Protecting India s Trade Liberalization? Tariff Reform, Antidumping and Safeguards

Transcription

1 Protectng Inda s Trade Lberalzaton? Tarff Reform, Antdumpng and Safeguards Chad P. Bown Brandes Unversty Patrca Tovar Brandes Unversty 30 October 2007 Prelmnary and ncomplete: please do not quote, cte or crculate JEL No. F13 Abstract Inda undertook a major unlateral lberalzaton to ts tarff polces n the 1990s. Over the subsequent decade, Inda has also transformed from beng a non-user to beng the most frequent user of two of the WTO system s major exceptons to lberal trade.e., antdumpng and safeguard polces. Ths paper addresses two questons wth potentally mportant mplcatons for other areas of economc research: whether Inda s ncreasng resort to such polces s economcally mportant (.e., whether Inda s smply replacng hgh tarffs wth new antdumpng and safeguard mport restrctons); and whether the resort to such new polces s related to the magntude of Inda s tarff reform. Frst, we use a newly provded data source to comprehensvely examne Inda s product-level applcaton of antdumpng and safeguard mport polces over the perod. We then provde structural estmates from a Grossman and Helpman (1994) model that s modfed to examne determnants of Indan antdumpng and safeguard use. We fnd that larger tarff cuts between 1990 and 1997 are assocated wth an ncreased lkelhood of antdumpng and safeguard protecton over Robustness checks from a reducedform specfcaton confrm ths relatonshp and suggest that t s economcally mportant a one standard devaton ncrease n the sze of the tarff cut away from the mean ncreases the predcted probablty by almost 50% of new antdumpng or safeguard use. Fnally, we nterpret the mplcatons of our results for research lookng at the effect of Indan mport polcy reform on Inda s economc development, as well as the queston of whether use of such polces may have allowed Inda to sustan ts overall movement toward lberalzaton. Keywords: Trade Polcy, Inda, Tarff Reform, Antdumpng, Safeguards Bown: Department of Economcs and Internatonal Busness School (Brandes Unversty). Correspondence: Department of Economcs, MS021, Brandes Unversty, Waltham, MA , USA tel: , fax: emal: cbown@brandes.edu, web: Tovar: Department of Economcs and Internatonal Busness School (Brandes Unversty). Correspondence: Department of Economcs, MS021, Brandes Unversty, Waltham, MA , USA tel: , fax: , emal: tovar@brandes.edu, web Bown acknowledges fnancal support from the World Bank. The authors thank Olver Cadot, Jean-Mare Grether, and Marcelo Olarreaga for sharng data. All remanng errors are our own.

2 1 Introducton Perhaps because of ts sze, subsequent growth, and pattern of successes, the Indan economc reforms of the 1990s are an mportant area of research. Nevertheless, wthn the study of Inda s broader economc reforms, a relatvely nascent lterature has begun to examne Inda s trade lberalzaton experence and the relatonshp between the pattern of dramatc mport market openng and other fundamental changes n the Indan economy. 1 At the same tme that Inda was undertakng a transformatve reform to ts mport polcy regme va the dramatc reducton of mport tarffs and radcal dsmantlng of GATT/WTO nconsstent non-tarff barrers to trade such as mport lcensng and quanttatve restrctons Inda also transformed from beng a non-user of mport-restrctng polces such as antdumpng and safeguards to becomng the WTO system s foremost user of these alternatve forms of mport protecton. Nevertheless, the potental lnk between Inda s mport tarff reform n the early 1990s and the subsequent Indan ncrease n use of antdumpng and safeguards s not yet well examned nor understood. The foremost goals of ths paper are to provde an emprcal approach to examnng the use of such polces, a thorough examnaton of the Indan experence of antdumpng and safeguards use from a newly avalable dataset, and an understandng of the mplcatons of ths use for the sze and composton of Inda s overall trade lberalzaton reform. As a research topc, any lnk between Inda s tarff reform and the use of new forms of mport protecton va antdumpng and safeguards s mportant for a number of reasons. Frst, f Inda smply replaced ts pre-reform structure of mport tarff protecton wth new (and potentally more costly) nstruments of protecton such as antdumpng and safeguards, studes could be msmeasurng the effect of mport market access lberalzaton n the 1990s that s beng used n other areas of research examnng Inda s more fundamental economc development e.g., research on growth, poverty, productvty and 1 Examples of recent studes of Inda examnng such lnks nclude the relatonshp between lberalzaton and ndustry/frm productvty (Krshna and Mtra, 1998; Topalova, 2004), poverty (Topalova, 2005), the demand for labor (Hasan, Mtra, and Ramaswamy, 2007) as well as chld labor (Edmonds, Pavcnk and Topalova, 2007). 1

3 labor demand. Second, s there a relatonshp between Inda s tarff reform program of the early 1990s and ts subsequent resort to antdumpng and safeguards over the latter half of the decade and nto the 2000s? One of the arguments for ncluson of these sorts of exceptons nto multlateral trade agreements lke the WTO are that they allow a country lke Inda a safety valve to escape from certan oblgatons and temporarly protect certan ndustres (that have mposed poltcal pressure) whle mantanng an overall level of openness n the economy. Ths paper attempts to address a number of questons related to Inda s tarff reform and ts subsequent resort to antdumpng and safeguard polces. Frst, we provde a comprehensve examnaton of the Indan antdumpng and safeguard data we llustrate the targeted sectors and products, the targeted foregn exporters, and the types and duraton of measures mposed. Then we provde two formal emprcal approaches for estmatng the relatonshp between Inda s mport tarff lberalzaton undertaken n the early 1990s and ts subsequent use of antdumpng and safeguards polces. We begn the formal emprcal analyss by estmatng the Grossman and Helpman (1994) poltcal economy model of determnants of Indan antdumpng and safeguard use over the early 2000s. Estmates of the protecton for sale model sutably modfed to allow for product-level requests for new antdumpng and safeguard protecton suggests an economcally mportant relatonshp between the depth of the mport tarff cut between 1990 and 1997 and Indan use of antdumpng between 2000 and As a robustness check, we also provde evdence of ths relatonshp va estmates from a reduced-form model of Indan antdumpng and safeguard ntatons over exporter-product pars. The rest of ths paper proceeds as follows. Secton 2 descrbes the nsttutonal backdrop of Inda s tarff reform n the 1990s and the subsequent rse of ts use of antdumpng and safeguards. Secton 3 llustrates how we modfy the standard Grossman and Helpman (1994) model to estmate the relatonshp between Indan use of antdumpng and safeguards and ts tarff reform as well as descrbng the data we use n the estmaton. Secton 4 presents our baselne estmates from the Grossman and Helpman model and frst round of senstvty analyss. Secton 5 presents an alternatve estmaton 2

4 framework of a reduced-form estmaton exercse further documentng the robustness of our results of the lnk between Inda s tarff cuts and subsequent resort to antdumpng and safeguards. Secton 6 concludes. 2 Inda s Tarff Reform and ts Antdumpng and Safeguards Laws 2.1 Trade Lberalzaton n Inda Between 1947 and the late 1980s, Inda followed an nward-orented development strategy. It was characterzed by mport protecton, complex ndustral lcensng requrements, sgnfcant nterventon n fnancal markets and government ownershp of heavy ndustry (Cerra and Saxena, 2002). Internatonal trade was sgnfcantly restrcted by hgh tarffs and non-tarff barrers, whch ncluded mport lcensng, state monopoly of some mports and exports ( canalzaton ), government purchases that favored domestc producers, and restrctons on mports by ntermedares. A combnaton of external shocks n the late 1980s and early 1990s led to larger macroeconomc mbalances. Increased mport costs due to hgh ol prces, a decrease n remttances from Indan workers because of the conflct n the Mddle East, weak demand n export markets, a deteroraton of the fscal poston and the current account defct, hgh external debt, and rsng poltcal uncertanty led to a loss of confdence by nvestors and captal outflows. The loss of nternatonal reserves contnued and ended n a severe balance of payments crss. 2 In August of 1991, Inda requested a stand-by arrangement from the Internatonal Monetary Fund. Among the condtons for the arrangement was that Inda had to mplement major structural reforms, ncludng trade lberalzaton, fnancal sector reform and tax reform. Before the reform, n , the mport-weghted average of tarffs was 87 percent, the smple average was 128 percent, and some tarffs were over 300 percent. Moreover, non-tarff barrers (especally quanttatve restrctons) affected mports of 65 percent of all products and 90 percent of manufactures (Srnvasan, 2001). 2 See Cerra and Sweta (2002) for a dscusson of the factors that led to the crss. 3

5 The reform ncluded a sgnfcant reducton n the average level as well as the dsperson of tarffs. The maxmum tarff fell from 355 percent n to 150 percent n and 30.8 percent n The weghted average decreased from 87 percent n to 24.6 percent n , although t then gradually ncreased to reach 38.5 percent n The ncrease concded wth a sgnfcant lftng of quanttatve restrctons (Narayanan, 2006), and t was possble because Inda s tarff bndngs from the Uruguay Round were set at much hgher levels than the appled rates (Srnvasan, 2001). 3 The smple average tarff rate fell from 128 percent n to 34.4 percent n It then ncreased to 40.2 percent n but contnued decreasng after that. In , the smple and weghted averages of tarffs were 29 percent (Narayanan, 2006). As reported by Topalova (2004), there was a sharp decrease n tarffs n most ndustres from 1991 to 1992 the sharpest reducton n average tarffs and ther dsperson took place from 1991 to Quanttatve restrctons on most mports have been elmnated. In 1991, most quanttatve restrctons on ntermedate and captal goods were removed, and the lst of goods subject to quanttatve restrctons was reduced sgnfcantly (although t was stll long) to nclude manly consumer goods and agrcultural products. 4 The Uruguay Round agreement, sgned n 1994, requred the elmnaton of quanttatve restrctons. Inda s quanttatve restrctons expred on Aprl 1 st, As a result of the reforms, total trade as a percentage of GDP ncreased from an average of 13 percent n the 1980s to almost 19 percent n The volume of exports and mports also ncreased sgnfcantly snce the early 1990s (Topalova, 2004). Lastly, and mportantly, the frst antdumpng case was ntated n Inda mposed bndngs on 62 percent of the tarff lnes of ndustral products. Before the Uruguay Round, only 3 percent of tarff lnes had bndngs (Natonal Board of Trade, 2005). 4 Accordng to estmates by the World Bank, the share of mports from all sectors ncluded n ther study and covered by NTBs decreased from 95 percent n to 62 percent n and 24 percent n (Srnvasan, 2001). 5 A small number of quanttatve restrctons permtted under Artcles XX and XXI of the GATT reman on grounds of health, safety and moral conduct (Narayanan, 2006). 4

6 2.2 The Use of Antdumpng Table 1 shows that n the perod from January 1995 to June 2006, Inda was the top ntator of antdumpng cases, wth 448, followed by the Unted States (366), The European Communty (345), Argentna (209) and South Afrca (199). Inda was also the country wth the hghest number of antdumpng measures mposed. The legslaton on antdumpng was ntroduced n Inda n 1985, and follows the WTO regulatons. In 1998, a separate dvson the Drectorate-General of Ant-dumpng and Alled Dutes (DGAD) was created wthn the Department of Commerce, whch had more staff and dedcated resources to manage antdumpng complants and recommend antdumpng dutes. Narayanan (2006) also reports that n the 1990s the government frequently and publcly nformed the domestc ndustres about the avalablty of antdumpng (and safeguard) measures. Dutes are leved by the Mnstry of Fnance. For an nvestgaton to be ntated, accordng to the WTO agreement an applcaton needs to be made by on or behalf of the domestc ndustry, whch means that t must be supported by 50 percent of the domestc producers of the ndustry who are expressng ther opnon, and by 25 percent of the total domestc producton of the good. Aggarwal (2002) ndcates that although the Indan antdumpng law follows the WTO standards, n practce only the latter condton s verfed by the authortes. The applcants must provde certan nformaton about producton and prces, as well as evdence of dumpng and njury or threat of njury, and a causal lnk between both. In contrast to some countres such as Canada, Australa and the EU, the Indan antdumpng law does not requre that publc nterest be taken nto consderaton by the DGAD. The prelmnary fndng s normally made wthn 150 days of the date of ntaton. Prelmnary dutes are mposed for a perod not longer than sx months, extendable to nne months under certan crcumstances. The fnal fndng must be made wthn one year of the ntaton of the nvestgaton, extendable up to 18 months. Dutes must be leved wthn three months of the date of publcaton of the fnal fndng (Natonal Board of Trade, 2005). They are mposed for fve years and are subject to annual revews (Aggarwal [2002]). 5

7 The behavor of the number of antdumpng ntatons by Inda over tme s shown n Fgure 1. The frst antdumpng cases were ntated the year after the 1991 reforms. As Inda started the process of trade lberalzaton the use of antdumpng took off, and t presents an ncreasng trend untl Our data on antdumpng ntatons covers the perod from 1992 to As shown n Table 2, 374 antdumpng cases were ntated by Inda durng that perod. In addton, accordng to the WTO (2006), 28 cases were ntated n 2005 and 20 n the frst half of Among the cases ncluded n our data, n 289 of them a prelmnary duty was mposed. Ths number represents 92 percent of the number of ntatons wth non-mssng data on prelmnary decsons (314). In 288 cases a fnal duty was mposed, representng 83 percent of the number of ntatons wth non-mssng data on fnal decsons (348). In 26 cases no evdence of dumpng was found and n 33 cases no njury was found. Only 10 cases were wthdrawn or termnated. These fgures ndcate that not only s the number of antdumpng ntatons by Inda hgh, but ther rate of success s very hgh as well. 7 Frms that fle a petton would thus expect a fnal measure to be mposed wth hgh probablty, and a prelmnary duty to be leved n almost all cases, provdng at least temporary protecton. Ffty countres were named by Inda. The country most frequently named was Chna (70 tmes), followed by the EU (33), Tawan (28), South Korea (26) and the Unted States (20). We can also ask whch sectors have made use of antdumpng. Table 3 reports the number of ntatons by sector for the case of manufacture products (only sectors wth non-zero ntatons are ncluded). The most frequent user of antdumpng has been by large the ndustral chemcals ndustry, wth 214 ntatons. Next are ron and steel (36), other chemcals (18), machnery except electrcal (17) and machnery electrc (14). The ndustral chemcals sector was also the most frequent recever 6 Indan antdumpng data s taken from the Global Antdumpng Database (Bown, 2007). The workng paper accompanyng the database descrbes the detaled data, but n short, the data for Inda was taken drectly from what the Drectorate General of Antdumpng and Alled Dutes n the Mnstry of Commerce publcly reported n The Gazette of Inda 7 For example, based on the data contaned n Table 1, the rato of the number of measures mposed to the number of ntatons s hgher n Inda than n any of the other top 10 ntators, except Turkey. 6

8 of fnal antdumpng measures, wth 173 fnal measures mposed. It was followed by ron and steel (25), other chemcals (17), machnery electrc (13) and textles (9). Perhaps due to Inda s hgh use of antdumpng over the last decade, recently three countres (the EU, Tawan and Bangladesh) have sought consultatons under the WTO Dspute Settlement Understandng concernng Inda s antdumpng decsons. They have ponted to varous shortcomngs regardng the procedures followed by the DGAD. 8 Furthermore, from January 2005 to June 2006, 124 antdumpng cases have been ntated aganst Inda, surpassed only by the number of ntatons aganst Chna, Korea, the U.S., Tawan and Japan (WTO, 2006). 2.3 The Use of Safeguards As can be seen n Table 3a, from January 1995 to June 2006, Inda was the most frequent ntator of safeguard nvestgatons, wth 15 ntatons. 9 It was followed by Jordan (12), Chle (11), Turkey (11), and the Unted States (10). Moreover, Inda mposed the largest number of safeguard measures durng the same perod (8). The domestc law dealng wth the mplementaton of the Agreement of Safeguards was enacted under Secton 8B and Secton 8C of the Customs Tarff Act, The procedures were outlned n the Customs Tarff (Identfcaton and Assessment of Safeguard Duty) Rules, 1997, and Customs Tarff (Transtonal Products Specfc Safeguard Duty) Rules, A Drector General of Safeguards (DGS) under the Department of Revenue of the Mnstry of Fnance was apponted to receve the pettons and conduct the nvestgatons requred for the mposton of a safeguard duty See, for nstance, Natonal Board of Trade (2005) and Narayanan (2006) for detals on those cases. Also, see Aggarwal (2002) for a dscusson of the ambgutes and shortcomngs n Indan procedures to determne dumpng and njury, whch may lead to bases and dscreton n the decsons. 9 Table 1 uses data from WTO (2007). Bown (2007) regstered 16 ntatons by Inda. 10 Secton 8C regulates the mposton of safeguard dutes on any product mported from Chna for whch ncreased mports are causng or threatenng to cause market dsrupton to the domestc ndustry. 11 The nformaton about safeguard legslaton and procedures provded n ths secton was obtaned from the Drectorate General of Safeguards webste: 7

9 An applcaton for an nvestgaton to be ntated can be made by any producer, trade representatve body, frm or nsttuton, whch s representatve of the domestc ndustry. The DGS must nvestgate the exstence of serous njury or threat of serous njury due to ncreased mports. The DGS should then submt ts fndngs to the Central Government, along wth a recommendaton regardng the amount and duraton of the duty that would be adequate to remove the njury or threat of njury to the domestc ndustry. The nvestgaton should be completed wthn eght months of the date of ntaton. The duty can be mposed for a perod of up to four years, extendable to up to ten years. Imposton of a safeguard measure also requres the ndustry to take actons to adjust to mport competton. The levy of dutes takes effect from the date of publcaton of the notfcaton mposng the duty. Furthermore, the DGS must revew the need for contnued mposton of the duty and make the correspondng recommendatons to the Central Government. When the perod of mposton of the duty exceeds three years, the revew must be done no later than at the md term of the mposton. Any revew shall be concluded wthn eght months. The frst safeguard nvestgaton by Inda was ntated n As shown n Table 2, 8 ntatons led to the mposton of a safeguard measure between 1997 and Ths represents 67 percent of the number of cases for whch data on measures was not mssng (12). In only one case no njury was found, and one case was termnated. These numbers ndcate a rate of success lower than n the case of antdumpng nvestgatons (see prevous secton). Table 3 presents the number of safeguard ntatons and measures by ndustry (only ndustres wth non-zero ntatons are shown). The most frequent user of safeguards has been by far the ndustral chemcals sector, wth 9 ntatons (64 percent of the total number of ntatons n manufactures) and 6 measures (75 percent of the total number of measures). As noted before, ths ndustry was also the most frequent user of antdumpng n Inda. It s followed by other chemcals, wth 2 ntatons and 2 measures. 12 Food products, paper and products and machnery except electrcal account for the 12 We should note that the WTO (2006) reports that, snce 1995, chemcal products were the most frequent subject of safeguard nvestgatons n the world. 8

10 remander, wth one ntaton each (and no measures reported). All these ndustres were also ntators of antdumpng cases. 2.4 Economc Importance of Antdumpng and Safeguard Users Over the perod , ndustral chemcals (the top antdumpng and safeguard user) had the hghest mport value among all ndustres, representng 15 percent of all Indan manufacturng mports, and reachng almost 20 percent n some years. Other major antdumpng and safeguard users are also large mporters (see Table 3). These shares are lkely to be underestmated because we would expect mports to be lowered by the use of antdumpng and safeguards. Nonetheless, they ndcate that those polces can potentally have mportant economc dstortonary effects. At the 6-dgt HS product level, we found that the share of manufacturng mports between that were n products affected by antdumpng or safeguard ntatons s 9 percent when the weghts are mports from 1991 (the share of all mports s 6 percent). When the weghts are the average of mports from , the share of manufacturng mports between that were n products affected by antdumpng or safeguard ntatons s 13 percent (the share of all mports s 9 percent). 13 Ths renforces the mportance of studyng Inda s use of antdumpng and safeguards. 3 Econometrc Model, Estmaton Strategy and Data 3.1 Econometrc model Our econometrc model s based on the Grossman and Helpman (1994) (henceforth GH) model of trade protecton. GH explctly model government-ndustry nteractons to derve from frst prncples a set of testable predctons about the determnants of protecton, and ther model has become the leadng poltcal economy model of trade protecton. They consder a small open economy n whch there s a numerare good produced only wth labor, and n non-numerare goods produced wth labor and a specfc factor. The 13 In the same perod, the share of tarff lnes n manufactures for whch there was an AD or SG ntaton s 5 percent (the share of all tarff lnes s 4 percent). 9

11 owners of each specfc factor may organze themselves nto lobbyng groups, and smultaneously offer the government a contrbuton schedule that maps each polcy the government mght choose nto a campagn contrbuton level. In the second stage, the government selects the trade polcy vector to maxmze a weghted sum of contrbutons and socal welfare. The model provdes the followng equaton for the equlbrum tarffs: t I α a + α L = (1) L z ε where t s the ad valorem tarff; I s an ndcator varable that equals one f the sector s organzed nto a lobby and zero otherwse; α L denotes the fracton of the populaton that owns some specfc factor; a s the weght that the government places on socal welfare relatve to poltcal contrbutons; z s the equlbrum rato of domestc output to mports; andε s the absolute value of the elastcty of mport * demand defned over the world prce as follows: = m ( p )( p m ( p )) mports of good, and ε, where n turn p and p * denote the domestc and world prce of good, respectvely. 14 m denotes If tarffs cannot be ncreased beyond certan levels, as t happened when Inda commtted to reduce ts tarffs under the agreement wth the IMF, antdumpng (AD) or safeguard (SG) dutes could be used as substtutes for tarffs. In that context, as the GH model predcts an equlbrum level of protecton gven by equaton (1), the AD/SG duty would be equal to the dfference between such level and the exstng tarff, as follows: 14 To obtanε from the elastcty defned over domestc prces, e, that we use n the estmaton, we would need to p p + t * dvde the latter by = (1 ). However, snce output s measured at domestc prces whle mports are measured at world prces, we also need to dvde z by ( 1+ t ), whch s equvalent to sayng that we can drectly use e nstead of ε n equaton (1) n the estmaton. 10

12 I α z t L τ = (2) a + α L ε whereτ s the AD/SG duty and t s the appled mport tarff. We want to test whether, n addton to respondng to the poltcal economy determnants of protecton modeled by GH, AD and SGs were used n Inda to compensate producers from the tarff reductons that took place under the trade lberalzaton process, and thus we add the tarff change to the rght-hand sde of equaton (2). The equaton that we estmate s therefore gven by: τ z z + = β β I β β t β tarffchange µ ε (3) ε where β 1 = 1 ( a + α L ) > 0, β 2 = α L ( a + α L ) < 0, β = 1 0 and µ s the regresson error term < For organzed sectors, protecton ncreases wth ( z ε ); whle t decreases wth ( z ε ) n the case of unorganzed sectors. The magntude of the devaton from free trade (n ether drecton) s thus hgher when ( ) z ε s hgher, because a larger output means the beneft from protecton s hgher for the lobby, and the welfare cost from protecton s lower the lower are the volume of mports and the elastcty of mport demand. The GH model also predcts that β + β 0. Fnally, we want to test the hypothess 1 2 > that β 4 s negatve; that s, sectors wth hgher tarff reductons are expected to obtan more protecton n the form of AD/SG measures. 15 The error term s ncluded to capture potental measurement error n the varables and other factors (not accounted for n the model) that may nfluence the determnaton of trade polcy. 11

13 3.2 Data Tarffs, antdumpng and safeguard polces Tarff reductons n Inda took place mostly between 1990 and Therefore, we use the tarff change from 1990 to 1997 to estmate the effects of trade lberalzaton on the use of future AD/SG polcy. The perod for whch we have complete data after 1997 s ; hence, our benchmark estmaton uses AD/SG protecton mposed n as the dependent varable. Our unt of observaton s the 6-dgt Harmonzed System (HS) level. Indan mport tarff data at the 6-dgt HS level for varous years over the perod s avalable from the WTO s Integrated Database avalable n WITS. AD/SG protecton s measured as the AD/SG coverage rato. We use data at the exporter-product level to calculate the coverage ratos at the product level, wth product-specfc nformaton on Inda s AD/SG use derved from Indan government sources as descrbed n the Global Antdumpng Database (Bown, 2007). The coverage rato s calculated as the fracton of mports of a product for whch there was an AD/SG measure mposed, where the weghts are the average of mports of the product n 1999 and 2000 comng from each exportng country. 16 We also complement the baselne specfcaton by estmatng the model on a varable defned as the AD/SG-ntaton coverage rato, for whch we used AD/SG ntatons nstead of measures mposed Import data, producton, elastctes, and poltcal organzaton The data for other varables used to estmate the GH model on Indan data are derved from a number of other sources. Frst, data on producton and elastctes at the 3-dgt ISIC level for Inda are taken from the World Bank s Trade and Producton database (Ncta and Olarreaga (2007). Imports at the 6-dgt HS level are taken from the UN s Comtrade database made avalable through WITS. 16 If a product was mported n 2001 but not n 1999 and 2000, the weghts are mports n We do not have consstent and unform data on AD dutes mposed by Inda. We should pont out that the use of coverage ratos has the potental problem that t may understate or overstate protecton; however, they are consdered the best avalable measure of NTBs when more detaled data s mssng. Goldberg and Magg (1999) and 12

14 The poltcal organzaton varable that we use comes from Cadot et al. (2003). They use an teratve procedure n whch a standard GH equaton s frst estmated for the case of Inda wthout dstngushng between organzed and unorganzed sectors. The resduals from ths estmaton are then used to rank ndustres and reclassfy those wth hgh resduals as organzed. A new estmaton s performed and the process s repeated untl the sum of squares s mnmzed, wth the cutoff value used to reclassfy an ndustry as organzed beng determned by a search grd. As a robustness check, we propose an alternatve method for classfyng whether a sector s organzed for the purposes of recevng AD protecton. We dentfy a sector as beng organzed f producers of that good have ever fled an AD case. We beleve that the requrement to fle a case and follow the necessary procedures for that purpose provdes a drect way to dentfy organzed sectors, whch does not face the problem that PAC contrbutons (whch have been used to test the GH model for the U.S.) may understate or overstate traderelated nfluence actvtes. Note fnally that, wth the excepton for the tarff change varable descrbed above, we use the average values of the rght-hand sde varables n 2000 and 2001 as regressors. 3.3 Estmaton Strategy The protecton varable s censored bellow zero and we have potentally endogenous varables some of them enterng nonlnearly on the rght hand sde, whch nclude the output to mport rato, the elastcty, the organzaton varable and the appled tarff. In addton, the organzaton varable and the elastctes may be measured wth error. Gven these, the methodology we apply s a Tobt estmaton combnng the Smth-Blundell (1986) and the Kelejan (1971) approaches. 18 That methodology requres us to regress the Gawande and Bandyopadhyay (2000) also use the NTB coverage rato as the dependent varable n ther tests of the GH model. 18 Gawande and Bandyopadhyay (2000) and Gawande, Krshna and Robbns (2006) also use ths procedure, although the frst only reports the two-stage least square results. We should pont out that although we cannot take the elastcty to the left-hand sde of the protecton equaton as done by Goldberg and Magg (1999) (see equaton (3)), the elastcty estmates that we use have much greater precson, wth nearly all of them beng sgnfcant at least at the 5 percent level. Gawande and Bandyopadhyay (2000) and Mtra (2002) also leave the elastcty on the rght-hand sde. 13

15 rght-hand-sde endogenous varables and ther nonlnear transformatons on the nstruments (by least squares) and nclude the resduals from these regressons as addtonal varables n the orgnal equaton for protecton. 19 The nstruments can nclude the exogenous varables, as well as ther quadratc terms and cross-products. We use as nstruments manly ndustry characterstc data and our choce s motvated by prevous tests of the GH model. The varables used to nstrument for the poltcal organzaton varable nclude the number of establshments (a measure of concentraton), value added per frm, the share of output sold as ntermedate goods, the captal stock and the number of employees. The nstruments for the output to mport rato nclude factor shares, such as the share of captal, sklled labor, land, and natural resources; and the captal-labor rato. Snce varables that affect mports also affect the elastcty of mport demand, these are also used as nstruments for the elastcty. Fnally, gven that tarffs n the GH framework are also determned by the organzaton, output-mport and elastcty varables, we use ther nstruments as nstruments for the appled tarff Emprcal Results 4.1 Baselne estmates Table 4 presents summary statstcs for the relevant varables. The results of our benchmark Tobt estmaton are reported n column 1 of Table 5. We fnd that poltcally organzed sectors do get more AD/SG protecton than unorganzed ones, as predcted by the GH model. β 1, the coeffcent of I ( ε ) z, s negatve and sgnfcant at the 10 percent level, whle 2 β, the coeffcent of ( z ε ) postve and sgnfcant at the 1 percent level. However, the sum of those coeffcents s negatve, n, s 19 Includng the resduals corrects for endogenety n the correspondng varables and all the coeffcents become consstent. If the resduals are statstcally sgnfcant we can reject the null hypothess that the varables are exogenous. 20 Some of these data derves from Ncta and Olarreaga (2007) and some were provded by Marcelo Olarreaga, Olver Cadot and Jean-Mare Grether. 14

16 contrast to the model s predcton. The coeffcent of the tarff, β 3, s postve but not sgnfcant, and the coeffcent of the tarff change from , β 4, s negatve (as expected) and sgnfcant at the 1 percent level. Therefore, we fnd evdence that the sectors whch expermented larger tarff reductons durng the trade lberalzaton perod ( ) are also the ones that receved hgher protecton n the form of AD or SG dutes afterwards. 21 Ths s a potentally mportant result as t ndcates that at least part of the trade lberalzaton undertaken by Inda was reversed wth the later use of contngent protecton. In addton, t suggests that the prevous research on trade lberalzaton n Inda that focuses only on tarff reductons suffers from msmeasurement that may affect ther results. In partcular, the postve relatonshp between the sze of the tarff reducton and subsequent use of AD and SGs that we fnd ndcates that there s less dsperson n the actual reducton of protecton across products than n the tarff-only data that those studes have used. 22 Column 2 shows the results when we use AD or SG ntatons as the dependent varable. Some authors have shown that an AD ntaton can have a restrctve effect on trade even f no measure s mposed (see for example Fnger [1981]). The results are qualtatvely smlar and also quanttatvely close to those dscussed above. In column 3 we present the results that we obtan when we modfy our benchmark estmaton by usng the poltcal economy varable obtaned wth our crteron. As mentoned earler, we classfy a sector as organzed f t ever fled an AD case. 23 In ths case we fnd that both, the coeffcent of I ( ε ) and the coeffcent of ( ) z z ε are sgnfcant at the 1 percent level. In addton, we fnd that the sum of ther coeffcents s postve, as predcted by the GH model. The coeffcent of the tarff s now negatve and close to 1 (once the varable s properly scaled by 100), as expected, although not statstcally sgnfcant. Fnally, the coeffcent of the tarff change s agan negatve and sgnfcant at the 21 The coeffcents of the resduals are not reported but none of them was sgnfcant at the 10 percent level, meanng that we cannot reject the exogenety of our regressors once we nclude those resduals. 22 We also found that 91 percent of the AD/SG ntatons from were n ndustres n whch the standard devaton of the tarff cut was larger than the medan for all ndustres. 15

17 5 percent level. However, we should pont out that the resdual of the frst rght-hand sde varable s statstcally sgnfcant, ndcatng that we can reject the exogenety of that regressor. 4.2 Senstvty analyss In ths secton we dscuss the results of some addtonal robustness tests. Frst, we use the tarff change from nstead of , snce Topalova (2004) argues that from 1990 to 1996 Inda s trade polcy was largely affected by exogenously mposed targets and restrctons, whle startng n 1997 external pressure was almost nonexstent. The results are qualtatvely smlar and also quanttatvely very close to those already shown. The results are also robust to usng the appled tarff from other years rather than the average from 2000 and We used the tarff from 2000, from 1997 (we do not have tarff data for 1998 or 1999), and also from None of the results are affected by that. In column 4 of Table 5 we present the results when the dependent varable s only the AD measure coverage rato (that s, we exclude safeguards). The results are smlar to those of column 1, except that the frst varable, I ( ε ), s not sgnfcant (the probablty was 0.102). z As a fnal senstvty analyss, we dd the estmaton for the perod 1997 to 1999, usng the tarff change from 1990 to The results are shown n Table 6. Column 1 shows the estmates when the dependent varable s the AD/SG measure coverage rato. The coeffcents of I ( ε ) and ( z ε ) z are postve and negatve, respectvely (as predcted by the model) and both are sgnfcant at the 1 percent level. Nevertheless, ther correspondng frst-stage resduals were statstcally sgnfcant. The appled tarff s agan postve but not sgnfcant, and the tarff change s now postve and sgnfcant. 24 The latter would ndcate that products that receved smaller tarff reductons untl 1996 also receved hgher AD/SG protecton durng However, we thnk that ths s not an deal perod to study the effects of 23 We should menton that out of the sectors that we classfy as organzed followng ths crteron, about 60 percent were also organzed under the Cadot et al. (2003) classfcaton. 16

18 trade lberalzaton n Inda because durng that perod tarffs on many products were actually rased. 25 That s, we may have that durng that perod both tarffs and AD and SGs were ncreasngly used. Table 6 also shows the results wth the alternatve specfcatons descrbed already for the perod. Agan, the results do not change much from what we already descrbed, except that when we use our classfcaton of organzed and unorganzed ndustres (column 3) the tarff change becomes not sgnfcant and, n contrast to what we found for the perod, when we use only AD measures as the dependent varable (column 4), the coeffcent of I ( ε ) s sgnfcant at the 1 percent level. z 5 Alternatve Estmaton Framework and Results 5.1 Probt model and predctons We also estmated a Probt equaton n whch the dependent varable s a bnary varable that s equal to one f there was an AD/SG ntaton for a 6-dgt HS product comng from a certan exportng country. Ths allows us to explot varaton across exporters as well. Our estmaton n ths secton does not follow a specfc theoretcal model; nstead, we nclude as explanatory varables some varables that have been found to affect AD ntatons n prevous studes. We want to study the effect that tarff reductons had on the ntaton of AD and SG cases. We focus ntally on AD/SG cases ntated n , and use as regressors mostly varables from We nclude the value of mports, the current appled tarff (average from ), output, the number of employees, the number of establshments, the elastcty of mport demand, a dummy varable that equals one f there was an AD or SG ntaton on that productexporter par before 2000, and the tarff change from We expect hgher mports to ncrease the probablty of ntatng an AD/SG nvestgaton. A hgher tarff s expected to be negatvely related to the probablty of ntatng a case, as t ndcates that the product currently has a hgher protecton level. A hgher output s expected to be postvely related to 24 We use the appled tarff from 1997, as we do not have tarff data for 1998 or We also tred usng the average of the tarffs from 1997 and 2000 and the results dd not change much. 25 For example, durng ths perod the percentage of products wth tarff ncreases was actually hgher than the percentage of products wth tarff reductons (see also Topalova [2006]). 17

19 the probablty of ntatng an AD/SG case, as t means that the producers have more to gan from protecton and also may have more resources to support the ltgaton costs assocated wth an AD nvestgaton process. The number of employees may be used as a proxy for poltcal nfluence, and s also expected to have a postve mpact on the probablty of ntatng an nvestgaton. The number of establshments s nversely related to concentraton n the ndustry (whch s n turn lkely to affect the ablty to overcome the free-rder problem), and t s therefore expected to reduce the probablty of an ntaton. The elastcty of mport demand n absolute value s drectly related to the deadweght loss assocated wth protecton, and thus we expect a hgher elastcty to reduce the probablty of ntatng a case, as long as producers perceve that a measure would be less lkely to be mposed gven ts larger socal cost. Snce the actual and not the absolute value of the varable s used n the estmaton, a postve sgn for ts coeffcent would be expected. The tarff change s expected to have a negatve coeffcent, as a larger tarff reducton would ncrease the ncentve for the producers to fle a case n order to seek alternatve protecton n the form of AD or SG measures. Fnally, we would expect the coeffcent of the dummy varable of a prevous ntaton on the product-exporter par to have a negatve sgn. However, we should pont out that a postve sgn s possble because AD and SG cases are sometmes defned at the 8-dgt (or even more dsaggregated) HS product level, and thus t s possble to have an ntaton on a dfferent subgroup of, say, 8-dgt HS products that belongs to the same 6-dgt HS product category even f an nvestgaton on another subgroup was already ntated n the past. 5.2 Estmates from the probt model Column 1 n Table 7 reports the results of the frst specfcaton. The coeffcent of mports has the expected sgn but s not statstcally sgnfcant. The current tarff has a negatve coeffcent, as expected, and t s sgnfcant at the 10 percent level. 26 The coeffcents of output, the number of employees and the number of establshments all have the predcted sgn and they are sgnfcant at the 1 percent level. The 26 Ths dffers from the postve and not statstcally sgnfcant coeffcent obtaned n our benchmark estmaton of the GH model. 18

20 elastcty has a sgnfcant but negatve coeffcent. The coeffcent of the tarff change s negatve and sgnfcant at the 1 percent level, ndcatng that hgher tarff reductons durng the trade lberalzaton perod ncrease the probablty of ntatng an AD/SG nvestgaton afterwards. Ths renforces the results that we obtaned n the prevous secton. The ndcator varable of a prevous AD or SG nvestgaton has a postve and sgnfcant coeffcent, whch s lkely due to the reason dscussed above. We should menton that we also expermented wth usng mport growth nstead of mports but ts coeffcent was not sgnfcant (the other results were not affected). When we used the mport penetraton rato, however, ts coeffcent was sgnfcant at the one percent level (and postve) once fxed effects were ncluded. 27 Column 2 adds exporter and ndustry fxed effects to the specfcaton from column 1. We can see that once fxed effects are added, mports become statstcally sgnfcant at the 10 percent level. The results on the other varables stll hold, except for the number of employees and the number of establshments, whose coeffcents change sgns, although the coeffcent of the latter becomes not sgnfcant. Usng the specfcaton from column 2, we should pont out that the predcted probablty of an AD/SG ntaton s when the estmated coeffcents are evaluated at the mean value of each explanatory varable. Havng noted ths, a 1 percentage pont ncrease n the tarff reducton between 1990 and 1997 would ncrease the probablty of ntatng an nvestgaton by (or approxmately 1 percent of the predcted probablty value). Gven the large tarff reductons that took place n Inda durng that perod, these estmates are economcally sgnfcant. We also have that a one standard devaton ncrease n the tarff reducton mples a predcted ncrease n the probablty of an nvestgaton by , whle the predcted probablty of an nvestgaton when evaluated at the means of the varables s , as already mentoned. 19

21 5.2 Senstvty analyss In columns 3 and 4, we show the results obtaned when we use as the dependent varable a bnary varable that equals one f there s an AD ntaton only, wthout and wth fxed effects, respectvely. The results are very smlar to those shown n columns 1 and 2. The man dfferences are that mports become sgnfcant at the 5 percent level when fxed effects are ncluded, and that the tarff s sgnfcant at least at the 5 percent level. 28 Furthermore, usng the results from column 4, we obtan that a one standard devaton ncrease n the tarff reducton mples a predcted ncrease n the probablty of an AD nvestgaton by , as compared to a predcted probablty value of an AD ntaton (when evaluated at the means of the varables) of Agan, ths result s economcally sgnfcant. Columns 5 and 6 contan the results assocated wth an addtonal robustness test. We had measured the tarff change as the dfference n tarff levels n 1997 and 1990, and now we measure t as the dfference between those tarffs (each scaled by 100) dvded by one plus the average of the tarffs (ths new varable s denoted t97_90p). The results are robust to ths alternatve way of measurng the tarff change. The only dfference s that the current tarff becomes not statstcally sgnfcant, but the coeffcent of the tarff change s stll negatve and sgnfcant at the 1 percent level. Addtonally, we ncluded also the dfference between the appled and the bound tarff rate (as a percentage of the appled rate) n the rght-hand sde of our benchmark specfcatons of columns 1 and 2. We would expect ths varable to have a negatve coeffcent, snce a lower dfference ndcates smaller room for tarff ncreases and may thus lead to seekng AD or SG protecton nstead. Its coeffcent was negatve and t was statstcally sgnfcant at least at the 10 percent level. However, ths result s not always robust to the way of measurng the dfference (level versus percentage, etc.). Furthermore, snce the average appled tarff from may be endogenous to the exstence of an AD or SG ntaton durng that perod, we also estmated our benchmark specfcatons n the followng ways: ) excludng ths varable, and ) excludng t but replacng t wth the tarff from an 27 The only other result that changes s that the coeffcent of the current tarff becomes not sgnfcant. 20

22 earler year (1997 or 1992). Ths does not affect the results for the remanng varables (and the tarff from 1992/1997 was negatve but not sgnfcant). As a fnal robustness check, Table 8 shows the results obtaned from smlar specfcatons to those ncluded n Table 7, but when we do the estmaton for the 1997 to 1999 perod. The dfferences are, frst, that n ths case the appled tarff has a postve and sgnfcant coeffcent. Moreover, the coeffcent of tarff change (from 1990 to 1996) s now postve. These results are actually smlar to those obtaned n our estmaton of the GH model for the same perod. The underlyng explanaton s that the perod was specal n that t ncluded tarff ncreases for several products and thus sgnaled a perod of a complementary relatonshp between ncreased tarff protecton as well as ncreased antdumpng and safeguards protecton, as explaned earler. We also ran specfcatons smlar to those of Table 8 but at the product level and havng the tarff change between 1997 and 2000 as the dependent varable (usng OLS) and found that the tarff change from had a negatve and sgnfcant coeffcent, ndcatng that the products that had larger tarff reductons n were gettng hgher tarff ncreases (or at least smaller reductons) n , whch may explan why we do not fnd more AD/SG use n those products n Sector-Level Analyss We also preformed our benchmark Probt estmaton of AD/SG ntatons n but nteractng the tarff change wth ndustry (ISIC 3) dummes. For three ndustres the coeffcent of ths varable s sgnfcant (at least at the 10 percent level). In all three, t s negatve. Those ndustres are textles (sgnfcant at the 10 percent level), ron and steel (sgnfcant at the 1 percent level) and machnery electrc (sgnfcant at the 5 percent level). The man excepton s ndustral chemcals, the top AD and SG user, for whch the coeffcent of the tarff change was postve but not sgnfcant. However, we should pont out that the three sectors that seem to be drvng our results are mportant AD and SG users, 28 In these specfcatons the AD/SG prevous-ntaton ndcator s replaced wth one based on prevous AD ntatons only. 21

23 accountng for 27 percent of the total number of products wth at least one ntaton n The results are smlar when we add exporter fxed effects. To determne why we do not fnd the negatve and sgnfcant relatonshp between the tarff change from and AD and SG use n for ndustral chemcals, we consdered dfferent potental explanatons. Frst, we nvestgated whether the products wth larger tarff reductons n ths sector obtaned tarff ncreases later, whch could help explan why we do not fnd more AD/SG use by them. We ran an OLS regresson of the tarff change from on the tarff change from and the other varables (as the one descrbed n secton 5.2) for ndustral chemcals only, but we found no evdence supportng ths hypothess as the varable was not sgnfcant (and postve). 30 Second, we re-dd the Probt estmaton of AD/SG ntatons n but replacng the tarff change wth a dummy equal to one f there was a tarff reducton, nteracted wth ndustry dummes. We wanted to fnd out f n ths sector products for whch the tarff was decreased used more AD/SG (even though the sze of the reducton may not matter). We found that the coeffcent of the dummy was negatve and sgnfcant for ndustral chemcals, ndcatng that products wth tarff reductons n actually used less AD/SG n Fnally, we explored the possblty that tarff lberalzaton n ths sector took place later than n others, and maybe because of ths the sector reacted dfferently regardng the use of contngent protecton. We dd ths by examnng average tarff changes by ndustry over tme. Agan, we dd not fnd enough evdence to support ths hypothess. Therefore, we conclude that the reason behnd the lack of relatonshp between tarff reductons durng the lberalzaton perod and subsequent use of AD and SG polces by ths sector remans an underlyng puzzle. Perhaps t s related to the enforcement of an nternatonal cartel, whch has been suggested before as an explanaton for why ndustral chemcals s a very frequent user of these polces across many countres. 29 We use 2000 as we do not have tarff data for

24 6 Concluson Inda undertook a major unlateral lberalzaton to ts tarff polces n the 1990s. Ths feature of ts economc reform, when coupled wth ts transformaton from beng a non-user to beng the most frequent user of WTO sanctoned antdumpng and safeguard polces rases a number of mportant questons that ths paper seeks to address. We explot a newly avalable data source to provde evdence that Inda s resort to antdumpng and safeguard polces n s related to the sze of Inda s tarff reform the larger the tarff reform, the more lkely was the ndustry to seek these new forms of protecton. Ths s confrmed va structural estmates from a Grossman and Helpman (1994) model that s modfed to examne determnants of Indan antdumpng and safeguard use as well as a reduced form specfcaton that s able to explot a larger sample of data. Fnally, our results have mportant mplcatons for understandng (and measurng) the sze and scope of Inda s mport market access reforms n the 1990s, as such trade polcy reforms are an ncreasng area of study wthn the focus on determnants of Inda s economc development. 30 The same result obtans for ths sector f we do a regresson for all sectors together nteractng the tarff change wth ndustry dummes. 23