Beyond the standard model: How trade agreements

reduce political protectionism

Goran Lazarevski

Columbia University

Abstract

The standard Grossman-Helpman/Bagwell-Staiger model of trade agreementsstates that their purpose is to prevent terms-of-trade manipulation and as-signs trivial role to special interest politics and the resulting politically mo-tivated protectionism. Despite its popularity, the standard model remainsinconsistent not only with observed trade policy, giving rise to so calledtrade policy puzzles, but also with the practitioners’ understanding of tradetreaties, according to which concerns over special interests are central in thenegotiations process. By assuming that subsidies have additional politicalcost beyond their monetary cost, I show in this paper how internationalagreements result in the reduction of political protectionism through thecrucial role of exporting lobbies in the negotiations process. In addition, themodel resolves three prominent puzzles in the literature: (i) the terms-of-trade puzzle, (ii) the anti-trade bias puzzle, and (iii) the inefficient redistri-bution puzzle. I motivate my key assumption using survey data on US voters’trade policy preferences, particularly voters’ asymmetric belief that subsidieswould negatively affect the average voter through fiscal adjustments. Finally,by using sector-level data on US agricultural trade policy, I find empiricalsupport for the main theoretical predictions of the model.

Keywords: Trade Agreements, Standard Model, Protectionism,Terms-of-Trade Puzzle, Anti-Trade Bias Puzzle, Inefficient redistributionpuzzle, Agricultural subsidies, Trade policy, Median voter

1. Introduction

The GATT1 (WTO) agreement has resulted in unprecedented worldwidetrade liberalization since the end of the Second World War. Understand-ing the reasons that led to this historic transformation is one of the cen-tral tasks for academic trade economists. Most of them regard the well-established work of Grossman and Helpman (1997) and Bagwell and Staiger(1997,2004) as the conventional wisdom when it comes to explaining tradeagreements. These models share so much of their internal logic – indeedGrossman-Helpman can be interpreted as a micro-founded version of Bagwell-Staiger – that I will refer to them simply as ”the standard model”. Yet thestandard model, despite its popularity, is inconsistent with certain key fea-tures of actual trade agreements and the trade policies that we observe inreality. Moreover, there is a fundamental disagreement between the for-mal theory and most trade practitioners’ understanding of trade agreements,which include practicing economists, trade negotiators and diplomats whoare involved first-hand in crafting and negotiating trade treaties.2 While thestandard model supports the notion that trade agreements prevent terms-of-trade manipulation and considers special interest politics to be irrelevant,practitioners consider the latter as the key driving factor behind trade agree-ments.

This paper aims to bridge that gap by introducing a key assumption inthe standard model, that will nevertheless radically alter the purpose of tradeagreements and bring it in line with the practitioners’ view, while at the sametime resolving three popular puzzles that render the standard model incon-sistent with reality: (i) the terms-of-trade puzzle, (ii) the anti-trade biaspuzzle, and (iii) the inefficient redistribution puzzle. In particular, I assumethat trade policies associated with increased fiscal outlays (i.e. subsidies)incur political cost to the government, in addition to their monetary cost.The additional political cost can be motivated by the fact that democrati-cally accountable governments are also concerned with the well-being of themedian voter, and not just the well-being of the average voter (i.e. aggregate

1General Agreement on Tariffs and Trade (1947) was succeeded in 1995 by the WorldTrade Organization (WTO).

2For in-depth discussion of the inconsistencies between the standard model and thepractitioners’ understanding refer to Regan (2015), whose paper greatly informs the dis-cussion that follows.

2

social welfare).3 Applying this assumption in the micro-founded frameworkof Grossman-Helpman aids the formal exposition of my argument and allowsfor empirical testing, which this paper also does.

According to the standard model, the sole purpose of trade agreementsis (or should be) to prevent governments from engaging in terms-of-trade(ToT) manipulation. If governments set trade policies unilaterally they willtry to improve their terms-of-trade by setting ”optimal tariffs” in the caseof imports and ”optimal export taxes” in the case of exports, thereby im-posing a ToT externality on their trading partner. The government benefitsfrom this terms-of-trade improvement only via the revenue it collects, sothe government’s concern with trade-tax revenue is absolutely central to thestandard model. The incentive to manipulate the ToT exists irrespective ofthe internal politics and any influence by special interest groups. A tradeagreement enables cooperation between two governments who agree not toimpose ToT externality on each other thereby maximizing their joint surplus.Special interest politics plays no role whatsoever in the motivation for tradeagreements according to the standard model, even when these special inter-est groups are actively lobbying for the terms of the agreement. This is notto say that lobby groups have no effect on the final negotiated levels of tradeprotection - they obviously do, but the trade agreement does not reduce po-litical protectionism, i.e. the component of the tariff (or subsidy) that is dueto lobbying efforts remains the same before and after the agreement. Onlythe component of the tariff that is due to ToT manipulation is eliminated asa result of the trade treaty. But as much as the Grossman-Helpman modelmakes this distinction explicit by additively separating the two componentsin the expression for the unilateral equilibrium trade policy, Grossman (2016)conflates them in arguing that the criticism levied at the model is a matter ofsemantics. The confusion arises due to not distinguishing between cause andeffect. Every tariff, irrespective of the causal motivation for setting it, willhave both a protective effect on the import-competing industry, as well as aneffect on the terms-of-trade. But whether the tariff is motivated by politicalprotectionism, or terms-of-trade manipulation, or both, depends on whetherthe government is aiming to protect a politically influential industry, or raiserevenue by improving the terms-of-trade effect, or both.

3The justifications for this assumption, including strong support from survey data onvoters’ beliefs and preferences, are discussed in more detail in the theoretical section.

3

Contrary to the standard model, in the practitioners’ view of trade agree-ments there is very little emphasis on the governments’ concern for the trade-tax revenue and through it, on the terms-of-trade. While there may be a ToTexternality, it is by no means the central goal of trade agreements to eliminateit. But rather, the start of trade negotiations changes the political balanceof forces, without which there cannot be a reduction in political protection-ism. Due to the sudden mobilization of (at least some segment of) exporterinterests, the practitioners’ story goes4, the government will want to lowerits tariffs if in exchange for losing support from its import-competing inter-ests, it gains support from its exporting interests because it has succeeded inreciprocally expanding market access. Similarly, in my model the exportinginterests are able to mobilize around lobbying for reducing the trading part-ner’s tariffs once the negotiations open, because tariff reductions do not incuradditional political cost, as opposed to the high-cost lobbying effort requiredfor obtaining export subsidies5. Crucially, trade negotiations do not alterthe balance of political forces by affecting their ability to organize (prettyad-hoc assumption), but rather their potency by giving them a policy optionpreviously unavailable in the non-cooperative context.

It is perhaps surprising that a model that makes special interest poli-tics essentially irrelevant for trade treaties has become the dominant modelfor explaining them. Moreover, because ToT manipulation is the drivingmotivation behind trade agreements, it predicts that small countries thatcannot affect their terms-of-trade would have no reason to ever enter suchagreements, which clearly hasn’t been the case for the 164 member states ofWTO6. Furthermore, the standard model generates three prominent puzzles

4Regan (2015) cites Pauwelyn (2008), Hudec (1992) pp.314-316, Krugman (1997) p.118,Destler (2005) pp.17, 253-254.

5Unlike the two-sector Bagwell-Staiger model, Lerner symmetry does not apply in themany-sector Grossman-Helpman model, so exporters will want to lobby only for exportsubsidies in the unilateral context. Applying the logic of my model to Bagwell-Staigerwould require the additional assumption that exporters would *not* lobby for reducingdomestic tariffs in the domestic import-competing sector even though that would affect theprice of their product through Lerner symmetry. Regan (2014) discusses few justificationsfor such an assumption.

6There is a class of so called commitment models that explain why countries, includingsmall economies, facing an irreversible costly decision may want to join an existing inter-national treaty, but they cannot explain how such a treaty would reduce existing tariffs,nor the reciprocal nature of multi- and bilateral agreements.

4

in the political economy of trade7. If trade agreements are meant to preventterms-of-trade manipulation, that means they should prohibit export taxesand reductions in export subsidies. Instead, export taxes are not regulatedwhereas reducing export subsidies is not only allowed, but it is mandatory- export subsidies are fully banned under WTO. Moreover, even if countriesare allowed to manipulate ToT for their exports through imposing taxes onthem, they almost never do it. This is known as the terms-of-trade puzzle.The anti-trade bias puzzle arises from the observation that the net effect ofunilateral trade policies in the real world is to contract trade, whereas Levy(1999) shows that in a symmetric, two-country Grossman-Helpman world theeffect of lobbying would be to encourage net trade promotion8. The thirdconundrum is the inefficient redistribution puzzle which finds puzzling theprevalent use of trade policies as a means to redistribute income to specialinterests groups given that there are other more efficient ways to do it (such asproduction subsidies or lump sum transfers). A simple extension of the G-Hmodel to include production subsidies will render trade policies unattractiveas a redistribution mechanism. The status of certain policy instruments inWTO also presents a challenge for the standard model. Voluntary export con-straints (VER’s) and freely given import quotas were widely used before and,therefore, are strictly regulated by WTO even though they negatively affectthe terms-of-trade. Thus their regulation in treaties can only be explainedby dominant political protectionist motivations, while ToT motivations mustplay a minor role, if any.

Assuming additional political cost for subsidies resolves all of the abovementioned puzzles by leading countries to rely more on unilateral tariffs forprotectionism than on comparatively costlier export and production subsi-dies. Trade negotiations serve as a forum for reconfiguring the balance ofpolitical forces by allowing organized special interests to negotiate on andinfluence their trading partners’ policies, which is not possible in the non-cooperative context. Because this political reconfiguration through cooper-ation empowers the interests that are in favor of liberalization, trade agree-ments result in reduction of political protectionism. In the case of trade

7For more in-depth discussion of the puzzles, see Rodrik (1995).8This puzzle is sometimes also known as the export subsidy transfer puzzle, because

the theory implies that a country’s export subsidies would be only partially countervailedby the trading partner’s import tariffs, in effect transferring income to the foreign import-competing sector.

5

policies which directly affect prices, export lobbies acquire a new avenue forinfluencing their sale price (reducing foreign tariffs), which is much morecost-effective than trying to raise export subsidies at home. Thus the as-sumption of politically costly subsidies is crucial. In the case of trade policiesfor quantitative restrictions (VER’s, embargoes, quotas), the only avenue forexporters to affect these policies is to influence them directly through thetrading partner, which is only possible under trade negotiations and cooper-ation9. Therefore, the assumption of politically costly policies is not neces-sary in this case and the politically protectionist motive prevails even if thesepolicies deteriorate the terms-of-trade. Finally, because the ToT motive issecondary to the political motive, even small countries benefit directly fromcooperation on trade.

There are several papers in the literature that modify the Grossman-Helpman model by introducing non-standard assumptions with the aim ofresolving some of its contradictions. Tovar (2009), and Freund and Ozden(2004) introduce loss aversion and reference dependence of the agents’ pref-erences in a standard Grossman-Helpman framework in order to explain ob-served trade protection dynamics: its persistence over time and the findingthat industries experiencing losses are more likely to receive protection. To-var’s (2009) model resolves the anti-trade bias puzzle provided that the lossaversion coefficient is high enough and the country undergoes liberalization,so that declining import-competing sectors have the most to gain from orga-nized lobbying. While this model captures salient features of the dynamicsof trade protection10, it does not explain how trade agreements result in re-duction of political protectionism, nor does it adequately address the otherpuzzles raised in the literature. Similarly, Limao and Panagariya (2006) in-troduce concern for inequality in the government’s objective function whichcould generate the desired anti-trade bias, but leaves the core internal logic ofthe standard model intact. Ethier (2007) considers the implications of assum-ing a lower government weight on trade-tax revenue (and thereby the ToT

9I do not formally model this result here, but this is implied by the logic of the argument.10Most empirical estimates of Grossman-Helpman find a significant constant term ex-

plaining trade protection levels. Ederington and Minier (2008) show that this findingcontradicts the Grossman-Helpman model because it cannot be explained by extraneouspolitical factors and amounts to deviations from welfare-maximizing behavior. Policy per-sistence might be a promising candidate for explaining this term, though this is not thesubject of this paper.

6

motivation) in the Grossman-Helpman model and shows that the only wayto eliminate both the terms-of-trade puzzle and the anti-trade bias puzzle isto impose the unrealistic assumption of zero weight attached to governmentrevenue. He correctly points out that this is not so much problematic becauseToT concerns should necessarily matter11, but because the Nash equilibriumhe gets, which consists of prohibitive tariffs driving imports to zero, is asingularity and not the limiting case as the weight on revenue goes to zero.

I test the model using data on US agricultural protection from before theimplementation of the Uruguay Round Agreement on Agriculture (1995). Iemploy a non-linear censored IV-Tobit regression to directly estimate theparameters of the model and test its predictions. The estimation results forthe US support the modified G-H model, while I find little evidence for theoriginal G-H specification. Across all specifications the results reject the nullhypothesis that there is no additional political cost to subsidies against thealternative that this coefficient is negative (greater cost). I find no evidencethat terms-of-trade motivations are ever significant, which is in accordancewith the practitioners’ understanding. Moreover, the implied estimates forthe government weight on social welfare are one of the lowest in the litera-ture, whose unrealistically high estimates have presented a puzzle in empiricaltests of the G-H model. Gawande and Hoekman (2006) also test the originalG-H model using US agricultural data, which allows them to exploit bothvariation in export subsidies and import tariffs. However, there are severalproblematic issues with their method, which this paper resolves. Becausethey don’t have data on export supply elasticities, they set them equal to2 for all sectors with no explanation. Another issue is that they use datafrom 1999 for the export subsidies which was after the entry into force ofURAA (1995), when they became subject of international regulation. Theidentification equations used however are based on the assumption of no co-operation. Lastly, even though they find evidence in the data for differentialgovernment preference over different policy instruments, they lump togetherproduction subsidies with export subsidies (and non-tariff measures) with-out any theoretical justification in the model. My approach addresses theseobjections, so its estimates are arguably more reliable.

11As Regan (2015) says: ”In the classic studies of United States trade policy from theHawley-Smoot period to the present, there is not one word to suggest that tariffs were evermotivated in even the slightest degree by the desire for tariff revenue.”

7

The paper proceeds as follows: First, I present the theoretical and empiri-cal underpinnings of the model and analyze the non-cooperative equilibrium.Next, I consider the equilibrium trade agreements that would result as conse-quence of cooperation between governments. This section also shows formallyhow trade agreements result in the reduction of political protectionism andhow they resolve the puzzles discussed. In the empirical section, I give de-scription of the data used, the estimating procedure and I discuss the results.The final section concludes.

2. Model

My approach follows closely the modeling framework of Grossman-Helpman(1997), which is considered to be the best micro-founded exposition of thestandard model. However I introduce a key assumption in which I departfrom the standard model: Based on analyzed survey data, I assume thatpolicies associated with fiscal outlays are valued differently from policies as-sociated with fiscal revenues in the government’s political support function,which I show has far-reaching implications regarding the fundamental pur-pose of trade agreements.

I consider two open economies trading with each other with similar un-derlying economic and political structure. Therefore I will describe in detailthe Home country, while it is understood that the same setup applies to theForeign country. There are N identical households, each of which maximizesan additively separable utility function:

U = cZ +n∑i=1

ui(ci) (1)

where cZ is the consumption of the numeraire good that doesn’t generateconsumer surplus, whereas ci denotes the consumption of goods i=1,2,...,n.The functions ui(·) are differentiable, increasing and strictly concave. Letdi(pi) be the demand (per capita) for good i12, while pi denotes its inter-nal domestic price. The offshore price is given by πi and the ad-valoremtariff/subsidy by τi, and therefore pi = τiπi.

13 If we define E to be the house-hold’s total spending and S(p) ≡

∑i ui(di(pi))− pidi(pi) the total consumer

12di(·) can be obtained from the first-order conditions from consumer maximization asthe inverse of u′i(·).

13Similarly, for the Foreign country: p∗i = τ∗i πi.

8

surplus that it derives from consumption of all non-numeraire goods, givena vector of goods prices p, then the household’s indirect utility V can bewritten as:

V (E, p) = E + S(p) (2)

All goods are produced competitively. The numeraire good sector usesonly labor (at constant returns to scale), which we assume is always in suf-ficient supply such that this good is always produced. The labor units arechosen WLOG, such that the wage rate is set to 1. The other goods are alsoproduced at constant returns to scale, but utilizing two inputs: labor anda sector-specific input, that is supplied inelastically. Let Πi(pi) be the rent(profits) that the owners of the specific factor i earn14. Then, by the envelopetheorem, the industry supply will be given by:

yi(pi) = Π′i(pi) (3)

The government can only tax or subsidize imports and exports of non-numeraire goods, and it can collect tax revenue or redistribute tax receiptsto the citizens in lump sum terms in order to finance its trade policies. Thus,τi > 1 signifies either an import tariff or an export subsidy, while τi < 1 iseither an import subsidy or an export tax. The sector-specific revenue ri willbe:

ri(τi, πi) = (τiπi − πi) (di(τiπi)N − yi(τiπi)) (4)

From now on, we normalize the total population N=1 without loss of gener-ality.

In the international context, the world market for all goods clears, whichallows us to solve for the off-shore market-clearing price πi(τi, τ

∗i ) for all i:

di(τiπi)− yi(τiπi) = y∗i (τ∗i πi)− d∗i (τ ∗i πi) (5)

The political economy framework utilizes the common agency approachby Grossman-Helpman. We assume that individuals can only claim (partial)ownership of a specific factor of at most one type. Lobby formation is exoge-neously determined at the industry level, and each industry lobby consists ofall owners of a particular specific factor type. Thus, each individual can be-long to at most one lobby and will have a direct stake in the trade policy that

14Πi(·) can be obtained from profit maximization at the industry level.

9

pertains to that lobby. We assume that the specific factor owners of goodi constitute αi share of the total population, which reflects their interest intrade policies as consumers. In addition, they earn li income from their ownlabor. Then the joint welfare of lobby i will be given by:

Wi(τ, τ∗) = li + Πi(τi, τ

∗i ) + αi

(∑i

ri(τi, τ∗i ) + S(τ, τ ∗)

)(6)

The lobbies express their political demands through campaign contribu-tion schedules Ci(τ, ·): they offer to contribute to the incumbent politician fi-nancial funds that depend on the particular trade policies implemented. Theobjective function that the lobby aims to maximize by adjusting its contri-bution schedule will be the expression in (6) net of this realized contribution.Under the non-cooperative (unilateral) regime of policy determination, thecontribution schedules Ci(τ ; τ ∗) can be conditioned only on the τ domesticpolicies implemented, taking the foreign policies τ ∗ as given. Whereas underthe cooperative bargaining regime, Ci(τ, τ

∗) can be conditioned on both theτ domestic and τ ∗ foreign trade policies. In both cases, the lobbies act simul-taneously taking as given the schedules of all other lobbies in both countries.The lobbies can influence only the politician in their respective country andtheir schedules cannot be observed by the foreign politician15.

Once the various lobbies have set their contribution schedules, in the sec-ond stage the incumbent politician chooses the optimal trade policy vector –either unilaterally or through international cooperation with other sovereigns– by maximizing a political welfare function G, which is a weighted averageof the sum total of campaign contributions, the social welfare W and a po-litical cost term P .

G =∑i

Ci(τ, ·) + a (W (τ, τ ∗) + P (τ, τ ∗)) (7)

W (τ, τ ∗) = l +∑i

Πi(τi, τ∗i ) +

∑i

ri(τi, τ∗i ) + S(τ, τ ∗) (8)

To motivate the introduction of the term P , I assume that the governmentis democratically accountable and responsive to distributional concerns in the

15As a consequence, lobbies do not set their schedules strategically with respect to theForeign government. See Grossman-Helpman (1997) for justification of this assumption.

10

sense that it attaches a certain weight aM on the welfare of the median voter16

WM , in addition to the aggregate social welfare weighted with a factor a117.

Furthermore, we assume for simplicity that the median voter also internalizesthe social welfare with a weight of a2, but she also dislikes policies associatedwith additional fiscal outlays, because she believes that the government wouldraise taxes on ordinary citizens to finance them. On the other hand, policiesassociated with increased fiscal revenues, such as tariffs and taxes, do notenter symmetrically in the median voter’s welfare, because she doesn’t believethat the increased fiscal revenues would translate into any direct benefit forher, whether via decrease in personal taxes or direct government spending,or at least not to the same degree as the subsidies. In terms of the model,the term P captures the political cost of trade policies that the governmentwould incur due to how the median voter believes she is affected by thegovernment’s taxation policies18. We assume for computational simplicitythat there is no perceived political benefit to policies associated with fiscalrevenues, which corresponds to the extreme case when none of the traderevenue raised translates into palpable benefits for the median voter. As forthe political cost of subsidies, it is logical to make it proportional to thesize of the subsidy (by a factor of δM), which in turn is proportional to thetaxation burden to voters, depending on their beliefs. Thus the political costterm P can be written as:

P (τ, τ ∗) = δM∑i

1(ri(τi, τ∗i ) < 0) ri(τi, τ

∗i ) (9)

Writing out the government political support function G using the expressionfor WM = a2W +P , it is easy to obtain (7) by introducing the reduced formcoefficients a→ a1 + aMa2 and δ → aM δM

a1+aMa2.

16The median voter is a proxy for the vast majority of voters.17The government in Grossman-Helpman is only concerned with aggregate welfare and

there are no distributional considerations.18Note that due to the political nature of the cost, it is not required that the government

actually behaves in conformity with voters’ beliefs, though these beliefs are consistent witha redistribution scheme that works to the benefit of either marginalized or privileged socialstrata and at the detriment of the majority of voters in the middle. Moreover, the politicalcost does not necessarily correspond to the economic cost on the median voter. Politically,she considers the policies in isolation and develops a rule-of-a-thumb attitude towardseach type based on her beliefs about the government’s taxation strategy, whereas theactual economic cost depends on the combined fiscal effect of the entire set of policies.Observations O3 and O6 in Appendix A lend further support to this assumption.

11

The assumptions on voter beliefs are supported by my own survey dataon voters’ beliefs and preferences conducted on a representative sample of620 respondents from the USA via AmazonTurk. Respondents were askedto consider a small open industry in which they hold no direct stake eitheras consumers nor as employees. Only 17% of respondents believe that theadditional fiscal revenue raised through taxing an exporting industry wouldbenefit the average voter, however the reverse doesn’t hold - if the govern-ment decided to give production subsidies to that same industry, 80% believethat the average voter would foot the bill, either through increased taxes orspending cuts19. Similarly for an importing industry, only 21% of respondentsbelieve that the revenues generated through increased tariffs would directlybenefit them, but if production subsidies were granted, 79% say that theaverage voter would be negatively affected due to fiscal adjustments20. Thisasymmetry in beliefs regarding the fiscal implications of trade policy reveals ageneral public distrust of government interventions in trade and boils down toan inherent preference for small government, hence the assumption that tradepolicies that translate into increased government spending incur additionalpolitical cost for every dollar spent. Analysis of the survey data furthermoreshows that voters’ self-reported trade policy preferences largely match thetheoretically predicted policy preferences in a partial equilibrium small openeconomy setting under the above welfare assumptions, and, moreover, theycontradict the policy preferences implied by the Grossman-Helpman assump-tion that only aggregate social welfare matters. Appendices A and B delveinto more detail on the survey and how it supports the key assumption of thepolitical cost of subsidies. This assumption is also in line with the practition-ers’ conventional wisdom (Regan, 2015) and the larger cultural and politicalcontext in which implicit subsidies through tax cuts (tariffs) are easier toobtain than explicit transfer subsidies. Dunkel and Roessler (2012), for ex-ample, argue that governments are reluctant to offer production subsidies asan alternative to tariffs, even when the subsidies would be more efficient.

Under the non-cooperative trade war regime, governments set their tradepolicies unilaterally, ignoring the impact that their actions have on the othercountry. In accordance with Grossman-Helpman, an equilibrium responseby a country to a policy choice by its trading partner can be defined as follows:

19See O2 in Appendix A20Appendix A: O5

12

Definition 1: Let τ ∗ be an arbitrary trade policy vector chosen by For-eign. Then an equilibrium response to τ ∗ consists of a set of feasible21 con-tributions {Co

i } and a trade policy vector τ o such that:

a) τ o = argmaxτ

∑iC

oi (τ ; τ ∗) + a(W (τ, τ ∗) + P (τ, τ ∗))

b) For all organized lobbies i, there does not exist an alternative optimaltrade policy vector τ̃ i and an alternative feasible C̃i(τ ; τ ∗) strictly pre-ferred by the lobby, i.e. that satisfy:

(i) τ̃ i = argmaxτ

C̃i(τ ; τ ∗) +∑

j 6=iCoj (τ ; τ ∗) + a(W (τ, τ ∗) + P (τ, τ ∗))

(ii) Wi(τ̃i, τ ∗)− C̃i(τ̃ i; τ ∗) > Wi(τ

o, τ ∗)− Coi (τ o; τ ∗)

Consequently, the Nash Equilibrium of this game will consist of two pol-icy vectors τ and τ ∗ that are equilibrium responses to each other. To solvefor it, we proceed as in Grossman-Helpman (1997) 22. First we apply theBernheim-Whinston (1986) theorem separately for each country taking theexternal prices as given in order to obtain the equilibrium response. TheBernheim-Whinston theorem states the optimization conditions23that sup-port the subgame perfect Nash Equilibrium between the lobbies and thegovernment within each country.

There are many possible contribution schedules that constitute equi-librium responses to the above game, but we restrict our focus to the socalled globally truthful equilibrium, which arises when lobbies truthfullyreveal their valuation for all potential policies, i.e. the contribution sched-ules Ci(τ, τ

∗) mirror the curvature of the lobby’s utility. Globally truthfulcontributions are a weakly dominant strategy for the lobbies. In this case, alobby i would get the same payment for all policies (τ, τ ∗) that induce pos-itive contributions Ci > 0. Thus, deciding the contribution schedule boilsdown to deciding by what scalar Bi > 0 the lobby wants to shift its utility.

21Feasible contributions are those that the lobby can afford, i.e. non-negative offers thatdo not exceed the aggregate income of the lobby’s members.

22For details about this solution approach, refer to Grossman-Helpman (1992) andGrossman-Helpman (1997).

23The theorem states that (given τ∗) {{Ci}i, τo} is SPNE iff:1o. Ci(τ ; τ∗) are feasible ∀i2o. τo maximizes the government objective function G(τ, τ∗) under (7)3o. τo maximizes the joint gov+lobby objective function G+ (Wi − Ci) ∀i4o. ∀i ∃τ i (vector), such that it maximizes G(τ, τ∗)− Ci(τ ; τ∗)

13

Mathematically, the truthful contributions can be written as:

Ci(τ, τ∗) = max{Wi(τ, τ

∗)−Bi, 0} (10)

Substituting (8), (9) and (10) in (7), taking the first-order condition withrespect to τi (holding τ ∗i constant) and rearranging, we get an implicit ex-pression for the equilibrium policy best response24:

τ − 1 =(I − αL)

a(1 + δr) + αL

y

(−m′)π+

aδra(1 + δr) + αL

m

(−m′)π+

m∗

τ ∗m∗′π(11)

Here, we define I to be an indicator for whether the sector is orga-nized into a lobby or not, αL is the total share of sophisticated consumersthat firm owners represent, δr ≡ δ 1(ri < 0) and m(p) ≡ d(p) − y(p) andm∗(p∗) ≡ d∗(p∗) − y∗(p∗)25denote Home and Foreign’s import demand (ifpositive) or export supply (if negative) respectively. This expression decom-poses the resulting trade policy intervention into three effects. The first termrepresents the political protectionism from the lobby. The third term repre-sents the terms-of-trade effect from imposing tariffs on imports or taxes onexports and also equals by definition the elasticity of import demand (exportsupply). It is completely independent from the existence of lobbies (specialinterest politics) and stems from the government’s ability to raise revenue byimproving its terms of trade. These two terms are the exact same terms asin Grossman, Helpman (1997). The second term comes from the impositionof the additional political cost function, and only appears if the trade policyoutcome results in costly fiscal outlays.

To analyze this equilibrium vis-a-vis Grossman-Helpman’s standard model,consider the reduced form special case they also focus on, in which (with-out loss of generality) Home imports from Foreign and both countries haveconstant trade elasticities ε > 1 and ε∗ > 0 respectively26:

m(p) = m0(p)−ε (12)

m∗(p∗) = −m∗0(p∗)ε∗

(13)

24The index i is dropped henceforth, because it is understood that the derivation is forsector i, and the same approach applies for all sectors.

25Consistent with standard notation, m′ = dm(p)dp and m∗

′= dm∗(p∗)

dp∗ . Also note that m′

and m∗′

are always negative.26m0 and m∗0 are both positive constants.

14

The off-shore price π can be expressed explicitly as a function of thedecision trade policies by substituting the above expressions in the market-clearing condition (5).

π(τ, τ ∗) =

(m0

m∗0

)1/(ε+ε∗)(1

τ

)ε/(ε+ε∗)(1

τ ∗

)ε∗/(ε+ε∗)(14)

Then from expression (11) we derive the trade policies for the Home andForeign country when both sectors are organized into lobbies:

τ

(1− (1− αL)

a+ αL

y(τπ)

εm0(τπ)−ε

)= 1 +

1

ε∗(15)

τ ∗(

1− (1− α∗L)

a∗(1 + δ∗r) + α∗L

y∗(τ ∗π)

ε∗m∗0(τ ∗π)ε∗+

a∗δ∗ra∗(1 + δ∗r) + α∗L

1

ε∗

)= 1− 1

ε(16)

The importing country (Home) will impose an import tariff because theright-hand side of (11) is always positive: both the import-competing lobbyand the government want to tax imports thereby raising revenue, and there-fore no additional political cost is incurred. However, for the exporting coun-try (Foreign), both export subsidies and export taxes are possible, dependingon whether the political support the government gets from the export lobbyon the one hand is stronger than the additional political cost of subsidizingthat industry compounded with the terms-of-trade effect on the other.

Figure 1 depicts three characteristic equilibria under the non-cooperativetrade war regimes: J denotes the Johnson equilibrium first derived by John-son (1953) when governments are only concerned with maximizing socialwelfare without any political influence27. The ad-valorem trade policies inthis case are simply the inverse trade elasticities. The Grossman-Helpman(1997) equilibrium is reproduced on the left diagram and denoted by NE asthe intersection of the two countries’ best response curves when δ → 0, i.e.there is no additional political cost. The right diagram represents the tradewar Nash equilibrium implied by the model from the intersection of the bestresponse policies as given by (15) and (16).

The effect of the negative political cost term (when it applies) wouldbe to decrease the equilibrium policy response τ ∗ for any given τ .28 Thus,

27The Johnson equilibrium corresponds to the limiting case when a→∞ and a∗ →∞28The political cost term in (16) causes the intercept with the y-axis to decrease to (1−

15

Figure 1: Non-cooperative Nash equilibria in the original G-H model vs. this model.

depending on the value of τ , Foreign’s best response could fall into threeranges: for low enough τ , the optimal policy response would be an exporttax so δ∗r = 0; for τ ∈ (τ1, τ2) it’s optimal not to intervene at all (cornersolution); whereas if τ > τ2 the optimal policy would be an export subsidygiven by (16) when δ∗r = δ. If we denote by xNE the ratio of non-cooperativeequilibrium policies τ

τ∗, it is easy to see that xNE in this model is rotated

clockwise relative to the one in Grossman-Helpman (1997), here representedby the ray from the origin OS. If the political cost of subsidies δ is ”highenough”, Foreign’s best response will decrease29 causing xNE to rotate furtherclockwise until it reaches 1 and thus the model will generate the anti-tradebias that we observe in reality, namely that the net effect of unilateral tradepolicies is to contract trade (i.e. xNE > 1), rather than expand it as the

1ε )(1 +

a∗δ∗ra∗(1+δ∗r )+α

∗L

1ε∗ )−1, because y∗(p∗)

m∗(p∗) ≡ z(ττ∗ ) is increasing in τ

τ∗ so it approaches 0 as

τ → 0. At the same time, as τ∗ →∞, the term in brackets must go to 0, which determines

the asymptotic behavior of Foreign’s best response by solving:(1−α∗L)

a∗(1+δ∗r )+α∗L

1ε∗ z(

ττ∗ ) =

1 +a∗δ∗r

a∗(1+δ∗r )+α∗L

1ε∗ . It approaches asymptotically a value of τ∗/τ that is unambiguously

lower than the equivalent Grossman-Helpman benchmark due to the effect of the politicalcost term and there is no restriction on how low it could get provided that δ is sufficientlylarge.

29See previous footnote.

16

Grossman-Helpman (1997) model implies30.

Cooperative regime: Trade agreements

Under a cooperative regime, national governments are allowed to nego-tiate and implement an agreement on trade policies (τ, τ ∗) and compensateone another with a transfer payment R, which is to be distributed equallyamong the public. The lobbies are again allowed to lobby their domesticgovernments, but in this case they can condition their contribution scheduleson the full policy vector that would be the result of the trade negotiation.Assume for simplicity that firm owners constitute a negligible share of so-phisticated consumers, or equivalently that firms can only lobby for the tradepolicies directly affecting their sector, therefore setting αL = α∗L = 0. Irre-spective of the particular bargaining procedure, the bargaining outcome willbe such that no government will be able to increase its own welfare withoutlowering the welfare of its trading partner (also known as politically efficientoutcome)31. Therefore an equilibrium trade agreement will always result inpolitically efficiency. More formally we can write (as in G-H (1997)):

Definition 2: An equilibrium trade agreement consists of a set of feasiblecontributions {Co

i } and {C∗oi } and a pair of policies (τ o, τ ∗o) such that:

a) (τ o, τ ∗o) = argmax(τ,τ∗)

a∗G(τ, τ ∗) + aG∗(τ, τ ∗)32

b) For all organized domestic lobbies i ∈ L, there does not exist an al-ternative optimal trade policy vector (τ̃ i, τ̃ ∗i) and an alternative feasibleC̃i(τ, τ

∗) strictly preferred by the lobby, i.e. that satisfy:

(i) (τ̃ i, τ̃ ∗i) = argmax(τ,τ∗)

a∗(C̃i(τ, τ∗)+

∑j 6=iC

oj (τ, τ ∗))+a

∑j C∗oj (τ, τ ∗)+

aa∗(W (τ, τ ∗) + P (τ, τ ∗) +W ∗(τ, τ ∗) + P ∗(τ, τ ∗))(ii) Wi(τ̃

i, τ̃ ∗i)− C̃i(τ̃ i, τ̃ ∗i) > Wi(τo, τ ∗o)− Co

i (τ o, τ ∗o)

c) For all organized foreign lobbies i ∈ L∗, there does not exist an alternativeoptimal trade policy vector (τ̃ i, τ̃ ∗i) and an alternative feasible C̃∗i (τ, τ ∗)strictly preferred by the lobby, i.e. that satisfy:

30Levy (1999) has shown that in a symmetric, two-country Grossman-Helpman worldthe effect of lobbying is to encourage net trade promotion.

31Grossman-Helpman (1997) discuss Rubinstein (1982) as one possible bargaining equi-librium that implements a politically efficient outcome.

32Note the cross-matching of coefficients with national welfares necessitated by the factthat the transfers are to be redistributed to the public whose welfare the politicians valuedifferently.

17

(i) (τ̃ i, τ̃ ∗i) = argmax(τ,τ∗)

a∗∑

j Coj (τ, τ ∗)+a(

∑j 6=iC

∗oj (τ, τ ∗)+C̃∗i (τ, τ ∗))+

aa∗(W (τ, τ ∗) + P (τ, τ ∗) +W ∗(τ, τ ∗) + P ∗(τ, τ ∗))(ii) W ∗

i (τ̃ i, τ̃ ∗i)− C̃∗i (τ̃ i, τ̃ ∗i) > W ∗i (τ o, τ ∗o)− C∗oi (τ o, τ ∗o)

This two-country game has a structure equivalent to a setup in which asingle world government maximizes the objective function given in a), takingas given the contributions from various interest groups from both countries.Therefore, the same solution approach utilizing the Bernheim-Whinston the-orem for a single economy can be applied. Because of the existence of non-convexities in the objective function (through δr(τ, τ

∗)), we split the domaininto four regions in order to find the optimal solution:

(i) R1: ∀(τ, τ ∗) : τ > 1 ∩ τ ∗ < 1

(ii) R2: ∀(τ, τ ∗) : τ < 1 ∩ τ ∗ > 1

(iii) R3: ∀(τ, τ ∗) : τ > 1 ∩ τ ∗ > 1

(iv) R4: ∀(τ, τ ∗) : τ < 1 ∩ τ ∗ < 1

It can be shown that the optimal trade agreement cannot fall in R3 or R4,because the first-order conditions with respect to τi and τ ∗i on these domainsare inconsistent with each other. The logic behind this result is the following:When we move the vector of trade policies along a certain ray from the originxi ≡ τi

τ∗i, this results in a welfare transfer from one country to another while

keeping internal prices fixed. At the same time, moving towards the originstarting from R3 and away from the origin starting from R4 would reducethe additional political cost P that the subsidies-paying government incurs.Thus, starting from R3 or R4, it will always be possible to increase jointwelfare by reducing trade interventionism and the political cost associatedwith it.

After some manipulation, the first order conditions for R1 and R2 respec-tively33 can be expressed in terms of the policy ratio x and therefore do notuniquely determine the trade policy vector:

x0 − 1 =y∗(x0)

a∗m∗(x0)e∗− y(x0)

am(x0)ex0 (17)

33The two FOC’s with respect to τi and τ∗i (for any i) are linearly dependent and theyreduce to a single FOC expressed as a function of x. There is one such equation for R1given by (17) and one for R2 given by (18).

18

xδ

(1 + δ(1 +

1

e)

)−(

1 + δ(1 +1

e∗)

)=

y∗(xδ)

a∗m∗(xδ)e∗− y(xδ)

am(xδ)exδ (18)

In the above expressions, e and e∗ signify the import and export elastici-ties, which are constant in the reduced form model considered here. x0 andxδ determine two ”peak rays”34, possible candidates for the global maximumrepresenting the optimal trade agreement, provided that they fall under theirassigned domains: x0 < 1 and xδ > 1. There are three possible case scenar-ios. If both x0 < xδ < 135, then xδ ∩ R236 is the optimal solution, underwhich trade would be stimulated with export and/or import subsidies. Ifx0 < 1 < xδ, then free trade (1,1) is the optimal (corner) solution. And fi-nally, if both x0 > 1 and xδ > 1, then x0∩R1 is the optimal trade agreementallowing for tariffs and export taxes.37

Obviously, if free trade is the outcome of the trade negotiations, the tradeagreement will eliminate all sources of political protectionism, beyond just re-moving the terms-of-trade externality. But trade agreements reduce politicalprotectionism even if they don’t necessarily result in free trade. To see thatthis is indeed the case, I will make use of the concept of politically optimalpolicies. Bagwell and Staiger (2004) define ”politically optimal” policies asthose that the governments would set non-cooperatively if they didn’t try tomanipulate the terms-of-trade. They represent the protectionist componentof the unilateral tariff. In the standard model, Bagwell and Staiger show,the politically optimal policies are always politically efficient, i.e. they areexactly what governments would agree to if they were to negotiate a tradeagreement. In other words, if governments were not motivated by the terms-of-trade motive when they set trade policies unilaterally, there would be noscope for any trade agreement to happen. Therefore, the entire purpose oftrade agreements in the standard model is to eliminate the terms-of-tradeexternality.

34Equations (17) and (18) do not typically have unique solutions, but the local maximaof interest correspond to the smaller of the two solutions.

35If both x0 < 1 and xδ < 1, then x0 < xδ, because the intersection of the left-hand side

expressions of (17) and (18) is x̄ = 1+1/ε∗

1−1/ε > 1. This property also guarantees continuity

of the optimal policy ratio x on the entire parameter domain.36I use the same notation for x to denote both the value of the slope and the vector

points corresponding to that ray {∀(τ, τ∗) : ττ∗ = x}, depending on context.

37Note that we didn’t have to check separately for corner solutions along the bordersτ = 1 and τ∗ = 1, because their values are already captured by the ”indifference rays” x.

19

Figure 2: Equilibrium trade agreements and politically optimal policies in the originalG-H model vs. this model.

In this model however, the politically optimal policies are never politicallyefficient. So even if governments ignore the terms-of-trade motive, therewould still exist scope to negotiate a trade agreement and further reduce tradeprotection, that arises from the mobilization of exporter interests to lobbymore cost-effectively for cutting foreign tariffs rather than obtaining domesticsupport in the form of export subsidies. If we subtract the politically optimal(PO) policies given by (19) and (20)38, divide by τPO, and compare the

38These expressions are obtained by eliminating the ToT term from (11) for Home andForeign.

20

resulting expression with (17) and (18), we conclude that they are neverequal i.e. they are never politically efficient.

τPO − 1 =y

a(−m′)π(19)

τ ∗PO − 1 =1

a∗(1 + δ)

y∗

(−m∗′)π− δ

(1 + δ)

τ ∗POe∗

(20)

Figure 2 shows graphically the effect of a trade agreement in a vec-tor policy space. In the example given with appropriate parameter choice,x0 < 1 < xδ, implying that free trade (FT ) is the optimal trade agreement.Relative to the non-cooperative equilibrium (NE), the trade agreement low-ers the importer’s tariff. Compared to the politically optimal policies (PO),the trade agreement lowers tariffs and export subsidies further, owing tothe mobilization of exporters to lobby for tariff reductions instead of exportsubsidies. We can utilize the same graph to analyze the model under thestandard Grossman-Helpman assumptions, here denoted by the subscript”GH”. In this case, the ray x0 represents the set of all possible optimal tradeagreements, which clearly includes the politically optimal policy POGH . Theconclusion is that once the terms-of-trade motive is eliminated for unilateraltrade policies (move from NEGH to POGH), governments have no incentivesto negotiate a trade agreement39.

Now we turn to the resolution of the terms-of-trade puzzle, whichrequires that the exporting country should have no incentive to manipulateits terms-of-trade by implementing export taxes or cutting export subsidies(unless the terms-of-trade motive truly is dominant over the political motive).Formally, this can be formulated as the requirement that the trade agreementpolicy vector cannot be ”north” of Foreign’s best response. In the case whenx0 < 1, a sufficiently high value for δ would place xδ in R1, which wouldimply that free trade is the optimal trade agreement. If the protectionistmotive dominates over the terms-of-trade motive in the exporting countrywhen the importing country practices free trade i.e. if τ ∗BR(τ = 1) > 1,then Foreign’s best response correspondence would pass right through freetrade (1,1), so the requirement would be satisfied. Conversely, if the terms-of-trade motive dominates and τ ∗BR(τ = 1) < 1 (which is very rare in reality)

39Agreeing to implement another policy along x0 does not constitute a qualitativelydifferent trade agreement, but rather represents a welfare transfer from one governmentto another, which is already fully captured by the transfer payment R.

21

then the trade agreement could and should stipulate provisions limiting theuse of export taxes. In the case when x0 > 1, x0 ∩ R1 will be the optimaltrade agreement, which similarly would call for trade treaty provisions for(incomplete) restriction of export taxes, provided that the terms-of-trademotive heavily dominates in the exporting country s.t. 1 < τ

x0< τ ∗BR(τ) for

some τ ≥ 1.

Inefficient redistribution puzzle

The inefficient redistribution puzzle highlights the fact that governmentstypically use inefficient policies in order to redistribute income to specialinterest groups even when other more efficient policies are available. In thecontext of trade, the puzzle applies to the use of tariffs instead of productionsubsidies or lump sum transfers40. I argue in this paper that subsidies incuradditional political cost to the incumbent government. This cost may inprinciple reverse the preference ranking of redistribution policies, favoringmore indirect measures such as tariffs and tax cuts over export subsidies,production subsidies and lump sum transfers.

To illustrate the argument analytically, I extend the model from the pre-vious section to also include production subsidies as a possible policy instru-ment, but for simplicity I restrict it to a small organized import-competingsector that cannot influence its terms-of-trade. If t is the production subsidyand τ the tariff, the prices that the producers and consumers receive are:pS = π + τ + t and pC = π + τ respectively. Production subsidies, like allsubsidies, incur additional political cost δ as a share of the funds received, sothe additional political term in the government objective function becomes:P (τ, t) = δy(pS)t. Solving as before, we obtain the first-order conditions fort and τ respectively:

t =

(1− αL − aδa+ αL + aδ

)y

y′−(

a+ αLa+ αL + aδ

)τ (21)

τ = −(

1− αLa+ αL

)y

m′+

(a+ αL + aδ

a+ αL

)y′

m′t (22)

40Dixit (1985) lays out the argument about the inefficiency of trade policy as a re-distirbution mechanism, while Ederington and Minier (2008) confirm the finding in thecontext of the Grossman-Helpman (1992) model by showing that production subsidieswould always be preferred to tariffs or export subsidies.

22

Thus we can solve explicitly for the two policies, which will be used si-multaneously as long as (1− αL)(−d′) > aδ(−m′):

t =y

y′(1− αL)d′ − aδm′

(a+ αL + aδ)d′(23)

τ =−aδy

(a+ αL)d′(24)

If the political cost of subsidies δ is high enough s.t. (1 − αL)(−d′) <aδ(−m′), then tariffs will be strictly preferred to production subsidies (t=0):

τ = −1− αLa+ αL

y

m′(25)

Finally, if δ = 0, then we get the familiar result that only productionsubsidies will be used as the less inefficient redistirbution instrument:

t =1− αLa+ αL

y

y′(26)

3. Data

In order to test the model we need to exploit variation in exporter sup-port. Export subsidies for industry have been banned by the GATT sinceits establishment, but agricultural export and production subsidies were un-regulated until the entry into force of the Uruguay Round Agreement onAgriculture on January 1st, 1995. Therefore, I use US data on average ex-port subsidies in the period 1986-1990 from the country’s notifications tothe WTO during the Uruguay Round negotiations. On the import protec-tion side, I use the ad valorem equivalents of non-tariff measures41 in 1999at the HS 6-digit level which are not regulated by WTO, as calculated byKee, Nicita and Olarreaga (2009) using Leamer’s (1990) comparative advan-tage approach. Domestic production support data is available either as AMS(Aggregate Measure of Support) 1986-1988 average from USA’s self-reportedcommitments on agricultural products to WTO, or as PCST (Producer Sin-gle Commodity Transfer) 1986-1990 average as calculated by OECD42.

41NTM’s include price and quantity control measures, technical regulations, monopolis-tic measures, such as single channel for import, etc.

42For detailed discussion of the differences between these two measures refer to Effland(2011).

23

The rest of the data used comes from a variety of sources. Trade flows datafrom 1991 is form the UN Comtrade database. Agricultural output data from1990 comes from the FAOSTAT database available online at the UN Foodand Agriculture Organization website43. I use Kee, Nicita and Olarreaga’s(2008) estimates of import demand elasticities (for the period 1988-2001)at the HS 6-digit level, and Broda, Limao and Weinstein’s (2006) exportsupply elasticity estimates for the period 1994-2003, which are estimated atthe HS 4-digit level. Aggregation44 is done at the level of author-defined”food groups” which roughly correspond to HS4-level products, but in someinstances were optimally expanded in order to better match the data at handon export subsidies, as well as the organizational structure of production (forexample, milk, dairy, beef and hides were all joined to form cattle farming).

Campaign contributions by PAC’s from 1990 obtained by the Center forResponsive Politics were used as a proxy for political organization by sec-tors45. In empirical estimations of the Grossman-Helpman model the problemof endogenous regressors (import penetration ratio and political organiza-tion) arises due to reverse causality, as argued by Trefler (1993), which isalleviated by the choice of appropriate instruments. I instrument for politi-cal organization using data on the farm size averages (at sectoral food grouplevel) for total wages paid and total employment obtained from the QuarterlyCensus of Employment and Wages (1990), as well as average total sales andharvested land area from the Agricultural Census (1992). The AgriculturalCensus also provides data on the value of capital invested (in terms of ma-chinery, land and buildings) which along with total farmland can be used asan instrument for the import penetration ratio.

The farm averages are independent of the scale of production and areproxies for industry concentration, which is a key determinant of the capacity

43In the few instances when only quantity output data was available, prices were acquiredseparately and deflated for the appropriate year, so that the gross production value ofoutput could be calculated.

44Aggregation is done using the TRI (Trade Restrictiveness Index) method developed byAnderson and Neary (1994; 2003), which aims to keep the home country welfare constantwhen aggregating. In the few cases (cattle, poultry and bee-keeping) when export elastic-ities of major subsectors are missing, we use the simple average method of aggregation.

45Assignment to food groups was based on CRP’s own industry classification, as wellas Beaulieu and Magee’s (2004) PAC-SIC correspondence for campaign contributions byvarious food industries.

24

of agricultural producers to organize46. Generally speaking, politicallyorganized groups may find it easier to lobby against anti-trust regulationwhich would have a reverse impact on industry concentration. However, thisobjection carries much less weight in the case of agriculture, because thesector has been exempt from the application of antitrust laws since 192247.On the other hand, total farmland and the value of the capital stock investedin the product sector capture the comparative advantage and therefore areappropriate regressors for the import penetration ratio, at least during theshort run time span this paper considers.

Because the data comes from a wide variety of sources, merging it is nottrivial. For this purpose the following concordance tables were used: HS-SIC-NAICS from Pierce, Schott (2009), FCL-HS from FAOSTAT, HS’07-HS’92from UN Stats. The HS-SIC and HS-NAICS matchings are based on 1992,the year used in the Agricultural Census data48. In case of multiple matchesfor the same HS 6-digit product, each match is associated with a weight (like-lihood of match) based on the number of HS 10-digit product lines associ-ated with that particular HS6-SIC (or HS6-NAICS) match. Similarly, a foodgroup may match to multiple HS 4-digit products, so we define weights basedon trade volume (imports plus exports) to be associated with each match.Because of the manageable number of food groups, manual assignment wasused whenever possible to improve upon the algorithmic assignment.

4. Estimation and results

The basis for structural estimation of the model under a non-cooperativeregime is equation (11). Rewriting it in terms of elasticities, we’d get:

τi − 1

τi=

(θi − αL)

a(1 + δ) + αL

ziei− aδ

a(1 + δ) + αL

1

ei+

1

τie∗i> 0 (27)

If the above expression is negative, then the optimal policy according to themodel is either free trade as a corner solution, import subsidies or export

46Trefler (1993b) also argues that industry concentration exogenously determines polit-ical organization and suggests several possible regressors.

47See CapperVolstead Act (P.L. 67-146), the Co-operative Marketing Associations Act(7 U.S.C. 291, 292)

48If a match doesn’t exist for 1992, then use matches for the next available year that’sclosest to 1992.

25

taxes. Because we don’t observe any import subsidies and export taxes inthe data, this implies that the model should be estimated using a Tobitregression with a lower boundary of 0.

The political participation index θi is usually treated in the literature andin the original Grossman-Helpman (1992) paper as a dummy variable, indi-cating whether the sector is politically organized or not. Bombardini (2008)and Gawande, Magee (2012) extend the G-H model to provide theoreticalmicro-foundations for treating θi as an index of partial political organizationtaking values between 0 and 1. In this paper I adopt the latter approach andI define θi as follows:

θi =

{ xiyiχ̄

: if xiyi≤ χ̄

1 : if xiyi> χ̄

Here, xi stands for the campaign contributions donated by sector i, while yiis its gross output. χ̄ is a cutoff value to be chosen that determines the levelof campaign spending above which a sector is considered fully organized andinvolved in the political process49.

In order for the set of instruments used Z to be valid, we need to makesure that they are relevant, which can be inferred from the value of the first-stage F-statistic using the Stock, Yogo (2002) critical values50. The otherrequirement for instrument validity is that they are exogenous, for which weconduct an overidentifying restrictions test using the Amemiya-Lee-Neweyminimum χ2 statistic.

There are two competing approaches for dealing with the terms-of-tradeterm. We can put it on the left-hand side to merge it with the dependentvariable, however in this case we would not be able to conduct an IV Tobitestimation of the model, but only purely OLS as in the standard Grossman-Helpman. The second approach and the one I pursue is to leave the terms-of-trade as a regressor meant to capture market power by specifying a particularfunctional form. I follow Broda, Limao and Weinstein’s (2006) and consider

49The sectors with the highest campaign contributions per dollar of revenue are tobaccoand the sugar industry, which are expected to lobby disproportionately more in order toinfluence regulation connected with the negative externalities associated with these sectors:obesity and smoking-related diseases. Therefore, it would be prudent to set χ̄ below theirxy ratios, for example the ratio of the sector that is third in ranking. I check multiplevalues of χ̄ for robustness.

50The Stock-Yogo values refer to linear 2SLS regressions, so may not be entirely appro-priate as guidelines for non-linear IV structural estimation.

26

a linear and a log specification51. This approach is also advantageous becauseit will parse out any terms-of-trade motivations in setting trade policy fromother considerations of lobbyist influence and political cost.

Formalizing the discussion hitherto, the estimating equations of the modelhave the following form:

τ ∗i − 1

τ ∗i= β0 + β1

ziθiei

+ β2ziei

+ β31

ei+ β4f(

1

e∗i) + εi

ziθiei

=∑j

γ1jZji + u1i

ziei

=∑j

γ2jZji + u2i (28)

τi − 1

τi=

{τ∗i −1

τ∗i: if

τ∗i −1

τ∗i> 0

0 : ifτ∗i −1

τ∗i≤ 0

Note that the underlying theory deals with perfectly competitive sectorswhich can be either exporting or importing. In reality, each agriculturalsectors both exports and imports, even for rather narrowly defined goods.Therefore, as a rule we get two observations out of every sector52. We excludethose observations that don’t make for ”convincing” importing (or exporting)sectors by setting a cutoff for the import (export) penetration ratio z > 100.53

The results for the baseline specification54are given in Table 1. The setof instruments includes: average sales per firm, average wages paid per firm,harvested land area, value of land and buildings capital per worker and valueof machinery capital per worker. They pass the overidentification test for

51As the authors write: ”We use a log specification to minimize the influence of theoutliers. The other motive for using the log specification is that the estimation procedurefor the elasticities cannot yield non-positive estimates. Thus the distribution of estimatesis skewed with positive deviations from the median vastly exceeding negative ones in mag-nitude. However, the density function of the log of the inverse export elasticity estimateshas a pattern quite similar to a normal density plot.”

52In fact, importing and exporting sectors are sometimes demarcated differently due todata availability, so they don’t necessarily match one-to-one.

53Gawande and Hoekman (2006) use a cutoff of z > 30 to eliminate observations, buthere that value would exacerbate the low power problem.

54In the baseline specification χ̄ = 30, which causes the distribution of θ to be centeredaround 0.7.

27

Table 1: IV-Tobit baseline specification; Estimating (28) with χ̄ = 30 cutoff for thepolitical organization variable.

28

exogeneity55 and they can be considered relevant with 0.25 maximal biasrelative to the OLS estimator at the 5% significance level56 The results showthat across all specifications the political cost term is negative as predictedby the theory, remains significant and explains a large share of the variation.This is evidence for the existence of additional political cost associated withthe imposition of subsidies, though it is by no means conclusive, as therecould be other models in which own sector elasticity enters the estimatingequation. In regressions (1), (4) and (7) we find empirical support for theassumption that firm owners constitute a negligible share of the population(αL = 0). There is also evidence to assume all sectors are fully organized s.t.θi = 1,∀i (regressions (2), (5) and (8)), but there is no evidence to reject bothof these assumptions (regressions (3), (6) and (9)). Across all specifications,there is no evidence that the terms-of-trade motivation has any effect ontrade policy once we control for the political cost term. This finding is inaccordance with the practitioners’ understanding of international agreementsand at odds with the literature which finds such an effect, namely Broda,Limao, Weinstein (2006). However, even when the estimates are insignificant,their signs are in accordance with the theory, so there could be an issue oflow power as there are only 32-34 observations.

Table 2 gives the estimates from a simple Tobit regression without instru-ments. The qualitative findings are very similar to the IV Tobit approach,but the quantitative estimates for all coefficients are significantly lower. Con-sequently, the implied structural parameters differ as well: δ is 0.1 instead of0.6-0.8, whereas the weight on social welfare a is five times larger. Therefore,the inclusion of the political cost term reduces the estimate for a to valuesin the low range of what previous studies have found57, though a value of 58is still perplexingly high.

The estimating results might be biased due to the existence of an ad-ditional policy instrument, such as production subsidies. The previous lit-erature58 has treated domestic support as equivalent to non-tariff measures,

55p-value=0.956The first-stage F statistics are 5.82 and 9.35, so they are above the relevant Stock-

Yogo critical values. However, if we are interested in the maximal size bias of a 5% Waldtest, then the instruments are somewhat ”weak”, because the critical value is 6.79.

57Previous studies estimate a from 24 (Eicher-Osang (2002)) all the way up to 3175(Gawange-Bandyopandhyay (2000)).

58Gawande, Magee (2012), Gawande, Hoekman (2006)

29

Table 2: Tobit (no instruments) specification with χ̄ = 30 cutoff for political organization.

aggregating them additively in the dependent variable to capture overall pro-tection without any theoretical basis. But as we saw in the theoretical sectionof this model, allowing for production subsidies leads to a substantially dif-ferent estimating equation for the non-cooperative case in which tariffs andproduction subsidies are conditionally dependent on one another. Extendingthe theoretical model to the large country case, with pS = π(τi, τ

∗i , ti, t

∗i )τt,

the first-order condition for τ , conditional on the production subsidy t, canbe expressed as:

τi − 1

τi=−(θi − αL)tia(1 + δr) + αL

ziei

+

(a+ αL + aδ

a+ αL + aδr

)zie

Si

ei(ti−1)−aδr − aδzi(ti − 1)

a(1 + δr) + αL

1

ei+

1

τie∗i(29)

Here eSi stands for the elasticity of producer supply, for which unfortunatelythere is no data available, so we proxy for it with the export supply elasticitybut aggregate them using gross output shares as weights. Incorporatingproduction subsidies in the estimation changes the political cost term andadds an additional term conditional on the production subsidy. The resulting

30

estimating system of equations59 is:

τ ∗i − 1

τ ∗i= β0 + β1

ziθitiei

+ β2zitiei

+ β3Di − zi(ti − 1)

ei+ β4

zieSi (ti − 1)

ei+ β5f(

1

e∗i) + εi

ziθitiei

=∑j

γ1jZji + u1i

zitiei

=∑j

γ2jZji + u2i (30)

τi − 1

τi=

{τ∗i −1

τ∗i: if

τ∗i −1

τ∗i> 0

0 : ifτ∗i −1

τ∗i≤ 0

Table 3 shows the results from the IV-Tobit estimation in (30), whilethe estimation in Table 4 treats all regressors as exogenous. We can con-clude that the inclusion of the production subsidy conditional term mattersin most specifications, however the sign is opposite of what the theory pre-dicts - rather than being policy substitutes, production subsidies and tradeprotection measures seem to be policy complements. This is most likely dueto omitted variable bias - an omitted factor that increases both trade pro-tectionism and domestic support will result in a positive bias. Or it couldalso be a problem of mis-measurement, since we proxy for production supplyelasticities with export elasticities60. Otherwise, the results are qualitativelyand quantitatively similar to those under the baseline specifications (withoutproduction subsidies). Most importantly, β3 is still negative and significant,so political cost matters even after controlling for production subsidies.

Finally, Table 5 shows the results from a traditional IV estimation of theGrossman-Helpman model, which rests on the assumption that δ = 0. Test-ing for the subsample of import sectors gives results that are quantitativelyand qualitatively in accordance with the previous literature. Nevertheless due

59To minimize degrees of freedom, we impose an approximating restriction: a(1 + δr)→a, which is very reasonable given the point estimates for a and δ. Nevertheless, we also runspecifications with added interaction terms for the third (pol. cost), fourth (conditional)and fifth (ToT) term. The interacting term Di equals 1 if the sector is exporting, and 0otherwise.

60Export supply elasticity is higher than production supply elasticity and to the extentthat this measurement error is correlated with the dependent variable, it could bias thecoefficient upward.

31

Table 3: IV Tobit specification, conditional on production subsidies; Estimating (30) withχ̄ = 30 cutoff for the political organization variable θ.

32

Table 4: Tobit (no instruments) specification, conditional on production subsidies, withχ̄ = 30 cutoff for θ.

33

Table 5: IV specification, standard Grossman-Helpman model.

to low power, the lobbying coefficients are not always significant.61 Omittingthe political cost term in the estimation renders the terms-of-trade coeffi-cient significant, however its negative sign and small magnitude are contraryto what the theory predicts, so it’s unclear how much importance to attachto this finding, if at all any. If we run the estimation for the entire sampleof importing and exporting sectors, none of the coefficients are significant.

I check the robustness of the results from multiple angles. First I con-sider different cutoffs62 χ̄ ∈ {10, 20, 30, 150} for the definition of the politicalorganization variable θ, both as a binary and partial variable between 0 and1. The political cost coefficient is negative and significant across all IV-Tobitspecifications and falls in the range (-0.47,-0.08). As before, there is evi-dence for the assumptions that either all sectors are organized or that firmowners constitute negligible part of consumers, but no evidence to rejectboth. Another robustness check pertains to the aggregation approach for thead-valorem equivalents of the non-tariff measures which determine the de-

61To increase the power of the test, I exclude harvested farmland from the set of instru-ments which maximizes the number of observations.

62The choice of cutoffs is based on natural breaks in the data.

34

pendent variable. In place of the TRI method, I use simple averaging, whichit turns out doesn’t affect the results substantially63. Finally, using alter-native measures for the agricultural domestic support, whether the OECD’sPCST measure or GATT’s AMS, also does not affect my conclusions fromthe estimations that take into account production subsidies.

5. Conclusion

In this paper I outlined how the standard model of international agree-ments that has been in use for over 20 years is at odds with facts about tradepolicy observed in practice and with the practitioners’ understanding thatspecial interest politics motivate the signing of trade treaties. I gathered sur-vey data that shows how voters’ beliefs regarding fiscal adjustments necessaryto accommodate government interventions in trade can impregnate deficit-generating policies with additional political cost beyond their monetary costas long as the government is in any degree democratically responsive to itsconstituents. I showed how adding this assumption of politically costly tradesubsidies to the standard model can explain how trade agreements resultin the reduction of political protectionism, as opposed to merely preventingterms-of-trade manipulation, and how it resolves three prominent puzzles inthe political economy of trade literature. I then tested the empirical predic-tions of the model on US data on agricultural trade policy and found thatthe political cost term implied by the assumption is negative and significantacross (almost) all specifications.

Understanding correctly the purpose of trade agreements matters notonly because it’s valuable in itself to have a coherent idea about how theworld works, but also because it could well have practical relevance for tradedisputes and the trade negotiations process. If trade negotiators for examplewere to take the standard model seriously, they would have to withdrawfrom trying to convince their trading partner to reduce tariffs further beyondthe elimination of the optimal tariff component. If judges in trade disputestook the standard model seriously, they would be unbothered by a countryraising its tariffs to protect a domestic industry as long as the terms-of-trade remained unaffected, because international agreements after all wouldbe designed solely to prevent ToT manipulation. We would have lived in a

63The point estimates are slightly higher.

35

world with significantly higher levels of trade protection than the one in whichwe actually live in. But fortunately practitioners do not take the standardmodel seriously. To be taken seriously the literature needs to close the gapbetween theory and reality and I believe this paper contributes to that goal.

36

References

[1] Anderson, James E., and J. Peter Neary. ”Measuring the restrictivenessof trade policy.” The World Bank Economic Review 8.2 (1994): 151-169.

[2] Bagwell, Kyle, and Robert W. Staiger. An economic theory of GATT.No. w6049. National Bureau of Economic Research, 1997.

[3] Bagwell, Kyle, and Robert W. Staiger. The economics of the world trad-ing system. MIT Press, 2004.

[4] Beaulieu, Eugene, and Christopher Magee. ”Four simple tests of cam-paign contributions and trade policy preferences.” Economics & Politics16.2 (2004): 163-187.

[5] Bernheim, B. Douglas, and Michael D. Whinston. ”Menu auctions, re-source allocation, and economic influence.” The quarterly journal of eco-nomics 101.1 (1986): 1-31.

[6] Bombardini, Matilde. ”Firm heterogeneity and lobby participation.”Journal of International Economics 75.2 (2008): 329-348.

[7] Broda, Christian, Nuno Limao, and David Weinstein. Optimal tariffs:The evidence. No. w12033. National Bureau of Economic Research, 2006.

[8] Destler, Irving M. American trade politics. Columbia University Press,2005.

[9] Dixit, A.K. (1985), ”Tax policy in open economies”, in: A.J. Auerbachand M. Feldstein, eds., Handbook of public economics, vol. I (North-Holland, Amsterdam).

[10] Dunkel, Arthur, and Frieder Roessler, 2012. The Ranking of Trade Pol-icy Instruments Under the GATT Legal System, in Andrew Guzmanand Joost Pauwelyn, International Trade Law, 2nd Edition (WoltersKluwer), 222-226.

[11] Ederington, Josh, and Jenny Minier. ”Reconsidering the empirical ev-idence on the GrossmanHelpman model of endogenous protection.”Canadian Journal of Economics/Revue canadienne d’conomique 41.2(2008): 501-516.

37

[12] Effland, Anne. ”Classifying and Measuring Agricultural Support.” Iden-tifying Differences Between the (2011).

[13] Eicher, Theo S., and Thomas Osang. ”Protection for sale: An empiricalinvestigation: Comment.” (2002).

[14] Ethier, Wilfred J. ”The theory of trade policy and trade agreements: Acritique.” European Journal of Political Economy 23.3 (2007): 605-623.

[15] Freund, Caroline L., and Caglar Ozden. ”Loss aversion and trade pol-icy.” (2004).

[16] Gawande, Kishore, and Usree Bandyopadhyay. ”Is protection for sale?Evidence on the Grossman-Helpman theory of endogenous protection.”Review of Economics and statistics 82.1 (2000): 139-152.

[17] Gawande, Kishore, and Bernard Hoekman. ”Lobbying and agriculturaltrade policy in the United States.” International Organization 60.03(2006): 527-561.

[18] Gawande, Kishore, and Christopher Magee. ”Free Riding and Protectionfor Sale1.” International Studies Quarterly 56.4 (2012): 735-747.

[19] Grossman, Gene M., and Elhanan Helpman. Protection for sale. No.w4149. National Bureau of Economic Research, 1992.

[20] Grossman, Gene M., and Elhanan Helpman. ”Trade wars and tradetalks.” Trade and Tax Policy, Inflation and Exchange Rates. SpringerBerlin Heidelberg, 1997. 171-214.

[21] Grossman, Gene M. ”The purpose of trade agreements.” Handbook ofCommercial Policy 1 (2016): 379-434.

[22] Hudec, Robert E. ”Circumventing Democracy: The Political Moralityof Trade Negotiations.” NYUJ Int’l L. & Pol. 25 (1992): 311.

[23] Johnson, Harry G. ”Optimum tariffs and retaliation.” The Review ofEconomic Studies 21.2 (1953): 142-153.

[24] Kee, Hiau Looi, Alessandro Nicita, and Marcelo Olarreaga. ”Importdemand elasticities and trade distortions.” The Review of Economicsand Statistics 90.4 (2008): 666-682.

38

[25] Looi Kee, Hiau, Alessandro Nicita, and Marcelo Olarreaga. ”Estimatingtrade restrictiveness indices.” The Economic Journal 119.534 (2009):172-199.

[26] Leamer, Edward E. ”Latin America as a target of trade barriers erectedby the major developed countries in 1983.” Journal of DevelopmentEconomics 32.2 (1990): 337-368.

[27] Krugman, Paul. ”What should trade negotiators negotiate about?.”Journal of Economic Literature 35.1 (1997): 113-120.

[28] Levy, Philip I. ”Lobbying and international cooperation in tariff set-ting.” Journal of international Economics 47.2 (1999): 345-370.

[29] Limo, Nuno, and Arvind Panagariya. ”Inequality and endogenous tradepolicy outcomes.” Journal of International Economics 72.2 (2007): 292-309.

[30] Pauwelyn, Joost. ”New trade politics for the 21st century.” Journal ofInternational Economic Law 11.3 (2008): 559-573.

[31] Pierce, Justin R., and Peter K. Schott. Concording US harmonized sys-tem categories over time. No. w14837. National Bureau of EconomicResearch, 2009.

[32] Regan, Donald H. ”Explaining Trade Agreements: The Practitioners’Story and the Standard Model.” World Trade Review 14.03 (2015): 391-417.

[33] Rodrik, Dani. ”Political economy of trade policy.” Handbook of inter-national economics 3 (1995): 1457-1494.

[34] Rubinstein, Ariel. ”Perfect equilibrium in a bargaining model.” Econo-metrica: Journal of the Econometric Society (1982): 97-109.

[35] Stock, James H., and Motohiro Yogo. ”Testing for weak instruments inlinear IV regression.” (2002).

[36] Tovar, Patricia. ”The effects of loss aversion on trade policy: Theory andevidence.” Journal of International Economics 78.1 (2009): 154-167.

39

[37] Trefler, Daniel. ”International factor price differences: Leontief wasright!.” Journal of political Economy 101.6 (1993): 961-987.

[38] Trefler, Daniel. ”Trade liberalization and the theory of endogenous pro-tection: an econometric study of US import policy.” Journal of politicalEconomy 101.1 (1993b): 138-160.

40

Appendix A. Survey questions and responses

The following survey on voters’ beliefs and policy preferences was admin-istered online via Survey Monkey, an online survey development tool, on aninitial sample of 1065 respondents from the United States. Respondents wererecruited through Amazon Mechanical Turk, a marketplace for on-demandhuman intelligence workforce, by providing a monetary incentive of $0.5 forevery successful completion of the survey. To ensure quality control, sub-missions which were completed in less than 250 seconds were excluded fromthe sample. The final sample consisted of 620 respondents for the questionsconsidering an exporting industry, and 309 respondents for the questionsconsidering an importing industry64.

Appendix A.1. Part 1

1. How old are you?

2. What is the highest degree of education you have obtained?

(a) High school or less - 165(b) College (BA) - 330(c) Postgraduate degree (Master’s, MPhil, PhD, professional degree)

- 125

3. How familiar are you with economics?

(a) Not very familiar - 63(b) I haven’t had any formal instruction in economics, but I am broadly

familiar with economic concepts - 220(c) I have taken only an intro-level economics course - 302(d) I have graduated with a minor or major in economics - 35

4. How do you identify politically when it comes to economic policies?(Choose the option that most closely corresponds to your convictions)

(a) Progressive/Socialist (very left-wing) - 33(b) Liberal/Democrat (left-wing) - 186(c) Independent, leaning liberal (slightly left-wing) - 122(d) Independent, leaning conservative (slightly right-wing) - 118(e) Conservative/Republican (right-wing) - 135(f) Libertarian (very right-wing) - 26

5. Which of the following best describes your voting behavior?

64Due to a technical glitch, many respondents weren’t able to complete both sections.

41

(a) I vote in (almost) every election. - 367(b) I vote mostly in Presidential elections, but typically not in Midterm

elections. - 168(c) I hardly ever vote. - 69(d) I don’t have the right to vote, but I would if I did. - 15(e) I don’t have the right to vote and I wouldn’t if I did. - 1

6. Consider a certain US industry in which you are not employed, thatconsists of many companies making identical products that you don’tneed and never buy for personal use. The products however sell for $500per product and are also exported abroad. For each of the followingpolicies that the federal government considers implementing for thisparticular industry, indicate your personal stance as a voter. (Stancesrange in order: Strongly support, Support, Neutral, Oppose,Strongly Oppose)

• Tax raise: Government levies $100 tax on every product sold - 14,129, 158, 211, 108

• Tax cut: Government cuts the companies’ taxes by $100 for everyproduct sold - 25, 115, 141, 232, 107

• Production subsidy: Government gives companies $100 for everyproduct sold, whether at home and abroad - 16, 80, 132, 243,149

• Export subsidy: Government gives companies $100 for every prod-uct sold abroad - 19, 87, 140, 228, 146

7. For the industry described in question 6, rank the following policies inorder of your personal preference. You are allowed to assign the sameranking to multiple policies. (Rankings range in order from: 1 (mostpreferred), 2, and 3 (least preferred).)

• Production subsidy: Give companies $100 for every product sold,whether at home and abroad - 138, 235, 247

• Tax raise: Levy a $100 tax on every product sold - 228, 147, 245

• Export subsidy: Give companies $100 for every product sold abroad- 137, 253, 230

8. Suppose the government decided to levy taxes on the industry describedin question 6. What do you believe the government is most likely todo with this expected increase in tax revenue?

42

(a) It will spend it in a way that will benefit the average voter - 82(b) It will spend it in a way that will not benefit the average voter -

400(c) It will save it to reduce government debt - 113(d) It will use it to reduce taxes on the average voter for the same

amount - 259. Now suppose the government decided to give subsidies of $100 per

product sold in the industry described in question 6. How do youbelieve the government is most likely to pay for this expected expense?(a) It will raise money through additional taxes on the average voter

- 306(b) It will borrow money from investors - 62(c) It will cut spending elsewhere at the expense of the average voter

- 191(d) It will cut spending elsewhere, but the average voter will not be

negatively affected - 61(e) It depends on whether the subsidies are given only for exported

products, or for any product sold (please explain how your answerdiffers in each case) - 0

10. Suppose the industry described in question 6 wasn’t subsidized at allin 2015. In 2016 the government gave companies subsidies for exportsat 20% their sales price, and repealed them again in 2017 back to zero.Suppose that nothing else changed. Do you believe that you as a voterwould be better off in 2017 than in 2015? Please briefly explain yourreasoning why. (Reasons answers not included.)(a) I’d be better off in 2017. - 131(b) There is no reason why in either year I would be better off than

in the other. - 360(c) I’d be better off in 2015. - 129

Appendix A.2. Part 2

11. Consider a certain US industry in which you are not employed, thatconsists of many companies making identical products that you don’tneed and never buy for personal use. The products however sell for $500per product and they also need to be imported from abroad to satisfydomestic demand. For each of the following policies that the federalgovernment considers implementing for this particular industry, indi-cate your personal stance as a voter. (Stances range in order: Stronglysupport, Support, Neutral, Oppose, Strongly Oppose)

43

(a) Tax cut: Government cuts sales taxes by $100 for every productpurchased - 17, 74, 79, 96, 43

(b) Production subsidy: Government gives US companies $100 forevery product sold at home - 23, 57, 86, 85, 58

(c) Tax raise: Government levies $100 sales tax on every productpurchased - 10, 62, 92, 96, 49

(d) Import tariff: Government levies $100 tax (tariff) on every prod-uct imported from abroad - 26, 82, 86, 77, 38

12. For the industry described in question 10, rank the following policiesin order of personal preference. You are allowed to assign the sameranking to multiple policies. (Rankings range in order from: 1 (mostpreferred), 2, and 3 (least preferred).)

• Import tariff: Levy a $100 tariff on every product imported fromabroad - 117, 109, 83

• Production subsidy: Give US companies $100 for every productsold - 90, 100, 119

• Tax raise: Levy a $100 sales tax on every product purchased - 49,123, 137

13. Suppose the government decided to levy tariffs or taxes in the industrydescribed in question 10. What do you believe the government is mostlikely to do with this expected increase in government revenue?

(a) It will spend it in a way that will benefit the average voter - 52(b) It will spend it in a way that will not benefit the average voter -

177(c) It will save it to reduce government debt - 66(d) It will use it to reduce taxes on the average voter for the same

amount - 13(e) It depends on whether the revenue is raised through tariffs on

imported goods or through taxes on all sold goods (please explainhow your answer differs in each case) - 1

14. Now suppose the government decided to give production subsidies inthe industry described in question 10. How do you believe the govern-ment is most likely to pay for this expected expense?

(a) It will raise money through additional taxes on the average voter- 150

(b) It will borrow money from investors - 43

44

(c) It will cut spending elsewhere at the expense of the average voter- 94

(d) It will cut spending elsewhere, but the average voter will not benegatively affected - 22

15. Suppose the industry described in question 10 didn’t tax imports atall in 2015. Then the government raised the import tariff to 20% in2016, and lowered it again back to zero in 2017. Suppose that nothingelse changed. Do you believe that you as a voter would be better off in2017 than in 2015? Please briefly explain your reasoning why. (Reasonsanswers not included.)

(a) I’d be better off in 2017. - 64(b) There is no reason why in either year I would be better off than

in the other. - 207(c) I’d be better off in 2015. - 38

Appendix A.3. Key observations

O1: On average, for the exporting industry respondents oppose produc-tion subsidies the most (-0.69 score on a scale -2 to 2), followed by exportsubsidies (-0.64), tax cuts (-0.45) and raising taxes (-0.43). 46% of respon-dents strictly prefer taxing the industry over production and export subsi-dies, compared to 36% who strictly prefer production or export subsidiesover taxes. Both left-leaning and right-leaning respondents oppose produc-tion and export subsidies, but left-leaning respondents are a lot less likelyto oppose increasing taxes, while right-leaning respondents are less likely tooppose tax cuts.

O2: Respondents overwhelmingly (83%) believe that the governmentwould not use revenue from taxing an exporting industry to directly benefitthem, either through increased spending or reducing personal taxes. Con-versely, 80% believe that if the government needed funds to subsidize pro-duction in the exporting industry, it would obtain them at the expense ofthe average voter, either through increased taxes or spending cuts.

O3: 58% of respondents believe that imposing and repealing export subsi-dies leaves voters as well off as before the policy. The rest are equally dividedbetween feeling they’d be worse off and feeling they’d be better off. Whenasked to justify their preference, none of these respondents identifies the fiscaleffect on voters caused by the reversal of policy as the reason. This suggeststhat on average the preference for subsidies exhibits path-independence anddepends only on the level of the subsidy, as we assumed in the model.

45

O4: On average, for the importing industry respondents oppose taxes onconsumption the most (-0.36 score on a scale -2 to 2), followed by productionsubsidies (-0.32), tax cuts (-0.24) and imposing tariffs (-0.06). 48% of respon-dents strictly prefer taxing imports over production subsidies, compared to34% who strictly prefer production subsidies over tariffs. 52% of respon-dents strictly prefer taxing imports over taxing consumption, compared to24% who strictly prefer consumption taxes over tariffs. Both left-leaningand right-leaning respondents oppose production subsidies, and their stanceon taxes reflects the same political divisions as in O2, however right-leaningvoters are more favorable to tariffs than leftists, and actually slightly supportthem (+0.01).

O5: Respondents overwhelmingly (79%) believe that the governmentwould not use increased revenues from trade policies to directly benefit them,either through increased spending or reducing personal taxes. Conversely,21% believe that if the government granted production subsidies to the im-port competing industry, it would finance them at the expense of the averagevoter, either through increased taxes or spending cuts.

O6: 67% of respondents believe that imposing and subsequently repealingimport tariffs leaves voters as well off as before the policy. 37% of the restfeel they’d be worse off than before, but none of them identifies the fiscaleffect on voters caused by the reversal of policy as the reason behind theirpreference. This suggests that on average the preference for taxes (tariffs)exhibits path-independence and depends only on the level of the tax, as weassumed in the model.

46

Appendix B. Theoretical predictions for policy preferences

The motivating assumption in this paper is that the median voter’s wel-fare is a weighted average of the aggregate social welfare (SS) and the politi-cal cost (benefit) of trade policies, whereby policies associated with increasedfiscal outlays incur higher political cost for every dollar spent (δO) than thepolitical benefit of policies associated with increased fiscal revenues per dollarof revenue raised (δR)65:

W = aSSS + 1(r < 0)δOr− + 1(r ≥ 0)δRr+ (B.1)

To examine the reasonableness of this assumption, we derive the theoreticallypredicted preferences over trade policies of a hypothetical agent with theabove described welfare function in the context of a small open economy,and then compare them with the actual voters’ preferences as measured bythe survey. We limit our analysis and survey to the case of a small openeconomy in which the respondent (agent) holds no direct stake either asa consumer nor as an employee and is exclusively invested in his role as avoter in order to abstract from any superfluous effects such as terms-of-trade,consumption and income effects, which would only complicate the analysisand make it difficult to draw any inferences.

PolicyTax onproduction

Productionsubsidy

Exportsubsidy

Revenue (outlay): r xQT −xQS −x(QS −QD)Social surplus: SS -C -A -A-BAgent: W δRxQT − aSC −δOxQS − aSA −δOx(QS −QD)− aS(A+B)Agent: sign -/+ - -

Table B.6: Calculated changes in welfare (surplus) due to implementation of surveyedpolicies in a partial equilibrium exporting industry. The monetary magnitude of all threepolicies is fixed at x per unit of output.

Table 6 summarizes the welfare effects for the agent from implementingany of the three surveyed policies (for a given magnitude $x per unit ofoutput66) in the partial equilibrium export industry depicted in Figure 3.

65In the model, I assume for simplicity that δO > 0 = δR, but a more realistic claimwould be limited to δO > δR > 0.

66In the survey x=$100.

47

Figure B.3: Social surplus effects in a partial equilibrium export industry for export sub-sidy, production subsidy and production tax all equal to x per unit of output.

Note that the preferences implied by considering only the changes in socialsurplus correspond to the preferences of the hypothetical aggregate-welfare-maximizing representative voter implied by the Grossman-Helpman model.

The first thing to note is that the admissible signs on the agent’s welfareeffect do not contradict the direction of the surveyed voters’ preferences sum-marized in O1 – voters on average oppose all policy interventions. However,the G-H welfare-maximizing agent poses a contradiction, because in theoryshe should be in favor of a tax cut, however in reality voters oppose it evenmore than the other deadweight-loss inducing policies. Therefore, whateverthe government motivations in the standard model may be, they exclude theprinciple of democratic accountability.

Unfortunately, we cannot infer any strict policy rankings without mak-ing parametric assumptions. If voters are to prefer taxing over subsidizingproduction, as O1 shows that they in fact do, then the following inequality

48

must hold:

δOx(QT (x)δRδO

+QS(x)) > aS(C(x)− A(x)) (B.2)

This inequality holds only if δO is high enough, i.e. higher than a certain criti-cal value δcO(x, aS), and it cannot hold for the Grossman-Helpman assumptionδO = 0.67 Thus, introducing voting and political cost considerations in theGrossman-Helpman lobbying approach reconciles it with the survey data.

Table 7 applies the same analysis as above to the case of the partialequilibrium import-competing industry depicted in Figure 4 for the threesurveyed policies: consumption tax, import tariff and production subsidy.

Figure B.4: Social surplus effects in a partial equilibrium import-competing industry fortariff, production subsidy and consumption tax equal to x per unit of output.

Just as for the exporting industry, the theoretically predicted signs matchthe stated preferences in the survey, while the fact that respondents on aver-age still oppose tax cuts is evidence against the idea of a welfare-maximizingrepresentative voter.

67Because C(x) ≥ A(x)∀x. This follows from the convexity of the supply function.

49

PolicyTax onconsumption

Productionsubsidy

Importtariff

Revenue (outlay): r xQD −xQS x(QD −QS)Social surplus: SS -B -A -A-BAgent: W δRxQD − aSB −δOxQS − aSA δRx(QD −QS)− aS(A+B)Agent: sign -/+ - -/+

Table B.7: Calculated changes in welfare (surplus) due to implementation of surveyedpolicies in a partial equilibrium import competing industry. The monetary magnitude ofall three policies is fixed at x per unit of output.

Comparing the welfare effects, it follows that for any given x, our agentstrictly prefers consumption taxes over tariffs, and if δR is ”high enough”,consumption tax is preferred over production subsidies as well. However,O4 shows that the surveyed respondents were on average most opposed toconsumption taxes. This suggests that the model may be missing a salientaspect of reality that could explain this preference ranking. It is possiblethat voters do not internalize the logic of the perfectly competitive marketas much as economists do, so they oppose consumption (sales) tax more thanthey would en equivalent tax on production due to a widely held belief thatit is primarily consumers who bear the burden of the sales tax. Anotherpotential explanation may be the static nature of the model: Whereas instatic models tariffs strictly reduce aggregate welfare, if there are learningspillovers in the protected industry, voters may be internalizing these gainsand calling for infant industry protection through higher tariffs, which wouldmake them preferable to consumption taxes despite being comparatively lessefficient.

50