Trade Policy of a Free Trade Agreement in the Presence of Foreign Lobbying

Transcription

1 Trade Polcy of a Free Trade Agreement n the Presence of Foregn Lobbyng Andrey Stoyanov y The Unversty of Brtsh Columba October 2007 Abstract Ths paper studes the e ect of foregn lobbes on trade polcy of a country whch s a member of a Free Trade Agreement (FTA). It uses a monopolstcally compettve poltcal economy model n whch the government determnes external tar s endogenously. The e ect of foregn lobbyng under the FTA s examned emprcally usng Canadan ndustry-level trade data that allow d erentatng of lobby groups by the country of orgn. The analyss suggests that the presence of foregn lobbyng has a sgn cant e ect on the domestc trade polcy. The heterogenety of foregn lobbes s also mportant: the presence of an organzed lobbyng group n an FTA partner country tends to rase trade barrers whle an organzed lobbyng group of exporters from outsde of the FTA s assocated wth less protecton. JEL Class caton: F12, F13, F14, D72 Key Words: Trade Polcy, Free Trade Areas, Foregn Lobbes I am especally ndebted to Matlde Bombardn for contnuous support and gudance. For helpful comments and suggestons, I thank Bran Copeland, Werner Antweler, and partcpants of the 2007 European Meetngs of the Econometrcs Socety, Canadan Economc Assocaton conference, Team for Advanced Research on Globalzaton, Educaton, and Technology conference, and semnar partcpants of the Unversty of Brtsh Columba for ther comments and useful dscussons. y Ph.D. candate, Department of Economcs, Unversty of Brtsh Columba, East Mall, Vancouver, BC, Canada V6T 1Z1. Emal: stoanov@nterchange.ubc.ca. 1

2 1 Introducton In the poltcal economy lterature a growng number of studes vew trade polcy as an endogenous outcome of lobbyng actvty by specal nterest groups. Several authors (Goldberg and Magg, 1999, Gawande and Bandyopadhyay, 2000) have con rmed that lobbyng ntensty by domestc rms s one of the man determnants of the cross-ndustry pattern of protecton. More recently, Gawande, Krshna, and Robbns (2006) also nd that lobbyng by foregn rms for trade barrers reducton has a sgn cant e ect on the structure of tar s across ndustres. However, f a country s a member of a regonal free trade agreement (FTA) and foregn rms can a ect the government s decson regardng trade polcy, t becomes necessary to dstngush foregn lobbyng from wthn and outsde of the FTA. Organzed foregn nterests wth preferental market access wll lobby for more protecton aganst other foregn rms, and the trade agreement may become more protectonst wth a strong lobby group n a prospectve FTA partner country. Actve foregn lobbyng under the preferental trade agreement may not only lead to an ncrease n trade barrers, but also make welfare-reducng trade agreements poltcally feasble. In ths paper I analyze the e ect of foregn lobbyng on domestc trade polcy when the country s a member of a preferental trade agreement usng Canadan post-nafta trade data. Gven the relatve sze of the US and Canadan economes, the e ect of US lobbyng n Canada wll be consderably larger than the e ect of Canadan lobbyng n the US. Therefore, focusng on Canadan data s partcularly advantageous for the emprcal analyss of foregn lobbyng under the FTA. Ths analyss reveals two man results. Frst, the actvty of foregn lobbysts n Canada s a sgn cant determnant of the Canadan trade polcy, and sectors n whch foregn rms wthout preferental market access are poltcally organzed tend to receve less protecton. Ths result supports the prevous ndng by Gawande, Krshna, and Robbns (2006) for the US. Second, NAFTA has an mportant e ect on the structure of foregn lobbes. The data con rm that foregn rms wth preferental market access lobby for more protecton just as domestc rms do. Ths result mples that an FTA wth an actve foregn lobby may become more protectonst and, under certan condtons, lead to a welfare reducton. Ths paper s the rst one that analyzes from theoretcal and emprcal ponts of vew the e ect of foregn lobbyng on domestc trade polcy n the presence of an FTA. The poltcal economy model presented n ths paper ncorporates a monopolstcally compettve market structure nto the Grossman and Helpman (1994) protecton for sale setup (henceforth GH) to analyze the role 2

3 of foregn lobbyng n the makng of natonal trade polcy. The framework s further extended by allowng for two types of foregn nterest groups, namely, lobbyng groups formed by rms from an FTA partner country and by rms from countres outsde of the FTA, aggregated nto the rest of the world (ROW). Ths d erentaton of foregn lobbes by market access s mportant n the presence of an FTA. When two countres jon a regonal trade agreement, grantng zero mport tar s to each other, rms from countres wth preferental market access wll lobby for more protecton under the FTA to lock t from competton from the ROW rms, whle rms from outsde of the FTA wll contnue to lobby for trade lberalzaton. 1 Ths results n heterogenety of foregn lobbyng under the FTA wth respect to ther lobbyng objectves, and mples that n a world where almost every country s a part of at least one FTA, a complete theory of the e ect of foregn lobbng on the natonal trade system should take ths heterogenety nto account. The model of foregn lobbyng n the presence of the FTA s tested usng Canadan post-nafta trade data. The emprcal analyss suggests a strong and statstcally sgn cant e ect of domestc and foregn lobbyng on Canadan trade polcy, and ponts to the mportance of dstngushng partner country lobbyng from ROW lobbyng. Usng data on lobbyng ntensty by sector and by country of orgn, ths paper ver es that the man predctons of the model are consstent wth the data. Frst, the mportance of both partner country and ROW foregn lobbyng under the FTA s con rmed n the Canadan data: whle the presence of the organzed domestc lobbyng group rases ndustry mport tar by 3-5% relatve to the unorganzed ndustry, organzed partner country lobbyng would rase t by 1-2%, and foregn lobbyng from the ROW would lower t by 2-3%. These results are economcally meanngful and con rm the man predcton of the model that foregn rms wth preferental market access behave just as domestc rms do, whch ntroduces an addtonal dstorton n the polcy makng process. Second, the emprcal evdence shows that even poltcally unorganzed sectors receve a postve level of protecton from the government of 2-4 percentage ponts of the ad-valorem tar, provdng support for the mperfectly compettve structure of the model. Addtonally, an mportant contrbuton of ths work to the poltcal economy lterature of trade s ts determnaton of more plausble values for the government s valuaton of poltcal contrbutons. In ths paper, the government s estmated to value poltcal contrbutons more than natonal welfare, whch s n sharp contrast wth prevous tests of the benchmark GH spec caton n whch governments were found to 1 Two consderatons should be taken nto account when partner country lobbyng for more protecton s consdered. Frst, the WTO tar bndng constrans the lobbyng opportuntes by the partner country rms; however, they may stll play an mportant role n the future and mpede multlateral trade lberalzaton n those sectors. Second, the WTO precludes countres from rasng ther tar s once the FTA s sgned. Yet foregn lobbes may oppose further tar reducton as descrbed above and use ant-dumpng and countervalng measures to gan protecton. 3

4 have stronger preferences for welfare (Goldberg and Magg (1999), Gawande and Bandyopadhyay (2000)). Ths result explans why relatvely small poltcal contrbutons may have strong polcy e ects. My emprcal results provde new perspectve on the e ect of regonal trade blocks on trade polces of member countres. Most of the lterature on endogenous trade polcy concludes that a country s more lkely to reduce ts external tar when t enters an FTA. Rchardson (1993), Bagwell and Stager (1997), Bohara, Gawande, and Sangunett (2004), Bond, Rezman, and Syropoulos (2004) show that a welfare-maxmzng government wll lower external mport tar s once an FTA s formed. By dong ths, the government restores part of the tar revenue lost due to the shft n mport demand from the ROW to the partner country rms. Ornelas (2005a,b) examned the poltcal economy of an FTA wthout foregn lobbyng usng an olgopolstc market structure n the GH model. He shows that FTA formaton weakens the lobbyng power of domestc rms because the elmnaton of tar s between FTA member countres shfts part of the tar rent from domestc rms towards rms from a partner country. Therefore, poltcal economy factors lower the potental bene t from protecton for home country rms, thus renforcng the welfare maxmzaton consderatons dent ed by Bagwell and Stager (1997) and reducng the government s ncentves for protecton. The model n ths paper, however, allows for cross-border lobbyng actvty, whch lessens the e ect ntroduced by Ornelas (2005a,b) snce the reducton n poltcal actvty by domestc rms s coupled wth an ncrease n contrbutons for protecton by the FTA partner country rms. As a result, n the presence of foregn lobbyng a country may n fact rase external tar s under the FTA when strong lobbyng by an FTA partner country puts extra pressure on the government for hgher trade barrers. Actvtes of foregn nterest groups n natonal polcy-makng have receved a growng attenton n the poltcal economy lterature. Many scholars have evaluated the ntensty of foregn lobbyng n the US and argue that t has hgh potental for polcy n uence. Mtchell (1995) found that foregn a lates n the US contrbuted 5.6% of total corporate poltcal contrbutons n , and 42% of them hred professonal lobbysts to promote ther nterests n Washngton. Hansen and Mtchell (2000) clam that, although foregn corporatons make lower poltcal contrbutons than domestc ones due to the exstng legal restrctons, they are just as ntensve as domestc corporatons wth respect to lobbyng actvty and lobbyng expendtures. Choate (1991) presents a lst of over 200 former US o cals (ncludng members of the presdent s admnstraton) who lobbed for the nterests of foregn corporatons, along wth the lst of the foregn clents they represent and the amounts pad. Whle t has been argued that foregn corporatons are qute n uental n US poltcs, very lttle 4

5 research has been done on ther e ect on trade polcy outcomes. In ther poneerng work, Gawande, Krshna, and Robbns (2006) demonstrate that foregn agents lobbyng expendture n the US s even greater than poltcal contrbutons by domestc corporatons, and that the elastcty of the US mport tar wth respect to foregn lobbyng s almost as bg as wth respect to the domestc one. Ther paper was the rst to show that foregn lobbyng s an mportant factor n the formaton of natonal trade polcy and argue that foregn lobbyng may be bene cal to the country s trade polcy as a counter-pressure to domestc nterests, helpng to reduce dstortonal trade barrers. In contrast, my paper suggests that the above argument s nvald n the presence of preferental trade agreements, when rms from a partner country prefer to mantan hgh dscrmnatory tar s for thrd-country mports. Snce most of the countres are members of at least one preferental trade agreement, ths paper demonstrates that cross-border lobbyng may stmulate more protectonst trade polces of FTAs and dsrupt multlateral trade lberalzaton. The rest of the paper s organzed as follows. Secton 2 provdes some evdence on the lobbyng actvtes of foregn rms n Canada. Secton 3 ntroduces the mod ed GH verson of the model that allows mperfectly compettve market structure and two groups of foregn lobbes, and motvates the emprcal methodology. Secton 4 descrbes the data and Secton 5 explans the estmaton procedure. Results are presented n Secton 6 and Secton 7 concludes. 2 The role of foregn lobbes n Canada To motvate the presence of foregn lobbyng n the theoretcal model and emprcal spec caton, we need to dscuss the man channels of foregn lobbyng actvty n Canada and ts relatve strength as compared to domestc lobbyng. Concerns about foregn n uence on Canadan polcy and the necessty of specal regulaton emerged after WWII. The man law that regulates the actvty of domestc and foregn lobbysts n Canada s the Canada Electons Act, ntroduced n Ths law regulates the amount of poltcal contrbutons by Canadan natonals to poltcal partes that could be used for poltcal advertsement campagns. It also explctly bans the use of poltcal contrbutons from foregners (s.217): No person or party shall accept or use contrbutons from a person who s not a Canadan ctzen..., corporaton or assocaton that does not carry on busness n Canada, foregn poltcal party, or foregn government. However, ths law su ered from a lack of transparency n the lobbyng process. On September 9, 1985, Canadan Prme Mnster Mulroney announced hs ntenton to ntroduce a legslaton to ensure that persons, who are approached by lobbysts for Canadan corporatons, assocatons and 5

6 unons, and by agents on behalf of foregn governments and other foregn nterests, would be clearly aware of who s behnd the representatons. As a result, a Lobbyng Regstraton Act (LRA) came nto force on September 30, Ths pece of legslaton ntroduced a de nton of a lobbyst and a requrement for lobbysts to regster wth the Lobbysts Regstrar. But more mportantly, t requres lobbysts to provde nformaton about the name and busness address of the organzaton that has a drect nterest n the outcome of the lobbysts actvtes on behalf of the clent, all of ts subsdares and corporate headquarters, f there s one. An amendment to the LRA, ntroduced n 1996, made ths nformaton publcly avalable. It also ntroduced a strct dsclosure of funds polcy appled to poltcal partes. Together wth the Canada Electons Act, the LRA made t d cult for foregn rms to lobby ther nterests n Canada drectly. Nevertheless, there are some loopholes n the legal restrctons, and poltcally actve foregn rms can stll n uence trade polcy outcomes n at least two ways. Frst, they can hre Canadan agents and consultants to lobby the executve branch on ther behalf and a ect polcy outcomes n a way that suts the nterests of foregn rms. Second, subsdares of foregn enterprses can make legal poltcal contrbutons wth ther own funds to defeat legslators who are unfrendly to ther nterests. Snce there are no restrctons on the share of foregn captal n the assets of a company that makes poltcal contrbutons, any local subsdary of a foregn corporaton can make poltcal donatons from ts own funds f t carres busness n Canada. Moreover, almost any bg foregn company that exports to Canada has an ndependent local sales department, whch s legally allowed to lobby for a reducton n trade barrers on the products mported by ts parent company nto Canada. Lobbyng e orts of such subsdares wll be counter to the e orts of domestc rms and, therefore, poolng all Canadan rms together regardless of ther ownershp may lead to msleadng results and estmaton problems. In the trade polcy lterature, corporate poltcal actvty s typcally measured by nancal contrbutons to canddates and poltcal partes, whle very lttle attenton has been pad to other means of a ectng polcy outcomes such as drect lobbyng. However, earler research on the effect of foregn companes on natonal polcy (Hansen and Mtchell, 2000) suggests that foregn corporatons prefer drect lobbyng to poltcal contrbutons not only because of legal restrctons on contrbutons, but also because of nformal legtmacy questons for poltcans wth respect to acceptng money from corporate sources wth foregn ownershp. Hansen and Mtchell found that because lobbyng s less vsble than contrbutons, foregners use t as ntensvely (and e ectvely) as domestc rms do. Gawande, Krshna, and Robbns (2006) establshed a smlar pattern: n the 1978 US electon cycle, foregn corporatons spent approxmately 40% more on lobbyng than ther US counterparts contrbuted to poltcal partes. For these reasons, lobbyng expendtures seem to 6

7 be a better measure of foregn poltcal nvolvement, especally n countres wth legal restrctons on poltcal contrbutons by foregners. Unfortunately, data on lobbyng expendtures by domestc and foregn corporatons are unavalable for Canada. To measure the foregn lobby ntensty n Canada, I collected the nformaton on the number of lobbysts o cally regstered wth the O ce of the Regstrar of Lobbysts, as s requred by the LRA. The Act de nes a lobbyst as an ndvdual who, for payment, undertakes to lobby on behalf of a clent and represents an organzaton n arrangng meetngs wth publc o ce holders, or communcate wth a publc o ce holder n an attempt to n uence the development of any legslatve proposal,... the makng or amendment of any regulaton,... the development or amendment of any program or polcy. Intutvely, the number of lobbysts hred by a company or ndustry should be hghly correlated wth the lobbyng expendture and hence can serve as a proxy for poltcal actvty. The lobbysts regstraton data s publcly avalable and s dscussed n more detal n Secton 4.2, but the followng gures llustrate the relatve mportance of foregn lobbyng n Canada. In , there were 1,032 o cally regstered lobbysts representng nterests of manufacturng rms regardng Canadan trade polcy, wth 47%, 26% and 27% of them actng on behalf of Canadan, US and ROW rms, respectvely. These numbers support prevous ndngs by Gawande, Krshna, and Robbns (2006) and demonstrate that foregn agents have actvely attempted to n uence formaton of domestc trade polcy. The objectve of the remanng part of the paper s to estmate the e ectveness of lobbyng by two nterest groups (n the partner country and n the ROW) and compare t to the e ect of domestc lobbysts on Canadan trade polcy. 3 The model The theoretcal model s based on the Grossman and Helpman (1994) poltcal economy model and presents several mod catons that allow for the presence of foregn lobbyng and facltate econometrc estmaton. In ther orgnal formulaton, Grossman and Helpman consdered a small open economy, whch leaves no room for foregn companes to lobby because pre-tar prces are xed. In ths work I develop and buld nto the GH setup a model of monopolstc competton wth d erentated goods to allow foregn rms to gan or lose from mport tar s. There are N products (ndustres) n the model and three countres: Canada (Home country), the US (FTA Partner country) and the ROW, denoted by H; P and ROW, respectvely. Industres are denoted by ndex 2 f1; :::; Ng and countres by j 2 fh; P; ROW g. There are n j rms n country j and ndustry. These rms are assumed to be symmetrc wthn the same country and ndustry,.e. they share the same cost structure and hence face the same demand functons and 7

8 charge the same prces. In total, there are n H + n P + n ROW d erent varetes of each product. A representatve consumer maxmzes a quaslnear utlty functon wth a constant elastcty of substtuton ndex nested nto a Cobb-Douglas functon: U = X 0 + NX! ln X =1 X = n H d H 1 1 x H + n P d P 1 1 x P + n ROW d ROW 1 x ROW 1! 1 (1) where X s an aggregate consumpton ndex for product,! s the share of product n the total consumer s expendture, x j s the demand for product produced n country j, > 1 s the elastcty of substtuton between varetes of product, 2 and d j s a country-wde taste (or qualty) parameter for product mported from country j. Maxmzng (1) subject to the standard budget constrant, we obtan the demand functons and an aggregate prce ndex for product : X =! (P ) 1 ; 1 (2) x j =! d j p j p j P P = n H d H (p H ) 1 + n P d P (p P ) 1 + n ROW! 1 (3) d ROW (p ROW ) (4) Frms wthn one country and sector are assumed to have the same constant margnal cost. Ths allows us to consder Canadan market ndependently from other markets,.e. prces on the Canadan market depend only on the demand elastcty, the ( xed) number of rms and the xed margnal cost structure. Denotng a spec c mport tar set by the home country government on mports of product from country j as j, we can wrte the pro t of a country j rm that produces product as: where q j s the quantty suppled and cj j = (pj c j j )qj (5) s the margnal costs. I assume that the number of rms s large enough to gnore the e ect of ther ndvdual prcng decsons on the ndustry prce ndex P,.e. each rm takes the prce ndex as gven. Knowng product demand functons (3), each rm sets the pro t-maxmzng prce as a markup over ts margnal costs: p H = c H ; p P = (c P + P ); p ROW = For convenence, solate costs from (6) and wrte down equlbrum pro ts (5) as: j = 1 p j qj 2 Substtuton elastcty between ndustres s assumed to be equal to unty. (c ROW + ROW ) (6), 8j (7) 8

9 The government chooses mport tar s to maxmze a weghted sum of natonal welfare W and poltcal contrbutons C: G( j ; C j ) = X C H + aw + b X C P + c X C ROW (8) where C H, C P and C ROW are ndustry-wde poltcal contrbutons from each country. Coe cent a s a weght that the government assgns to natonal welfare relatve to poltcal contrbutons. Coe cents b and c re ect the government s preferences for the US and ROW contrbutons, respectvely, over the contrbutons by domestc rms. As long as acceptng contrbutons from foregn rms nvolves rsk of reputaton loss or law nfracton, poltcans may prefer domestc contrbutons to overseas donatons thus both coe cents are presumably less than one. Frms n ndustry can organze themselves and form a group to lobby the local government for a change n trade polcy. 3 Frms wthn the FTA pay no mport tar s and hence lobby for more protecton, whle rms from other countres lobby for lower tar s for the opposte reason. The lobby representng ndustry of country j maxmzes ts welfare from obtanng protecton net of poltcal contrbuton: (W j C j ). As n Grossman and Helpman (1994), the equlbrum trade polcy s a soluton to a two-stage game. In the rst stage, knowng the government s objectve functon, each organzed lobbyng group provdes the government wth a schedule of poltcal contrbutons as a functon of mport tar. In the second stage, observng contrbuton schedules, the government sets trade polcy that maxmzes ts objectve functon (8). Grossman and Helpman (1994) show that for truthful contrbuton schedules 4 the optmal trade polcy s the one that maxmzes jont surplus of the government and organzed lobbyng groups. Let denote the share of the home country populaton enttled to the domestc ndustry pro ts, and I j denote an ndex varable that takes the value of one when ndustry n country j s poltcally organzed and zero otherwse. The jont welfare functon then takes the form: = NX =1 I H W H + aw + b NX =1 I P W P + c NX =1 I ROW W ROW (9) where W H = n H H + (T R + CS) s welfare of the domestc ndustry gross of poltcal contrbutons, T R and CS are total tar revenue and consumer surplus, respectvely, W j = n j j ; j 2 fp; ROW g s gross welfare of foregn ndustres from exports to the home county market, 3 Wth the number of rms n the sector beng lmted by the endowment of sector-spec c captal, rms n each ndustry have an ncentve to form a lobby group and seek for protecton from foregn competton. Here I gnore the free-rdng problem wthn each sector. See Bombardn (2005) for an extensve dscusson of rm-level contrbuton decson. 4 Contrbuton s de ned as truthful f t re ects the true preferences of lobbyng group for any possble polcy outcome. 9

10 and W = P (nh H ) + T R + CS s natonal welfare. Takng the rst order condton of the jont welfare functon wth respect to the ROW mport tar rate and rearrangng t, one obtans the expresson for the equlbrum trade polcy: 5 ROW " p ROW = 1 + ( 1) P p P s P + a 1 s H I H s H + (10) a + a + + bip 1 s P + cirow a + a + 1 s ROW 1 where s j = nj pj xj P X the left-hand sde of (10), denotes the share of country j rms on the Canadan market for product. On s the ad-valorem tar on the ROW mports, whch s ROW =p ROW multpled by the prce elastcty of demand for the ROW mports ". Therefore, as n the benchmark GH model, trade protecton s nversely related to the mport demand elastcty. 6 The rst term on the rght-hand sde s negatve: the model predcts that wth more d erentated varetes wll receve mport subsdy. Ths result s a drect consequence of monopolstc competton model wth spec c mport tar. 7 The second element on the rght-hand sde shows the postve relaton between the FTA external and nternal tar s and re ects a tar complementarty e ect: f the tar rate for the partner country s hgh, t s optmal for the government to rase the external tar as well. 8 Intutvely, an ncrease n the wthn-fta tar rate causes a declne n mports from the ROW, and tar revenue collected on the ROW mports s hgher for hgher ROW. Ths, n turn, rses mports from the partner country, that generates more tar revenue for hgher partner country tar rate. The tar complementarty e ect s proportonal to the market share of the partner country rms s P stronger f the partner country and the ROW exports are close substtutes. and s In contrast to the benchmark case, even for unorganzed ndustres protecton may stll be postve due to the mperfectly compettve market structure, as emphaszed by the thrd term, snce the coe cent a a+ s postve. Because the share of domestc rms on the market re ects ther ablty to capture protecton bene ts, the tar level s proportonal to s H and ncreasng wth. The fourth term s smlar to the benchmark GH model: a poltcally organzed domestc 5 Detals on dervaton of equaton (10) are provded n Appendx. 6 It should be noted that wthout the MFN rule, a set of equlbrum tar s for all mporters would be determned by a system of smultaneous equatons wth the number of equatons beng equal to the number of mportng countres. Wth the MFN and the FTA, the number of equatons goes down to two. However, under complete trade lberalzaton agreement, a wthn-fta tar s exogenously set to zero and the second term on the rght-hand sde of (10) vanshes. 7 In the model of monopolstc competton, the e ect of a spec c tar on prce s ampl ed by producer s markup. Therefore, for low the prce elastcty wth respect to tar s hgh and the gan consumer surplus from a subsdy outweghs the ncrease n government s expendture. 8 Ths result s consstent wth other studes, e.g. Bagwell and Stager (1997), Ornelas (2005a,b). 10

11 ndustry receves more protecton from the government. Moreover, the level of protecton s hgher f domestc and mported varetes are close substtutes and f the domestc sector s relatvely large, as the domestc lobby has more to gan from protecton n ths case. The fth term re ects the e ect of poltcal actvty by partner country rms on the natonal trade polcy. I p enters the equaton postvely, makng protecton more lkely n those sectors where partner country exporters are organzed nto lobbyng groups and where product varetes are closer substtutes. Smlarly to the domestc lobby, the e ect of partner country rms lobbyng on the mport tar s proportonal to ther market share. The last term s negatve and re ects the e ect of lobbyng e orts by the ROW rms to reduce protecton. As before, the scalng factor 1 re ects hgher motvaton by the ROW rms to lobby for trade lberalzaton when the degree of substtuton between varetes wthn a gven ndustry s hgh, but unlke domestc and partner country lobbyng, the ROW lobbyng ntensty declnes wth the market share. The ntuton behnd ths result s an ncreased damage from protecton for small ROW ndustres, and as a consequence these ndustres wll resst tar ncrease more ntensvely. 9 As n the GH model, domestc and partner country s lobbyng results n overprotecton and welfare reducton relatve to the rst-best outcome. The presence of an organzed foregn lobby from the ROW may help to (partally) restore the optmal level of mport tar s and rase natonal welfare. However, the presence of the ROW lobbyng alone causes underprotecton and s thus welfare-reducng. Therefore, the overall net e ect from the presence of the partner country lobbyng s lkely to lead to welfare reducton, whereas the overall welfare e ect of the ROW lobbyng actvty s unambguous: the e ect s postve f ROW rms counter-lobby aganst the e ort of domestc and partner country rms to rase protecton and negatve f ROW rms form a sngle organzed lobbyng group n the sector. 4 The data The emprcal secton of ths paper estmates the e ects of domestc, partner country and ROW lobbyng actvty on the Canadan post-nafta trade polcy. Prmarly, I use two measures for trade barrers: mport tar s and the share of mports that s subject to tar or non-tar trade restrctons. The US was treated as a Canadan FTA partner country, whle all other countres that have no preferental trade agreements wth Canada were aggregated nto ROW. The estmaton of 9 Ths result follows from the Cobb-Douglas utlty functon. Fxed product expendture shares mply that the mport tar mposed on one varety wll rase consumer s expendture on all varetes through aggregate prce ndex proportonally to ther market shares. Therefore, the hgher s the ROW market share (and the lower s the share of other varetes), the less harmful s the mport tar for the ROW exporters. 11

12 equaton (10) requres the followng data: the measure for trade protecton, mports by the country of orgn and by sector, domestc output by sectors, substtuton and prce elastctes, poltcal organzaton dummes, and three sets of nstruments for market shares. Ths study s conducted for 249 Canadan 6-dgt NAICS manufacturng sectors (NAICS 31-33) for the perod of Protecton measures and market shares Domestc manufacturng shpments data for 249 NAICS-6 ndustres are provded by Industry Canada. The values of Canadan mports, as well as customs dutes collected, were obtaned from Statstcs Canada at the HTS-10 level and aggregated to NAICS-6 usng the concordances tables from the Internatonal Trade Dvson of Statstcs Canada. In the orgnal formulaton, the GH model was meant to analyze the poltcal economy of mport tar formaton. However, tar rates are often argued to be an mperfect measure of trade protecton for the analyss of endogenous trade polcy formaton n the presence of WTO tar regulaton. Wth lmtatons on the magntude of tar s mposed by the WTO, organzed nterests would seek non-tar protecton from mport competton. The advantage of non-tar barrers (NTBs) as a measure of trade dstorton s ther unlateral adopton by d erent countres, as opposed to tar s that are set cooperatvely n WTO negotatons. Unless organzed ndustres can protect ther nterests durng WTO negotatons, NTBs are a preferred measure for trade protecton. However, measurement d cultes and the qualtatve nature of NTBs are mportant ssues, whch can be a serous problem n a small sample estmaton. In lght of ths, I used tar, NTBs and protecton coverage share as a measure of protecton. Advalorem tar rates were obtaned as the rato of aggregated duty collected by customs over the value of mports. 10 NTBs for Canadan mports were obtaned from the TRAINS database mantaned by UNCTAD, whch shows the proporton of mports that s covered by one or more qualtatve restrctons. These data were avalable at the HS-6 level and were aggregated nto NAICS-6 groups. In addton, the protecton share varable was constructed as the share of Canadan mports that s subject ether to the postve mport tar or NTBs. Descrptve statstcs for protecton measures and market shares are presented n Table 1. In 1997 the average tar rate, NTBs and protecton coverage ratos for the ROW mports were 4.8%, 18.2% and 77.5%, respectvely. Tar s and NTBs are hghly correlated both wthn and outsde of the FTA, whch mples that d erent measures of protecton are stll hghly complementary. 10 Therefore, tar measure controls for some non-tar dstortons as well, such as antdumpng or countervalng dutes. 12

13 4.2 Poltcal organzaton dummes Prevous studes that have tested the GH model emprcally used rm-level poltcal contrbutons to assgn the value for the poltcal organzaton dummy varable. 11 Although these data are avalable for Canada for 1997 and afterwards, ths paper uses a d erent approach. As was prevously mentoned, foregn corporatons prefer drect lobbyng to poltcal contrbutons because transparency of poltcal contrbutons may rase concerns about foregn nterference nto poltcal processes. Furthermore, snce d erent means of poltcal nvolvement are hghly correlated (Hansen and Mtchell, 2000), drect lobbyng seems to be an approprate measure for domestc poltcal actvty as well. In ths work, the degree of poltcal actvty n an ndustry s measured by the number of lobbysts representng the corporate nterests of that ndustry. The Lobbysts Regstraton Act (LRA) requres every ndvdual to regster at the Lobby Regstrar Canada f the person seeks a meetng or a phone call to any publc o ce holder regardng the development, mod caton or cancellaton of legslatve proposals, regulatons, publc polces and programs. The assumpton that poltcal contrbutons wll be ne ectve for the determnaton of trade polcy wthout such contact seems to be reasonable and, therefore, poltcal contrbutons should be followed up by a personal contact wth a polcymaker. For ths reason, the number of regstered lobbysts s used to measure rm-level lobbyng ntensty wthn an ndustry. The man advantage of ths data set s the large amount of detaled nformaton lobbysts are requred to submt. Ths ncludes nformaton on the busness address of a corporaton that bene ts from lobbyng, ts subsdares and headquarters, and the objectve of the meetng wth a publc o ce holder. Ths nformaton s very helpful n determnng the natonalty and ndustral a laton of lobbysts representng nterests of mult-product multnatonal corporatons. Another advantage of ths data set s that t gves a very narrow de nton of a lobbyst. Any person representng hs or her own nterests, and who s not beng pad for arrangng the meetng wth the publc o ce holder, s not oblged to regster. Ths removes nformaton on the very small rms. Large rms, whch have hgh lobbyng power and can e ectvely n uence the decsons of polcymakers, typcally use the servce of professonal consultants or corporate lobbysts, who are requred to regster. Frms were assgned a NAICS-6 ndustry code usng the Canadan Company Capablty database mantaned by Industry Canada. Assgnng an ndustry code to mult-product rms nvolves some degree of dscreton. For example, some rms n the automoble sector operate n more than ten NAICS-6 ndustres. Snce the number of such rms s relatvely small, I assgned prmary, sec- 11 e.g. Goldberg and Magg (1999), Gawande and Bandyopadhyay (2000), Bombardn (2005), Gawande, Krshna, and Robbns (2006). 13

14 ondary and tertary NAICS codes to such rms usng d erent nformaton sources: the Canadan Company Capablty database, the Federal Corporatons Regstry and the North Amerca Compustat database. The databases lsted above allow assgnng ndustry codes to US and ROW rms. The LRA also requres lobbysts to declare a subject-matter n respect to whch an ndvdual undertakes to communcate wth a publc o ce holder. In many cases, nformaton on the purpose of lobbyng actvty reported n the lobbyst regstraton form allowed me to attrbute a rm to a sngle (or a small number of) prmary NAICS code. Natonal a laton of each rm that a partcular lobbyst s representng was determned from two sources. Frst, the lobbyst regstraton form requres regstrants to provde the name and busness address of the parent corporaton and those subsdares whch drectly bene t from the lobbyng. Sometmes, lobbysts provde ncomplete nformaton and n ths case t was complemented by the nformaton from other databases mentoned prevously. Agan, qute often the natonalty of the rm was determned by the objectve secton of the lobbyst regstraton form. 12 There are two more advantages of usng lobbysts regstry data over usng poltcal contrbuton data. Frst, the necessty of reportng the objectve of the contact wth the publc o ce holders allows one to solate e ectvely those rms that lobby partcularly for a change n trade polcy. Ths s especally a problem for domestc lobbysts: on average, only one out of eght lobbysts, representng the domestc manufacturng sector, s concerned wth trade polcy. Therefore, poolng poltcal contrbutons by all domestc rms may cause serous measurement problems for the poltcal organzaton varable. Second, the man channel used by foregn rms to lobby ther nterest n Canada s through local subsdares, whch dstrbute the mported goods wthn Canada. Formally, these rms should be assgned to a trade sector code (NAICS 41-45) and dropped from the sample, but ths would substantally underestmate lobbyng e orts by foregn rms. For each trade rm concerned wth nternatonal trade polcy ssues I used the company pro le and lobbyng objectves nformaton to assgn an approprate manufacturng ndustry code and country. The amendment to the LRA announced n 1995 ntroduced several mportant re nements that made t more desrable to use post-1995 lobbyng data. Frst, for the purpose of transparency, the lobbysts connectons to the publc o ce holders were opened to publc scrutny, and the lobbysts regstry database became avalable for research purposes. Second, ths amendment extended the 12 For nstance, a lobbyst of the Japanese Automoble Manufacturers Assocaton of Canada regstered n Ontaro, Canada was attrbuted to the ROW on the bass of the meetng purpose to secure nternatonal trade for electrc equpment. 14

15 amount of nformaton that must be reported. Most mportantly, t made dsclosed nformaton more complete and relable. For the rst tme lobbysts were oblged to provde all the nformaton, and e ectve enforcement devces were ntroduced to encourage better complance. It extended the power of the Lobbysts Regstrar, whch was authorzed to seek clar caton of nformaton submtted. The regstrar was allowed to conduct an audt of provded nformaton and, when necessary, nvestgate the provded nformaton. Sanctons for volatng the LRA were also ncreased: every ndvdual who contravenes any part of the Act... s lable to a ne not exceedng twenty- ve thousand dollars. On proceedngs by way of ndctment, an ndvdual s lable to a ne of up to one hundred thousand dollars or to mprsonment for a term not exceedng two years, or to both (LRA, Secton 14). For these reasons, the data for poltcal actvty by rms were collected for the 1997 electon cycle and complemented wth the 1996 lobbyng data to take nto account any possble small lag n trade polcy response to lobbyng e orts. Smlarly wth regard to other studes, several thresholds for the number of lobbysts n an ndustry wll be set to determne the values of poltcal organzaton dummes. As a nal note on the lobbysts regstry data, I wll bre y dscuss the relablty of ths database. In 2001 an ndependent study of complance to the LRA was conducted by KPMG Consultng Inc. (2001). The already-regstered lobbysts were asked f they were aware of any non-complance behavor. Reported results ndcate that 50% of consultant lobbysts and 15% of corporate lobbysts were aware of non-regstered lobbyng, whle they evaluated the aggregate complance rate at 70% and 100%, respectvely. In general, complance was perceved to be hgh, although non-complance behavor s stll an mportant ssue. Descrptve statstcs for the number of lobbysts s provded n Table 2 and the dstrbuton of lobbysts across sectors s shown n Fgure The elastcty of substtuton To my knowledge, there are no studes to date that estmate substtuton elastctes for Canadan NAICS-6 ndustres, especally wthn a framework of monopolstc competton. In ths study, substtuton elastctes were estmated usng the approach by Feenstra (1994), recently appled by Broda and Wensten (2006) to a large set of US mported commodtes. Ths approach dent es supply and demand elastctes usng no nstrumental varables, and reles only on the assumpton of ndependent supply and demand dsturbances. The complete dervaton of the estmator s presented n Appendx C. It also ncludes a dscusson of a possble estmaton bas due to a small sample sze and a proposed soluton to that problem. 15

16 The summary statstcs and a hstogram of the elastcty estmates are presented n Table 3 and Fgures 2 and 3. Prce elastctes of the ROW mport demand were calculated usng (9a). As a robustness check, I estmated the substtuton elastctes for NAICS ndustres at varous level of aggregaton and ver ed that more aggregated commodtes are more d erentated: the average value of decreased from 5.85 to 5.34 and 4.56 whle movng respectvely from sx to ve and four dgts NAICS. As another robustness check, I estmated US elastcty of substtuton usng the same procedure for the same year and ndustry class caton. Presumably, Canada and the US should have smlar tastes, and varetes that are close substtutes n Canada should be close substtutes n the US as well. Ths suggeston s supported by Fgure 4, whch compares the percentage devaton from the mean for Canadan and US estmates and shows a very close relatonshp n the percepton of product characterstcs n Canada and US. 4.4 Instrumental varables In equaton (10), market shares are lkely to be determned smultaneously wth the tar rates and should be properly nstrumented. Tre er (1993) proposed to nstrument the mport penetraton rato wth ndustry factor endowments as the measure of comparatve advantage ndependent of the level of protecton. Followng ths approach, a lst of nstruments for the Canadan market share ncludes: the share of productve to non-productve workers, the captal stock n machnery and constructon, nventores, and the consumpton of fuel and electrcty. All of these data are provded by Statstcs Canada. The same lst of nstruments was constructed for the US market share n Canada usng the US Census data. To nstrument the ROW share n the Canadan market, two gravty varables were bult. The rst one s the dstance between Canada and the average exporter. For every product, the par-wse log-dstance between Canada and the exportng country was weghted by the share of ths country n the global export of the product. 13 The total exports by country and by sector were constructed usng the UNCTAD database. For the geographc dstance two measures were used: a smple geographc dstance between two captal ctes, and a dstance weghted by populaton densty and economc actvty wthn each country. These data were taken from the Centre d Etudes Prospectves et d Informatons Internatonales. The ratonale for usng ths dstance varable s the followng: f man producers of a partcular good are located far from Canada, the ROW share n the Canadan market s lkely to be small. The second gravty varable s a log of worldwde exports of a product by all countres ex- 13 Snce the exporter s share on the Canadan market s endogenous, I use the share on global market, assumng t s una ected by Canadan tar rate. 16

17 portng to Canada. The ntuton for ncludng ths varable as an nstrument s the same as for usng GDP n the blateral trade gravty model: the more the ROW countres produce (and export) worldwde, the more they wll export to Canada n partcular, and the larger the ROW share n the Canadan market should be. In prevous tests of the GH model, poltcal organzaton dummes were derved from poltcal contrbutons and treated as endogenous, because contrbutons are drectly related to tar s. In the data set used n ths paper there s no drect lnk between the mport tar rate and the number of lobbysts. However, I nstrument the poltcal organzaton varable as well for the reason of possble measurement error, such as assgnment to the wrong ndustry or mssng nformaton on the number of lobbysts. To nstrument the US and Canadan poltcal organzaton dummes, I use the nformaton on the total number of rms n an ndustry, shares of bg and medum rms, and the CR-4 concentraton rato. The ROW lobbyng ntensty s nstrumented wth the rato of exports by ROW rms to Canada relatve to ther worldwde exports n an ndustry. 5 Estmaton procedure Equaton (10) motvates the followng form of the estmaton equaton: Y = s h + 2 I h s h + 3 I p sp + 4I f [1 sf ] (11)! f Y = " = a a + ; 2 = 1 a + ; 3 = b a + ; 4 = p f c a + In equaton (10) the nverse elastcty was taken on the left-hand sde and both sdes were multpled by 1 because substtuton elastcty s lkely to be measured wth error. Usng (12), the four coe cent estmates of the reduced form (11) can be used to derve four structural parameters of the model. There are several ssues regardng the estmaton of (11) that should be addressed. Frst, the rght-hand sde varables of (11) nclude non-lnear combnatons of endogenous varables. To deal wth ths problem the set of nstrumental varables ncludes the ones descrbed n the Secton 4.4 above, ther square terms and nteractons. 14 Second, usng many nstruments relatve to the sample sze and to the number of nstrumented varables leads to a bas of 2SLS towards OLS results (Hansen, Hausman and Newey, 2006). The thrd problem wth a small rato of endogenous varables to the number of nstruments s a problem 14 Ths IV estmator was proposed by Kelejan (1971). (12) 17

18 of weak nstruments. When the correlaton between the endogenous varable and most of the nstruments s weak, conventonal asymptotcs may provde poor approxmatons for the reduced form estmates and test statstcs, and nferences on the sgn cance of coe cents may be very msleadng. Fnally, Goldberg and Magg (1999) pont to the problem of heteroskedastcty of the error term n (11), whch may be correlated wth the mperfectly measured elastcty of substtuton. Therefore, equaton (11) requres an estmator that wll be robust to the problems of many nstruments, many weak nstruments and heteroskedastcty. Several estmaton procedures that address these problems were proposed recently n the econometrcs lterature. 15 I used LIML wth Bekker (1994) standard error correcton. Hansen, Hausman, and Newey (2006) demonstrated that ths approach has better small sample propertes than 2SLS and s asymptotcally correct n the presence of many nstruments and many weak nstruments. It results n tests wth the correct sze and the tests can be made robust to heteroskedastcty through the use of conventonal robust covarance matrx estmators Results 6.1 Test of a benchmark GH model As a startng pont, I wll present the results on a GH verson of the model wth homogeneous goods to test how well the new data on Canada can t the benchmark model and compare ts performance wth the results of prevous emprcal studes. Snce n the benchmark model markets are perfectly compettve and mport supply s n ntely elastc, there s no reason for foregn rms to partcpate n lobbyng, and n the benchmark case I wll consder only the e ect of domestc lobbyng groups on the home country trade polcy. Followng Grossman and Helpman (1994), the optmal mport tar for a small open economy wth poltcally organzed sector-spec c factors of producton takes the followng form: " f f p f = X h + 1 X h I (13) a + M a + M where X h s a domestc value of shpments and M s a total value of mports. The model predcts that the nverse mport penetraton rato enters the equaton negatvely, whle the coe cent on ts nteracton wth poltcal organzaton dummy s postve. Table 4 represents estmaton results for equaton (13) usng tar s, NTBs and protecton share data as a measure of Canadan trade barrers. 15 Stock, Wrght, and Yogo (2002) and Andrews and Stock (2005) provde an excellent survey of ths lterature. 16 I also tred a correcton of standard errors proposed by Klebergen (2002), but results d er only margnally and I do not report them here. 18

19 Followng Goldberg and Magg (1999), several thresholds were used for the number of lobbysts n the constructon of poltcal organzaton dummes to verfy that the results are not drven by the way these dummes are assgned. In the rst column of Table 4 an ndustry s consdered to be poltcally organzed f t s represented by at least one lobbyst. For the second and thrd columns the threshold s two and three lobbysts, respectvely. Frst, consder the estmaton wth tar s as a protecton measure. The estmates of the regresson model (13) are of correct sgns across all spec catons: poltcally organzed sectors receve more protecton, whle protecton n unorganzed sectors s negatve and ncreases wth mport penetraton. The latter result s statstcally sgn cant at a 5% con dence level. In organzed sectors protecton declnes wth the mport penetraton, but ths result s not statstcally sgn cant. The model estmates are very robust to the way poltcal organzaton dummes are constructed. When trade barrers are measured wth NTBs and protecton share, results correspond closely to those obtaned for the tar equaton: the parameter estmates preserve correct sgns, but are estmated wth less precson when trade barrers are measured wth protecton share. Overall, the estmates of the structural model (13) usng Canadan data are generally n the lne wth the results of the studes by Goldberg and Magg (1999) and Gawande and Bandyopadhyay (2000). The estmates of the structural parameters of the model vary consderably across d erent measures of protecton. However, the varances of these parameters are very hgh and one cannot reject hypotheses that both and a are the same across all spec catons consdered. The fracton of the populaton represented by a lobby,, s estmated to be around 0:6 for the spec catons wth tar s, 0:2 for spec catons wth protecton shares, and greater than one for spec catons wth NTBs. In general, 95% con dence nterval for ncludes the whole [0; 1] nterval and the model does not allow one to obtan a precse measure for. Nevertheless, the obtaned results do not contradct the prevous estmates of by Goldberg and Magg (1999) and Gawande and Bandyopadhyay (2000), who estmated the share of the populaton represented by nterest groups to be around 0:85 and unty, respectvely. The estmates of the government s poltcal bas vary from 10 n the spec caton wth shares to 100 n spec caton wth tar s. The values of the parameter a greater then ten mply that the government assgns approxmately equal weghts to poltcal contrbutons and to a natonal welfare net of poltcal contrbutons, whch supports the results of studes that use US data. 17 Overall, the results of the GH model wth Canadan data are broadly consstent wth those 17 The government s objectve functon C + aw s equvalent to a 1C + a 2 (W C), n whch a = a 2 a 1 a 2, a 1 s the weght on poltcal contrbutons and a 2 s the weght on a welfare net of poltcal contrbutons. Therefore when a s much greater than one, a 1 = a+1 a a2 a 2. 19

20 obtaned by Goldberg and Magg (1999) and Gawande and Bandyopadhyay (2000) and other studes for the US. These results wll serve as a benchmark aganst whch the results of the monopolstc competton model wth foregn lobbyng and FTA partcpaton wll be compared n the next secton. 6.2 Estmaton results for the monopolstc competton model wth foregn lobby In ths secton I present the estmaton results for the poltcal economy model of trade wth monopolstc competton, FTA membershp and two groups of foregn lobbes. The results from the equaton (11) appear n Table 5, n whch several measures are used to measure trade barrers. Columns wth d erent numbers denote d erent threshold levels for constructon of the poltcal organzaton varable. Columns (1) and (2) report the results when an ndustry s assumed to be poltcally organzed f t has at least one and three lobbysts, respectvely. In column (3) a country s j ndustry s organzed f t s represented by at least three lobbysts and accounts for strctly more than one thrd of a total number of lobbysts n that ndustry. Ths measure was constructed to exclude sectors wth a postve but small number of lobbysts relatve to the whole ndustry. Frst, note that for any measure of protecton and poltcal organzaton, the coe cent 1 s postve and almost always statstcally sgn cant, mplyng that domestc ndustres receve a postve level of protecton regardless of poltcal economy factors. Ths s consstent wth the predcton of the model that the welfare-maxmzng government always nds t worthwhle to protect home country producers aganst competng mporters when domestc and foregn products are close substtutes and markets are mperfectly compettve. The postve level of protecton for unorganzed sectors s n contrast to the benchmark GH model and nds strong support n the data. As the theory predcts, among poltcally unorganzed sectors protecton ncreases wth the share of domestc rms on the market. When protecton s measured wth tar s, the pont estmate for 1 n the most preferred spec caton n terms of the log-lkelhood functon (column (3)) s 0:21, whch converts to the welfare-maxmzng ad-valorem mport tar of 2:2% for an average Canadan ndustry, gven the average Canadan market share, prce and substtuton elastctes of 0:66, 5:32 and 5:83, respectvely. In terms of the optmal level of NTBs and protecton share, the welfaremaxmzng NTB coverage for the average ndustry s estmated to be 8:9% of total mports from outsde of the FTA, whle the welfare-maxmzng share of mports subject to any trade restrcton s 17:9%. The e ect of a poltcally organzed domestc lobbyng (coe cent 2 ) s always estmated to be postve and very sgn cant, ndependently of the constructon of the poltcal organzaton 20

21 dummy and the measure of trade dstorton. Everythng else beng equal, actve domestc lobbyng n the ndustry leads to a hgher level of protecton and ths e ect s sgn cant and robust across all spec catons. The presence of a poltcally organzed domestc lobby tends to ncrease mport tar s by 5:4% for the average ndustry, the NTB coverage rato by 29:7%, and the protecton share by 31:7%. The novel results of ths secton are the estmates of coe cents 3 and 4. The coe cent 3 measures the e ect of the FTA partner country s lobbyng and s always estmated to be postve, although n spec catons wth tar s t s only margnally sgn cant. The e ect of the ROW lobbyng (coe cent 4 ) s always negatve and sgn cant at 5% except for two spec catons wth NTBs. Accordng to these results, the presence of poltcally organzed US ndustres n Canada leads to hgher Canadan mport barrers and the ROW lobbyng e ort s negatvely correlated wth the Canadan protecton measures. Thus, t seems safe to conclude that, whle our results do not allow researchers to get a precse estmate of 3, they strongly support the hypothess of foregn lobby d erentaton wth respect to market access mpled by the FTA and predcted by the theoretcal model of Secton 3. The pont estmates of coe cent 3 n spec catons wth the best data t are 0:345 (column (3) for tar s), 3:033 (column (3) for NTBs) and 5:041 (columns (2) for protecton share). These estmates mply that a poltcally organzed ndustry n the FTA partner country tends to ncrease a country s average mport tar by 1:2%, average NTB coverage by 10:5% and average protecton coverage by 17:4%. Smlarly, the presence of a poltcally organzed group of exporters from outsde of the FTA tends to decrease the average Canadan mport tar by 3:4%, the average NTB by 9:8% and the average protecton share by 34:9%. These numbers are economcally plausble and provde addtonal support to the estmates of the model. Estmates of equaton (11) allow one to derve the values of the structural parameters of the model. As expected, absolute magntudes of coe cent 4 are always less than 2, whch mples that ROW lobbyng s relatvely less mportant for Canadan polcymakers than lobbyng by domestc rms (c < 1). Ths result s not surprsng gven the legslatve restrctons on contrbutons by foregn rms n Canada. In spte of that, polcymakers valuaton of contrbutons from abroad s stll postve and sgn cantly d erent from zero. The estmates of the government s preferences towards poltcal contrbutons from the FTA partner country (parameter b) depend on the protecton measure that US rms are tryng to a ect. In lobbyng for tar s, partner country contrbutons are estmated to be 35 80% less e ectve than domestc contrbutons, whle n lobbyng for NTBs the e ectveness of domestc and partner country lobbyng s approxmately equal. Fnally, when trade dstortons are measured by protecton share, the government s valuaton of poltcal contrbutons from the FTA partner country relatve to domestc ones rses to 1:63, but the hypothess that b 1 21

22 can never be rejected. At the same tme, US contrbutons seem to be more mportant than contrbutons from the ROW. Dependng on the spec caton, the government s valuaton of US contrbutons s estmated to be two tmes the valuaton of contrbutons from other countres. Among possble reasons are greater proxmty of the US and Canadan nancal systems and a more sophstcated mxture of asset structure that makes t more d cult to dstngush between Canadan and US rms (relatve to the dstncton between Canadan and ROW rms). The estmates of the populaton share organzed nto lobbyng vary substantally across d erent spec catons. In tar equatons, the parameter s always greater than one, but the standard error s very large and 95% con dence ntervals almost always overlap wth the [0; 1] nterval. In equatons wth NTBs and protecton shares, pont estmates for are also very mprecse and n all spec catons one cannot rule out the possblty that s negatve. 18 As for the relatve mportance of natonal welfare for the government, the parameter a s estmated to be wth reasonable degree of precson as compared to the benchmark GH model. It mples that when organzed nterests lobby for a change n tar polcy, the government s valuaton of poltcal contrbutons s three to ve tmes hgher than the valuaton of natonal welfare net of contrbuton. The result that polcymakers are drven mostly by poltcal contrbutons they receve from d erent lobby groups sharply contrasts wth results obtaned prevously n a perfectly compettve setup: Gawande and Bandyopadhyay (2000) estmated a to be over 3,000; n Goldberg and Magg (1999) ts value s around 70. My own estmates for Canada fall n the range (see Secton 6.1). However, Gawande and Bandyopadhyay (2000) recognzed that hgh estmates of a contradct the emprcal evdence that welfare loss from protecton s always greater than the amount of poltcal contrbutons polcymakers receve n exchange for protecton. Intutvely, a < 1 s what we should expect to get n the emprcal model, because an ncrease n pro t from a 1% ncrease n tar (.e., the maxmum contrbuton that the government could get for protecton) s always lower than the correspondng decrease n welfare. Therefore, n order to observe any postve tar above the welfare maxmzng level n the equlbrum, the government should put a stronger emphass on contrbutons relatve to welfare and thus a < 1 s a desrable feature of the emprcal model. Estmates of the reduced form (11) shown n Table 5 have a meanngful economc nterpretaton for an average Canadan ndustry. Gven the average market shares of the three groups of rms, prce and substtuton elastctes, one can back up the welfare maxmzng level of protecton and 18 It should be ponted out that that measurng s a standard problem n the lterature and ts estmaton proved to be d cult. 22

23 the average e ect of each lobby group on Canadan trade polcy. Table 6 presents the average and margnal e ect of each lobby group on d erent measures of protecton. Margnal e ects are calculated as the e ect of a partcular lobbyng group on an ndustry wth average characterstcs, and are dscussed earler n ths secton. Average e ects are calculated as a smple average of the tted values of the dependent varable mpled by the model estmates. As a result, the average Canadan ndustry receves 1:33% mport tar s, 7:31% NTB coverage and 7:81% protecton share from lobbyng by domestc rms; 0:21% tar s, 1:87% NTBs and 4:26% protecton share from lobbyng by partner country rms; 0:48% tar s, 1:38% NTBs and 6:24% protecton share from lobbyng by ROW rms. These results are economcally reasonable and statstcally sgn cant. 6.3 Robustness Ths secton explores the robustness of the prevous results to the treatment of foregn lobbes operatng n Canada. Table 7 represents a senstvty analyss of the benchmark model results when one cannot d erentate between domestc and foregn lobbyng actvty and attrbutes all or part of poltcally actve foregn rms to domestc ndustry lobbyng. Table 7 only shows the estmates of the equaton n tar s. The rst three columns replcate the results for the benchmark GH model (wth a Canadan lobby only) from Table 4 to provde a standard for comparson. The next three columns report spec catons n whch US rms that are poltcally actve n Canada are treated as Canadan. Interestngly, the results reman qualtatvely the same as n the benchmark case, wth only a small change n parameter estmates and a small ncrease n the lkelhood functon. In the last three columns of Table 7 all foregn rms that are poltcally actve n Canada are treated as Canadan, and they show that the results are not robust to the treatment of the ROW lobbyng. Coe cents on Xh M I are based towards zero and measured wth a larger error, re ectng the preferences of organzed foregn lobbes for protecton removal. Ths result s preserved when other protecton measures are used and should not be surprsng gven con ctng nterests of domestc and ROW lobbyng groups. Table 8 shows how robust the results of the monopolstcally compettve model are to the excluson of one or two foregn lobby factors. As n the prevous secton, I only consder the tar equaton here for two reasons. Frst, results are qualtatvely the same when the NTB or protecton share are used to measure trade dstortons. Second, equatons n tar s gve hgher precson and t the data better. The rst three columns replcate the results from Table 6 for comparson. The estmaton results n Table 8 suggest that ncludng the e ect of the US and ROW sgn cantly mproves the t of the model. However, to a large extent, ths s due to the ncluson of the ROW lobby: the lkelhood rato test always rejects the model wthout ROW lobbyng. At the same tme, 23

24 omttng the lobbyng actvty by US rms reduces the log-lkelhood, but the lkelhood rato test fals to reject the model wthout US lobbyng. Another mportant mplcaton of Table 8 s that the estmates of 1 and 2 and thus the structural estmates of the model are not a ected serously when foregn lobbyng s excluded. The second set of robustness checks repeats the same thought experment used for the benchmark model, by assumng that one cannot dstngush domestc from foregn lobbyng actvty. In columns A and B of Table 9 the US rms that are poltcally actve n Canada are treated as Canadan. Furthermore, n column A the lobbyng actvty by ROW rms s gnored. Fnally, n column C all foregn rms that are poltcally actve n Canada are assumed to be Canadan. Results of Table 9 suggest that the nablty to dstngush Canadan lobbyng from US lobbyng tends to overestmate the e ect of domestc lobbyng and ts average e ect of mport tar, but does not make a bg e ect on the model s t, unless ROW lobbyng s excluded. However, the estmates of Table 8 are not robust to the treatment of the ROW lobbyng ntensty: not only does the varance of estmators ncrease, but also the estmates of 2 are based downwards. Ths result s not partcularly surprsng gven the fact that the lobbyng objectves of ROW rms are opposte to those of home country rms. At the begnnng of ths paper I noted the concern that the nablty to separate the e ect of foregn lobbyng from domestc may lead to msleadng results even when a proper set of nstruments s used. Ths concern nds strong support n the data. 6.4 The comparson of the benchmark wth monopolstc competton GH models To test two competng versons of the GH model estmated by nstrumental varables, I used a J-Test for non-nested hypothess testng (Davdson and MacKnnon (1993), p. 388). The testng procedure requres the same dependent varable, so for the purpose of ths secton both sdes of (11) were multpled by 1, and 1 was moved to the rght-hand sde. The alternatve benchmark spec caton (13) remans unchanged. As n Davdson and MacKnnon (1993), let f(s j ; Ij ; ) be a LIML regresson model for the monopolstc competton spec caton (11), and let g(s h ; Ih ) be a LIML regresson model for the benchmark spec caton (13). The valdty of the benchmark GH model can be tested usng an augmented LIML regresson eg = g(s h ; Ih ) + 1 b f, where b f s constructed from tted values of the LIML estmaton of (11). The benchmark GH model s rejected n favor of the monopolstc competton model f the test for 1 = 0 s rejected. In a smlar way, the monopolstc competton model s rejected n favor of the benchmark GH f n the regresson model e f = f(s ; I j ; ) + 2 bg the hypothess 2 = 0 s rejected. 24

25 The J-test results for three measures of protecton are presented n Table 10. They suggest that nether model can be rejected as a correct one when protecton s measured wth tar s. However, the benchmark GH model s always weakly rejected n favor of the model wth monopolstc competton as the monopolstcally compettve spec caton of the model adds more explanatory power. The second part of each secton of the table dsplays J-test results for the benchmark model and monopolstc competton wth only domestc lobbyng, usng the same set of nstruments. Agan, nether model can be rejected, but these results suggest that smply takng nto account the mperfectly compettve structure of the market mproves the explanatory power of the Protecton for sale model. When protecton s measured wth NTBs, J-test results ndcate that the benchmark model s rejected n favor of the monopolstc competton model wth foregn lobbes and weakly rejected n favor of the model wthout foregn lobbes, whle the monopolstc competton verson s never rejected. Ths mples that the monopolstcally compettve model developed n ths paper captures more nformaton than the model wth perfect competton. 7 Concluson The man objectve of ths paper s to study the e ect of foregn lobbes and FTA formaton on a country s trade polcy. Ths paper accomplshes ths goal by modfyng the orgnal GH model through the ntroducton of a monopolstcally compettve market structure to allow for the presence of two types of foregn lobbes: lobbes from the FTA partner country and from the ROW. The paper shows that sgnng a free trade agreement wth a bg country leads to a welfare-reducng ncrease n mport tar s n the presence of foregn lobbyng, n partcular lobbyng by an FTA partner country. The emprcal results suggest that the GH model wth monopolstc competton has more explanatory power than ts orgnal verson. For Canada, the e ect of foregn lobbyng s statstcally sgn cant and s n lne wth theoretcal predctons of the model: presence of an organzed lobbyng group n the FTA partner country tends to rase mport tar s, whle an organzed lobbyng group of exporters from other countres s assocated wth lower mport tar s. A major bene t of the approach used n ths paper s a more plausble set of values obtaned for the government valuaton of poltcal contrbutons. Contrary to prevous studes, the result that the government values poltcal contrbutons more than natonal welfare s very persstent and explans why relatvely small poltcal contrbutons have sgn cant e ect on trade polcy through lobbyng. 25

26 Appendx A: Dervaton of the Equlbrum Trade Polcy The equlbrum trade polcy s derved by maxmzng the sum of the government and lobbyng groups welfare functons (9): = NX =1 I H W H + aw + NX =1 I P W P + NX =1 I ROW W ROW W H = n H H + (T R + CS) - welfare of the domestc ndustry W j = n j j W = P (nh H, j = fp; ROW g - welfare of the partner country and ROW ndustry from trade ) + T R + CS - natonal welfare Before dervng the optmal trade polcy, calculate the responsveness of equlbrum prces and quanttes to the change n the ROW tar. An ncrease n the ROW tar wll not a ect equlbrum prces of domestc and partner country rms drectly, but wll ncrease equlbrum quanttes q j through an ncrease n the aggregate prce ndex P. Frms from the ROW wll have a drect negatve e ect on equlbrum quantty through an ncrease n consumer prce (6) and an ndrect postve e ect through an ncrease n the aggregate ndustry prce f = ( 1) f for j = H; P ROW = 1 p ROW + ( 1) ROW (2a) For future convenence, de ne the share of country j rms on Canadan market for the product : s j = nj pj xj P X = nj pj xj! (3) = n j dj p j P! 1 (3a) The e ect of the ROW tar rate on ndustry prce ndex can be ROW ROW = n ROW d ROW P (3a) p ROW = 1 1 srow P p ROW Now we can calculate the e ect of the ROW tar change on welfare terms n the objectve functon (9). The f = 1 nrow x ROW s a common factor that I want to solate n every equaton. Tar revenue: T R = X n P P x P + n ROW ROW x ROW (4a) (5a) 26

27 Consumer ROW ROW ROW n P P ( = n P P + n ROW x ROW 1)q P P X + 1 ( 1) P p P s P + 1 U (1) = X 0 + X + ROW p ROW + ROW p ROW + n ROW ROW ROW + ( 1)s ROW = (! ln X 1)s ROW (6a) Indrect utlty: V (2) = Y CS = P! (ln! 1) Pro t functons: P! + P! ln P! ROW! P = Y + P! (ln! 1) ROW ROW P! ln P (7a) = nrow n j q ROW ROW p ROW (7;1a) = nj p j ( 1) ROW = 1 s ROW n ROW (7) = ROW 1 qrow + nrow p ROW h s ROW + ( X s ROW = 1 n j pj P X ROW ; for j = H; ROW = 1 (2a;4a) = = s ROW ROW (8a) Substtutng (6a), (7a) and (8a) nto the dervatve of (9) wth respect to the ROW tar rate ROW + = (a + ) ( 1) P p P (I H + a) 1 s H + bi P s P ROW p ROW 1 s P + ci ROW 1 ( 1)s ROW + s ROW ROW Rearrangng and solatng the ROW tar rate on the left-hand sde we the obtan optmal trade polcy: ROW " p ROW = ( 1) P p P s P 1 + IH + a a + 1 s H + bip a + 1 Where " s the prce elastcty of demand for the ROW mports: s P + cirow a + 1 = 0 s ROW 1 (14) ROW p ROW q ROW ROW p ROW q ROW = ( 1)s ROW 27

28 Appendx B: Estmaton of the Elastctes of Substtuton To smplfy notaton, tme ndex s omtted n ths secton. We start wth the descrpton of supplydemand system. From equaton (3), total demand for the mports of product mported from country j equals: x j =! d j p j p j P! 1 (3) Snce usng expendture shares nstead of quanttes reduces measurement error, the above equaton s transformed nto shares: s j = pj xj P X = pj xj! = d j p j P! 1 (1b) Takng logs and tme d erencng, we obtan the new demand equaton: ln s j = ( 1) ln P ( 1) ln p j + "j (2b) The error term n (1b) s lkely to be correlated wth the market shares due to smultanety n determnaton of market shares and product prces. To model ths smultanety, followng Feenstra (1994) let the supply equaton for varety j of good to take the form: p j = exp(j )(xj )e where e s the nverse supply elastcty and j s a technology parameter. Takng logs and tme d erencng we obtan a mod ed supply equaton: ln p j = e ln(x j ) + j (3b) To derve the supply equaton n the form of expendture shares, substtute the expresson for quantty suppled from (1b) nto (3b): ln p j = e e + 1 ln(!j ) + e e + 1 ln(sj ) + j ; j = 1 e + 1 j (4b) To facltate estmaton t s convenent to elmnate product-spec c xed e ects from supply and demand equatons through takng the d erence n (2b) and (4b) wth respect to some reference country k: Where ex j = ( ln x j es j = ( ep j = 1)ep j + e" j e e +1 es j + e j (5b) ln k ) for any varable x. Although n the model I aggregated all non- NAFTA countres nto the ROW, n ths secton I treat the output of each country as a separate 28

29 product and assumed that the number of varetes equals to one (Canadan product) plus the number of mportng countres. Wth the assumpton that the ROW rms are symmetrc, treatng product of each country as a separate varety does not a ect predctons of the model. Therefore, after d erencng, the number of cross-sectons n equaton (5b) equals to the number of countres that mport product nto Canada. For consstent estmaton of substtuton elastcty parameters, the assumpton of ndependent error terms n (5b) s crtcal,.e. technologcal factor s assumed to be ndependent of the taste for varety parameter: 1 T T!1 Assumpton 1: p lm P t e j e" j = 0 for 8 product Wth ths assumpton, solate error terms on the rght-hand sde of (5b) and multply them through: Several authors 19 1 = Y j = 1 X 1 j + 2 X 2 j + u j u j = ( Y j = ep 2 j X 1 j = es2 j X 2 j = ep jes j 1) 1 e" j e j e (e + 1)( 1) ; 2 = e 1 (e + 1)( 1) (6b) (7b) showed that a consstent estmates of 1 and 2 can be obtaned by the IV estmator, where nstruments are country ndcators. across varetes for each product: Y = 1 X X 2 + u To demonstrate ths result, append (6b) The total number of observatons n (8b) s T C, where T s the number of perods n the sample and C s the number of countres that mport product nto Canada. Introduce T C C matrx of nstrumental varables where each column j s an ndcator varable for the mportng country j, such that vector e j s a T -dmensonal vector of 1 s: 2 3 e e Z = e C Then, for the xed number of mportng countres Z s a set of vald nstruments as T! 1. Frst, 1 P Assumpton 1 guarantees that p lm t Z 0 u = 0 holds for any product. The product 1 T Z 0 u s T T!1 19 For example Angrst (1991) and Angrst, Imbens, and Krueger (1999). (8b) 29

30 the C -dmensonal vector wth elements 1 T e j e" j, each approachng zero under the Assumpton 1. Independence of supply and demand error terms mples that nstruments Z are uncorrelated wth 1 P the error n (8b). Feenstra (1994) also showed that p lm t Z 0 X has a full column rank under a T T!1 techncal restrcton that the demand and supply curves should vary across countres,.e. for each product there should be some varance ether n taste or n productvty parameters across varetes. 1 P Therefore, the rank condton for dent caton s sats ed. Together wth p lm t Z 0 u = 0, ths T T!1 condton guarantees that the IV estmator wth nstrument matrx Z s consstent. To make the estmator dentcal to the one used n Feenstra (1994) and Broda and Wensten (2006), pre-multply equaton (8b) wth Z(Z 0 Z) 1 Z 0 to obtan: Y = 1 X X 2 + u (9b) where upper bars denote tme averages. The OLS estmator on (9b) s dentcal to the IV estmator on (8b), thus, a consstent estmator of the reduced form parameters 1 and 2 can be obtaned by averagng equaton (7b) over tme and estmatng t wth the weghted OLS for each product XC XC t j XjX 0 j t j XjY 0 j A (10b) j=1 Weghts t j, the number of years country j was exportng product to Canada, are used to compensate for the d erent number of observatons (number of years) for each mportng country. Even though estmator (10b) s consstent as T goes to n nty, several authors pont to ts bad performance n small samples where Assumpton 1 may not hold. 20 The data set used n ths work covers the perod from 1988 to 2004, whle the average number of mportng countres per product equals 80. The number of cross-sectons s, probably, too large relatve to the number of years to rely on asymptotc theory. The problem of a small sample can be solved usng a correcton method smlar to the Deaton (1985) error-n-varables estmator. Let Y and X to be the unobserved populaton means, and Y and X to be ther sample estmates. Assume the followng sample structure: 0 1 Y A Y X X The normalty assumpton s not crtcal here. sample moment matrx for X, and XY equaltes must hold: ; j=1 y xy xy 1 A Let to be a moment matrx for X, XX a a cross moment matrx for Y and X. The followng E( XX ) = + 20 Stager and Stock (1997) and Bound, Jaeger, and Baker (1995). 30

31 E( XY ) = + xy The OLS estmator for can be wrtten as follows: b = (X 0 X ) 1 (X 0 Y ) = ( XX )( XX xy ) = (X 0 X C ) 1 (X 0 Y C xy ) (11b) The d erence between ths estmator and WLS estmator (10b) s that the sample counterparts for varances n xy and can be calculated usng all T C observatons, and estmator (11b) s consstent for C! 1 holdng T xed. Fnally, the estmator I used s the WLS estmator (10b) wth the small sample correcton (11b): 0 XC b t j XjX 0 j j=1 1 j b A 1 0 C j=1 t j X 0 jy j 1 b xy A (12b) And b xy and b j are the sample counterparts for xy and : b j = 1 P tj t j 1 b xy;j = 1 t j 1 k=1 X 0 jk X j P tj k=1 X 0 jk X j X jk Y jk X j Y j The matrx of covarates X also ncludes a constant term to address the problem of measurement error n prces. On the left-hand sde of (6b) we have second moment for prces, whch equals to the varance of true prces plus the varance of the measurement error. Thus, the constant term wll absorb the measurement error n prces. Snce market shares are not correlated wth unt values errors, explanatory varable wll not be a ected by the measurement error. Therefore, ncluson of a constant term n X makes the estmator (12b) robust to the measurement error n prces. Once we obtan the vector of estmates b, we can derve the structural parameter of nterest from (7b): 2 = 1 + (13b) 2 + q To avod the magnary values, the condton must hold. In fact, ths condton was sats ed every sngle tme the estmate for was statstcally sgn cant. 31

32 References Andrews, D. W., and J. H. Stock (2005): Inference wth Weak Instruments, NBER Techncal Workng Papers 0313, Natonal Bureau of Economc Research, Inc. Angrst, J. D. (1991): Grouped-data estmaton and testng n smple labor-supply models, Journal of Econometrcs, 47(2-3), Angrst, J. D., G. W. Imbens, and A. Krueger (1999): Jackknfe Instrumental Varables Estmaton, Journal of Appled Econometrcs, 14(1), Bagwell, K., and R. W. Stager (1997): Multlateral Tar Cooperaton Durng the Formaton of Free Trade Areas, Internatonal Economc Revew, 38(2), Bekker, P. A. (1994): Alternatve Approxmatons to the Dstrbutons of Instrumental Varable Estmators, Econometrca, 62(3), Bohara, A. K., K. Gawande, and P. Sangunett (2004): Trade dverson and declnng tar s: evdence from Mercosur, Journal of Internatonal Economcs, 64(1), Bombardn, M. (2005): Frm Heterogenety and Lobby Partcpaton,. Bond, E. W., R. G. Rezman, and C. Syropoulos (2004): A strategc and welfare theoretc analyss of free trade areas, Journal of Internatonal Economcs, 64(1), Bound, J., D. Jaeger, and R. Baker (1995): Problems wth Instrumental Varables Estmaton When the Correlaton Between Instruments and the Endogenous Explanatory Varable Is Weak, Journal of the Amercan Statstcal Assocaton, 90(430), Broda, C., and D. E. Wensten (2006): Globalzaton and the Gans from Varety, The Quarterly Journal of Economcs, 121(2), Choate, P. (1991): Agents of In uence: How Japan Manpulates Amerca s Poltcal and Economc System. NY: Knopf. Davdson, R., and J. MacKnnon (1993): Estmaton and Inference n Eeconometrcs. New York, Oxford Unversty Press. Deaton, A. (1985): Panel data from tme seres of cross-sectons, Journal of Econometrcs, 30(1-2),

33 Feenstra, R. C. (1994): New Product Varetes and the Measurement of Internatonal Prces, Amercan Economc Revew, 84(1), Gawande, K., and U. Bandyopadhyay (2000): Is Protecton for Sale? Evdence on the Grossman-Helpman Theory of Endogenous Protecton, The Revew of Economcs and Statstcs, 82(1), Gawande, K., P. Krshna, and M. J. Robbns (2006): Foregn Lobbes and US Trade Polcy, The Revew of Economcs and Statstcs, 88(3), Goldberg, P. K., and G. Magg (1999): Protecton for Sale: An Emprcal Investgaton, Amercan Economc Revew, 89(5), Grossman, G. M., and E. Helpman (1994): Protecton for Sale, Amercan Economc Revew, 84(4), Hansen, C., J. Hausman, and W. Newey (2006): Estmaton wth Many Instrumental Varables, Dscusson paper, workng paper. Hansen, W., and N. Mtchell (2000): Dsaggregatng and Explanng Corporate Poltcal Actvty: Domestc and Foregn Corporatons n Natonal Poltcs, Amercan Poltcal Scence Revew, 94(4), Kelejan, H. (1971): Two-Stage Least Squares and Econometrc Systems Lnear n Parameters but Nonlnear n the Endogenous Varables, Journal of the Amercan Statstcal Assocaton, 66(334), Klebergen, F. (2002): Pvotal Statstcs for Testng Structural Parameters n Instrumental Varables Regresson, Econometrca, 70(5), KPMG Consultng Inc. (2001): Study on Complance under the Lobbysts Regstraton Act, Prepared for O ce of the Ethcs Counsellor. Ottawa. September 14, Mtchell, N. (1995): The Global Polty: Foregn Frms Poltcal Actvty n the Unted States, Polty, 27(3), Ornelas, E. (2005a): Endogenous free trade agreements and the multlateral tradng system, Journal of Internatonal Economcs, 67(2), (2005b): Rent Destructon and the Poltcal Vablty of Free Trade Agreements, The Quarterly Journal of Economcs, 120(4),

34 Rchardson, M. (1993): Endogenous protecton and trade dverson, Journal of Internatonal Economcs, 34(3-4), Stager, D., and J. H. Stock (1997): Instrumental Varables Regresson wth Weak Instruments, Econometrca, 65(3), Stock, J. H., J. H. Wrght, and M. Yogo (2002): A Survey of Weak Instruments and Weak Ident caton n Generalzed Method of Moments, Journal of Busness & Economc Statstcs, 20(4), Trefler, D. (1993): Trade Lberalzaton and the Theory of Endogenous Protecton: An Econometrc Study of U.S. Import Polcy, Journal of Poltcal Economy, 101(1),

35 Fgure 1: Dstrbuton of the number of lobbysts across sectors Fgure 2: Dstrbuton of the number of lobbysts across sectors 35

36 Fgure 3: Dstrbuton of substtuton elastctes Fgure 4: Comparson of Canadan and US estmates for the elastcty of substtuton 36