Protection and jobs: explaining the structure of trade barriers across industries

Journal of International Economics 61 (2003) 19–39www.elsevier.com/ locate/econbase

P rotection and jobs: explaining the structure of tradebarriers across industries

*Scott BradfordEconomics Department, Brigham Young University, Provo, UT 84602,USA

Received 29 October 2001; received in revised form 11 April 2002; accepted 16 May 2002

Abstract

This paper develops a model of protection and tests it using US data, including newprotection measures. We find that protection in an industry increases with its employmentbut not with its level of output. We also find that lobbying entails significant transactionscosts. We have limited evidence that industry characteristics, such as the number of firmsand geographical concentration, affect protection. Nested tests imply that assuming lump-sum rebating of import revenues or rents is justified. The results also suggest that US policymakers weight a dollar of campaign contributions about 15% more heavily than a dollar ofnational income. 2002 Elsevier B.V. All rights reserved.

Keywords: International trade; Protection; Political economy

JEL classification: D72; F13

1 . Introduction

Protection persists despite its weighty costs, and many studies have investigatedits causes. Still unresolved is the question of why some industries receivesignificantly more protection than others. Shedding light on what explains thevariation of trade barriers across industries may suggest more effective strategiesfor reducing protection (see Helpman, 1995; and Richardson, 1993). Also,

*Tel.: 11-801-422-8358; fax:11-801-422-0194.E-mail address: [email protected](S. Bradford).

0022-1996/02/$ – see front matter 2002 Elsevier B.V. All rights reserved.doi:10.1016/S0022-1996(02)00077-6

mailto:[email protected]

20 S. Bradford / Journal of International Economics 61 (2003) 19–39

incorporating the politics of trade is likely to improve trade theory (for example,see Cassing and Hillman, 1986).

Within the theoretical literature, Hillman (1982) pioneered the political supportapproach to trade barriers, in which policy makers choose protection levels tomaximize support from special interest groups and society at large. Grossman andHelpman (1994) extended this work by developing a framework that explicitlymodels lobbying. Many have built insightfully on their foundation (Maggi andRodriguez-Clare, 2000, among others). So far, however, the theoretical literaturehas tended to neglect the impacts of employment levels and of industrycharacteristics, such as number of firms, on protection levels. Also, almost all suchmodels imply that protection increases with output.

There are numerous econometric analyses of the structure of protection across1industries. Formal modeling does not guide most such studies, making the results

hard to interpret. Recent work, however, has advanced this literature by basingregressions strictly on theory. For instance, Goldberg and Maggi (1999) andGawande and Bandyopadhyay (2000) examine the Grossman–Helpman (GH)model empirically and confirm the main GH prediction—that protection isincreasing in the ratio of output to imports. They also find that jobs and industrycharacteristics do not significantly affect protection levels. They do not, however,explicitly model these factors, only adding them to the regressions in an ad hocway.

This paper provides additional evidence on the pattern of protection acrossindustries by estimating a structural equation derived from a new model. We alsouse alternative measures of protection that we believe are more trustworthy thanprevious ones. In the model (presented and discussed in Sections 2 and 3), policymakers choose protection levels to maximize votes, producers and import rentseekers lobby for rents, and workers and consumers vote according to protection’seffect on their economic well-being. The model implies that protection increaseswith workforce size and decreases with lobbying costs (as proxied, for instance, bythe number of firms). The predicted effects of output and imports are ambiguous.This paper’s primary results are in Section 4, which takes the model to the dataand confirms its main predictions. The results also imply that the governmentweights a dollar of campaign contributions about 15% more heavily than a dollarof consumer surplus and that each $1000 reduction in consumer surplus results inthe loss of one vote.

1Baldwin (1984) and Rodrik (1995) survey this large literature. Trefler (1993) is especiallynoteworthy. Our model below combines two of the informal models that Baldwin (1984) summarizes:the pressure group model and the adding machine (voting) model. (These are the two models thatGawande (1998) finds to be most important.)

S. Bradford / Journal of International Economics 61 (2003) 19–39 21

2 . The model

2 .1. The economic structure

nConsider a small economy whose consumers maximizeu 5 x 1o u (x ). x0 i51 i i 0

is the numeraire and is traded freely at a price of 1. Eachu is differentiable andi

strictly concave. The population is normalized to 1. Consumer surplus for eachnon-numeraire good iss( p )5 u [d ( p )] 2 p d ( p ), where p is the price andi i i i i i i i

2d ( p ) is demand.i i

For production, consider a modified version of the specific factors model. Thereare constant returns to scale and perfect competition. Capital is specific, and laboris mobile. Unlike the standard model, though, we follow Davis (1998) and Brecher(1974) and assume that each sector’s wage is fixed above market-clearing and

3each worker’s reservation wage, causing involuntary unemployment. Thus,L 5id ¯ ¯L (w /p ), whereL is the amount of labor in industryi, andw is the fixed wage.i i i i i

d 49SinceL , 0, dL /dp .0. The total reward to specific capital depends only oni i i

the price:p 5p ( p ), with p9. 0.i i i5The government chooses trade barriers of any type for each industry . No matter

the barrier, it raises the domestic pricep above the fixed world price, denoted byiwp . We assume no lump-sum rebating of trade tax revenues or import rents.i

Instead, lobbies compete for these.

2 .2. The political structure

Import-competing producers form lobbies. We also allow for a second type of6lobby. Since non-tax protection generates rents for importers , we assume that

importer lobbies form, one per sector. (Lobbies may also vie for trade taxes. SeeBhagwati and Srinivasan, 1980.) We assume that the effect of lobbying on anylobby’s consumer surplus is negligibly small. Thus, welfare for each producer

2This consumption structure follows GH.3The working paper, available upon request, shows one way to endogenize rigid wages. As long as

the wage is rigid and leads to unemployment, all of the results below hold. Allowing wages to movemight be interesting further work, though Revenga (1992) shows that employment is much moreflexible than wages in the short run.

4Labor is not specific in this model. Laid off workers can move to any sector. Employed workers dofeel attached to their sectors, since the alternative is a period of unemployment. Davidson et al. (1999)discuss how unemployment blurs the distinction between Heckscher–Ohlin and specific factors models.

5We do not analyze why protection is chosen over more efficient tools, taking as given thatprotection is ubiquitous. Rodrik (1986) and Mitra (2000) model the choice of protection over subsidies.

6See Krueger (1974) and Maggi and Rodriguez-Clare (2000) for models with such lobbying.


Plobby is simply profits,w ( p )5p ( p ), while welfare for each importer lobby isi i i iIM wthe value of import rents,w ( p )5 ( p 2 p )m ( p ), wherem is the quantity ofi i i i i i i

imports.Lobbying consists of making contributions. We assume that each lobby engages

in a bilateral Nash cooperative game with the government, so that total surplus is7maximized, with the division of the surplus indeterminate. Each lobby takes

prices in all other sectors as given. Most models assume that contributions arefrictionless transfers, but we allow for transactions costs. Contributions receivedby the government from the producer and importer lobbies, respectively, are

P P P IM IM w IM PC ( p )5 b [p ( p )2B ] and C ( p )5 b [( p 2 p )m ( p )2B ], where bi i i i i i i i i i i i i i iIM 8 P IMand b (both # 1) are lobbying friction coefficients , andB and B are thei i i

rents that the lobbies retain in the bargain. In this Nash set-up, the price chosendoes not affect theseB terms.

The government has the following objective function:

nd w P IM wV( p)5O[(L ( p )2 L )1 c[C ( p )1C ( p )] 1 a(s ( p )2 s )] (1)i i i i i i i i i i

i51

w wwhereV( p) is total votes;L and s are free trade levels of employment andi i

consumer surplus, respectively;c is the fraction of a vote that a dollar ofcontributions buys; anda is the votes lost per dollar of lost consumer surplus.

d wThe first term,L ( p )2L , gives the number of votes won from workers ini i i

protected industries. Specifically, each worker hired due to protection switchesfrom voting against the government (because he or she was unemployed) to voting

9for it. Thus, employment is assumed to override price changes in determining howworkers vote. The second pair of terms captures contributions induced byprotection; the government uses these funds to ‘buy’ votes at a rate ofc votes per

10dollar. The last pair of terms captures lost support from consumers who facehigher prices, with the number of votes lost directly proportional to the loss ofconsumer surplus.

7See Helpman (1995) for a discussion of why a series of bilateral bargains might be a morereasonable approach than that of a single multilateral menu auction game.

8The lobbying frictions are exogenous and thus compatible with efficient bargaining: the parties stillend up on the Pareto frontier, even though frictions may affect the position of the frontier.

9The results do not depend on havingeach worker switch her or his vote. As long as some fractionswitch, the results go through. Thus, we could express the model in terms of probabilistic voting. Anextension of this framework would be to model more explicitly who will switch. Note that partyaffiliation and the ideology of the government (eg, whether it is ‘free trading’ or not) do not play rolesin this model.

10See Potters et al. (1997) for a model of how campaign spending buys votes.


11The government chooses the price in each sector so as to maximize votes.Taking the derivative ofV with respect to a representative price,p , yields:i

P IMw ¯ ¯ ¯ ¯e L 1 c(b y 1 b m )2 a(y 1m )*p 2 p (L, p),i i i i i i i ii i˜ ]]] ]]]]]]]]]]]P 5 5 (2)i IM*p ¯e L 1 cb e mi (L, p),i i i (m, p),i i

where, in each sectori, e is the elasticity of labor demand,e is the(L, p),i (m, p),iw¯elasticity of import demand (.0), y 5 y p is the value of output at free tradei i i

w 12¯prices, andm 5m p is the value of imports at free trade prices.i i i

3 . Discussion and implications of the model

Three main features distinguish our model from most of the literature: treatingworkers as a separate source of support, instead of lumping them in withconsumers as a whole; not assuming lump-sum rebating of trade barrier revenue orrents; and allowing for lobbying frictions. In order to capture these forces whilepreserving tractability, we have assumed that all potential lobbies organize andthat the impact of lobbying on that lobby’s consumer surplus is negligible.

To help in interpreting Eq. (2), consider how each of the three main featuresaffects it. TheL terms in Eq. (2) capture the independent influence of workers oni

protection. Without rigid wages and unemployment, these terms would disappear,and the equilibrium equation would be

P IM¯ ¯ ¯ ¯c(b y 1 b m )2 a(y 1m )i i i i i i˜ ]]]]]]]]P 5 .i IM ¯cb e mi (m, p),i i

This resembles more closely GH-type models, in which protection depends on theratio of output to imports, the weight that contributions get relative to consumersurplus (c vs. a), and the elasticity of import demand.

Nevertheless, this expression is still more complicated because it does notassume lump-sum rebating of trade barrier revenue or rents and because there are

11In doing so, we follow Peltzman (1976) and Baldwin (1987). We could easily reformulate themodel to have policy makers maximize ‘power’ or ‘wealth’. See Becker’s comments on Peltzman.Also, although democratic governments only need a simple majority of votes to stay in office,super-majorities have value because they make it easier for governments to implement their overallagenda.

12P̃ maps protection onto the [0, 1) interval. This formulation follows GH and Goldberg and Maggi.i

Also, we assume no negative protection, which would imply import subsidies. This appears to beinnocuous since all the industries in our sample get positive protection. For notational convenience, wehave suppressed the dependence of labor demand, output, imports, and the two elasticities on the price.


IMlobbying frictions that vary by industry. To assume rebating would setcb 5 a,i

which means that the revenues or rents get the same weight as consumer surplus.This would reduce Eq. (2) to

P ¯e L 1 (cb 2 a)y(L, p),i i i i˜ ]]]]]]P 5i ¯e L 1 ae m(L, p),i i (m, p),i i

with unemployment and

P P¯(cb 2 a)y (cb 2 a)yi i i i˜ ]]] ]]]P 5 5i ¯ae m ae m(m, p),i i (m, p),i i

without.P IMIf contributions are assumed to be frictionless transfers, then theb and bi i

terms in any of the above equilibrium expressions would equal 1. Applying thischange to the previous equation, so that all three of the main features are dropped,gives

(c 2 a)yi˜ ]]]P 5 .i ae m(m, p),i i

Adopting the GH convention of lettingc 5 11 a, further reduces this to

yi˜ ]]]P 5 .i ae m(m, p),i i

This is the expression for protection that comes out of their framework (withseparate bilateral bargains) when all industries lobby and each lobby’s consumer

13surplus is ignored. Our model, therefore, is an alternative to the GH model butnot a generalization of it: we have abstracted from certain features of their modelwhile adding new elements. The model implies several ceteris paribus results forimport-competing industries.

14Result 1: Sectors with more workers receive more protection. The intuition forthis is straightforward. More workers in an industry means more potential votesfor the government when it imposes protection. The empirical literature finds arobust connection between workforce size and trade barriers. By incorporatinginvoluntary unemployment, this model differs from others in providing a theoret-ical backing for the claim, accepted by most observers, that jobs do matter.

Result 2: Sectors with higher elasticities of labor demand receive more

13Referring to Helpman (1995), all industries being organized means that the equation on p. 22 ofthat article applies to all industries, and abstracting from changes in lobbies’ consumer surplus meansthat a 5 0.j

14 d˜ ¯P is monotonically increasing inL . IncreasingL may also decreasee , but, as long asL (w /p)i i i (L, p),i

is decreasing, this secondary effect will not outweigh the direct effect of increasingL.


protection. Holding all other variables fixed, industries that hire many workers inresponse to a price increase also provide many votes if granted protection and thuswill receive more of it. As with result 1, this is unique, as far as we know, to thetheoretical literature.

PResult 3: Sectors with lower lobbying transactions costs (higher b ) receivei

more protection. Lower transactions costs imply that the producer lobby will havemore clout with the government, since the government will actually receive moreof the resources that the lobby dedicates to lobbying. Such transactions costs areprobably closely connected to the extent of free riding, which, in turn, is probablyrelated to such variables as industry concentration and the number of firms. Thus,we expect protection to be correlated with such variables. This result springs fromBecker’s idea (1983) that the pressure applied by interest groups may not actuallyequal the amount of resources that they devote to such lobbying. We have chosenexogenous friction coefficients as a reduced form operationalization of this idea,but more explicit modeling of such transactions costs would be interesting futurework.

Result 4: Sectors with higher levels of output receive more protection only ifPcb . a. Large industries have more resources to contribute to politicians in orderi

Pto acquire more rents. If, however,c or b is quite low (or both are), meaning thati

contributions are not valued enough or that lobbying costs are high (or both), thenlarger industries may get less protection. In this case, large industries cannotmuster enough contributions to counteract the large amount of consumer surplusthat protection in those industries would wipe out. This result goes against theconventional theoretical wisdom that protection increases with output. Muchempirical work, however, finds the opposite. Maggi and Rodriguez-Clare (2000) isanother formal model which allows for the possibility that protection can bedecreasing with output. Their result arises not from lobbying frictions but fromdistortionary taxation.

Result 5: Import rent seekers want neither free trade nor autarky but someintermediate level of protection that maximizes their rents. To see this, note whatwould happen to the price if all other special interests were removed from the

Pgame, i.e., ifa, b , and e were all set equal to 0. Then,i (L, p),i

1˜ ]]P 5 ,i e(m, p),i

which is the expression for the maximum revenue tariff, or, more generally, themaximum rent trade barrier. Import rent seekers want the price that maximizes

15rents. Unlike with producers, higher prices do not necessarily benefit them.

15This requires thate .1. Even if e is less than or equal to 1 at world prices, as the price(m, p)i (m, p)i

rises,e will exceed 1 at some point, as long as there is an upper limit on the price that consumers(m, p)i

are willing to pay for imports.


Result 6: Sectors with lower levels of imports receive more protection ifIMcb #a; otherwise, the connection between protection and imports isi

16 IMambiguous. Intuitively, if cb #a, then the clout of import rent seekers isi

smaller than that of consumers. In this case, industries with more imports get lessprotection, because more imports means that protection for such an industry willresult in a larger loss of consumer surplus. If, on the other hand, import rent

IMseekers have more clout than consumers so thatcb .a, then the fact that importi

rent seekers want an intermediate level of imports creates an ambiguous con-nection between imports and protection. The model in Maggi and Rodriguez-Clare(2000), like ours, has import rent seekers, so that results like[5 and[6 could bederived from their framework.

Result 7: Sectors with lower elasticities of import demand receive moreprotection. Since more elastic demand leads to greater deadweight loss whenprices are propped up, more elastic import demand leads to less protection. Thisaccords with all GH-type models and is implicit in other frameworks, as Helpman(1995) shows.

4 . Empirics

We now turn to the task of empirically testing the model’s predictions. Section4.1 develops an empirical model based on the theoretical model. Section 4.2briefly describes the data. Section 4.3 presents and analyzes the regression results.

4 .1. An econometric model

In Eq. (2), production, imports, and labor demand are all endogenous with17respect to the level of protection. Goldberg and Maggi (1999) and Trefler (1993)

instrument for the ratio of imports to output using industry-level factor shares. Wefollow them and instrument for these variables in the same way. The presumptionis that factor shares are correlated with imports and output but not with the price.Since labor is endogenous (because it is mobile), we do not include labor shares inthe instruments. We also instrument for labor using non-labor factor shares. For agiven technology, the amount of labor demanded will depend on the amount of

18other factors present.PAs mentioned above, we expectb to be a function of variables that reflect thei

extent of transactions costs in lobbying. The Trefler data set has four such

16The derivation is available upon request.17We assume that both elasticities are constant around equilibrium.18The factor instruments are physical capital, inventories, cropland, pasture land, forest land, coal,

petroleum, and minerals. The data is from Trefler (1993). The main results are robust to the choice ofinstruments, although a shorter list of instruments reduces the standard errors, as one would expect.


variables: 4-firm concentration ratio, geographical concentration, number of firms,and the unionization rate. We do not fully develop a model of the relation betweenthese variables and lobbying transactions costs. Instead, we specify a simple linear

P P P Prelation betweenb and each of the variables:b 5 b 1 b x , wherex is thei i 0 1 ij ij

value of thejth lobbying cost variable in industryi. The data will then tell uswhether, within the theoretical framework developed, these proxies for lobbyingcosts influence protection. Note that we are only trying to proxy for lobbyingeffectiveness. We are not trying to measure directly the amount of lobbying sincethis is not determined in our Nash set-up.

We do not proxy for importer lobbying frictions since we do not have theIMneeded data. Instead, we assume thatb is constant across sectors.i

Thus, our econometric model can be written as:

P P IM´LAB 1x([b 1b LOB ]OUT 1b IMP )2a(OUT 1 IMP )i 0 1 ij i i i i˜ ]]]]]]]]]]]]]]]]]]]P 5i IM

´LAB 1xb (ELAS )(IMP )i i i

1 u (3)i

¯where LAB is the number of workers (L ); OUT is the value of output (y ); IMPi i i i i

¯is the value of imports (m ); LOB is one of the lobbying costs variables (x ); andi ij ijP P IM

a, b , b , b , x, and´ are parameters to be estimated or fixed, corresponding to0 1P P IMa, b , b , b , c, and e , respectively.0 1 L, p

Notice that the elasticity of labor demand has become a parameter (´). This isbecause we do not have data on the industry-level elasticities of labor demand forour sample. We estimate the model using non-linear two-stage least squares

19(NL2S; see Amemiya, 1983).

4 .2. The data

We use mid-1980s US data for 191 SIC 4-digit industries. This choice ofcountry and time period stems from the fact that the endowments data needed forthe instruments are only readily available for the US in 1983 (Trefler, 1993).

20We use new, industry-level measures of protection from 1985. Specifically,

19The dependent variable falls within the [0, 1] interval but is neither truncated nor censored. Thereare no zeros in our data, indicating no truncation. Also, we have dropped no industries because theirprotection level was below zero, indicating no censoring.

20These protection measures are nominal, even though specific capital owners care about effectiveprotection, which could, theoretically, differ substantially from nominal protection. Unfortunately, it ismost difficult to calculate effective protection. The standard measures assume no substitutability amonginputs and thus overstate true effective protection. Some researchers have tried to overcome thisproblem, but there are no reliable estimates of effective protection for the 191 sectors. In the end, itappears to make little difference. According to data from Deardorff and Stern (1984), the correlationbetween nominal and effective protection for 18 2-digit sectors in the US was 0.99. (It was 0.93 for theEU and 0.87 for Japan.)


detailed price data from a sample of six OECD countries were used to constructtariff equivalent price gaps that capture all kinds of barriers to trade. See AppendixA for an overview. Bradford (forthcoming) provides further details and discusseswhy they are probably more trustworthy than other commonly used measures,such as NTB indices and unit value comparisons.

The employment data also come from Trefler and are 1983 US data. These datawere adjusted to account for intra-industry trade. Since some output from almostall industries gets exported, we multiplied the number of workers by the ratio ofnon-exported production to total production, to arrive at an estimate of the numberof import-competing workers for that sector.

The imports and exports data are from the Feenstra data set, and the output dataare from the Bartelsman, Becker, and Gray data set. Both data sets are from 1983and are on the NBER web site. As with employment, we adjusted outputdownward to reflect import-competing production. We did so by subtractingexports from output. These data sets give the value of imports and output atcurrent (protected) prices. We converted these to values at world prices by dividingthe current value by the ad valorem protection rate.

The lobbying cost variables—4-firm concentration, geographical concentration,number of firms, and unionization rates—are all from Trefler and are from 1983,as well.

Like Goldberg and Maggi (1999) and Gawande and Bandyopadhyay (2000), wetake the import demand elasticity data from Shiells et al. (1986). These estimatesare considered to be the best available at the level of disaggregation used in thisempirical analysis. For a few industries, the elasticity estimates were positive, andwe dropped these from the sample. Since the elasticity data are estimated, we usedthe errors-in-variables correction presented in Gawande (1997) to ‘purge’ theelasticities data. Summary statistics for all variables are shown in Table 1.

4 .3. Results

4 .3.1. Estimating the modelThe equation for protection is homogeneous of degree 0 ina, x, ande, meaning

21that we must peg one of these parameters in order to estimate the model.Doubling the units in which we convert dollars to votes (a andx) and doubling

˜the elasticity of labor demand will have no impact onP.We pega and let the regressions generate results fore andx. These latter two

parameters indicate whether jobs and contributions, respectively, play importantroles in the protection game. What value should be chosen fora? Given thate, theelasticity of labor demand, is a parameter that others have estimated, we choosea

such that the estimate fore comes out in a reasonable range. Estimates of labor

21This stems from not normalizing the weight on either consumer surplus or contributions to be 1, asGH and others have done.


Table 1Summary statistics (191 US Industries, 1983)

MEAN MEDIAN MIN MAX S.D.

Regression variables

Wp* 2 p˜ ]]Protection:P 5 0.188 0.138 0.000999 0.627 0.150p*

Employment (thousands): LAB 53.2 24.6 1.59 994 101

Production ($ million, valued at world prices): OUT 4360 1920 42.1 165,000 12,800

Imports ($ million, valued at world prices): IMP 443 144 0.0129 16,224 1410

Elasticity of import demand (corrected):e or ELAS 1.62 1.33 0.221 3.78 0.876m, p

Lobbying cost variables:

4-firm concentration ratio: LOB 0.373 0.350 0.0300 0.940 0.1881

Geographical concentration: LOB 0.691 0.692 0.300 0.996 0.1552

Number of firms (scaled by output): LOB 0.277 0.150 0.00155 2.10 0.3513

Unionization rate: LOB 0.333 0.308 0.0630 0.754 0.1264

Underlying datap*]Tariff equivalent: 1.28 1.16 1.001 2.68 0.295wp

Raw employment (thousands) 57.4 26.0 2.20 999 106

Raw production ($ million, valued at domestic prices) 5640 2690 73.1 183,000 14,600

Raw imports ($ million, valued at domestic prices) 509 187 0.0167 17,500 1510

Exports ($ million) 424 124 0 10,400 1150

Alternative data for robustness checks

NTB/Tariff index: as is 1.15 1.05 1.00 2.00 0.249

NTB/Tariff index: doubled 1.30 1.10 1.00 3.00 0.498

NTB/Tariff index: tripled 1.45 1.15 1.00 4.00 0.747

Uncorrected elasticities 2.00 1.07 0.0420 23.9 2.65

demand elasticities at the industry level in Hammermash (1986) range from 0.20to 1.03. It turns out that, in the regressions below, settinga at 0.001 generates

22point estimates fore that range from 0.30 to 0.77. This value ofa implies that23each $1000 drop in consumer surplus results in the loss of one vote.

P P IMIt also turns out that we also need to peg one ofb , b , b , and x :0 1

Multiplying all of the b ’s by a positive constant and dividingx by the sameIM P P˜constant also would not affectP. We pegb and generate results forb andb .0 1

Unlike with a, however, there is no empirical work that sheds light on reasonableIM IMvalues forb . Thus, the equations were estimated 10 times, withb set equal to

IMall multiples of 0.1 ranging up to 1. (Recall thatb is bounded by 0 and 1.) Thesign and significance of all three estimated parameters are robust to all choices of

22Increasinga by a factor of 10 increasese by a factor of 10. Thus, settinga equal to 0.01 wouldlead to estimates fore between 3.0 and 7.7. Similarly, a 10-fold reduction ina would decrease theeestimates by a factor of 10.

23Changinga changes only the point estimate fore, not affecting the substantive results below.


IM IMb . We will focus on the results forb 50.9, because it seems likely that,despite some transactions costs associated with importer lobbying, the majority of

IMcontributions from importers are not dissipated. Picking lower values forb ,meaning higher transactions costs, would unequivocally strengthen all the conclu-sions below. We discuss the implications of assuming no transactions costs

IM(b 51) in footnote 24.

4 .3.2. Results for the unrestricted modelThe first four columns of Table 2 show the results of estimating the equation

with each of the four lobbying cost variables. In each case, the point estimate fore

is significantly positive at the 5% level or better, indicating that, ceteris paribus,industries with a greater number of workers do receive more protection. Thus,regressions stemming from a structural model that accounts for job concernsconfirm the widely held belief that workforce size has a positive independentinfluence on protection.

The results forx provide unambiguous evidence that lobbying contributionsinfluence protection, just as votes do. The point estimate is significantly positive in

Table 2Non-linear 2-stage least squares estimation of the general model

Parameter Lobbying cost variable used

4-Firm Geographical Number Union- Noneconcen- concen- of firms izationtration tration rateEstimate Estimate Estimate Estimate Estimate(t-stat.) (t-stat.) (t-stat.) (t-stat.) (t-stat.)

e 0.351** 0.433** 0.562*** 0.558** 0.299***(1.90) (2.05) (2.47) (1.80) (2.45)

[x 0.00117*** 0.00116*** 0.00115*** 0.00115*** 0.00117***

(4.49) (4.04) (3.78) (3.13) (5.12)P [[

b 0.857*** 0.860*** 0.868*** 0.868*** 0.854***0

(6.41) (4.95) (4.36) (7.56) (5.99)P

b 0.00120 0.00227 20.00581** 0.009361

(0.496) (1.08) (21.89) (1.02)Pseudo-SSR 1.4307 1.4767 1.2333 1.3302 1.3442Variance of the residuals 0.04641 0.04757 0.04272 0.05090 0.046

Wp* 2 p IM˜ ]]Dependent variable:P 5 . Number of observations: 191.a is set equal to 0.001.b is setp*

equal to 0.9. Standard errors are robust White.*,**,*** Significant at the 10%, 5%, or 1% level, respectively. All are 1-tailed tests.[ Asterisks andt-stats refer to whetherx .a. If this is so, then contributions get more weight than

consumer surplus.[[ P PAsterisks andt-stats refer to whetherb ,1. Given the results forb , this implies that there are0 1

transactions costs in lobbying.


all cases. Also, the estimate forx is significantly greater than the pegged value ofa in each case. This implies that contribution dollars are more valuable to policy

24makers than consumer surplus dollars. The point estimate ofx indicates thatpoliticians find contributions to be about 15% more valuable in terms of how manyvotes they can buy than is consumer surplus. This estimate seems more reasonablethan that of Goldberg and Maggi (1999), which implied a value of about 2%.

The results for the lobbying parameters indicate significant transactions costsP Passociated with lobbying:b is significantly less than 1, and the estimate forb is0 1

P Pnever significant and large in absolute value. Thus,b 1b LOB is always0 1 j

significantly less than 1 for all values of the four LOB variables. This casts doubtj

on the standard assumption in lobbying models that contributions are frictionless.Of the four lobbying cost variables, only the number of firms is significant (at the5% level). The point estimate is so low, though, that the estimated impact of thisvariable on lobbying transactions costs is quite small. For instance, halving thenumber of firms from its mean would only reduce transactions costs by about

250.6%. Thus, the one political variable that is statistically significant has littleeconomic significance.

Given these weak results for the lobbying variables, we re-estimated theP Pequation without one. Thus, we assumed thatb simply equals a constant,b .i 0

The result of this model is in the fifth column of Table 2. As with all the otherP Pregressions,´,x, and b are significantly positive. Notice also thatb is0 0

IMsignificantly less thanb , which has been set at 0.9. This implies thattransactions costs for producer lobbies exceed those for importers. The estimate of0.854 implies that producer transactions costs are 14.4% (120.854), given theassumption that import transactions costs are 10% (120.1).

Is protection increasing in output? The GH model says unequivocally that it is,while Goldberg and Maggi (1999) and Gawande and Bandyopadhyay (2000) have

26found empirical evidence for this proposition. Recall from Result 5 above that, in

24 IM IMThis is the only conclusion that does not hold for all calibrations ofb . In particular, ifb 51,thenx is not significantly greater thana when using geographical concentration, number of firms, or

IMunionization. The reason that the result weakens when we setb at a higher level is that imputingmore influence to importers for a given amount of protection means that producers’ contributions needto receive less weight in order to best estimate the model with the given data. When using 4-firmconcentration,x is significantly greater thana at the 10% level, and, with no lobbying costs variables,x is significantly greater thana at the 5% level.

25Number of firms is measured in firms per one million dollars, and the mean for this variable isP P Pˆ0.28. Since the estimate for the producer lobbying coefficient is given byb 5b 1b NUM (wherei 0 1 i

PNUM is the number of firms), the estimate forb evaluated at the mean value of 0.28 isi iP0.8681(20.00581)(0.28)50.8664. Cutting the mean in half increases the estimate forb to 0.8681iP(20.00581)(0.14)50.8672. Thus, the estimate for transactions costs, which equals 12 b , is reducedi

from 0.1336 to 0.1328, a 0.6% reduction.26 IMThese papers also imply that protection decreases with imports, but, since we do not estimateb ,

we cannot test this proposition. Hillman (1982) shows that output and protection can move in oppositedirections when world prices change.


P Pour framework, protection increases in output ifcb . a. Thus, if x(b 1i 0P

b LOB ) is significantly greater thana, that constitutes empirical evidence that1 ijP Pprotection increases in output. Imposing the restrictionx(b 1b LOB )5a does0 1 ij

not significantly worsen the fit, for all five specifications. Thus, we have noevidence that protection is increasing in output. While policy makers do valuecontributions more than consumer surplus, the presence of transactions costsapparently diminishes producer influence to the point where contributions fromlarge producers cannot outweigh the large consumer costs inflicted by protectingthose large producers.

4 .3.3. A restricted model without import rent seekersOur modeling has sought to account for possible lobbying by import rent

seekers, since casual observation suggests that this occurs. It turns out, though, thatwe can ignore import rent seeking and assume lump-sum rebating because doingso does not significantly worsen the fit.

As discussed above, imposing lump-sum rebating changes the prediction to:

P ¯e L 1 (cb 2 a)yL, p i i i˜ ]]]]]]P 5 (4)i ¯e L 1 ae mL, p i (m, p),i i

The econometric equation becomes:

P P´LAB 1 [x(b 1b LOB )2a ]OUTi 0 1 ij i˜ ]]]]]]]]]]]P 5 1 u (5)i i´LAB 1a(ELAS )(IMP )i i i

Once again, we need to peg eitherx or one of thebs, as well asa. Thus, we setP 27

b equal to the point estimates from the previous results in Table 2.0

We report the results for this restricted model in Table 3. Quasi-likelihood ratiotests (Gallant and Jorgenson, 1979) for each of the five pairs of models neverreject the restricted model, even at the 10% level. In this simpler model, theestimates fore are strongly significant in all cases (at the 1% level), confirmingthat jobs matter. As before, the estimate forx is significantly positive and issignificantly greater than the calibrated value ofa, confirming that politicians doindeed value contributions more than wealth spread across the populace. We alsoget stronger results for the lobbying cost variables: geographical concentration andunionization rate join number of firms in being significant at the 5% level, with thepredicted signs. As before, though, the point estimates are so small that thesevariables have limited economic significance.

Since it has been standard procedure to use NTB indices to measure protection,let us compare our results with such empirical studies. Gawande andBandyopadhyay (2000), Gawande (1998), and Lee and Swagel (1997), all find a

27 PThe results are robust to the choice ofb .0


Table 3Non-linear two-stage least squares estimation of the model without import rent seekers

Parameter Lobbying cost variable used

4-Firm Geographical Number of firms Unionization Noneconcentration concentration rateEstimate Estimate Estimate Estimate Estimate(t-stat.) (t-stat.) (t-stat.) (t-stat.) (t-stat.)

e 0.616*** 0.712*** 0.754*** 0.769*** 0.477***(4.25) (4.33) (4.59) (3.94) (4.38)

[x 0.00116*** 0.00116*** 0.00115*** 0.00114*** 0.00117***

(72.2) (49.4) (238) (28.9) (635)P

b 0.00364 0.00482** 20.00768*** 0.0143**1

(1.25) (1.88) (22.51) (1.69)

Pseudo-SSR 1.4980 1.5185 1.2685 1.3553 1.4705Reject this NO NO NO NO NOmodel for the unrestricted one? (5% significance)

Wp* 2 p P˜ ]]Dependent variable:P 5 . Number of observations: 191.a is set equal to 0.001.b is set0p*

equal to the point estimates shown in Table 2. Standard errors are robust White.*,**,*** Significant at the 10%, 5%, or 1% level, respectively. All are 1-tailed tests.[ Asterisks andt-stats refer to whetherx .a. If this is so, then contributions get more weight than

consumer surplus.

positive significant relation between workforce size and NTB protection, whileGoldberg and Maggi (1999) and Trefler (1993) find a positive but insignificantrelation. So, using a new model and data, we confirm the NTB literature’spresumption that protection increases with the number of jobs at stake. As for thelobbying variables, though, our results diverge more sharply from the literature. Inparticular, our significant, positive results on geographical concentration andunionization are unique. Most papers find no significant relation, while Ray (1981)and Gawande and Bandyopadhyay (2000) get significant, negative coefficients ongeographical concentration and unionization, respectively. We find robust evidencethat protection decreases with number of firms; only Trefler (1993) examines this,and he finds no significant correlation. Finally, while we find no connectionbetween four-firm concentration and protection, Marvel and Ray (1983) and Ray(1981) find a negative relation, while Trefler (1993) and Gawande (1998) find apositive relation.

4 .3.4. Robustness checksWe checked the robustness of these results in five ways. (1) We used an

alternative protection measure based on the NTB indices used in many otherpapers. (2) We tried a quadratic, instead of linear, functional form for lobbyingcosts. (3) We experimented with different sets of instruments. (4) We estimated the


system using limited information maximum likelihood (LIML), instead of NL2S.(5) We used uncorrected elasticities, instead of those corrected for errors-in-variables.

NTB coverage ratios are fraught with problems, but, as a check on whether ournew measures drive the results (which still would not invalidate our results, if ourmeasures are superior), we estimated the equation using the NTB measures. These

28data are censored at zero (53% of the observations ), however, and trying toincorporate some kind of Tobit correction into our non-linear framework liesbeyond this paper. So, zeros in the NTB data were replaced with tariff rates.Nearly all of these rates are small, so the effect of this modification was to replacezeros with small numbers. Another problem with the coverage ratios is that theyare bounded above at 1, which, in effect, limits the maximum protection rate to100%. We thus followed Goldberg and Maggi and considered three different cases:using the NTB data as is; doubling the data, so that the maximum tariff equivalentbecomes 200%; and tripling the data, so that the maximum becomes 300%.

Using the NTB/ tariff data as is, with no upward scaling, workforce size is onlysignificant in the unrestricted model when the number of firms is the lobbying costproxy. In the restricted model, workforce size is significant in all cases exceptwhen using unionization rate. We reject this streamlined model, however, whenusing 4-firm concentration, geographical concentration, or no lobbying costvariable. So, in the end, workforce size is only significant when using number offirms. Thus, using the NTB/ tariff data as is gives mixed results, but this isprecisely the data that is most dubious.

Doubling the NTB/ tariff data yields more robust results. Tripling the dataimproves the results even more. When the data is doubled, all the results arerobust, with the following exceptions. We reject the streamlined model (no importrent seekers) when using geographical concentration or no lobbying cost proxy.Workforce size is not significant under the geographical concentration spe-cification. Also, among the political variables, neither geographical concentrationnor unionization rate is significant (at the 5% level). When the data are tripled, wereject the streamlined model when no lobbying cost proxy is used, but, otherwise,all the results carry through.

Using a quadratic specification for the lobbying cost variables does not affectour conclusions that workforce size matters and that assuming lump-sum rebatingis justified. In only one case out of eight (four regressions each for the full modeland the restricted model), did the quadratic term come out significant at the 5%level (restricted model, 4-firm concentration). The conclusions regarding thesepolitical variables were largely unchanged.

We also experimented with different instruments and found no change in themain results. Using no instruments, however, did undermine the results. We need

28The fact that such a high fraction of industries show zero protection, when each one of them haspositive tariffs, casts further doubt on the usefulness of this NTB data.


to correct for endogeneity. We also estimated the system using LIML and got verysimilar results. Using LIML was problematic, however, because we rejectednormality of the residuals. Dropping outliers so that normality was not rejectedalso produced very similar results. Finally, we ran the regressions using un-

29corrected elasticities, and this did not affect any of the main results.

5 . Conclusion

We have developed a protection model that incorporates the incentives ofpoliticians to win support through creating or preserving jobs, as well as throughcampaign spending. In fact, we may think of job preservation as a direct way, and

30campaign spending as an indirect way, of winning votes. The theory predicts thatprotection for an industry will be increasing in the number of workers in thatindustry and that there is no necessary connection between protection and eitheroutput or imports. In particular, if industries cannot organize well enough,protection may be decreasing in output. Also, the impacts of both imports andoutput on protection are conditioned by workers’ ability to influence policy makersthrough voting. The result also implies that protection is decreasing in lobbyingcosts.

The empirical results provide strong evidence that protection increases with the31number of workers in that industry. We also do not find that protection increases

with output. We have provided some evidence that variables that affect lobbying(geographical concentration, unionization rates, and, especially, the number offirms) affect protection levels. We find that there are significant transactions costsin lobbying: contributions are not frictionless transfers. The empirics also implythat one does not need to replace the lump-sum rebating assumption with a formalspecification of the role that import rent seekers play in the protection game.Finally, this work implies that politicians place about 15% more weight on a dollarof campaign contributions than on a dollar of consumer surplus.

Our framework can be modified to test other hypotheses. For instance, to testwhether there is a sympathy motive for protection, we could letc depend on skill

29 2 2The adjustedR for the regressions range from 0.02 to 0.05. Using adjustedR to measuregoodness-of-fit, however, is problematic because 2-stage least squares does not minimize the sum ofthe squared errors. It minimizes the errors after projecting them onto the instruments matrix. Theminimized value of this objective function is the pseudo-SSR that we have reported.

30Goldberg and Maggi (1999) also find that ‘‘there is some evidence that factors linked tounemployment may affect protection through channels different than the ones suggested by the G–Htheory.’’ They go on to say: ‘‘This suggests that it might be fruitful to . . . allow for . . .unemployment and examine the impact that this has . . . .’’ We have done precisely this.

31This contrasts with the claim made in Grossman and Helpman (1994) that ‘‘our formula suggeststhat only two variables (the elasticity of import demand and the ratio of domestic output to imports)should explain the cross-industry variation in protection levels.’’


Plevels or industry growth, just as we have letb depend on lobbying costi

variables. It would also be useful to see whether this paper’s conclusions hold withmore recent data or data from other countries or both. On the theoretical side,modeling the labor market and unemployment more explicitly within a politicaleconomy model such as this may prove quite fruitful. For instance, such modelingwould make it possible to explicitly incorporate the influence of wages onprotection into the analysis. While much work remains to be done, we hope thatthis paper has shed light, and will stimulate further research, on the complexconnections among protection, jobs, voting, and lobbying.

A cknowledgements

I thank Donald Davis, Jeffry Frieden, Gene Grossman, Elhanan Helpman, DaleJorgenson, Hiro Lee, Aaron Tornell, and David Weinstein for their help. Specialthanks go to Arye Hillman, Val Lambson, Robert Lawrence, and three anonymousreferees. I thank Daniel Trefler for sharing his data and Kishore Gawande for datahelp.Various members of the USC and BYU economics faculty have made helpfulcomments on this work. I have also been helped by comments from participants inthe International Economics seminar and International Economics and PoliticalEconomy workshops at Harvard University. Any errors are mine alone.

A ppendix A. New protection measures

Nations protect their industries in many ways. Aside from standard instruments,health and safety standards, labeling laws, certification requirements, biasedgovernment procurement, burdensome customs procedures, and threats can allrestrict trade. Thus, measuring protection is not straightforward. Here, we brieflydescribe our method. See Bradford (forthcoming) for more details and a discussionof other methods.

We infer protection levels from price gaps. The philosophy is that internationalbarriers to arbitrage should be considered barriers to trade. This implies that, afteraccounting for shipping costs between countries, a price gap for equivalent goodsin different countries indicates protection (even if policies not explicitly designedto impede trade are responsible).

The underlying data come from the OECD, which collects carefully matchedretail prices in order to calculate Purchasing Power Parity estimates. The datacover 124 traded final goods categories (103 household goods and 21 capitalgoods) and are from 1985. All prices were converted to US dollars using 1985market exchange rates.

We need to convert these consumer prices to producer prices to measure


32protection. We did so using data on distribution margins. We have such data, forsix countries—Australia, Canada, Japan, the Netherlands, the UK, and the US—which comprise our sample. Thus,

cpijp ]]p 5 ,ij 11mij

p cwherep is the producer price of goodi in country j, p is the consumer priceij ij33(from the OECD data), andm is the margin.ij

We infer protection by comparing the domestic producer price to the landed(world) price of the foreign good. We infer the world price by using data on exportmargins and international transport costs, as follows. By adding the export marginsto the producer prices, we calculated the export price for each product in each

e p ecountry. This price is given byp 5 p (11 em ), wherep is the export price ofij ij ij ij

good i for country j, and em is the export margin of goodi for country j. Theij

common world price was then found by adding the international transport cost towthe lowest export price in the samplep 5 p (11 tm ), where, for each goodi,i iM i

w e ep is the world price,p 5min(p , . . . , p ) is the minimum of the 6 exporti iM i1 i6

prices, andtm is the international transport margin.i

We then used the ratio of each country’s producer price to the world price as apreliminary protection measure:

ppij]ppr 5 .ij Wpi

These measures will be biased downward if each country has substantial barriersto imports. For such goods, the calculated world price exceeds the true worldprice. At the same time, if just one of the countries has no barriers in that good,then these measures will not be biased downward, since, in this case, prices in thefree trading country will approximate world prices. With fairly free traders such asAustralia, Canada, and the US in the sample, we believe that the low priceapproximates the world price the great majority of the time. Nevertheless, we usedata on trade taxes to correct, at least partially, for the possible downward bias.The final measure of protection ispr 5max(ppr , 11 tar ), where tar is theij ij ij ij

tariff rate for goodi in country j. We simply use the fact that trade taxes provide alower bound on protection. After this correction, these measures will only bebiased downward if all countries in the sample have non-tariff barriers against therest of the world.

32These cover retail trade, wholesale trade, transportation costs and taxes collected by retailers.33m is the fraction by which the consumer price exceeds the producer price. Thus, if the consumerij

price is 25% higher, thenm is 0.25.ij


R eferences

Amemiya, T., 1983. Non-linear regression models. In: Griliches, Z., Intriligator, M. (Eds.). Handbookof Econometrics, Vol. 1. North-Holland, Amsterdam.

Baldwin, R.E., 1987. Politically realistic objective functions and trade policy. Economics Letters 24,287–290.

Baldwin, R.E., 1984. Trade policies in developed countries. In: Jones, R.W., Kenen, P.B. (Eds.).Handbook of International Economics, Vol. 1. North-Holland, Amsterdam.

Becker, G.S., 1983. A theory of competition among pressure groups for political influence. QuarterlyJournal of Economics 98, 371–400.

Bhagwati, J.N., Srinivasan, T.N., 1980. Revenue seeking: a generalization of the theory of tariffs.Journal of Political Economy 88, 1069–1087.

Bradford, S.C., forthcoming. Paying the price: final goods protection in OECD countries. Review ofEconomics and Statistics.

Brecher, R.A., 1974. Minimum wage rates and the pure theory of international trade. Quarterly Journalof Economics 88, 98–116.

Cassing, J.H., Hillman, A.L., 1986. Shifting comparative advantage and senescent industry collapse.American Economic Review 76, 516–523.

Davidson, C., Martin, L., Matusz, S., 1999. Trade and search generated unemployment. Journal ofInternational Economics 48, 271–299.

Davis, D.R., 1998. Does European unemployment prop up American wages: national labor markets andglobal trade. American Economic Review 88, 478–494.

Deardorff, A.V., Stern, R.M., 1984. The effects of the Tokyo round and the structure of protection. In:Baldwin, R.E., Krueger, A.O. (Eds.), The Structure and Evolution of Recent US Trade Policy.University of Chicago Press, Chicago.

Gallant, A.R., Jorgenson, D.W., 1979. Statistical inference for a system of simultaneous, non-linear,implicit equations in the context of instrumental variables estimation. Journal of Econometrics 11,275–302.

Gawande, K., 1997. Generated regressors in linear and non-linear models. Economics Letters 54,119–126.

Gawande, K., 1998. Comparing theories of endogenous protection: Bayesian comparison of Tobitmodels using Gibbs sampling output. Review of Economics and Statistics 80, 128–140.

Gawande, K., Bandyopadhyay, U., 2000. Is protection for sale? Evidence on the Grossman–Helpmantheory of endogenous protection. Review of Economics and Statistics 82, 139–152.

Goldberg, P.K., Maggi, G., 1999. Protection for sale: an empirical investigation. American EconomicReview 89, 1135–1155.

Grossman, G.M., Helpman, E., 1994. Protection for sale. American Economic Review 84, 833–850.Hammermash, D.S., 1986. The demand for labor in the long run. In: Ashenfelter, O., Layard, R. (Eds.),

Handbook of Labor Economics. North-Holland, Amsterdam.Helpman, E., 1995. Politics and trade policy. NBER Working Paper[5309.Hillman, A.L., 1982. Declining industries and political-support protectionist motives. American

Economic Review 72, 1180–1187.Krueger, A.O., 1974. The political economy of the rent-seeking society. American Economic Review

64, 291–303.Lee, J.-W., Swagel, P., 1997. Trade barriers and trade flows across countries and industries. Review of

Economics and Statistics 79, 372–382.Maggi, G., Rodriguez-Clare, A., 2000. Import penetration and the politics of trade protection. Journal

of International Economics 51, 287–304.Marvel, H.P., Ray, E.J., 1983. The Kennedy round: evidence on the regulation of international trade in

the United States. American Economic Review 73, 190–197.


Mitra, D., 2000. On the endogenous choice between protection and promotion. Economics and Politics,12.

Peltzman, S., 1976. Toward a more general theory of regulation. Journal of Law and Economics 19,211–248.

Potters, J., Sloof, R., van Winden, F., 1997. Campaign expenditures, contributions, and directendorsements: the strategic use of information and money to influence voter behavior. EuropeanJournal of Political Economy 13, 1–31.

Revenga, A.L., 1992. Exporting jobs? The impact of import competition on employment and wages inUS manufacturing. Quarterly Journal of Economics 107, 255–284.

Ray, E.J., 1981. The determinants of tariff and nontariff trade restrictions in the United States. Journalof Political Economy 89, 105–121.

Richardson, M., 1993. Endogenous protection and trade diversion. Journal of International Economics34, 309–324.

Rodrik, D., 1986. Tariffs, subsidies, and welfare with endogenous policy. Journal of InternationalEconomics 21, 285–296.

Rodrik, D., 1995. What does the political economy literature on trade policy (not) tell us that we oughtto know? In: Grossman, G.M., Rogoff, K. (Eds.). Handbook of International Economics, Vol. 3.North-Holland, Amsterdam.

Shiells, C.R., Stern, R.M., Deardorff, A.V., 1986. Estimates of the elasticities of substitution betweenimports and home goods for the United States. Welfwirtschlaftiches Archiv, 497–519.

Trefler, D., 1993. Trade liberalization and the theory of endogenous protection. Journal of PoliticalEconomy 101, 138–160.

Protection and jobs: explaining the structure of trade barriers across industries

Documents

Transcript of Protection and jobs: explaining the structure of trade barriers across industries