Trade, Turnover, and Tithing 1

Transcription

1 Trade, Turnover, and Tthng 1 Chrstopher Magee *, Carl Davdson **, and Steven J. Matusz ** Revsed October 2002 Abstract Ths paper develops a poltcal economy model to test the proposton that the effect of nternatonal trade on the dstrbuton of ncome s systematcally related to the extent of labor-market turnover. The model reveals that trade benefts the abundant factor and hurts the scarce factor regardless of where those factors are employed f turnover s hgh, but benefts factors n exportng ndustres and harms those n mport-competng ndustres when turnover s low. We test these predctons usng data on campagn contrbutons gven by ndustry-specfc poltcal acton commttees to congressonal representatves who subsequently voted for or aganst trade-lberalzng legslaton. We fnd emprcal evdence n favor of the model s predctons. 1 We thank Danel Hamermesh, Steve Magee, Keth Maskus, two anonymous referees and Jonathan Eaton for helpful comments on earler versons of ths paper. * Department of Economcs; Bucknell Unversty; Lewsburg, PA 17837; e-mal: cmagee@bucknell.edu. ** Department of Economcs; Mchgan State Unversty; East Lansng, MI 48824; emal: Davdson davdso4@msu.edu; Matusz matusz@msu.edu.

2 1. Introducton One of the man themes of nternatonal economcs s that trade relatonshps have profound mplcatons for the domestc dstrbuton of ncome. Whle there s no queston that changes n trade polcy create wnners and losers, the dentty of the wnners and losers largely depends on the degree to whch factors of producton can move between sectors. The two polar extremes are emboded n the n the Heckscher-Ohln-Samuelson (HOS) model, where factors are assumed to be perfectly moble between sectors, and the Rcardo-Vner model (a.k.a. Specfc Factors model) where some factors of producton are assumed to be completely mmoble. One of the fundamental results of the HOS model s the Stolper-Samuelson theorem, whch demonstrates that the economy s abundant factor benefts from trade lberalzaton, even f employed n the declnng mport-competng sector, and the economy s scarce factor s harmed by trade lberalzaton, even f employed n the expandng export sector. By contrast, analyss of the Rcardo-Vner model reveals that factors that are trapped n the mport-competng sector are harmed by trade reform regardless of relatve abundance, whle factors fortunate enough to be ted to the export sector beneft. 2 Attempts to test these two theores have met wth lmted success. Magee (1980) tested ther predctons by explotng the fact that they have dfferent mplcatons for lobbyng actvty n the Unted States. The Stolper-Samuelson theorem predcts that captal, an abundant factor n the U.S., should gan from lberalzaton whle low-sklled labor, a scarce factor n the U.S., should lose. Consequently, low-sklled labor and captal should have polar opposte vews wth regard to trade polcy even when both are 2 The welfare mpact of trade reform on moble factors s ambguous, dependng on ther preferences. 1

3 employed n the same ndustry. On the other hand, f captal and labor are both ted to ther sector, then the Rcardo-Vner model predcts that captal and labor groups wthn each ndustry should share the same vew on trade polcy ssues. Magee showed that lobbyng behavor on the 1973 Trade Reform Act was consstent wth the Stolper- Samuelson theorem n only 2 of 21 ndustres. The Rcardo-Vner model fared much better. In 19 ndustres labor and captal lobbed for the same type of trade polcy. Irwn (1996) also found evdence favorng the predctons of the specfc factors model n the 1923 Brtsh electon for Parlament, where the man ssue was whether or not to adopt tarff protecton. He concluded that the man determnants of votng behavor n each dstrct were the ndustry and occupatonal characterstc of the county. Other research has tended to support the Stolper-Samuelson theorem. For example, Rogowsk (1987) argues that the theorem can be used to explan the lobbyng coaltons that have formed n many developed countres snce Beauleu (1998, 2001) and Balstrer (1997) fnd support for HOS n the votng preferences of Canadans wth respect to NAFTA, GATT, and the Canadan-US Free Trade Agreement of Scheve and Slaughter (1998) offer smlar evdence based on the vew of trade polcy held by Amercans. Fnally, Beauleu and Magee (2001) fnd that both the ndustry and the factor that PACs represented nfluenced the pattern of ther contrbutons to supporters of NAFTA and GATT n the US. The factor that the group represents appears to be more mportant than the ndustry, however, partcularly for captal. 3 3 Beauleu and Magee (2001) argue that snce the Magee (1980) and Irwn (1996) studes focus on votes that could have been overturned wthn a decade, what they are pckng up s the voters short-term concerns. In contrast, the other studes focus more broadly on overall vews of trade polcy that are lkely to be governed by long-run concerns. They conclude, as do Leamer and Levnsohn (1995) that ths group of results taken as a whole ndcates that the HO model does a good job explanng the lnk between trade and factor rewards n the long-run whle the Rcardo-Vner model s more approprate for the short-run. 2

4 The fact that the evdence s so mxed should not be too surprsng. These two models embody the two most extreme assumptons that can be made about factor moblty. In realty, factors are quas-fxed, movng between sectors n response to changes n factor rewards. Recognzng ths, a number of authors n the 1970s, most notably Mayer (1974), Mussa (1974, 1978), and Neary (1978), developed models wth mperfect factor moblty n whch both short-run specfc factors and long-run Heckscher-Ohln labor markets are relevant for worker preferences concernng trade polcy. Lobbyng behavor then depends on factors that determne whch tme horzon s most mportant to each factor n each ndustry (e.g., tme preference and age profle). Casual observaton also suggests that the two models should have dffculty explanng the movement of wages, partcularly those of low-wage workers, whose labor market experence bears lttle resemblance to that modeled n the HOS or RV settngs. These workers typcally cycle between perods of employment and unemployment, often fndng t dffcult to obtan new jobs quckly. Moreover, these workers frequently encounter sgnfcant adjustment costs when swtchng sectors due to search costs, the costs of retranng and the non-trval amount of tme they may spend unemployed. Ths experence contrasts wth a fundamental assumpton emboded n the HOS and RV models that factors are fully employed at all tmes. The models developed by Mayer, Mussa, and Neary also mantan the assumpton of full employment and gnore the adjustment costs that come hand-n-hand wth resource allocaton. 4 Snce recent papers by Jacobson, LaLonde and Sullvan (1993a, 1993b), Trefler (2001), Kletzer (2001) and Davdson and Matusz (2001b) suggest that these adjustment costs may be sgnfcant, t s 4 An excepton s Mussa (1978) n whch adjustment costs assocated wth changng the stock of captal n a gven sector are taken nto account. Labor faces no adjustment costs when swtchng sectors. 3

5 mportant to take them nto account when assessng the lnk between trade and the dstrbuton of ncome. Buldng on the tradton establshed by Mayer, Mussa, and Neary; Davdson, Martn, and Matusz (1999) recently extended the HOS model to allow for labor market turnover and showed that many of the model s canoncal results were altered. In ther model, labor and captal are treated as quas-fxed n the sense that dsplaced factors must search for new producton opportuntes once a job dssolves. Thus factors face employment rsk and the rate at whch jobs are created and destroyed plays a role n determnng the allocaton of resources. In such a settng, any change n trade patterns creates unemployment and generates adjustment costs. The result s a more nuanced vew of the lnk between trade and the dstrbuton of ncome. The pcture that emerges from the DMM model has features that derve from both the HOS and RV models. In partcular, when labor market turnover s modeled, the mpact of trade lberalzaton on factor rewards s made up of a convex combnaton of Stolper-Samuelson and Rcardo-Vner forces. Stolper-Samuelson forces domnate n sectors wth hgh labor market turnover, whle the Rcardo-Vner forces domnate n sectors that are characterzed by low turnover. Intutvely, f jobs are dffcult to fnd but durable once obtaned (that s, f turnover s low), then a worker s attachment to the sector wll be strong. In ths case, the dffculty of fndng reemployment and the durablty of current employment creates an attachment that makes workers act as f they have sector specfc sklls. On the other hand, f a sector s characterzed by hgh turnover n the sense that jobs are easy to fnd or do not last long once secured, then the worker s attachment to that sector wll be weak. In ths case, the return to those workers wll vary 4

6 wth trade polcy as f they were perfectly moble across sectors. One of the man conclusons of the DMM model s that the lnk between trade and the dstrbuton of ncome should be dependent on job turnover, whch vares wdely across ndustres. 5 In ths paper, we test the lnk between ndustry turnover and trade preferences. 6 Toward that end, we frst develop a smple model of trade wth factor market turnover n the sprt of the DMM model. In contrast to DMM, who focus on the steady-state propertes of ther model, we examne the tme path of factor rewards along the adjustment path that takes the economy from ts orgnal short-run equlbrum to ts new steady state. We then add on a model of the poltcal process n whch lobbyng coaltons provde contrbutons to canddates n order to nfluence electon outcomes and trade polcy decsons made by representatves. 7 In our emprcal work, we combne data on PAC contrbutons wth the Davs, Haltwanger, and Schuh (1996) data on job creaton and job destructon n US manufacturng ndustres to examne how the pattern of campagn contrbutons vares across ndustres and factors of producton. We use the data to undertake both non-parametrc and parametrc tests of propostons that are suggested by our model. Both the model and the emprcal work suggest that labor 5 One possble way to vew ths result s that when the Mayer, Mussa, and Neary approach s extended to allow for employment rsk the dfference between the short-run and long run s blurred and the lnk between trade and the dstrbuton of ncome becomes more complex. 6 Recent emprcal work by Goldberg and Magg (1999) estmates Grossman and Helpman s (1994) theoretcal model relatng ndustry characterstcs to the cross-ndustry structure of tarffs. In that analyss, lobbyng s an ntermedate step n the chan of causaton. Our focus s more narrow, usng observed lobbyng actvty to nfer preferences over trade polces that are held by nterest groups. 7 Davdson, Martn, and Matusz (1999) make no attempt to examne the nteracton between trade-polcy preferences and poltcal nsttutons n order to predct lobbyng behavor. As Mayer (1984) and others have shown, however, dfferent poltcal nsttutons can lead to very dfferent poltcal behavor for a gven set of trade-polcy preferences. Thus, as Rodrk (1995) emphaszes, poltcal behavor s the endogenous outcome of the nteracton between underlyng trade-polcy preferences and exstng poltcal nsttutons. 5

7 market turnover plays an mportant role n the determnaton of lobbyng actvty amed at nfluencng trade polcy. The remander of the paper dvdes nto four sectons. In the next secton, we provde a general equlbrum model of trade wth labor market turnover n whch nterest groups donate funds n order to nfluence trade polcy. Secton 3 descrbes the data, and the emprcal tests are presented n secton 4. We conclude the paper n secton The Model In ths secton we modfy the standard two-sector, two-factor general equlbrum model to ncorporate factor-market turnover and show how the lnk between changes n output prces and changes n factor prces depends on turnover. In partcular, the effect of a prce change on the welfare of each factor wll resemble a weghted average of Stolper-Samuelson effects and Rcardo-Vner effects, wth the weghts dependng on the ndustry s turnover rates. We then provde a model of nterest group lobbyng to complete the lnk between labor market structure, trade preferences and poltcal outcomes. The underlyng trade model s a smplfed verson of the one presented n Davdson, Martn, and Matusz (1999). The model of nterest group lobbyng bulds on earler work by Magee, Brock, and Young (1989) and Grossman and Helpman (1994). A. Trade, Turnover and Income We assume that there are two goods, an exportable (X) and an mport-competng good (M). Both goods are produced accordng to a constant-returns-to-scale technology 6

8 usng captal (K) and labor (L). Both captal and labor are nfntely lved. Each factor earns the value of ts margnal product when employed and nothng when unemployed. 8 Factor markets are dynamc n the sense that captal and labor cycle through perods of employment and unemployment. We assume that jobs are created and destroyed accordng to Posson processes, where s the job break-up rate and λ s the job acquston rate n sector. To keep the analyss smple, we assume that the break-up and job acquston rates wthn a sector are the same for both captal and labor. b 9 We begn by descrbng the envronment that confronts workers. Let w () t be the wage pad n sector at tme t. Consder an unemployed worker who s searchng for a job n ths sector and assume that the economy s n a steady state so that w () t = w for all t. Let ρ represent the subjectve dscount rate. The dynamc programmng soluton reveals that the searcher s expected ncome dscounted over the nfnte future, defned as V LS ( w ), s proportonal to a weghted average of the wage earned when employed and zero, the ncome earned when unemployed: 1 λ ρ ρ + λ (1) VLS ( w ) = w + b Smlarly, the expected lfetme ncome of a worker who s employed n sector, defned. as V LE ( w ), s proportonal to a weghted average of the current wage and the ncome earned upon separaton (zero). In ths case, 8 For ease of exposton, we refer to captal as well as labor as ether beng employed or unemployed, rather than rates of capacty utlzaton, dle captal, and so on. We also use the word job to refer to a unt of employed captal as well as to a unt of employed labor. 9 In our emprcal analyss, we capture turnover by the use of job destructon rates. Snce we do not have data on captal turnover, we assume that the job destructon rate assocated wth a gven sector apples to both the captal and labor used wthn that sector. 7

9 (2) VLE ( w ) w = 1 ρ + λ ρ ρ + λ + b The dfference between (1) and (2) s that because of dscountng, the worker places more weght on the ncome that s currently beng earned and less weght on the future. Unemployed workers search n the sector that offers the hghest value of dscounted ncome. Dversfed producton, where unemployed workers search n both sectors, can only be sustaned n a steady-state f V. ( X ) LS ( M M w = V w ). In order to use analytc devces famlar from the standard general equlbrum producton model, we assume that λ = λ and b b for = X, M. 10 The value functons are then the same n = both sectors, so we can smplfy notaton by suppressng the superscrpts. Dversfed producton then requres that X LS w X = wm n a steady state. 11 By symmetry, smlar relatonshps hold for captal. Lettng r represent the steady-state return to captal n sector, the values of dscounted ncome for searchng and employed captal are gven n (3) and (4): V (3) KS ( ) r = 1 λ r ρ ρ + λ + b 1 ρ + λ ρ ρ + λ + b (4) VKE ( r ) = r 10 None of our qualtatve results depend upon ths assumpton. 11 More generally, each unemployed factor consders three aspects when decdng upon where to search for a job. The most obvous aspect s, of course, the wage (or rental rate) that wll be receved when employed. The two remanng aspects are the expected duraton of tme searchng before fndng a job and the expected length of the spell of employment once the job s secured. Gven our assumpton that turnover s characterzed by a Posson process, the expected duraton of search before successfully fndng employment s 1 λ, and the expected duraton of a job s 1 b. All other thngs equal, unemployed factors wll be drawn to search n the sector wth the lower expected duraton of unemployment, the hgher expected duraton of a job, and the hgher wage (return to captal). Assumng that turnover rates are the same across sectors neutralzes these latter two consderatons. 8

10 Integratng the zero-proft condtons wth the above dscusson allows us to solve for the steady-state wage and rental rates n the usual way. Fgure 1, drawn under the assumpton that the export sector s relatvely captal ntensve, llustrates the unque soluton for w 0 and r 0. When trade s lberalzed, as shown n Fgure 2, there s a reducton n the domestc prce of the mport-competng good. Eventually the economy settles at a new steady state characterzed by factor prces w T and r T. Due to lberalzaton, the steadystate value of the wage rate falls relatve to both product prces, and the steady-state value of the return to captal ncreases relatve to both product prces. Ths result s, of course, the magnfcaton effect dentfed by Stolper and Samuelson. Factor markets are sluggsh, however, and adjustment to the new steady state does not occur nstantly. Intally, both w and r fall by the same proporton as the reducton n the prce M M of the mport-competng good, and all labor and captal that had been searchng for work n the mport-competng sector moves to the export sector. Because the mportcompetng sector s labor ntensve compared wth the export sector, the pool of searchers n the export sector becomes more labor ntensve, mplyng that employment n that sector wll also begn to become more labor ntensve. As t does so, the wage n the export sector begns to fall whle the return to captal ncreases. In fgure 2, the transton path s shown as a movement down along the zero-proft curve for the export sector. If the reducton n factor prces s not too large, captal and labor employed n the mportcompetng sector at the tme of lberalzaton wat untl there s an exogenous separaton before movng to the export sector to search. Snce we have assumed that the break up rate s the same for captal and labor, the employment of both shrnks proportonately, so 9

11 that the captal ntensty of ths sector remans unchanged. Wth unchanged captal ntensty, w and r also reman unchanged. M M Eventually, the wage n the export sector falls to the level of the wage n the mport-competng sector, though the return to captal s stll hgher n the export sector than n the mport-competng sector. Labor s now ndfferent between searchng for work n the mport-competng sector or the export sector, but captal stll prefers to search n the export sector. Upon separaton, all captal but only some of the labor that had been employed n the mport-competng sector moves to search n the export sector. Both sectors become more labor ntensve, mplyng that the wage rate (now equal across sectors) contnues to fall, whle the return to captal ncreases n both sectors. In Fgure 2, wages and rental rates move down along the zero proft curve n the mport-competng sector n parallel to our slde down along the zero proft curve n the export sector, all the whle mantanng equalty between w and w. The economy reaches the new steady state (wth wages and rental rates n both sectors of X M w T and r T ) at tme T. The tme paths of real wages and the real returns to captal are llustrated n Fgure 3, n whch Π (Π ) represents the tarff-dstorted (free trade) prce ndex. The td bold lne represents the tme path of real factor prces for a low turnover ndustry whle the dashed lne shows the transton path for a hgh turnover ndustry. Observe that the economy reaches the new steady state n fnte tme, wth the hgher turnover ndustry takng less tme to reach ths new long-run equlbrum. ft If the economy only asymptotcally approached the new steady state, the return to captal would forever be hgher n the export sector than n the mport-competng sector, and therefore the export sector would absorb the entre stock of captal. Wthout any captal, the margnal product of captal n the mportcompetng sector would become nfnte, thereby creatng a logcal nconsstency snce the return to captal n the export sector s fnte. The analytcal soluton for the transton path s avalable upon request. 10

12 Fgure 3 reveals that labor ntally employed n the mport-competng sector (the upper left graph) s harmed by lberalzaton, whle captal ntally employed n the export sector (the lower rght graph) clearly benefts. In contrast, the mpact of lberalzaton on the real ncomes of labor ntally employed n the export sector and on captal orgnally employed n the mport-competng sector s ambguous. At frst, labor employed n the export sector s better off snce the real wage n ths sector ncreases whle losses do not occur untl later. The stuaton s reversed for captal ntally employed n the export sector, where the losses are up front and the gans are delayed. Gven a partcular dscount factor, the net mpact on the real ncomes of labor groups n exportng ndustres and captal groups n mport-competng ndustres hnges on how fast the economy reaches the new steady state. Hgher turnover rates speed the adjustment to the new steady state, shortenng the tme that labor ntally employed n the export sector enjoys hgher real ncomes, and reducng the tme that captal ntally employed n the export sector suffers lower real ncomes. Nether the ntal nor the termnal steady state levels of real ncomes depend on turnover rates. The standard HOS model s a specal case (wth a job acquston rate that approaches nfnty) of the more general formulaton presented here. In ths case, adjustment s mmedate because ndvduals could always qut a job that pays a low wage and mmedately fnd a job n the hgh-wage sector. Smlarly, the model nests a specalzed verson of the specfc factors model (where there are no moble factors) as a specal case when the job acquston and job breakup rates are both zero. Wth ntermedate levels of turnover, t s clear from Fgure 3 that labor groups n exportng ndustres wll support trade lberalzaton only f the turnover rate for labor s 11

13 suffcently low. In fact, there s a crtcal level of turnover below whch exportng labor groups wll support lberalzaton, and above whch they wll oppose lberalzaton. A smlar crtcal value for turnover can be defned for nterest groups representng mportcompetng captal owners such that the owners wll support trade lberalzaton f the ndustry turnover rate s hgher than the crtcal value and oppose lberalzaton f the turnover rate s lower than the crtcal value. B. A Smple Model of Tthng In order to examne how nterest group trade preferences wll be reflected n ther lobbyng behavor, we ntroduce a smple model of lobbyng. Assume for the moment that canddates choose ther polcy postons pror to nterest groups gvng campagn contrbutons as n Magee, Brock, and Young (1989). Interest groups then donate funds to help ther preferred canddates get elected. For the sake of exposton, suppose that the trade ssue n queston s whether or NAFTA should be passed n the House. Suppose further that nterest group gets expected PDV of lfetme ncome V f NAFTA passes and V f t does not pass. N ( N ) The group s expected ncome net of lobbyng expendtures s R = γvn + ( 1 γ ) V( N ) c j j, where γ s the probablty that NAFTA passes the House, and cj represents nterest group 's contrbuton to canddate j. The group sets ts contrbutons so as to maxmze expected ncome, or t chooses c j so that R γ (5) = ( VN V( N ) ) 1 = 0 c c j j when c j > 0, 12

14 R γ and sets cj = 0 when = ( VN V( N ) ) 1 < 0. Let θ be the probablty that any c c j j one vote n the House s decsve n determnng the outcome of the bll approvng NAFTA. In that case, the effect of the nterest group contrbuton on the lkelhood that γ p j NAFTA passes s: = θ ( N j N j ), where p j s the probablty canddate j wns c c j j electon, N j s the probablty canddate j votes for NAFTA, and N-j s the probablty that canddate j s opponent would vote for NAFTA. Substtuton nto (5) reveals that an nterest group gves money to a canddate untl: p j (6) θ ( N j N j )( VN V( N ) ) 1 = 0 when c j > 0. c Rearrangng (6) we have: j (7) ( N j N j )( VN V( N ) ) =. p j 1 θ c j Snce the rght-hand-sde of (7) s postve, the two terms n parentheses on the left-handsde must be the same sgn. Equaton (7) thus reveals that nterest groups favorng (opposng) NAFTA wll tend to gve money to canddates who are more lkely to vote for (aganst) NAFTA than ther opponents. Furthermore, t s reasonable to assume, as Magee, Brock, and Young (1989) do, that contrbutons receved by a canddate have dmnshng margnal effects on the probablty that the canddate wns electon. Thus, as c j rses, p c j j falls, and the rght-hand-sde of (7) rses. Wth canddates polcy postons predetermned, t must be the case that c rses wth V. In other j N V ( N ) 13

15 words, nterest groups that have a larger payoff from seeng NAFTA passed (defeated) wll gve more money to each canddate who supports (opposes) NAFTA. In realty, nterest groups may gve campagn contrbutons wth an eye toward nfluencng more than just the outcome of electons. Grossman and Helpman (1994), for nstance, set up a model n whch contrbutors take electon outcomes as gven and donate funds to nfluence the government s choce of trade polcy. Interest groups may also gve money to buy unobserved servces (such as rewrtng legslaton and blockng unfavorable blls n commttees) that representatves can provde for them. Hall and Wayman (1990) provde emprcal evdence that campagn money encourages House commttee members to spend more tme and effort on polcy ssues that are mportant to the PAC. We now expand our lobbyng model to nclude these consderatons. Consder a lobby group whose ncome net of contrbutons s j N ( N (8) S + γv + ( 1 γ ) V c, R = ) j j j where S j s the monetary value of servces provded to PAC by representatve j. Ths term can be consdered to nclude all non-trade consderatons of an nterest group n donatng money. Assumng that campagn contrbutons nfluence both the lkelhood that representatve j wns electon and the probablty that she votes for NAFTA, the mpact of contrbutons on the probablty that NAFTA passes Congress s γ p j N j (9) = θ ( ( N j N j ) + p j ) c c c j j The rght-hand-sde of (9) ncludes two terms that correspond to what Grossman and Helpman (1996) refer to as the electoral and the nfluence motves for gvng campagn contrbutons. The frst term n brackets measures the effect of helpng a favored j 14

16 canddate wn, whle the last term represents the effect of nfluencng the polcy choce of N a canddate lkely to wn ( c j j s the effect of a dollar on the probablty that canddate j supports NAFTA). Dfferentatng (8) wth respect to c j and substtutng (9) nto the result reveals that a PAC wll contrbute money to canddate j untl Sj p j N j (10) + θ ( ( N j N j ) + p j )( VN V( N ) ) = 1 when c j > 0 c c c j j j S c j j p j N j + θ ( ( N j N j ) + p j )( VN V( N ) ) < 1 for c j = 0. c c j j The electoral motve for gvng money suggests that PACs favorng NAFTA wll gve money to those canddates most lkely to vote for the bll s approval, whle the nfluence motve encourages such PACs to gve the largest contrbutons to canddates wth a 50% probablty of votng for NAFTA, snce the margnal mpact of a dollar of contrbutons s maxmzed n ths stuaton. Wth both motvatons beng consdered, a PAC wll gve the most money to canddates wth a greater than 50% chance of votng n the PAC s favored manner on the trade bll. Thus, on average, each PAC s pattern of contrbutons across all canddates wll reveal ts preference on the trade bll. Equaton (10) also reveals that other consderatons play a role n PAC contrbuton decsons, such as the ablty of the poltcan to provde valuable servces to the nterest group, and the lkelhood that the canddate wns electon. Based on (10), we can rewrte the contrbuton decson as N j Sj (11) cj = F( N j N j,,, p j ), c c j j 15

17 where F s a functon that solves (10). Ths functon depends on the nterest group s preference on NAFTA (V n 10). The Grossman and Helpman (1994) model, N V( N ) estmated by Goldberg and Magg (1999), s a specal case of (11) where electons and non-trade servces provded to PACs are assumed away. C. Estmaton Issues In emprcally estmatng (11), we face the problem that the underlyng probablty a canddate s elected and hs ablty to provde servces to the PAC are unobserved. We take two approaches to solvng ths problem. In the frst approach, we assume that a representatve s poston on lberalzaton s unrelated to hs ablty to provde a PAC wth servces and to hs chance of wnnng electon. We can thnk of a large number of canddates for offce, some support NAFTA whle others do not. If the NAFTA proponents and opponents are equally capable of provdng servces to the PAC, there s no reason why the PAC would buy servces from the NAFTA supporters more often than t bought servces from NAFTA opponents. In that case, the share of PAC j s total campagn contrbutons that are gven to NAFTA supporters do not depend on the last two terms n (11). On average, PACs who support NAFTA should gve a larger share of ther contrbutons to legslators who are n favor of NAFTA compared wth PACs who oppose NAFTA. Ths predcton wll be true as long as electoral motves play any role n PAC contrbuton decsons. If the electoral motve plays no role, the two groups of PACs would each gve ther money to the same group of undecded legslators. Let V = V be the nterest group s net gan from NAFTA (postve for N V( N ) supporters of the agreement and negatve for opponents), and let T denote the turnover 16

18 n ndustres represented by the group. Suppose that the nterest group s net gan from NAFTA depends on the factor of producton t represents, ts ndustry s net export poston, and ndustry turnover n the followng manner: (12) = A K + A X + A T + A K T + A X T A UL. V The varable K = 1 f PAC represents the nterests of captal and = 0 otherwse, K X = 1 f PAC represents exportng ndustres on average and X = 0 for mportcompetng ndustres, and UL measures the extent of unsklled labor n the ndustres represented by PAC. The Stolper-Samuelson theorem suggests that captal groups n the U.S. wll beneft from NAFTA on average ( A 1 > 0 ). The specfc factors model predcts that exportng ndustres wll beneft from trade lberalzaton ( A 2 > 0 ). The new nsght of our model s that Stolper-Samuelson effects wll be stronger n hgh turnover ndustres ( A 4 > 0 ) whle the ndustry s net export poston wll matter more n low turnover ndustres ( A5 < 0 ). 13 We do not observe the nterest group s net gan or loss from NAFTA, V, but the lobbyng model reveals that electoral consderatons push PACs to gve more money, on average, to poltcans who are lkely to vote n ther favored manner on lberalzaton. Thus, we estmate (12) usng the share of contrbutons gong to NAFTA supporters as a proxy for the nterest group s net gan from the trade deal. Snce the underlyng 13 As a referee ponted out to us n an earler verson of ths paper, sklled workers mght have dfferent trade preferences than unsklled labor (wth sklled labor havng the same preferences as captal), and sector-specfc turnover rates mght be correlated wth the skll composton of the sector. If hgh turnover ndustres are ntensve n the use of unsklled labor, then we would see labor and captal opposed to each other n hgh turnover ndustres but not n low turnover ndustres. Ths would lead to our observed correlatons, but the mechansm would be dfferent. We account for ths possblty by ncorporatng the fracton of the workforce classfed as unsklled or sem-sklled as a control varable n our regressons. 17

19 preferences are lnear functons of the parameters but the dependent varable s lmted to the range [0, 1], we estmate (12) as a lnear regresson wth censorng above and below. The second method of dealng wth unobserved servces and probabltes of electon s to estmate (11) drectly wth proxes for the unobserved varables. We use representatves membershp on mportant commttees, ther postons of power wthn congress, and ther party as proxes for a legslator s ablty to provde servces to nterest groups, and we use terms n offce as a proxy for the probablty a canddate wll wn reelecton. The predcted probablty that legslator j votes for NAFTA s employed as a proxy for the frst two terms n (11). In order to facltate the estmaton of (11), we assume a partcular functonal form for F. Let contrbutons from PAC to canddate j be determned by (13) c = A ' Y + B' Z + N ( C' Z ), j j j where Y j s a vector of the canddate characterstcs descrbed above that measure the legslator s ablty to wn electon and to provde servces to the PAC. The terms B' Z and C' Z nclude characterstcs of the nterest group that determne ts preferences on NAFTA. As n equaton (12), we assume that these terms take the followng form: (14) B' Z = B1K + B2 X + B3T + B4KT + B5 X T + B6UL (15) C' Z = K + C X + C T + C K T + C X T + C UL. C When combned wth our lobbyng model, the Stolper-Samuelson theorem suggests that captal groups n the US wll gve more money to supporters of NAFTA on average ( C 1 > 0 ) than to NAFTA opponents. The specfc factors model predcts that exportng ndustres wll contrbute more heavly to supporters of trade lberalzaton than 18

20 to opponents (C 2 > 0 ). Our model predcts that we should observe larger Stolper- Samuelson effects n hgh turnover ndustres ( C 4 > 0 ) and larger specfc factors effects n low turnover ndustres ( C 5 < 0 ). There are two complcatons n estmatng (13). Frst, the dependent varable s censored from below at zero because c j = 0 when the left hand sde of (10) s less than one. Thus, estmatng (13) requres a Tobt specfcaton. Second, the trade vote measure s endogenous snce t may be affected by the campagn contrbutons that a canddate receves from nterest groups. In order to deal wth ths endogenety, we use an nstrumental varables technque. Frst, we estmate an equaton predctng how representatves wll vote on NAFTA based on the model n Baldwn and Magee (2000). Then we use the predcted probablty that canddate wll vote for the trade bll n place of N j n (13). In order to check the robustness of the results, we use three dfferent trade lberalzaton measures: the NAFTA, the GATT Uruguay Round Agreement, and both blls jontly. The next secton descrbes the data used n the emprcal tests. 3. Data Table 1 presents the defntons, sources, and means of the varables used n the emprcal tests performed n secton 4. The measure of ndustry turnover used n ths study was compled by Davs and Haltwanger (1992), and Davs, Haltwanger, and Schuh (1996). These authors calculated the change n the number of jobs lost n shrnkng establshments (for job destructon) and the change n the number of jobs ganed n growng establshments (job creaton) relatve to the employment base wthn the ndustry. The job destructon measure for sector s n tme perod t s 19

21 yet yet 1 (16) JD st =, Y + Y e E st yet < yet 1 st 2 st 1 where yet s employment n establshment e, Y st s total employment n sector s, and Est s the set of establshments n sector s at tme t. Whle these data are referenced n the lterature as measurng gross job flows, they are n fact measures of the net change n establshment sze over one year. Davs and Haltwanger (1992) dscuss several dfferent measures of job turnover based on ther data on changes n establshment sze. Ths paper uses the average job destructon rate as defned n (16) between 1988 and 1992 as the measure of ndustry turnover snce the job destructon rate s closely ted to the noton of job securty n our model, though we experment wth alternatve specfcatons of turnover dscussed n Davs and Haltwanger (1992) n order to explore the robustness of our results. In order to lnk poltcal acton commttees to the ndustry they represent, we use a data set from the Center for Responsve Poltcs (CRP) that places 217 manufacturng PACs nto groups of 4-dgt SIC ndustres. Usng descrptons of each company and unon avalable on the nternet, we are able to dentfy the 2, 3, and 4-dgt SIC ndustry afflatons of 202 other corporate and labor PACs that gave money to House members who voted on the blls enactng the NAFTA or Uruguay Round agreements. These poltcal acton commttees are dentfed as representng ether captal or labor nterests based on the Federal Electon Commsson classfcaton of each PAC as a corporate or labor group. In total, the data set conssts of 42 labor and 377 corporate PACs The job destructon rate has a value of -2 for plant deaths. Plant brths are not ncorporated n ths measure because t s a measure of job loss, not job gans. 20

22 Each nterest group s classfed as representng mport-competng or exportng nterests based on the net trade poston of the PAC s ndustres of orgn. The PAC net export poston equals one f the ndustres total exports were greater than mports over the perod , and t equals zero otherwse. 15 Under ths defnton, the data set ncludes 226 mport-competng nterest groups and 193 exportng PACs. The trade flow data used to make these calculatons are taken from the NBER US mports and exports data sets ( that are descrbed n Feenstra (1996, 1997). Because sklled labor and unsklled labor may dffer n ther support for lberalzaton, we nclude a measure of the skll level of workers n each ndustry, taken from the Current Populaton Survey (CPS), n estmatng the PAC contrbuton equatons. The CPS dvdes workers nto fve categores: engneers/scentsts, whte collar workers, sklled labor, unsklled labor, and semsklled labor. The skll level of workers represented by a PAC s measured by the fracton of workers (over the perod ) that are classfed as unsklled or semsklled n the ndustres lnked to the PAC. The Federal Electon Commsson provdes nformaton on the contrbutons each PAC gves to every canddate n the House of Representatves. In ths paper, we examne three dfferent measures of whether the contrbutons were gven prmarly to supporters of trade lberalzaton. These measures are the share of contrbutons that were gven n 1992 to representatves who voted for NAFTA (mean=0.60), the share gven to canddates votng to approve the GATT Uruguay Round (mean=0.73), and the share gven to supporters of both NAFTA and the GATT blls (mean=0.51). About 54% of representatves voted for NAFTA, 67% voted for the GATT bll, and 46% voted for both 15 Usng an alternatve measure, ndustry net exports as a share of domestc consumpton, n the emprcal estmaton does not alter the basc results. 21

23 trade blls. Ths paper also examnes the factors that affect PAC j s contrbutons to canddate. The mean contrbuton was over $1300 when an nterest group gave money to a representatve. Because the typcal PAC gave money to only 7% of all canddates, however, the average contrbuton from a PAC to a representatve s $95 n our data set. Poltcs n Amerca: 1994, edted by Duncan (1994) provdes nformaton on other factors that may nfluence the amount of PAC money canddates receve: membershp on key commttees, postons of leadershp n the party and n commttees, terms n offce, and party afflaton. 4. Emprcal Tests Table 2 provdes non-parametrc tests of two predctons from the model n secton 2. In hgh turnover ndustres, we should observe a large dfference between captal and labor groups n the fracton of contrbutons gven to NAFTA supporters. Low turnover ndustres should not exhbt ths large dfference as the Stolper-Samuelson effects are less mportant. Low turnover ndustres, unlke hgh turnover ndustres however, should reveal a large dfference between mport-competng and exportng PACs n the fracton of contrbutons gven to NAFTA supporters. Table 2 presents the fracton of PAC contrbutons gven to congressonal representatves who voted for NAFTA, for GATT, and for both blls. Lobby groups representng the nterests of captal owners n low turnover ndustres, for example, gave almost 61 percent of ther contrbutons to representatves who ultmately voted n favor of NAFTA. The results n Table 2 provde strong support for the model s predctons. As one would expect, labor groups gave a smaller share of contrbutons to representatves who 22

24 voted n favor of free trade than dd corporate nterest groups n every case examned. Consstent wth the model presented n ths paper, labor groups n hgh turnover ndustres revealed much greater opposton to free trade n ther contrbuton patterns than dd labor groups n low turnover ndustres. Meanwhle, the contrbuton patterns of captal groups show that captal PACs n hgh turnover ndustres are more supportve of free trade than captal PACs n low turnover ndustres. The mportant result for the model n ths paper s that n hgh turnover ndustres, captal groups gave a sgnfcantly larger fracton of ther contrbutons to NAFTA supporters, to GATT supporters, and to supporters of both trade blls, than dd labor groups. In low turnover ndustres, however, captal groups dd not gve sgnfcantly larger fractons of ther campagn donatons to supporters of trade lberalzaton than labor groups dd. The lower half of Table 2 examnes the model s predcton that the ndustry net export poston wll be mportant n determnng nterest group support for trade lberalzaton only n low turnover ndustres. The table shows that n low turnover ndustres, PACs representng exportng ndustres gave a sgnfcantly greater porton of ther contrbutons to supporters of trade lberalzaton than dd mport-competng PACs. In hgh turnover ndustres, however, the dfference between mport-competng and exportng PACs n ther contrbuton patterns was neglgble, as the model predcts. Table 3 presents the estmates of (12) usng the PAC s share of total contrbutons gong to NAFTA supporters (column 1), the share gong to GATT supporters (column 2), and the share gong to representatve who voted for both blls (column 3) as measures of PAC preferences on lberalzaton. An observaton represents a sngle poltcal acton commttee. The bottom half of the table ncludes ndustry fxed effects for each 2-dgt 23

25 SIC manufacturng ndustry n order to control for unobserved ndustry characterstcs that mght nfluence the degree of factor specfcty and therefore affect the PAC s contrbuton decsons. Our model predcts that the factor represented by the PAC should affect ts contrbuton patterns n hgh-turnover ndustres (where captal would be supportve of free trade whle labor would not be), but not n low-turnover ndustres. Note that the margnal effect of the captal varable on the nterest group s trade preference n (12) s V (17) = A1 + A4T. K The theory predcts that A1 = 0 and A 4 > 0. Furthermore, there should be a crtcal value of turnover (call t ( ) T K ) such that = 0 f T TK V 18. K > 0 f T > TK Table 3 provdes strong support for these predctons. The estmates of A 1 are small and nsgnfcantly dfferent from zero, whle the estmates of A 4 (the coeffcent on the nteracton term between turnover and captal) are postve n all specfcatons of the model, both wth and wthout ndustry fxed effects. The coeffcent estmates of A 4 are also statstcally sgnfcant (wth and wthout fxed effects) usng NAFTA and both trade blls as measures of the PAC s support for trade lberalzaton. Fgure 4 llustrates the estmated effect (and 95% confdence nterval) from (17) of the PAC s factor of producton on the share of ts contrbutons gong to 24

26 representatves who voted for both the NAFTA and GATT blls. 16 As the theory predcts, the factor of producton has a sgnfcant effect on support for lberalzaton only for PACs n ndustres wth suffcently hgh turnover levels. The estmated crtcal value beyond whch the factor of producton nfluences trade preferences s T = About 72% of the PACs n our sample have turnover levels hgher than ths crtcal value. As noted above, we used the job destructon rate as defned n (16) as our measure of job turnover. Ths measure s conceptually a better ft to our theoretcal model than some of the alternatve measures suggested by Davs and Haltwanger (1992), such as the sum or the mnmum of job creaton and job destructon. We checked for the robustness of these results by re-estmatng our equatons usng these alternatve measures of turnover and found that our results were strongly robust to alternatve specfcatons. Our model also predcts that the ndustry s net export poston should have an mpact on nterest-group behavor only for low turnover ndustres, and that mpact should dmnsh as turnover rses. From (12), the margnal mpact of the ndustry s net export poston on the share of contrbutons gven to free trade supporters s V (19) = A2 + A5T. X The model s predcton, then, s that A2 > 0, A < 5 0, and that there exsts a crtcal value of turnover (call t ( ) T X ) such that > 0 f T TX V 20. X = 0 f T > TK K 16 Fgures 4 and 5 both llustrate the results from the specfcaton that ncludes ndustry fxed effects. 25

27 The results n Table 3 are generally consstent wth ths predcton. As the model predcts, the estmates of A 5, the coeffcent on the nteracton term between turnover and the ndustry net export poston, are negatve n all sx specfcatons of the models. The coeffcent estmate s also statstcally sgnfcant usng the GATT Uruguay Round bll as the measure of support for trade lberalzaton. When ndustry fxed effects are added to the estmaton, the coeffcent estmates do not change very much, although the larger standard errors push the p-value of the estmate of A 5 above the 10% level n the GATT estmaton ( p = 0.16 ). Usng net exports as a share of domestc consumpton n place of the varable X n Table 3 provdes qualtatvely smlar results. Fgure 5 llustrates the estmated effect from (19) of the ndustry s net export poston on the share of contrbutons gven to GATT supporters. Among low turnover ndustres, the ndustry s net export poston has a sgnfcantly postve effect on the share of a PAC s contrbutons gong to representatves votng for GATT. For turnover levels beyond the crtcal value of T = 9. 1, however, ndustry net exports do not X nfluence a PAC s support for trade lberalzaton. Roughly 39% of PACs n our data set fall nto ths category. The concluson that the predcton n (20) s supported by the evdence n Table 3 s tentatve n two respects. Whle the coeffcents have the correct sgn n all sx regressons, the estmate of A 5 s statstcally sgnfcant n only one of the specfcatons of the model. Second, unlke our estmates based on (18), the sgnfcance of the estmate of A 5 n the GATT regresson s senstve to our use of the job destructon rate as the measure of turnover. Usng the sum of job creaton and job destructon or the mnmum of those varables, means that the coeffcent estmate n the GATT equaton s no longer 26

28 statstcally sgnfcant at the 10% level. The sgns of the coeffcent estmates reman correctly predcted by the model usng all of the dfferent measures of turnover. Tables 4A and 4B present the results of estmatng (13), (14), and (15), whch determne how contrbutons from PAC to representatve j vary wth the legslator s votes on trade blls. Table 4A uses the representatve s actual votes on the NAFTA (columns 1 and 2) and GATT blls (columns 3 and 4) as measures of hs or her preference on lberalzaton. Table 4B uses the predcted probabltes that the representatve voted for NAFTA and GATT, based on the emprcal model n Baldwn and Magee (2000), to account for the endogenety of legslators votng decsons. To control for unobserved ndustry, representatve, and PAC varables that affect the contrbuton choces, fxed effects are ncluded n each regresson. Columns 1 and 3 nclude ndustry and canddate fxed effects whle columns 2 and 4 control for ndustry and PAC fxed effects. 17 The unt of observaton s a PAC-representatve combnaton. The results n Table 4 reveal that, as our model predcts, Stolper-Samuelson consderatons are much stronger for hgh turnover ndustres than for low turnover ndustres. Whle captal groups tend to favor supporters of trade lberalzaton, the estmates of C reveal that the preference of captal groups for NAFTA and GATT 4 supporters s much weaker n low turnover ndustres than n hgh turnover ndustres. The estmates of C are postve n all eght specfcatons of the model (and statstcally 4 sgnfcant n seven of the specfcatons). Column 2 of Table 4A provdes a typcal comparson from among the regressons. A captal PAC wth the lowest level of turnover n our data gves $1218 more to a NAFTA supporter than to a NAFTA opponent. An 27

29 dentcal captal PAC wth the hghest level of turnover n our data set donates $2781 more to NAFTA supporters than to NAFTA opponents. Thus, the estmates n Table 4 present strong evdence n favor of the predcton that the Stolper-Samuelson effects of lberalzaton on nterest groups wll be stronger among hgh turnover ndustres. Our model s other predcton, that the ndustry net export poston s more mportant n determnng a PAC s trade polcy stance n low turnover ndustres, s also consstent wth the evdence n Table 4. The estmates of C 2 are postve n seven of eght specfcatons, suggestng that exportng ndustry PACs n low turnover ndustres tend to gve larger contrbutons to supporters of trade lberalzaton than to opponents. The exportng ndustry revealed preference for trade lberalzaton declnes as turnover ncreases, however, as the negatve estmates of C 5 reveal. The emprcal estmates n Table 4 reveal that C 5 < 0 n all eght specfcatons of the model, and sgnfcantly so n fve of them. The four specfcatons usng the GATT vote as a measure of trade preferences show coeffcent estmates of C that are sgnfcant at the 1% level. As the 5 model predcts, the hgher the turnover s n a PAC s ndustry, the smaller the effect of the ndustry s net export poston on ts contrbutons to NAFTA and GATT supporters. 5. Concluson Earler work by Davdson, Martn, and Matusz (1999) provded a theoretcal bass for lnkng ndustry turnover and nternatonal trade. Ths paper develops the model more fully to reveal that ndustry turnover can be used to dvde nterest groups nto those whose trade preferences should be determned by ther factor of producton and those 17 Some PACs represent multple ndustres. Because of the large number of dummy varables requred, t 28