pakistan's exports and number of measure have been proposed to stimulate exports. This paper points to barriers to trade with neighbours--policy barriers to trade with India and transportation barriers to trade with China--as major long-term obstacles to Pakistan's export expansion. The evidence presented in the paper suggests that transportation barriers to Pak-China trade are high and CPEC has significant potential to reduce these barriers and expand Pakistan's trade with China.
Calculation of Equivalent Tariff
In the Gravity model (1), the bilateral component, [[phi].sub.ij,t], represents the trade effect of bilateral trade costs for countries i,j in period t. Letting [te.sub.ij,t], denote the tariff equivalent for bilateral trade costs for the pair, we can relate the bilateral component to the equivalent tariff as
[[phi].sub.ij,t] = [(1 + [te.sub.ij,t]).sup.-[epsilon]], ... (A1)
where [epsilon] is the elasticity of trade with respect to trade costs. We can use (Al) and estimates of the trade elasticity to measure the tariff equivalence of the additional Pakistan-China trade costs. For a given period, let [[phi].sub.PC,t] represent the estimated bilateral component for Pakistan and China, and [te.sub.PC,t] the associated equivalent tariff. Let [[phi]'.sub.PC,t] be the value of the bilateral component in the absence of additional costs (that is, the value when the Pak-China dummy variable is set equal to zero) and [te'.sub.PC,t] the corresponding equivalent tariff. The difference log([[phi].sub.PC,t]) - log([[phi].sub.PC,t]) equals the coefficient of the PakChina dummy variable. Denoting this coefficient by [[beta].sub.PC,t] and using (A1), we have
[te.sub.PC,t] - [te'.sub.PC,t] [approximately equal to] log(1 + [te.sub.PC,t]) - log(1 + [te'.sub.PC,t]) = [[beta].sub.PC,t]/[epsilon] ... (A2)
Note that [te.sub.PC,t] - [te'.sub.PC,t] in (A2) represents the tariff equivalent of additional Pak-China trade costs in fractions and would be multiplied by 100 to express it in percentage points.
Anderson-van Wincoop Model
This model can be expressed as
[X.sub.ij,t] = [[Y.sub.i,t]/[[OMEGA].sub.i,t]] [[E.sub.j,t]/[[phi].sub.j,t]] [[phi].sub.ij,t], ... (A3)
where the expressions [[OMEGA.sub.]i,t], = [summation over (k)] [[[phi].sub.ik,t] [E.sub.k,t]/[[PHI].sub.k]] and [[OMEGA.sub.]j,t], = [summation over (k)] [[[phi].sub.jk,t] [E.sub.k,t]/[[PHI].sub.k]] are weighted averages of bilateral trade costs of the exporting and importing countries with the rest of the world and are called inward and outward multilateral-resistance terms.
Table A1 Import Regressions Variables OLS OLS EK-Tobit PPML Indist -1.093 *** -1.505 *** -1.917 *** -0.597 *** -0.00841 -0.00932 -0.0137 -0.0115 pak_india -1.043 *** -5.035 *** -7.064 *** -1.747 *** -0.143 -0.194 -0.194 -0.143 pak_china -0.914 *** 1.869 *** -3.068 *** -1.063 *** -0.203 -0.0919 -0.138 -0.0749 pak+afg 1.414 *** 1.971 *** 3.068 *** 2.976 *** -0.105 -0.127 -0.388 0.151 pak_Iran -0.197 0.928 *** 1.351 *** -0.228 -0.242 -0.261 -0.253 -0.208 contig 0.817 *** 0.490 *** 0.0975 0.404 *** -0.0369 -0.039 -0.0606 -0.0268 comlang_off 0.878 *** 0.824 *** 1.362 *** 0.0651 *** -0.017 -0.0177 -0.0258 -0.0267 comcol 0.892 *** 0.822 *** 0.906 *** 0.558 *** -0.0243 -0.0236 -0.0352 -0.0539 rta 0.867 *** 0.549 *** 0.818 *** 0.493 *** -0.0163 -0.0169 -0.0237 -0.0216 InreporterGDP 1.044 *** -0.0026 In_partnerGDP 1.264 *** -0.00257 reporter dummies No Yes Yes Yes partner dummies No Yes Yes Yes year dummies No Yes Yes Yes Observations 188,438 192,421 237,369 232,156 R-squared 0.68 0.758 0.861 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Table A2 Export Regressions Variables OLS OLS EK-Tobit PPML Indist -1.146 *** -1.597 *** -2.193 *** -0.629 *** (0.00858) (0.00938) (0.0140) (0.0105) pak_india -4.081 *** -4.818 *** -7.078 *** -1.469 *** (0.137) (0.189) (0.231) (0.151) pak_china -0.905 *** -1.744 *** -2.821 *** -0.674 *** (0.172) (0.143) (0.369) (0.0997) pak+afg 1.810 *** 1.745 *** 2.186 *** 2.951 *** (0.212) (0.212) (0.0369) (0.157) pak_Iran -0 733 *** 0.295 0.561 *** -0.0746 (0.213) (0.257) (0.241) (0.194) contig 1.132 *** 0.699 *** 0.0102 0.506 *** (0.0351) (0.0382) (0.0643) (0.0265) comlangoff 0.882 *** 0.867 *** 1.491 *** 0.136 *** (0.0172) (0.0177) (0.0261) (0.0267) comcol 0.824 *** 0.674 *** 0.978 *** 0.408 *** (0.0249) (0.0237) (0.0351) (0.0512) rta 0.745 *** 0.531*** 0.952 *** 0.590 *** (0.0167) (0.0170) (0.0247) (0.0216) InreporterGDP 1.300 *** (0.00294) In_partnerGDP 0.901 *** (0.00261) reporter dummies No Yes Yes Yes partner dummies No Yes Yes Yes year dummies No Yes Yes Yes Observations 165,853 169,083 216,036 211,445 R-squared 0.650 0.749 0.870 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
Ehsan U. Choudhri <email@example.com> is Distinguished Research Professor, Department of Economics, Carleton University, Canada.
Anderson, James E. and Eric van Wincoop (2003) Gravity with Gravitas. American Economic Review 93, 170-92.
Choudhri, Ehsan, Antonio, Marasco, and Ijaz Nabi (2017) Pakistan's International Trade: the Potential for Expansion Towards East and West. (Mimeographed).
Eaton, J. and S. Kortum (2001) Trade in Capital Goods. European Economic Review 45:7, 1195-1235.
Head, Keith and Thierry Mayer (2014) Gravity Equations: Workhorse, Toolkit and Cookbook. In G. Gopinath, E. Helpman, and K. Rogoff (eds.) Handbook of International Economics (Elseveir), Vol.4, Chap. 3.
Helpman, E., M. Melitzand, and Y. Rubinstein (2008) Estimating Trade Flows: Trading Partners and Trading Volumes. Quarterly Journal of Economics 123:2, 441-487.
Nabi, Ijaz (2013) A Growth Vent Anchored in History and Geography. In Rashid Amjad and Shahid Javed Burki (eds.) Pakistan: Moving the Economy Forward. Lahore School of Economics, Chapter 14.
Santos Silva, J. and S. Tenreyro (2006) The Log of Gravity. The Review of Economics and Statistics 88:4, 641-658.
(1) See Choudhri, Marasco, and Nabi (2017).
(2) More detailed estimates of the regressions including the regression for the traditional Gravity model are presented in the Appendix Tables A1 and A2.
Caption: Fig. 1. Pakistan's Aggregate Imports and Exports, 2004-13
Caption: Fig. 2. Share of Pakistan's Imports and Exports in GDP, 2004-13
Table 1 Estimates of Pak-China Dummy Variable Import Regressions Export Regressions OLS (non-zero obs.) -1.869 *** -1.744 *** (0.0919) (0.143) EK Tobit -3.068 *** -2.821 *** (0.138) (0.218) PPML -1.063 *** -0.674 *** (0.0749) (0.0997) Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Fig. 3. Sources of Pakistan's Imports UAE 18% Japan 15% Germany 9% China 9% Malaysia 5% Indonesia 4% Kuwait 4% India 4% Other countries 3% Saudi Arabia 3% USA 26% Note: Table made from pie chart. Fig. 4. Destinations of Pakistan's Exports UAE 15% Japan 11% Germany 8% China 7% Malaysia 6% Indonesia 4% Kuwait 3% India 3% Other countries 3% Saudi Arabia 2% USA 38% Note: Table made from pie chart.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Mahbub-ul-Haq Memorial Lecture; China-Pakistan Economic Corridor|
|Author:||Choudhri, Ehsan U.|
|Publication:||Pakistan Development Review|
|Date:||Dec 22, 2017|
|Previous Article:||Role of Productivity, Quality, and Innovation in Making CPEC Work for Pakistan.|
|Next Article:||Sovereign Development Toward a Grand Strategy for Pakistan.|