Can Duty Drawbacks Have a Protectionist Bias? Evidence from Mercosur

Evidence from Mercosur suggests that eliminating duty drawbacks for intra-regional exports would lead to increased counterlobbying against protection of intermediate products. Without the duty drawback, the common external tariff would have been an estimated 3.5 percentage points (25 percent) higher on average. Duty drawback (or rebate) systems reduce or eliminate the duties paid on imported intermediate goods or raw materials used in the production of exports. When a firm imports an intermediate product for use in the production of an export good, tariff payments on the imported intermediate good are either waived (duty drawback) or returned to the producer once the final product is exported (rebate). These incentive systems are often justified on the grounds that they tend to correct the anti-trade bias imposed by high tariff levels. The problem with this line of reasoning is that it assumes that tariffs are predetermined policy variables; if they were, the easiest way to reduce their anti-trade bias would be to eliminate them. But this is rarely done because existing levels of protection correspond to a political economy equilibrium difficult to modify in the presence of lobbying pressures. Cadot, de Melo, and Olarreaga show that in a political economy setting where tariffs and duty drawbacks are endogenously chosen through industry lobbying, full duty drawbacks are granted to exporters that use imported intermediate goods in their production. This in turn decreases their incentives to counterlobby against high tariffs on their inputs. Indeed, under a full duty drawback regime, tariffs on intermediate goods are irrelevant to exporters because they are fully rebated. In equilibrium, higher tariffs will be observed on these goods. Creating a regional trading bloc alters the incentives by eliminating duty drawbacks on intraregional exports, which leads to lower tariffs for goods that intraregional exporters use as inputs. Evidence from Mercosur suggests that eliminating duty drawbacks for intra-regional exports would lead to increased counterlobbying against protection of intermediate products. Cadot, de Melo, and Olarreaga estimate that without this mechanism, the common external tariff would have been 3.5 percentage points (25 percent) higher on average. This paper - a product of Trade, Development Research Group - is part of a larger effort in the group to study the political economy of trade policy. The research was funded by the Bank's Research Support Budget under the research project "The Anti-export Bias of Duty Drawbacks." The authors may be contacted at olivier.cadot@hec.unil.ch, demelo@ibm.unige.ch, or molarreaga@worldbank.org.


Non-Technical Summary
Reduction of the anti-export bias of existing trade policies in developing countries has been a key component of trade policy reform packages since the early eighties. A quasi universal instrument t o a c hieve this objective, has been the creation or improvement of duty-drawback systems and temporary admission regimes. The pervasiveness of duty-drawback systems is evident in a sample of 42 developing countries that have undertaken WTO Trade Policy Reviews, and in which all but three countries Benin, Hong Kong and Singapore were found to have in place some form of duty-drawback system.
Duty drawbacks or rebate systems reduce or eliminate the duties paid on imported intermediates or raw materials that are used in the production of exports. When a rm imports an intermediate product for use in the production of an export good, tari payments on the imported intermediate are either waived duty drawback system or returned to the producer once the nal product is exported rebate system.Thus, the objective of these mechanisms is to promote exports by partially or fully compensating exporters for the anti-trade bias of existing protection, since exporters have access to their imported inputs at world prices in spite of the existing levels of tari protection. These incentive systems are often justi ed on the grounds that they tend to correct the anti-trade bias imposed by high tari levels. The problem with this line of reasoning is that it assumes that tari s are predetermined policy variables. If such w as the case, the easiest way of reducing their anti-trade bias would be simply to eliminate them. The reason why this solution is rarely achieved is that existing levels of protection correspond to a political-economy equilibrium that is difcult to modify in the presence of lobbying pressures. A political economy approach, such as the one used in this paper is thus necessary to understand the rationale for such systems.
Our analysis is cast around the many recently-formed regional blocs that have partially or completely phased out duty-drawbacks on intra-regional exports. This type of policy change, which w e derive endogenously as a response to shifting government incentives, is likely to trigger a re-balancing of domestic power and incentives.
Using a common agency model of endogenous protection with intermediate goods, we show that duty-drawback and rebate systems decrease exporter incentives to lobby against protection on imported intermediate goods. Indeed, under a full dutydrawback regime, tari s on intermediates are irrelevant to exporters since they are fully rebated. This leads, ceteris paribus, to higher levels of protection on intermediate goods heavily used in export industries, penalizing non-exporting users of such goods. Note that whether duty-drawbacks systems are desirable from a welfare perspective remains an open question that we are planning to address in future research.
i We then analyze how the formation of a regional trading bloc alters these incentives. Intra-regional exporters may be, in terms of pro t levels, better o than before since they are now the bene ciaries of the area's external tari s, but at the margin, their incentive to lobby against intermediate-good protection rises as duty-drawback and rebate schemes are endogenously eliminated on intra-regional exports. In equilibrium, this results in a lower level of external protection for those intermediate goods that are used heavily in sectors where intra-regional exports are large. Indirect e ects of that type are at the very least consistent with the reduction in mfn tari s that has accompanied the New Regionalism" that has been observed by several authors.
We then try to see if the model's predictions are borne out in the case of the Common Market of the Southern Cone Mercosur which i s a n i n teresting case-study for at least three reasons. First, Mercosur members all had duty-drawback systems for exporters in place when they negotiated their Common External Tari cet in 1994 which will no longer be allowed for intra-regional trade, once convergence to the cet is achieved in December 2000. Second, Mercosur has been identi ed as one of the recent regional blocs satisfying the New Regionalism"'s characteristics that preferential tari reductions are accompanied by general mfn tari reductions. Third, recent w ork has shown that industry lobbying was an important determinant of Mercosur's cet.
Our approach to testing the political-economy h ypothesis explained above proceeds in two steps. First, we assume that the Mercosur's CET is endogenously determined through cooperative bargaining among its members, and there is evidence that this was indeed the case as will be discussed later. Then, using input-output tables to trace the use of imported intermediates in downstream industries, we test whether deviations from the optimal cet are correlated with the intensity of input use in downstream industries. The interest of the exercise is two-fold. First, it gives a statistical indication of the magnitude of the lobbying e ects attributable to the elimination of duty-drawback s c hemes. Second, it provides an indirect test of the common-agency model of endogenous protection, whose empirical predictions have been the object of some controversy.
We nd that during the negotiations for the Mercosur's cet, counter-lobbying against protection of intermediate goods increased following the elimation of dutydrawbacks for intra-regional exports. In the absence of this mechanism, we estimate that Mercosur's cet would have been on average 3.5 percentage points higher 25 percent higher.

Introduction
A k ey objective of trade reforms initiated since the early eighties was the reduction of the anti-export bias of existing trade policies in developing countries. For example, a component of the World Bank recommendations in their trade loans was the creation or improvement of duty-drawback systems and temporary admission regimes see Krueger and Rajapatirana, 1999. The pervasiveness of duty-drawback systems is also evident in a sample of 42 developing countries having undertaken WTO Trade Policy Reviews, for which Michalopoulos 1999 nds that all but three countries Benin, Hong Kong and Singapore have in place some form of duty-drawback system.
Duty drawbacks or rebate systems reduce or eliminate the duties paid on imported intermediates or raw materials that are used in the production of exports. When a rm imports an intermediate product for use in the production of an export good, tari payments on the imported intermediate are either waived duty drawback system or returned to the producer once the nal product is exported rebate system. 1 Thus, the objective of these mechanisms is to promote exports by partially or fully compensating exporters for the anti-trade bias of existing protection, since exporters have access to their imported inputs at world prices in spite of the existing levels of tari protection.
These incentive systems are often justi ed on the grounds that they tend to correct the anti-trade bias imposed by high tari levels. The problem with this line of reasoning is that it assumes that tari s are predetermined policy variables. If such was the case, the easiest way of reducing their anti-trade bias would be simply to eliminate them. The reason why this solution is rarely achieved is that existing levels of protection correspond to a political-economy equilibrium that is di cult to modify in the presence of lobbying pressures. Thus, it is di cult to understand the rationale for such systems or to get a complete picture of their incentive e ects in the absence of a political-economy approach, which is the one taken in this paper.
Moreover, many recently-formed regional blocs have partially or completely phased out duty drawbacks on intra-regional exports. 2 This type of policy change, which w e derive endogenously as a response to shifting government incentives, is likely to trigger a re-balancing of domestic power and incentives see Lawrence, 1999. This process and its e ect on the structure of the regional bloc's external tari s is the focus of our analysis.
Using a common agency model of endogenous protection Grossman-Helpman, 1994 with intermediate goods Cadot, de Melo and Olarreaga, 1997, we show that duty-drawback and rebate systems decrease exporter incentives to lobby against protection on their imported intermediate goods. Indeed, under a full duty-drawback regime, tari s on intermediates are irrelevant to exporters since they are fully rebated. This leads, ceteris paribus, to higher levels of protection on intermediate goods heavily used in export industries, penalizing non-exporting users of such goods.
The formation of a regional trading bloc alters these incentives. Intra-regional exporters may be, in terms of pro t levels, better o than before since they are now the bene ciaries of the area's external tari s, but at the margin, their incentive to lobby against intermediate-good protection rises as duty-drawback and rebate schemes are endogenously eliminated on intra-regional exports. In equilibrium, this results in a lower level of external protection for those intermediate goods that are used heavily in sectors where intra-regional exports are large. Indirect e ects of that type are at the very least consistent with the reduction in mfn tari s that has accompanied the new regionalism". Indeed, several authors Ethier, 1998a; Lawrence, 1999 have argued that one of the important c haracteristics of the new wave of regional trade agreements is that they are accompanied by simultaneous reductions in mfn tari s. 3 2 In the case of nafta, for example, Canadian exporters to the US market do not necessarily bene t from the full duty-drawback, but the level of drawback is determined by the mimimum amount b e t ween the tari revenue paid on their inputs and the tari revenue they avoid on their exports by bene tting from intra-Nafta free-trade.
3 Some authors, including Ethier 1998b or Freund 2000, have argued that the causality m a y If substantial enough, they might e v en partially explain it.
We then try to see if the model's predictions are borne out in the case of the Common Market of the Southern Cone Mercosur. Mercosur is an interesting casestudy for at least three reasons. First, Mercosur members all had duty-drawback systems for exporters in place when they negotiated their Common External Tari cet in 1994. And, as stipulated in Article 12 of mercosur cmc dec No. 10 94, these will no longer be allowed for intra-regional trade, once convergence to the cet is achieved in December 2000. Second, Ethier 1998a has identi ed Mercosur as one of the recent regional blocs satisfying the New Regionalism"'s characteristics that preferential tari reductions are accompanied by general mfn tari reductions Estevadeordal et al. 1999. Third, recent w ork Olarreaga et al. 1999 has shown that industry lobbying was an important determinant of Mercosur's cet.
Our approach to testing the political-economy h ypothesis explained above proceeds in two steps. First, we assume that the Mercosur's CET is endogenously determined through cooperative bargaining among its members, and there is evidence that this was indeed the case as discussed later. Then, using input-output tables to trace the use of imported intermediates in downstream industries, we test whether deviations from the optimal cet are correlated with the intensity of input use in downstream industries. The interest of the exercise is two-fold. First, it gives a statistical indication of the magnitude of the lobbying e ects attributable to the elimination of duty-drawback s c hemes. Second, it provides an indirect test of the political-support model of endogenous protection, whose empirical predictions have been the object of some controversy. 4 be reversed from unilateral or multilateral liberalization towards regional integration. Others, starting with Bhagwati 1993 have expressed the fear that, on the whole, the surge in RTAs is likely to diminish the incentives to engage in multilateral non-discriminatory tari reductions. For formal models that lead to this conclusion, see Levy 1997 andKrishna 1998. For a comprehensive review of the literature see Panagariya 2000. 4 Our empirical results indirectly con rm the prediction strenght of both the common-agency model of Grossman and Helpman 1994 and the Nash bargaining model of Maggi and Rodriguez-Clare 1998.
To a n ticipate our main result, we nd that during the negotiations for the Mercosur's cet, counter-lobbying against protection of intermediate goods increased following the elimation of duty-drawbacks for intra-regional exports. In the absence of this mechanism, w e estimate that Mercosur's cet would have been on average 3.5 percentage points higher 25 percent higher.
The remainder of the paper is organized as follows. Section 2 develops a politicaleconomy model a la Grossman and Helpman 1994 where, in the presence of imported intermediate products and duty-drawbacks, there are incentives to lobby for exporters. The characteristics and implications of a duty-drawback s c heme are studied and compared with the alternative o f n o s u c h incentive s c heme. Section 3 focuses on how incentives to lobby and the resulting endogenously determined tari structure is likely to change when two countries enter in a customs union cu. Section 4 applies the model to the case of Mercosur, where, the formation of the cu resulted in a removal of duty rebates for exports to the region. Section 5 concludes.

Tari drawbacks and lobbying
We explore exporter incentives to lobby against tari s on intermediates products in a Grossman-Helpman 1994 model, 5 to which w e add an intermediate good following Cadot, de Melo and Olarreaga 1997. 6 Consider, then, a small price-taking open economy, that produces 4 traded goods. Goods 0 and 1 are exported but nevertheless consumed at home while 2 and 3 are import-competing. Goods 0, 1 and 2 are nal goods while 3 is an intermediate that is also used in nal consumption. Good 0 is the num eraire. It is produced with labor only under Constant Returns to Scale crts; good 3 the intermediate is produced with labor and sector-speci c capital; goods 1 and 2 are produced with labor and sector-speci c capital, and with good 3. The focus of the model is on the interaction between sectors 1 exported nal good and 3 imported intermediate.
Technologies in sectors 1 and 2 are Leontief between intermediate consumption and value added, with value-added being generated with labor and sector-speci c capital. Omitting sector-speci c capital, value-added is an increasing and concave function of labor, f i `i, i = 1 ; 2; and output is given by x i a xed input-output coe cient. The tari on good i is t i ; export goods 0 and 1 are neither taxed nor subsidized.
In the three non-num eraire sectors, the presence of sector-speci c capital implies diminishing returns to labor and hence rents accruing to owners of sector-speci c capital, who are also the rms' residual claimants. These rents are a ected by trade policy: they are the reason for lobbying and the source of political contributions. The presence of a good produced under crts with labor only has the e ect of pinning down the wage rate, so that there is no interindustry rivalry on the labor market which simpli es considerably the model's structure. 7 Moreover, we assume that capital ownership is su ciently concentrated for lobbies to disregard the e ect of protection on consumer prices.
Together, these assumptions ensure that the only source of interindustry rivalry are input-output linkages. Given this supply-side structure, the political line-up is as follows: sector 3 lobbies for protection, sector 1 lobbies against the protection of sector 3, and sector 2 lobbies for its own protection and against the protection of sector 3. We treat the game as if lobbies 1, 2 and 3 were acting as non-cooperative principals vying for in uence over their common agent, the government, following the common-agency literature.
Consumers have identical tastes represented by a quasi-linear and additive utility function u = c 0 + uc 1 + uc 2 + uc 3 where c i stands for the consumption of nal good i and the function u has the usual properties. It follows that, given that good 1 is consumed but also exported hence not protected while goods 2 and 3 are imported, in equilibrium u 0 c 1 = p ? 1 , u 0 c 2 = p ? 2 1 + t 2 and u 0 c 3 = p ? 3 1 + t 3 , where p ? i is the world price of good i and t 1 and t 2 are ad-valorem tari s.
3 2 y 2 0; @ 2 @t 2 = p ? 2 y 2 0; 6 @ 3 @t 3 = p ? 3 y 3 0: Note that with a full duty-drawback i.e. when = 1, producers of good 1 become indi erent to the level of the tari on the intermediate good, i.e. @ 1 =@t 3 = 0. This does not apply, h o wever, to producers of good 2, who serve the domestic market and are consequently not eligible for the duty-drawback.
As consumers have identical quasi-linear preferences, social welfare is the sum of income and consumer surplus, income being itself the sum of labor income, industry pro ts and tari revenue. The tari -revenue term is complicated by the presence of the duty-drawback s c heme, which segments the intermediate-good market between sector-1 users, who are eligible for it, and sector-2 users, who are not. Given the availability of the scheme, sector-1 rms use only imported intermediate goods. In order to avoid a taxonomy of cases, we will assume that the domestic output of the intermediate good is not enough to cover the needs of sector 2 and nal users, who accordingly use a mixture of home-produced and imported intermediates all priced at p ? 3 1+t 3 . Let m 3 = c 3 + 1 y 1 + 2 y 2 ,y 3 stand for good 3's imports. Net tari revenue on good-3 imports is, after deduction of duty-drawback repayments, p ? 3 t 3 m 3 , 1 y 1 : Given this, welfare is: W = w`+ 1 t 1 ; t 3 ; + 2 t 2 ; t 3 + 3 t 3 + p ? 2 t 2 m 2 + p ? 3 t 3 m 3 , 1 y 1 i=0`i : Welfare terms are thus @W @t 2 = p ? 2 y 2 + p ? 2 m 2 + p ? 2 t 2 m 0 2 + u 0 c 2 c 0 2 p ? 2 , p ? 2 c 2 , p ? 2 1 + t 2 c 0 2 p ? 2 = p ? 2 t 2 m 0 2 0 7 for protection in sector 2, and @W @t 3 = p ? 3 y 3 , p ? 3 1 , 1 y 1 , p ?
3 2 y 2 + @ @t 3 p ? 3 t 3 m 3 , 1 y 1 for protection in sector 3: But note that c 3 + 1 y 1 + 2 y 2 = y 3 + m 3 ; and that @ @t 3 p ? @W @t 3 = p ? 3 t 3 m 0 3 + 1 , 2 1 y 0 1 : 8 Finally, the welfare e ect of the duty-drawback at rate is @W @ = @ 1 @ + @ @ p ? 3 t 3 m 3 , 1 y 1 = 0: 9 The absence of a welfare e ect of the duty-drawback re ects the fact that it is a pure transfer entailing no welfare loss, since it a ects neither the consumer price of good 1 nor its producer price, but only the price of an input in a Leontief production function. 8 Combining 6 with 7, 8, and 9 gives @G @t 2 = p ? 2 y 2 + at 2 m 0 2 ; 10 @G @t 3 = p ? 3 , y 3 , 1 , 1 y 1 , 2 y 2 + at 3 m 0 3 + 1 , 2 1 y 0 1 ; 11 @G @ = p ? 3 t 3 1 y 1 : 12 The function G is globally concave i n t 2 and t 3 , so the second-order condition holds for these instruments. We will assume in addition that interior solutions hold for both instruments since otherwise the whole problem of endogenous protection would become irrelevant. By contrast, 12 is clearly nonnegative no matter what, and strictly positive whenever t 3 0 the duty-drawback is irrelevant i f t 3 is zero. Therefore in equilibrium there is full drawback, i.e. = 1. Using this condition to simplify 11, the rst-order condition for t 3 reduces to y 3 , 2 y 2 + at 3 m 0 3 = 0 ; or t 3 = y 3 , 2 y 2 ,am 0 3 : 13 Finally, setting 10 equal to zero and solving for t 2 gives t 2 = y 2 ,am 0 2 : 14 Together, t 2 ; t 3 and = 1 de ne the initial pre-rta equilibrium. 9 Recall that sector 8 Note that this implies that duty-drawbacks cannot be justi ed on welfare grounds under the assumptions in this paper. Panagariya 1992 provides a more general model where duty-drawbacks have a m biguous e ects on welfare. 9 Note that we w ould have obtained similar optimal tari s had we assume that the government 1 is the export sector and that sector 2 along with sector 3 is an import competing sector. Hence sector 2 is not eligible for a duty-drawback. Compared to a model with no intermediates, what is new here is the negative term involving the output of sector 2 in the numerator of 13. This term re ects counter-lobbying by producers in sector 2 against protection in sector 3, because protection raises the price of the intermediate good and consequently hurts their pro ts. However, sector 1, is not active in this counter-lobbying, because the full duty-drawback shelters it from the cost of intermediate-good protection. The absence of counter-lobbying by sector 1; because it bene ts from full duty-drawback, tends ceteris paribus, to raise the equilibrium level of protection in sector 3 and therefore to hurt sector 2 and consumers. Thus, duty-drawbacks have a protectionist bias. 10 Note, incidentally, that the duty-drawback s c heme creates intra-industry trade even under perfect competition. The reason is that producers of good 1 will always choose to export all their output if doing so makes them eligible for the duty-drawback encouraging exports is indeed the scheme's objective while domestic consumption of good 1 will be entirely covered by imports. A country having a duty-drawback scheme will then both export and import good 1. export substantial amount of manufactures to each other see Yeats, 1998. It would also apply to countries in Asia that have duty-drawbacks and have the potential to export substantial amounts of manufactures to each other.
Suppose then that the governments of A and B agree on an e cient solution for the cet, and then bargain over how they share the bene ts of cooperation via monetary transfers shallow i n tegration" in the terminology of Cadot et al., 1999. 11 In this setup, let us rst examine under which conditions duty-drawbacks would be eliminated for intra-regional exports, and if so what are the consequences for external tari s.

Elimination of duty-drawbacks after CU formation
Article 12 of mercosur declaration cmc dec No. 10 94 stipulates that intra-regional exporters can no longer bene t from duty-drawbacks. To explain Mercosur's decision let us write the CU rst order condition for the optimal level of duty-drawback for intra-regional trade. 12 In view of the application in section 4, we use superscript M for Mercosur to denote the customs union. Recalling that country B absorbs all of country A exports of good 1 and using 12: @G M @ = @G A @ + @G B @ = p ? where G M is the CU's objective function given by the sum of A's and B's objective function and i is the cet in sector i. The rst term in 15 is as before the gain for the government of country A of imposing a duty-drawback. The second term is 11`S hallow i n tegration' boils down to setting up the problem as if a common agency was maximizing the sum of member-country government's welfare functions, whereas under`deep integration', there is a single agency in charge of trade policy that takes into account union-wide lobbying, so that unlike shallow i n tegration, country characteristics are aggregated. 12 It could be argued that allowing for duty-drawbacks within a CU makes little economic sense as this may create incentives for trade cross-hauling. The objective here is to show under which conditions it may also make little political-economy sense. now the loss for country B in terms of tari revenue of allowing exporters in A to bene t from duty-drawbacks. Indeed, the existence of a duty-drawbacks in intraregional trade shifts outwards the export supply of exporters in A, which reduces tari revenue for country B as it diverts from rest-of-the world imports.
Using 6 and noting that @y A 1 =@ = @ 2 A 1 =@ @ 1 , the right-hand-side of 15 is always negative, and therefore at the optimum there are no duty-drawbacks M = 0, if: where A S 1 stands for the price-elasticity of supply of good 1 in country A. T h us, if the weight given to social welfare in country B's objective function is su ciently high, then duty-drawbacks will be eliminated. The reason is that the gains in terms of contributions by producers of good 1 in A are not su cient to compensate for the loss in terms of tari revenue in B. Taking as an example the average tari in mercosur which is around 14 percent, and assuming an average price-elasticity of supply around 1, 16 implies that a B has to be larger than 8 for duty-drawbacks to be completely eliminated. Note that if import demand functions are non-concave, the second-order condition for the government's problem requires a B to be larger than 1. Estimates of a for the United States in the 1980s yield values between 50 and 88 see Goldberg and Maggi, 1999. Given that mercosur has eliminated duty-drawbacks for intra-regional trade, one can infer that the constraint in 16 was satis ed. The implications for cet levels are studied in the next section.

CET after duty-drawback elimination
Several trade patterns can be envisaged when the CU is formed. Since we are interested in situations where lobbying takes place, we assume that exporters of good 1 i n A sell in a protected market in B. This will be the case, if the following two assumptions are met.
First, let us assume that there is protection in sector 1, so that we can concentrate on the conditions when it will be pro table for A's producers to sell in B. Since duty-drawbacks are eliminated for intra-regional trade, this creates an incentive for exporters to re-direct their production from the region to the rest-of-the world. In such a situation the increase in counter-lobbying will not occur. In order to avoid this, we need to assume that producers always nd it more pro table to sell within the regional market with protection and paying duties on intermediate purchases, than selling on the world market at the world market price, but not paying duties on intermediate purchases. Letting 1 be the CET on good 1, as before, this requires that: Assumption 1, which is similar to the expression determining whether rms will sell in markets protected by rules of origin requirements see e.g. Krueger, 1993, guarantees that selling in the market protected by the cet is more pro table than selling in the world market even if it involves forsaking the bene t of the duty-drawback.
Second, we suppose that A and B's combined output in sector 1 is not enough to serve the entire cu demand at the equilibrium price. Again, this situation is most representative o f rtas among developing countries, at least among countries that do not di er too much in size like Argentina and Brazil. That is, letting c A 1 +c B 1 be the cu consumption of good 1 at price p ? 1  Thus, after the cu is formed, rms in sector 1 sell in a protected market the partner country's, which w as not the case before. This gives them a direct benet. However, they are no longer eligible for the duty-drawback s c heme so that they become sensitive to the rate of protection of the intermediate good. Through the truthfulness restriction, the degree of their sensitivity, given by the derivative of their pro t function with respect to the cet in sector 3, i.e. 3 ; determines the intensity of their counter-lobbying. 13 Turn now to the determination of the cet in sector 3. Recalling that dutydrawbacks are disallowed for intra-regional trade, i.e., that M = 0, the rst order condition to the cet  Protection for sectors producing nal goods would be given by an expression identical to 19, except that it would exclude the element that captures counter-lobbying on intermediate products.
Return to 19. To simplify, but also in view of empirical tractability for the application that follows, assume identical price e ects on import demand for both countries in sector 3, i.e. that m A0 3 = m B0 3 = 0: Also, choose units so that y A 3 + y B 3 = 1 . The assumption of identical price e ects is motivated by lack o f detailed information on import demand elasticities at a disaggregated level, whereas the normalization assumption is to avoid`size' e ects in the estimation. Let t A 3 and t B 3 stand respectively for the levels of tari s satisfying 13 for the home and partner countries respectively.
Then solving 19 for 3 ; and rearranging using 13 yields the expression that will be used in the estimation below: Expression 20 indicates that the cet is given by a w eighted average of the existing optimal tari s in both countries these are the rst two elements on the rhs of 20 plus a third term that denotes the increase in counter-lobbying by the export sector against tari s on the intermediate good due to the elimination of dutydrawbacks on intra-regional trade. The last term on the rhs of 20 is negative since 0, so its presence reduces the value of the cet. Note also that the increase in counter-lobbying would still be present in 20 if one did not impose a common import demand in each member country. 14 14 One can show that if government's objective when setting tari s is to maximize tari revenue, then the cet would have been given by an expression similar to the one in 20, i.e., The di erence is that the elimination of duty-drawbacks would have led to a higher cet rather than lower. Note that it would not be possible to endogenously explain the existence of duty-drawbacks in the pre-cu equilibrium if governments maximized tari revenue.
Note that when substituting t i 3 , i A ; B in 20, we are implicitly assuming that, when lobbying the government, lobbies do not take i n to account the second-round e ects of the adjustment in tari s on their production levels . This assumption is not only probably closer to reality than the alternative which w ould take second-round e ects into account, but also it is the only possible assumption in our empirical application on Mercosur. 15 Most importantly, to be able to substitute the existing tari s will avoid excluding from the empirical analysis some important determinants of the tari levels which cannot be directly measured.
To recapitulate, the simple political-economy model developed here shows that, if there is a possibility for exporters to obtain a duty-drawback, they will lobby to obtain the full drawback which is what is usually observed, and that duty-drawbacks tend to raise the equilibrium rate of protection on intermediate goods. More importantly for our purposes, the model shows that under plausible conditions, a cu will alter lobbying incentives, leading to pressures to eliminate duty-drawbacks and to reduce the external protection on intermediate goods.

Application to Mercosur
We use Mercosur data to check whether the model's predictions are consistent with observed outcomes. Beyond data issues and the reasons stated in the introduction, there are at least two practical reasons for our choice. First, steps towards putting in place a cet have gone much faster than in other recent rtas; recall that the agreement was signed in 1991 and that the cet came into e ect in 1995. Second, Mercosur is the only cu whose members previously had duty-drawbacks and which substantially eliminated barriers on internal trade. We discuss rst data and econometric issues, then turn to results.

The empirical model
To estimate the e ect of eliminating duty-drawbacks on the level of Mercosur's cet, we estimate a stochastic version of 20 on a cross-section of tari s. To estimate the expression, we need to calculate the third term in that expression which requires information on input-output relations and on production at a fairly disaggregated level isic 4 digit-level which includes 80 sectors. A table describing the characteristics of the 80 sectors used in the estimation is provided in the data appendix. The only Mercosur members for which industrial data was available at this level of disaggregation were Argentina and Brazil. Given that together they represent more than 85 percent of Mercosur production in any sector at the 3 digit isic classi cation level, the exclusion of Uruguay and Paraguay from the empirical analysis should not unduly a ect our results. 16 Moreover, in their study of the determination of Mercosur's cet, Olarreaga and Soloaga 1998 concluded that Brazil's political lobbying variables performed as well in explaining variations in Mercosur's cet on their own as those of the four members together.
In section 2, goods 1 and 2 were pure nal goods whereas 3 only was usable both in nal consumption and as an intermediate input. Now, let all goods be usable as nal goods and as intermediates in the production of other goods. We need then to rewrite slightly the coe cient z i j : The total demand for good j as an intermediate in the production of all other goods k k = 1 ; :::; 80, is P k jk y k , where jk is it's per unit requirement in the production of good k; t h us, where i j is i's share in the Mercosur's output of j. The higher is z M j ; the stronger is the counter-lobbying against protection of j, and consequently, in accordance with 19 the lower is, ceteris paribus, the level of good j's equilibrium external tari under the cet.
At the 4 digit ISIC classi cation level, Mercosur's cet is censored from above at its upper limit of 20 percent four sectors reach this upper bound at this level of disaggregation. To allow for censoring, we estimate 20 using a Tobit model; i.e. and ? j = a A a A + a B t A j + a B a A + a B t B j + 1 a A + a B z M j + j = 1 t A j + 2 t B j + 3 z M j + j 22 where subscript j stands for a sector, j = 1 ; :::80; j is the error term, and the expected signs are 1 0, 2 0 and 3 0 . In estimating 22, no constraints are imposed on the values taken by 1  where the expected signs are as before. The advantage of 23 is that it is no longer censored and can therefore be estimated by OLS or by w eighted least squares to control for potential group data heteroscedasticity. As discussed in Olarreaga, Soloaga and Winters 1999, tari s tend to be determined at the tari line level and not at the industry level. 17 This implies that if the error is determined at the tari line level and the number of tari lines in each industry is not the same as can be observed in the table in the data appendix the number of tari lines varies from 1 in ISIC 3131 to 501 tari lines in ISIC 3511, then the variance of the error term at the industry level will be negatively correlated with the number of lines in each industry, reducing the e ciency of our estimates. To correct for this potential heteroscedasticity, each observation at the industry level needs to be weighted by the number of tari lines in each industry i.e. multiplied by the square root of the number of tari lines in each industry.
However, a problem with the above correction is that it assumes that the errors at the tari line level are independently distributed. If observations within the same industry share a common unobserved determinant, then the above correction may reintroduce some heteroscedasticity. To test for this possibility, w e follow Dickens 1990 and when necessary, w e apply the correction he suggests which yields asymptotically e cient estimates. 18 The data appendix describes data sources and provides a table with the data used in the estimations. The cet is the one negotiated in Ouro Preto in 1994 by Mercosur members. Tari s or Argentina and Brazil are 1994 external o cial tari s corresponding to an fta situation as by then more than 95 percent of their internal trade was free of tari s. Trade data is the average of 1993, 1994 and 1995 whereas production data correspond to the industrial census of 1985 in Argentina and Brazil 17 If not we w ould not observe v ariation in tari levels within industries. 18 The test consists in verifying whether the error term of the weighted regression is correlated with the number of tari lines in each industry. If this is the case, then one estimates consistently the variance of the common and individual error components by regressing the error of the weighted estimate on a constant and on 1=n j , where n j is the number of tari lines in industry j. T o obtain asymptoticlly e cient estimates, one re-weighs the observations at the industry level using these variance estimates. updated to 1994 using industrial production indices provided by UNIDO. Table 1 reports results of the estimation of 22 in the rst column and of 23 in the second column. In both regressions, the coe cients have the expected sign and are statistically signi cant at the 99 percent level. 19 The regressors account for" around 60 percent of the cross-sectorial variation in Mercosur's cet.

Did counter-lobbying increase in Mercosur?
While it may be stretching the power of the model to interpret the values of 1 , 2 in terms of welfare weights, it is nonetheless worthwhile to note the following. Unconstrained estimates suggest that Argentina and Brazil give relatively similar weights to social welfare in their objective function. However, the constrained estimates suggest that Argentina's trade policy authorities are signi cantly more concerned by social welfare than Brazil's trade policy authorities b a A =b a B = 1 :27. These estimates are consistent with the fact that, on average, Argentina's pre-cet average tari is lower than Brazil's see sample data in the appendix.
For our purposes, however, the most important result is that in both regressions, the sector's share of sales to other sectors enters with a negative sign, suggesting that, indeed, as predicted by a political-economy approach to the determination of the cet , i n termediate goods producing sectors get less protection.
The results seem su ciently promising to use our estimates to calculate the marginal e ect of counter-lobbying on the cet from the coe cient estimates in table 1. The results of this exercise are shown in table 2. Let a z indicate the mean value of z M in the sample, and^ z the estimated marginal e ect of regressor z M , whose value is given in table 1. Then the change in the value of the mean value of the cet, is given by: =^ z: In our sample z = 0 :68.
A slight complication arises from the presence of censoring in our tobit estimate since the marginal e ects should not include observations for tari s on the upper bound, since these tari s could not be higher in the absence of counter-lobbying given the structure of the tari schedule. Therefore we need to weigh our estimated coe cients in the tobit regressions by the average probability of each observation being in the uncensored region. Given the low degree of censoring 5 percent in our data, this probability is relatively high and equal to 0.98. Thus, for the tobit equation our estimated marginal e ect is = 0 :98^ z. Table 2 reports results of the importance of counter-lobbying in terms of percentage points reductions in the average cet for our di erent estimations using this procedure and one standard deviation below and above the estimated coe cient. 20 The total cet reduction associated with counter-lobbying varies from 2:2 t o 4 :4 percentage points the average cet value in this sample is 13:9. Hence, according to the model, in the absence of counter-lobbying, Mercosur's cet would have been between 16:1 and 18:3. Taking the constrained estimates as reference, the average cet would have been 3.5 percentage points higher in the absence of increased counter-lobbying on intermediates' tari s. Table 3 indicates the top ve and the bottom ve industries where after the elimination of duty-drawbacks, counter-lobbying has led to the largest and smallest reductions in tari s, respectively. The largest increase in counter-lobbying on intermediate products tari s occured in industries 3699 non-metallic mineral products, 3692 cement, lime and platter, 3610 pottery, c hina and earthware, 3620 glass and 3691 structural clay. These tend to be sectors that are heavily used as intermediates in other sectors production and therefore which where more inclined to be subject to increase counter-lobbying on their tari s. The smallest increase in counterlobbying occured in industries 3220 wearing apparel, 3231 tanneries and leather 20 To give an example, the middle entry in row 1 of table 2, ,3:0 is obtained as follows: ,3:0 = 0:98,4:510:68 nishing, 3233 leather products, 3240 footwear and 3112 dairy products. These tend to be sectors which sell a little share of their output as input to other sectors. In non-metallic mineral products 3692, the tari could have been 5.4 percentage points higher in the absence of increased counter-lobbying, which represents more than 50 percent of the actual cet. A t the other end of the spectrum, for wearing apparel, the fall in tari was a low 0.1 percentage points, which represents around 0.5 percent o f the actual cet.

Concluding remarks
This paper used a political-economy framework to study the implications of dutydrawback s c hemes promoted by the World Bank in many developing countries as a mechanism mitigating the anti-trade bias of existing tari s for the incentives of export industries to lobby against upstream tari s on imported intermediates. In a model where duty-drawback s c hemes are jointly determined with tari s as part of a political-economy equilibrium, w e show that they reduce counter-lobbying incentives, leading, ceteris paribus, to higher tari rates on imported intermediates used heavily in export industries.
Moreover, we showed that the formation of a cu will endogenously lead to the elimination of duty-drawbacks for intra-regional exports. This re-creates an incentive for counter-lobbying by users of intermediate goods, resulting in lower external tari s on intermediate products, thereby formalizing a channel through which the new" regionalism may, in the terminology of Bhagwati, be a stepping stone rather than a stumbling block in the move t o wards greater integration of the world economy.
The model's predictions were tested and con rmed in the case of Mercosur. Our estimates suggest that the cet would have been on average 3.5 percentage point higher 25 percent in the absence of the increased counter-lobbying on intermediate products associated with the elimination of duty-drawbacks for intra-regional exports. Thus, the mechanism described above m a y partly explain the decline in external tari s associated with the new regionalism" de ned by Ethier and Lawrence. Although our results could no doubt have been derived from alternative political-economy models, they also provide indirect vindication of the empirical implications of common-agency and Nash-bargaining models of trade protection.

Yeats, Alexander 1998,`Does Mercosur's Trade Performance Raise Concerns
About the E ects of Regional Trade Arrangements?', World Bank Economic Review 1-28.   Common external tari data and external tari s were provided by the MERCO-SUR secretariat o cial tari s for 1995, announced in December 1994. Tari data are disaggregated at the 8-digit level of the harmonized system 9119 items and were converted to the 6-digit level by simple averages. To lter the data from the 6-digit harmonized system to the 4-digit isic classi cation we used a table provided by Jerzy Rozanski from the World Bank.
Trade data.
The sources are national accounts COMTRADE in US dollars. Data were averaged for 1993-95 and disaggregated at the 6-digit level of the harmonized system. To convert them to ISIC 4-digit we used the same lter as for tari data.
Industrial data.
The sources are the industrial censuses of Argentina and Brazil in 1985 for production data, and the GTAP database for Argentina and Brazil in 1995 for input-output coe cients. The industrial data at the 4-digit level was converted into 1994 values using a production index at the 3-digit level available at unido. Given that production data is denominated in domestic currency we converted them to 1993-95 US dollar values with the ratio of the average nominal GDP in Manufacture in 1993-95 from National Accounts to the total value added calculated from census gures to which w e also apply the production index. The data are disaggregated into 80 sectors corresponding to the 4-digit ISIC level. To convert the input-output data from the GTAP classi cation to the isic one, we used the tables provided by G T AP manuals.