Trade Liberalization, Antidumping, and Safeguards:

Evidence from India's Tariff Reform

Chad P. Bown†

Brandeis University Patricia Tovar‡

Brandeis University

This version: March 2008

Abstract

Economic theories of trade agreements identify a number of potential linkages between trade liberalization and the subsequent re-imposition of import protection under safeguard exceptions. To our knowledge, this paper is the first product-level study to explicitly investigate the link between liberalization and the subsequent re-imposition of such protection. We overcome endogeneity problems by exploiting tariff-cut heterogeneity across products within India, a country that underwent a major exogenous tariff reform program in the early 1990s and subsequently became a frequent user of new safeguard and antidumping import restrictions. First, we provide structural estimates from a Grossman and Helpman (1994) model modified to examine determinants of Indian antidumping and safeguard use, and we find that products with larger tariff cuts between 1990 and 1997 are associated with an increase in new import protection in the early 2000s. Second, estimates from a reduced-form model confirm these results and suggest that they are economically important – i.e., a one standard deviation increase in the size of the tariff cut away from the mean increases the predicted probability by almost 50% of such new protection. Third, we find heterogeneity in the size of the effect across sectors as tariff cuts are an important determinant of such new protection within the steel, iron and paper industries, but not within industrial chemicals, for which there is strong evidence of retaliatory or collusive effects. Combined, our results are consistent with the theory that access to such policy exceptions dilutes the potential commitment device role played by trade agreements. The results also provide one explanation for separate estimates in the literature that the magnitude of import reduction associated with India's use of antidumping is similar to the initial import expansion associated with its tariff reform. Finally, we interpret the implications of our results for the burgeoning research literature examining the effects of liberalization on India’s micro-level development.

JEL No. F13 Keywords: Trade agreements, commitment, India, tariff reform, antidumping, safeguards, retaliation, collusion ________________

† Bown: Department of Economics and International Business School, MS021, Brandeis University, Waltham, MA 02454-9110, USA tel: 781-736-4823, fax: 781-736-2269, email: cbown@brandeis.edu, web: http://www.brandeis.edu/~cbown/. ‡ Tovar: Department of Economics and International Business School, MS021, Brandeis University, Waltham, MA 02454-9110, USA tel: 781-736-5205, fax: 781-736-2269, email: tovar@brandeis.edu, web: http://www.brandeis.edu/~tovar/ . Thanks to Meredith Crowley, Steve Redding, Bernard Hoekman, Marcelo Olarreaga, T.N. Srinivasan, Hylke Vandenbussche, Jaimie de Melo, Giovanni Facchini, Sajal Lahiri, Kamal Saggi, Maurizio Zanardi and seminar participants at LSE, the University of Geneva and the MWIEG meetings in Michigan for helpful comments. Olivier Cadot, Jean-Marie Grether, and Marcelo Olarreaga also shared useful data. Bown acknowledges financial support from the World Bank and thanks the WTO for hospitality while a portion of this paper was being written. All remaining errors are our own, and any opinions expressed within are our own and should not be attributed to the World Bank or the WTO.

1

1 Introduction

Economic theorists have identified a central tension that occurs when a trade agreement includes

provisions for policy “exceptions” – typically referred to as “safeguards” – that allow for a government

to subsequently re-implement conditional import protection after trade liberalization occurs. On one hand,

Bagwell and Staiger (1990) illustrate how safeguards can play a positive role in maintaining a cooperative

trade agreement and relatively low tariffs in the face of unexpected shocks.1 On the other hand, an

important implication of a second major strand of the theoretical literature on trade agreements (e.g.,

Staiger and Tabellini, 1987; Maggi and Rodriguez-Clare 1998, 2007) is that ex ante inclusion of such a

safeguard exception can create time-consistency or commitment problems that make it difficult for a

government to implement even Pareto-improving trade liberalizing reform announcements ex post.

Despite important concerns these theories raise for understanding links between trade liberalization and

such import-restricting policy exceptions, economists have found it elusive to empirically test their

relevance in practice. There are a number of fundamental reasons for the lack of empirical progress, not

the least of which is concern over policy endogeneity that creates challenges for identification.

This paper provides a new approach that overcomes these endogeneity problems and allows us to

empirically investigate the link between trade liberalization and use of exceptions permitting new import

protection. First, our setting is the "natural experiment" created by India’s exogenously-mandated tariff

reform program of the 1990s. Focusing on a single country with exogenous tariff cuts allows examination

of the effect of the tariff cut treatment on subsequent response of new import protection.2 Second, we

1 In addition to Bagwell and Staiger (1990), other motivations for including ex ante safeguard provisions into a trade agreement are that it can provide insurance that encourages hesitant policymakers to liberalize. See Fischer and Prusa (2003) for one theoretical approach to modeling this relationship. See also Bagwell and Staiger (2005). 2 The first endogeneity concern is that a country's trade liberalization is typically not an exogenous event, but instead is part of a negotiated preferential or multilateral trade agreement. In such cases, endogenous factors may determine both the level of initial liberalization and subsequent resort to exceptions for new protection. A second endogeneity concern may arise if the trade liberalizing country is simultaneously negotiating the terms of the “exceptions” in the writing of the trade agreement – i.e., not only the question of whether to have any exceptions at all, but also the legal and economic evidentiary criterion that must be met in order to trigger the exceptions. This is also not of concern for our context as India’s accession to the WTO was part of the “Single Undertaking” which meant India would be subject to established GATT/WTO rules governing antidumping and safeguard exceptions.

2

focus on the product-level link between India’s tariff cuts and its subsequent resort to the liberal trade

policy “exceptions” of newly applied safeguard and antidumping trade restrictions, which themselves are

relatively substitutable forms of new import protection.3 Our approach is to use the Indian setting and

exploit cross-product variation to examine whether there is a link between size of the initial tariff cut and

the subsequent resort to such new import restrictions.

In addition to the exogeneity of its tariff reform, India is an excellent setting to test for this link

for a number of reasons. Subsequent to the initiation of its tariff reform program in 1991, India

transformed from being a non-user of policy exceptions such as antidumping and safeguards to becoming

the WTO system’s most frequent user (WTO, 2007a,b) of both types of import restrictions over the next

decade. Nevertheless, while the response to the Indian tariff reform program appears well timed with the

subsequent rise in filings and implementation of these safeguards and antidumping policy exceptions, is

there a product-level link? The top two panels of figure 1 illustrate suggestive evidence of the basic

relationship between the relative sizes of the 1990s tariff cuts and subsequent antidumping use for

products within two of India's major users of antidumping – the iron and steel as well as the paper

industries. The figures indicate that products that subsequently sought antidumping protection in the early

2000s, on average, started with higher tariffs and received larger tariff cuts over the 1990s.

Our econometric analysis investigates whether this suggestive evidence of a relationship between

the size of trade liberalization and subsequent resort to these policy exceptions is economically and

statistically important. Our formal approach proceeds in two steps. First, we begin by estimating the

structural Grossman and Helpman (1994) political economy model suitably modified to examine this new

3 Despite substantial legal differences between safeguards and antidumping, they have been shown in many contexts to be relatively substitutable instruments of import protection, given the lax enforcement rules regulating how these policies are implemented. See, for example, Bown (2004), Bown and McCulloch (2003) and also the discussion in Hoekman and Kostecki (2001). Nevertheless, our estimation approaches control for the most important differences (e.g., antidumping is country-specific and discriminatory, safeguards are nondiscriminatory) between them as we describe in substantial detail below. For comprehensive surveys of economic research in the antidumping literature see Blonigen and Prusa (2003) and for the safeguard literature, see Bown and Crowley (2005).

3

setting – i.e., determinants of Indian antidumping and safeguard use over the early 2000s.4 We examine

whether product-level requests for new antidumping and safeguard protection are based both on

determinants suggested by the prior literature and the determinant of interest for our analysis – i.e., the

depth of the product-level Indian import tariff cut between 1990 and 1997. Our results suggest structural

parameters broadly consistent with previous research applying the model to other countries and other

trade policy settings. Furthermore, from this modified setting we provide evidence of a significant

negative relationship between the size of the product-level trade liberalization undertaken between 1990

and 1997 and the subsequent resort to new protection in the early 2000s – i.e., the larger the good's initial

tariff cut, the more antidumping and safeguards protection the Indian producers of that good subsequently

demanded and received ex post. 5

Viewed from the theoretical literature on trade agreements, we interpret this evidence as

consistent with the implicit concerns raised by Staiger and Tabellini (1987) and Maggi and Rodriguez-

Clare (1998, 2007) – i.e., that India used new product-specific protection in the early 2000s to escape

from 1990s trade liberalization announcements that, ex post, it found too deep to sustain. We present

additional evidence in support of this relationship in a final section of the paper in which we investigate a

previously unexamined margin of the data on the duration of time that measures stay imposed. There we

4 The first papers to estimate structural versions of the Grossman and Helpman model on data for the United States include Goldberg and Maggi (1999) and Gawande and Bandyopadhyay (2000). While there are too many papers in the subsequent literature to cite here, Cadot, Grether, and Olarreaga (2003) is the first paper that we are aware of that applies the Grossman and Helpman model to estimating determinants of Indian import protection. Nevertheless their study does not examine the questions of interest of this paper - i.e., specifically whether the model can be used to understand determinants of a particular trade policy (antidumping and/or safeguards) as well as whether there is a relationship between demands for such forms of protection and the size of past trade liberalization. 5 Staiger and Tabellini (1999) is one of the few attempts of which we are aware to empirically test the commitment theory of trade agreements, using U.S. data on sectoral exclusions in the Tokyo Round of GATT negotiations in comparison to tariff responses under the U.S. safeguard law. On the other hand, there are some related papers closer to our approach but which use much more aggregated data and which also do not attempt to deal with the endogeneity issues that we have identified. For example, Crowley (2007) is a cross-country, macro-level study relating the subsequent number of safeguard cases that a WTO member initiated between 1995 and 2000 to a measure of the member's average tariff cut undertaken in the Uruguay Round. Feinberg and Reynolds (2007) is a similar cross-country approach which focuses on antidumping alone and is carried out at a very aggregated industry level. Our approach differs from these two studies along a number of different dimensions, including that it focuses on a single country in which the tariff cuts were arguably exogenous thus forming the basis for a better natural experiment, it is conducted at the product (6-digit Harmonized System) level, it examines both antidumping and safeguard use, and the estimates derive from both reduced-form and structural econometric models.

4

illustrate evidence that, within the set of Indian products receiving antidumping protection, there is also a

negative relationship between the length of protection under an antidumping measure and the size of the

1990s tariff cut. We thus find that "temporary" antidumping protection may be more likely to become

"quasi-permanent" protection, the larger was the product's original tariff cut.

The second step of the econometric analysis complements the structural Grossman and Helpman

approach by estimating a reduced-form model that exploits additional variation in our available data and

allowing us to address a number of related questions raised by the theoretical literature. First, the evidence

confirms the structural Grossman and Helpman model estimates of a negative relationship between the

size of the Indian 1990s tariff cut and the subsequent resort to antidumping protection in the early 2000s.

Second, we use this approach to interpret the economic size of this effect and to investigate the industry-

level source of this relationship. We find that the average effect is large – i.e., a one standard deviation

increase in the tariff cut away from the mean increases the predicted probability of new antidumping or

safeguard use by 50%.6 The negative relationship is driven by product-level variation within relatively

large Indian importing industries such as iron, steel and paper, which are also major Indian and global

users of antidumping.

Perhaps surprisingly, however, we find no evidence of a link between 1990s tariff cuts and

subsequent resort to antidumping by India's dominant sectoral user of antidumping – the industrial

chemicals sector – a result consistent with the suggestive evidence of the lower panel in figure 1. We

exploit characteristics of the export source of the Indian imported products and an additionally available

margin of the data to address the question of what does determine cross-product use of antidumping

within the industrial chemicals sector if it is not variation in the size of the tariff cuts. We find that India's

antidumping use across products within industrial chemicals targets imports from exporting firms in

6 Interpreting the size of our results for India suggests they are likely to have an economically important implication for trade flows as well. Indeed, our results that link trade policies (tariffs and antidumping/safeguards) over time provide evidence to confirm the implicit hypothesis presented in Vandenbussche and Zanardi (2006), whose gravity model estimates find that the trade decrease resulting from India’s antidumping policy is of the same magnitude as the trade increase that resulted from its earlier trade liberalization.

5

trading partners that previously targeted India's own chemical-producing exporters with antidumping.

This provides further evidence consistent with the long-held theory that antidumping is used by certain

industries as a retaliatory mechanism to enforce collusive, international market sharing arrangements.7

Our combined results allow us to confirm the results found in a variety of research settings – i.e., that

inclusion of trade policy exceptions such as safeguards and antidumping into trade agreements leads to

multiple political-economic motives for subsequent use by industries and policymakers.

Finally, we note that our identification of a link between India’s 1990s tariff reform and the

subsequent use of new forms of import protection via antidumping and safeguard policy is potentially

important for other areas economic research. A substantial literature has evolved that uses the size of the

exogenous Indian tariff cuts to examine the impact of trade liberalization on other fundamental and

microeconomic changes (poverty, productivity growth, labor demand, etc.) transforming the Indian

economy.8 Our results suggest that relying on only tariff cuts to proxy for trade liberalization in certain

Indian industries runs the risk of substantial mismeasurement.

The rest of this paper proceeds as follows. Section 2 describes the institutional setting of India’s

tariff reform in the 1990s and the subsequent resort to exceptions such as antidumping and safeguards. In

section 3 we modify the Grossman and Helpman (1994) structural model to estimate the relationship

between Indian use of antidumping and safeguards and its tariff reform. This section also describes our

data and presents our baseline estimates and first round of sensitivity analysis. Section 4 presents the

alternative reduced-form framework that both documents the robustness of our results and allows us to

explore additional questions. Section 5 concludes.

7 See Prusa (1992) and Hoekman and Mavroidis (1996), for example, for discussions. Recent papers finding evidence consistent with retaliatory effects on different samples of antidumping use data include Blonigen and Bown (2003), Prusa and Skeath (2002), Feinberg and Reynolds (2006) and Vandenbussche and Zanardi (2008). Note that none of these earlier empirical papers match antidumping use across countries at the actual level of product disaggregation (6-digit Harmonized System) that we have done here. 8 Examples of recent studies of India examining such links include the relationship between liberalization and industry/firm productivity (Krishna and Mitra, 1998; Topalova, 2004), poverty (Topalova, 2005), the demand for labor (Hasan, Mitra, and Ramaswamy, 2007) as well as child labor (Edmonds, Pavcnik, and Topalova, 2007).

6

2 India’s Tariff Reform, Antidumping, and Safeguards

2.1 Trade liberalization in India in the 1990s

Between 1947 and the late 1980s, India followed an inward-oriented development strategy. A

combination of external shocks in the late 1980s and early 1990s led to large macroeconomic imbalances,

and as a result, India requested a stand-by arrangement from the International Monetary Fund in August

of 1991. Among the conditions for the arrangement was that India had to implement major structural

reforms, including trade liberalization, financial sector reform and tax reform (Cerra and Saxena, 2002).

The trade reform started in 1991 and was completed within the export-import policy announced

in the government’s Eighth Plan in 1992, which outlined a program of tariff reductions for the next five

years on the basis of the 1991 agreement with the IMF (Pursell, Kishor, and Gupta, 2007). 9 The

government had to meet strict compliance deadlines, and it chose to implement the reform abruptly so as

to avoid the emergence of potential opposition and thus without time to analyze or debate its distributive

effects (Topalova, 2006). Such tariff reform characteristics point to its exogenous nature.

As additional evidence on the exogeneity of the tariff reductions, Edmonds, Pavcnik and

Topalova (2007) report a marked linear relationship between the pre-reform tariff levels and the tariff cuts

by industry – which we also confirm using our data – deriving from the fact that the IMF mandated a

reduction in both the tariff levels and their dispersion. Moreover, Topalova (2005) regresses the tariff

change on late 1980s industry characteristics, including factor shares, concentration, employment, wages,

productivity and others, and finds that tariff changes are not correlated with industry characteristics.

Prior to the IMF arrangement, the 1990-1991 Indian import-weighted average tariff was 87

percent, the simple average was 128 percent, and some tariffs were over 300 percent (Srinivasan, 2001).

The maximum tariff fell from 355 percent in 1990-1991 to 150 percent in 1991-1992 and 30.8 percent in

2002-2003. The weighted average tariff decreased from 87 percent in 1990-1991 to 24.6 percent in 1996-

9 Even though India was a member of the GATT, it did not participate in tariff-reducing GATT rounds (Edmonds, Pavcnik and Topalova, 2006). Topalova (2004) also describes these five-year plans as having been carried out largely as they were originally announced.

7

1997 before it gradually increased to 38.5 percent in 2001-2002.10 Finally, the standard deviation of tariffs

fell from 41 percent to 15 percent between 1991 and 1997-1998 (Hasan, Mitra, and Ramaswamy, 2007).

Because of the exogenous nature of India's IMF-mandated trade liberalization in the 1990s,

economists have used it as a "natural experiment" case study to test the impact of trade liberalization on

many different questions concerning fundamental microeconomic activity. However, one concern that we

examine is the extent to which this exogenous reduction in import tariffs is positively associated with the

subsequent re-application of new forms of import protection in India via WTO-permitted exceptions such

as the imposition of safeguards and antidumping import restrictions.

2.2 India's antidumping and safeguard policies and use

Table 1 documents how the pattern of new Indian antidumping initiations has evolved over the 1992-2004

period. India introduced its antidumping legislation in 1985 but did not initiate its first antidumping case

until 1992 and after its tariff reforms had begun. Furthermore, India enacted its domestic safeguard

legislation in 1997 and also initiated its first safeguard investigation that year. The use of antidumping in

particular accelerated in the late 1990s before reaching its peak in 2002.11 As table 1 illustrates, India

initiated 374 antidumping cases during that period. India imposed a final antidumping measure - e.g.,

typically an ad valorem or specific duty - in 288 of the investigations, representing 83 percent of the

10 The increase in applied tariffs after 1997 coincided with a significant lifting of quantitative restrictions (Narayanan, 2006) and was possible because India’s tariff bindings from the Uruguay Round were set at much higher levels than the applied rates (Srinivasan, 2001). The simple average tariff rate fell from 128 percent in 1990-1991 to 34.4 percent in 1997-1998 and then increased to 40.2 percent in 1998-99 but continued decreasing after that (Narayanan, 2006). 11 Our analysis draws on the publicly available Global Antidumping Database (Bown, 2007a) which provides detailed data on policy investigation outcomes, as well as products and exporting countries targeted by Indian use of antidumping between 1992 to 2004. The working paper accompanying the database describes the data in full detail. To summarize, the data for India was taken directly from what the Directorate General of Antidumping and Allied Duties in the Ministry of Commerce publicly reported in The Gazette of India http://commerce.nic.in/ad_cases.htm. The information on the duration of measures imposed was frequently supplemented by information India has made available to the WTO's Committee on Antidumping.

8

number of initiations with non-missing data on final decisions (348).12 Thus not only does India initiate a

high number of cases, but a very large majority of these cases result in the imposition of new trade

restrictions. India imposed final measures in 8 of the 12 safeguard cases with non-missing data during

this time period. Finally, India's use of both antidumping and safeguards went unchallenged by WTO

members through formal Dispute Settlement Understanding activity until December 2003, when the

European Union brought the first case against Indian antidumping (WTO, 2008). 13

Table 2 decomposes the Indian use of antidumping and safeguards over the 1992-2004 period for

industries within the manufacturing sector. The dominant user of antidumping and safeguards is industrial

chemicals, with 214 antidumping initiations and nine safeguard initiations. Other frequent users of

antidumping are iron and steel (36), other chemicals (18), machinery except electrical (17) and machinery

electric (14). Among industries that initiated safeguard investigations, each was also a user of India's

antidumping policy during this time period.

2.3 The economic importance of Indian antidumping and safeguard industry-level users

Are the industry-level users of these Indian policies economically important? The last column of table 2

presents information on the relative size of imports across sectors. Over the period 1992-2004, industrial

chemicals was not only the most frequent user of antidumping and safeguards within India, it also

competed with the largest value of imports among all Indian manufacturing industries, representing 15

percent of all Indian manufacturing imports. In some years industrial chemicals represented almost 20

12While we do not report it in the table, in 26 cases no evidence of dumping was found and in 33 cases no injury was found. Only 10 cases were withdrawn or terminated. Furthermore, in 289 of the 314 observations with non-missing information (92 percent), a preliminary duty was imposed implying that in almost all cases, petitioning firms received at least temporary protection from imports. 13 A contributing explanation to the high incidence of Indian industry "success" in antidumping and safeguard investigations (i.e., such a high share resulting in the imposition of final measures) is thus that India's use of antidumping and safeguards was not formally challenged by any trading partners under the WTO's dispute settlement provisions until December 2003. Nevertheless, in recent years Indian exporters have been increasingly targeted by other WTO members’ use of antidumping. WTO (2006) does report that its members initiated 124 antidumping cases against India between January 2005 to June 2006 alone. India as a target of foreign antidumping was only surpassed by cases against China, Korea, the U.S., Taiwan and Japan during this time period, despite India having a much smaller level of exports than these other countries.

9

percent of manufacturing imports, despite the potential trade destructive effects of the imposition of new

Indian antidumping and safeguard import restrictions. The other major industrial users of antidumping

and safeguards also face substantial competition from imports. One implication of this is the policy’s use

has potentially distorted incentives and activities in significant areas of the Indian economy.

Finally, when we match antidumping use and trade data at the 6-digit Harmonized System (HS)

level, we find that 9 percent of Indian manufacturing imports between 1992-2004 were in products

affected by antidumping or safeguard initiations when the weights are imports from 1991. When the

weights are the average of imports from 1992-2004, 13 percent of Indian manufacturing imports between

1992-2004 were in products affected by antidumping or safeguard initiations.14 While this serves as a

potential upper bound on the impact of India's use of antidumping on trade flows during this time period,

this reinforces the importance of examining India’s use of antidumping and safeguards in more depth.15

This also conforms with the aggregated gravity-model results of Vandenbussche and Zanardi (2006), who

estimate that India experienced a 10.2% annual reduction in imports as a result of its own antidumping

trade restrictions, which is of a similar magnitude as the annual average growth in its imports of 11.3% it

has been experiencing since the beginning of its trade liberalization reform in 1991.

3 The Grossman and Helpman Econometric Approach and Results

3.1 Econometric model

Our first econometric approach builds on the Grossman and Helpman (1994) model of trade protection.

Their approach has become the leading political economy model of trade protection as it begins from first

principles and derives a set of testable predictions about the determinants of protection based on

government-industry interaction. The model assumes a small open economy in which there is a numeraire

14 When measured as a share of all Indian imports, these figures are 6 percent and 9 percent, respectively. In the same period, the share of tariff lines in manufactures for which there was an antidumping or safeguard initiation is 5 percent and the share of all tariff lines is 4 percent. 15 This is an upper bound because antidumping investigations and measures are typically applied at the 8-digit level, and not all 8-digit products within a 6-digit HS category will necessarily be targeted.

10

good produced only with labor, and i = 1, …, n non-numeraire goods produced with labor and a specific

factor. The specific factor owners may organize into lobby groups and simultaneously offer the

government a contribution schedule that maps a government policy choice into a campaign contribution.

In the second stage, the government selects the trade policy vector to maximize a weighted sum of

contributions and social welfare. The model provides the following equation for equilibrium tariffs:

i

i

L

Lii

zaIt

εαα

⋅+−

= , (1)

where it is the ad valorem tariff; iI is an indicator variable that equals one if the sector is organized into a

lobby and zero otherwise; Lα denotes the fraction of the population that owns some specific factor; a is

the weight that the government places on social welfare relative to political contributions; iz is the

equilibrium ratio of domestic output to imports; and iε is the absolute value of the elasticity of import

demand defined over the world price as follows: ( ))()( *iiiiii pmppm′−=ε , where in turn im denotes

imports of good i, and ip and *ip denote the domestic and world price of good i, respectively.16

We apply the Grossman and Helpman model predictions to the case of India. Assume that

equation (1) holds for 1990 – i.e., the year prior to India's trade policy reform and thus the last year that

its tariffs were determined endogenously. Subsequent to the August 1991 IMF agreement, its tariff was

affected by an exogenous mandate, suggesting that by 1997, India's sector i applied tariff is given by:17

ii

i

L

Lii xz

aIt −⎟⎟

⎠

⎞⎜⎜⎝

⎛⋅

+−

=1997

1997,ˆ

εαα

(2)

16 To obtain iε from the elasticity defined over domestic prices, ie , that we use in the estimation, we would need to

divide the latter by )1(*iii tpp += . However, since output is measured at domestic prices while imports are

measured at world prices, we also need to divide iz by )1( it+ , which is equivalent to saying that we can directly use

ie instead of iε in equation (1) in the estimation. 17 As we describe in section 2, tariff reductions in India took place mostly between 1990 and 1997, and thus we focus on the tariff change from 1990-1997 to estimate the effects of trade liberalization on subsequent antidumping and safeguard use. Topalova (2006) provides evidence that the tariff reductions that were implemented in India during this period were largely exogenous to political economic determinants.

11

where xi is an exogenous term defined as the difference between what the 1997 tariff would have been (if

determined by the Grossman and Helpman model) and the actual 1997 applied tariff ( 1997,it ).

As table 1 indicates, India had become a relatively heavy user of antidumping by the early 2000s.

If India is exogenously constrained so that it cannot increase its applied tariffs, as arguably took place

when India committed to reduce its tariffs under the agreement with the IMF, antidumping or safeguard

duties could be used as a substitute policy instrument. Therefore, we hypothesize that the antidumping or

safeguard duty in 2000 becomes the difference between the unconstrained level that India would have

applied (under the Grossman and Helpman model) and the actual applied tariff, i.e.,

2000,20002000, ii

i

L

Lii t

zaI

−⎟⎟⎠

⎞⎜⎜⎝

⎛⋅

+−

=εα

ατ , (3)

where 2000,iτ is the antidumping or safeguard duty and 2000,it is the applied import tariff in 2000.

In order to obtain an expression for the antidumping or safeguard duty as a function of India's

1990s applied tariff change, we proceed as follows. First, add year (1990) subscripts to equation (1) to

obtain an expression for pre-reform applied tariffs as a function the Grossman and Helpman model's

determinants. Then, substituting equations (1) and (2) into (3), we obtain the following expression for the

antidumping or safeguard duty in 2000 as,

1990,1990

1997,1997

2000,2000

2000,ˆˆ

ii

i

L

Liii

i

i

L

Lii

i

i

L

Lii tz

aItxz

aItz

aI

+⎟⎟⎠

⎞⎜⎜⎝

⎛⋅

+−

−−−⎟⎟⎠

⎞⎜⎜⎝

⎛⋅

+−

+−⎟⎟⎠

⎞⎜⎜⎝

⎛⋅

+−

=εα

αεα

αεα

ατ

If we assume that ( ) ( )19901997 ))(())(( iiLLiiiLLi ezaIezaI αααα +−≈+− ,18 we can then rewrite

the previous equation as

iiiii

i

L

Lii xtttz

aI

−−−−⎟⎟⎠

⎞⎜⎜⎝

⎛⋅

+−

= )ˆ(ˆ1990,1997,2000,

20002000, εα

ατ ,

18 This amounts to assuming that the unconstrained tariff levels (i.e., in the absence of the trade liberalization that India had to implement) would have been similar in both years. Any difference that might exist that is constant across sectors would be captured in our estimation by the constant term.

12

which expresses the 2000 antidumping or safeguard duty as a function of the underlying Grossman and

Helpman structural determinants in 2000, the applied tariff in 2000, and the tariff change between 1990

and 1997. The equation that we estimate is therefore given by

iiiii

i

i

iii tttzzI μββ

εβ

εββτ +−++⎟⎟

⎠

⎞⎜⎜⎝

⎛+⎟⎟

⎠

⎞⎜⎜⎝

⎛×+= )ˆ(ˆ

1990,1997,42000,32000

22000

102000, , (4)

where 0)(11 >+= La αβ , 0)(2 <+−= LL a ααβ , 03 <β , 04 <β and iμ is the regression error term.19

Protection increases with ( )iiz ε for organized sectors and decreases in the case of unorganized sectors.

The magnitude of the deviation from free trade (in either direction) is thus higher when ( )iiz ε is higher,

because a larger output means the benefit from protection is higher for the lobby, and the welfare cost

from protection is lower the lower are the volume of imports and the elasticity of import demand. The

Grossman and Helpman model also predicts that 021 >+ ββ . Finally, the key hypothesis we want to test

is that 4β is negative – i.e., sectors with larger tariff reductions between 1990 and 1997 are expected to

obtain more protection in the form of higher antidumping or safeguard duties in 2000.

3.2 Data

3.2.1 Tariffs, antidumping and safeguard policies

Tariff reductions in India took place mostly between 1990 and 1997. Therefore, we use the tariff change

from 1990 to 1997 to estimate the effects of trade liberalization on the use of future antidumping or

safeguard (AD/SG) policy. The period for which we have complete industry-level data after 1997 is

2000/2001; hence, our benchmark estimation uses AD/SG protection imposed in 2000/2001 as the

19 The error term is included to capture potential measurement error in the variables and other factors (not accounted for in the model) that may influence the determination of trade policy. While the component of the exogenous term (xi) that was constant across sectors is captured by the constant term, the rest would go into the error term.

13

dependent variable.20 As table 1 documents, this is a useful time period to consider as it was during the

escalation period of India's antidumping use in particular.

We estimate the Grossman and Helpman model on a cross-section of data, and our unit of

observation is an imported product at the 6-digit Harmonized System (HS) level averaged over

2000/2001. Our dependent variable of antidumping and safeguards protection is defined as the AD/SG

measure coverage ratio. We use data at the exporter-product level to calculate the coverage ratio at the

product level, with product-specific information on India’s AD/SG use derived from Indian government

sources as described in the Global Antidumping Database (Bown, 2007a). We calculate the coverage ratio

as the fraction of total product imports for which an AD/SG measure was imposed, where the weights are

the average of imports of the product in 1999 and 2000 deriving from each exporting country.21 We also

complement the baseline specification by estimating the model on a variable defined as the AD/SG-

initiation coverage ratio, for which we used AD/SG initiations instead of measures imposed.22 Finally,

Indian import tariff data at the 6-digit HS level for various years over the 1990-2001 period is available

from the WTO’s Integrated Database available in WITS.23

20 We write this as 2000/2001 to highlight the fact that the model is treated as a cross-section, the explanatory variables are the average values for the 2000 and 2001 years, and the AD/SG indicators take on a value of one if a product faced a new investigation in either 2000 or 2001, since the investigation period can span multiple years. 21 Let the superscript c denote an exporting country, then the formula for the coverage ratio for import product i is:

( )[ ]( ) ( )[ ]∑∑ +×+ −+− cc

tic

ticc

ttic

tic

ti mmADSGmm 2/2/ ,1,1,,,1, , where m denotes imports, t (=2000) is the time subscript,

and cttiADSG 1,, + is an indicator that equals one if there was an AD or SG measure imposed on product i in t or t +1

against imports from a particular exporter c. Note that if 0,1, ==−c

tic

ti mm , we use ctim 1, + instead.

22 As India does not simply apply antidumping and safeguard measures in the form of ad valorem duties, but instead using complex schemes that include specific duties, price undertakings and minimum import prices, the underlying data does not have available a uniform measure of the size of new import protection. One potential solution is to construct ad valorem equivalents for these measures, but such an exercise is beyond the scope of this paper. We should point out that the use of coverage ratios has the potential problem that it may understate or overstate protection; however, they are considered the best available measure of NTBs when more detailed data is missing. Goldberg and Maggi (1999) and Gawande and Bandyopadhyay (2000) also use the NTB coverage ratio as the dependent variable in their tests of the Grossman and Helpman model. 23 Tariff data is available for 1990, 1992, 1996, 1997, 2000, 2001 and 2002.

14

3.2.2 Import data, production, elasticities, and political organization

The Indian data for other variables used to estimate the model derive from a number of sources. First, data

on production and elasticities at the 3-digit ISIC level are taken from the World Bank’s Trade and

Production database (Nicita and Olarreaga, 2007). Imports at the 6-digit HS level are taken from the

UN’s Comtrade database made available through WITS.24

As we do not have access to political campaign contribution data for Indian industries, we use

two different definitions to determine whether a given sector is politically organized. The first comes

from Cadot, Grether, and Olarreaga (2003) and is based on an iterative procedure in which they first

estimate a standard Grossman and Helpman equation on Indian data without distinguishing between

organized and unorganized sectors. They then use the residuals from this estimation to rank industries,

reclassifying those with high residuals as organized before performing a new estimation and repeating the

process iteratively until the sum of squares is minimized. They use a search grid to determine the cutoff

value used to reclassify an industry as organized.

As a robustness check, we propose an alternative method for classifying whether a sector is

organized for the purposes of receiving antidumping or safeguard protection. We identify a sector as

being organized if Indian producers of that product have ever filed an antidumping case. We believe that

the requirement to file a case and follow the necessary procedures provides a direct way to identify

organized sectors. This alternative approach also does not face the problem found in other empirical

applications that campaign contributions may understate or overstate trade-related influence activities.

Note finally that, with the exception for the tariff change variable described above, we use the

average values of the right-hand side variables in 2000 and 2001 as regressors. Table 3a presents

summary statistics for the relevant variables used to estimate the model.

24 We use the concordance files to associate HS products to ISIC industries made available in Nicita and Olarreaga (2007).

15

3.3 Estimation strategy

The dependent variable of antidumping and safeguard import protection in our model is censored below

zero. Furthermore, we have potentially endogenous variables—some of them entering nonlinearly—on

the right hand side, which include the output to import ratio, the elasticity, the organization variable and

the applied tariff. Finally, the organization variable and the elasticities may be measured with error. The

methodology we apply to address these concerns is a Tobit estimation combining the Smith-Blundell

(1986) and the Kelejian (1971) approaches.25 The methodology requires that we use least squares to

regress the right-hand-side endogenous variables and their nonlinear transformations on the instruments

and then include the residuals from these regressions as additional variables in the original antidumping

import protection equation.26 The instruments can include the exogenous variables, as well as their

quadratic terms and cross-products.

Our instruments are primarily industry characteristic data, and our choice is motivated by

previous tests of the model on other countries and trade policy settings. The variables used to instrument

for the political organization variable include the number of establishments (a measure of concentration),

value added per firm, the share of output sold as intermediate goods, the capital stock and the number of

employees. The instruments for the output to import ratio include factor shares, such as the share of

capital, skilled labor, land, and natural resources; and the capital-labor ratio. Since variables that affect

imports also affect the elasticity of import demand, these are also used as instruments for the elasticity.

Finally, given that tariffs in the Grossman and Helpman framework are also determined by the

25 Gawande and Bandyopadhyay (2000) and Gawande, Krishna and Robbins (2006) also use this procedure, although the first only reports the two-stage least square results. Although we cannot take the elasticity to the left-hand side of the protection equation as done by Goldberg and Maggi (1999) (see equation 4), the elasticity estimates that we use have much greater precision, with nearly all of them being significant at least at the 5 percent level. A number of papers adopt the approach of leaving the elasticity on the right-hand side, including Gawande and Bandyopadhyay (2000). 26 Including the residuals corrects for endogeneity in the corresponding variables and all the coefficients become consistent. If the residuals are statistically significant we can reject the null hypothesis that the variables are exogenous.

16

organization, output-import and elasticity variables, we use their instruments as instruments for the

applied tariff.27

3.4 Empirical results from the Grossman and Helpman model

The results of our Tobit estimation of determinants of the Grossman and Helpman model are reported in

table 4. We begin by following the Cadot, Grether, and Olarreaga (2003) classification of politically

organized Indian industries. Consider then column 1 which presents our baseline estimates of the

determinants of India's product-level use of antidumping and safeguards in 2000/2001. With respect to the

structural parameters of the model, we find evidence consistent with the theory that politically organized

sectors do receive more AD/SG protection than unorganized ones. In particular, the coefficient on

( )iii zI ε× (i.e., 1β ) is negative and significant at the 10 percent level, while the coefficient on ( )iiz ε

(i.e., 2β ) is positive and significant at the 1 percent level. However, the sum of those coefficients is

negative, in contrast to the model’s prediction.

Consider next our primary variables of interest on the relationship between India's applied tariff,

its recent trade liberalization experience, and its use of antidumping and safeguards in 2000/2001. First,

while the model predicts the effect of the applied tariff to be negative, the estimated coefficient of the

applied tariff in 2000/2001 (i.e., 3β ) is positive, though it is not significantly different from zero.

Nevertheless, the estimated coefficient of the 1990-1997 tariff change (i.e., 4β ) is negative and

statistically significant. Therefore, we conclude that there is evidence that the products which experienced

larger tariff reductions during the trade liberalization period (1990-1997) are also the ones that received

higher protection in the form of AD or SG measures in 2000/2001.28 This is a potentially important result

27 Some of these data are from Nicita and Olarreaga (2007) and others from Cadot, Grether and Olarreaga (2003). 28 The coefficients of the residuals are not reported but none were significant at the 10 percent level, meaning that we cannot reject the exogeneity of our regressors once the residuals are included.

17

that we explore in more detail below, as it indicates that at least part of the trade liberalization undertaken

by India was reversed with the later re-application of import-restricting measures through new protection.

The rest of table 4 presents a number of initial robustness checks on our results. For example, in

specification 2 we redefine the dependent variable. Instead of defining it as the coverage ratio of imports

affected by AD/SG measures, we define it as the coverage ratio of imports affected by the initiation of

AD/SG investigations, as prior research (e.g., Staiger and Wolak, 1994) has found the mere initiation of

an investigation can be sufficient to have a destructive effect on imports. The results are qualitatively

similar and also quantitatively close to those discussed above, which is not surprising given our

discussion in section 2 that such a large majority of Indian antidumping investigations result in the

imposition of new trade-restricting measures.

Column 3 presents a specification in which we modify the benchmark model and redefine the

indicator variable for whether an Indian industry is organized. Instead of using the Cadot, Grether, and

Olarreaga (2003) procedure, we classify a sector as being organized if it ever filed an AD case.29 In this

specification we find that both the coefficient of ( )iii zI ε× and the coefficient of ( )iiz ε are significant

at the 1 percent level. In addition, we find that the sum of their coefficients is positive, as predicted by the

model.30 The coefficient on the applied tariff is now negative, as expected, although not statistically

significant. Finally, the coefficient of the 1990-1997 tariff change is again negative and statistically

significant, as predicted by the theory.

In column 4 of table 4 we again redefine the dependent variable. Here we define the coverage

ratio as the share of imports affected by antidumping measures only (instead of antidumping and

safeguard measures), and the rest of the specification is identical to that presented in column 1. Most of

29 Note that out of the sectors that we classify as organized (unorganized) following this criterion, 60 percent (63 percent) were also organized (unorganized) under the Cadot, Grether, and Olarreaga (2003) classification. 30 However, we should point out that the residual of the first right-hand side variable is statistically significant, indicating that we can reject the exogeneity of that regressor. This is probably due to the use of our classification of organized industries, which is more correlated with the dependent variable than the one from Cadot, Grether, and Olarreaga (2003).

18

the results are unchanged from specification 1; the exception is that the coefficient on the first

variable, ( )iii zI ε× , is no longer statistically significant.

3.5 Additional sensitivity analysis to the Grossman and Helpman model

In this section we discuss the results of some additional robustness tests that are not reported in the table.

First, we use the tariff change from 1990-1996 instead of 1990-1997. This is motivated by Topalova's

(2004) argument that India’s trade policy reform between 1990 to 1996 was exogenous to domestic

political economy pressure as it was pressured by externally imposed targets, and that beginning in late

1997 some of this external pressure on tariff levels was relaxed so that domestic political economy may

have been able to affect the application of tariffs. When we substitute the 1990-1996 tariff change, our

results are quantitatively very close to those in the table. Second, we also checked whether our results are

sensitive to defining the 1990-1997 tariff change as a percentage instead of as the difference between two

levels, and the results are also unaffected. Third, the results are also robust to using an applied tariff from

years other than the average from 2000 and 2001. We used the tariff from 2000, from 1997 (we do not

have tariff data for 1998 or 1999), and also from 1992.31

The one area in which the results are sensitive is to examination of antidumping use during a time

period earlier than 2000/2001. As we detail in the appendix, we redid the estimation for the determinants

of antidumping and safeguard use during the 1997-1999 period, as a function of industry level data from

that period and a tariff change between 1990-1996. While the results of the model do change, we discuss

how these results change in intuitively appealing ways given other policy changes occurring in India

during those particular years.

31 We have also estimated the standard Grossman and Helpman model on this sample of data without augmenting it to include our two explanatory variables of interest - the applied tariff in 2000/2001 and the 1990-1997 tariff change. When we drop those variables, the results for 1β and 2β with the Cadot, Grether, and Olarreaga (2003) classification of an organized industry change only in that the first variable (I x z/e), which was significant at 10% becomes not significant. The results with our classification of industry organization are similar to those reported in the text.

19

4 Alternative Estimation Framework and Results

4.1 Probit model

The second step of our approach is to estimate an alternative model of determinants of India's product-

level antidumping use in 2000/2001. Our use of the Grossman and Helpman endogenous trade policy

model presented in the last section documents evidence of the empirical link between India's exogenous,

unilateral trade liberalization (i.e., 1990-1997 tariff change) and the subsequent product-level request for

and receipt of antidumping and safeguards protection in 2000/2001. Nevertheless, using the prior model

does present a limitation for examining antidumping. In particular, it does not fully exploit how our

available data can be used to explore additional questions about what forces drive this result as well as

other factors potentially affecting antidumping policy use.

In this section we therefore exploit an additionally available margin of the Indian data and

estimate determinants of an industry-level decision of whether to use antidumping protection against a

particular imported product from a particular exporter country.32 We use a binomial probit model to thus

estimate a reduced form relationship between political-economy determinants of antidumping protection

and a binary dependent variable that is equal to one if India faced initiation of an antidumping

investigation over a particular 6-digit HS product from a particular exporting country during 2000/2001.

This framework takes advantage of the fact that antidumping protection can be exporter specific – an

implication being that there may be foreign country-specific determinants (e.g., variation across exporter

sources) affecting the process.

This section thus differs from the first approach in that we do not estimate a structural model, but

instead we construct explanatory variables to proxy for political economy determinants of antidumping

use that prior researchers have found to affect the process when examining its use by other countries.

32 In this section we present the intuitive discussion of exporter-specific protection in terms of antidumping given that a safeguard is statutorily supposed to be applied across all exporters of a given product on an MFN basis. Nevertheless, in practice a safeguard can be applied in quite a discriminatory fashion as well (e.g., Bown and McCulloch, 2003). We confirm as a robustness check that including safeguard use does not substantially affect the results, controlling in the estimation for whether a particular exporting country was targeted by (or exempted from) each particular Indian safeguard import restriction.

20

Nevertheless, our primary focus continues to be an investigation of whether there is a link between India's

tariff reductions in 1990-1997 and the subsequent initiation of antidumping cases in 2000/2001.

4.2 Variable construction, additional data, and theoretical predictions

While the unit of observation is defined at the 6-digit HS product-exporting country level for India's

imported products in 2000/2001, for data availability reasons, our explanatory variables are constructed at

one of three levels of aggregation. Some determinants vary by product and exporter, some vary by

product only, and some are only available at the industry level.

Consider the potential determinants that vary by product and exporter. First, we use the 1999

value of 6-digit HS imports, expecting larger imports to increase the probability of initiating an

antidumping investigation. Second, we construct an indicator variable that equals one if the foreign

exporting industry had filed its own antidumping initiation against Indian exports in a 6-digit HS product

within the same 4-digit ISIC industry within the last five years. This variable is constructed from data in

the Global Antidumping Database and is designed to capture the potential for India's import-competing

industries that also export to be targeting foreign competitors with antidumping so as to retaliate against

being the target of antidumping use abroad. Third, we construct an indicator variable that equals one if

India had initiated an antidumping investigation on that product-exporter pair before 2000. This variable

may capture one of two competing effects. If our other variables are able to control for the fundamental

determinants of product-level pursuit of antidumping, we expect that the coefficient on this variable

would be negative, i.e., receipt of antidumping protection in the past decreases imports and the probability

that the industry needs new protection from the same exporter, ceteris paribus. However, a positive sign

on this coefficient may indicate that there is some product-specific component that is not otherwise being

captured through our other covariates that makes past users of antidumping more likely to request new

use. For example, this may occur given that antidumping and safeguard applications can occur at the 8-

digit level and there may be multiple 8-digit HS products within a single 6-digit HS category.

21

We have two additional variables that vary by product but not by exporter. The first is the applied

tariff averaged over 2000/2001. A higher tariff is expected to be negatively related to the probability of

initiating a case, as it indicates that the product currently receives a higher level of protection. Second, our

primary variable of interest is the product-level tariff change from 1990-1997. Once again, the tariff

change is expected to have a negative coefficient, as a larger tariff reduction would increase the incentive

for the producers to file a case in order to seek alternative protection in the form of AD or SG measures.

Finally, in our baseline regression we include a number of industry level (ISIC 3-digit) variables

that do not vary by exporter and for which there will be multiple 6-digit HS products: i.e., output, the

number of employees, the number of establishments, and the elasticity of import demand, all taken from

the Trade and Production database (Nicita and Olarreaga, 2007). A higher output is expected to be

positively related to the probability of initiating an AD/SG case, as it means that the producers have more

to gain from protection and also may have more resources to support the AD investigation costs. The

number of employees may proxy for political influence and is also expected to have a positive impact on

the probability of AD initiation. The number of establishments is inversely related to concentration in the

industry (which is likely to affect the ability to overcome the free-rider problem) and is therefore expected

to reduce the probability of an initiation. The elasticity of import demand—in absolute value—is directly

related to the deadweight loss associated with protection, and thus we expect a higher elasticity to reduce

the probability of initiating a case, as long as producers perceive that a measure would be less likely to be

imposed given its larger social cost. Since the actual and not the absolute value of the variable is used in

the estimation, we expect a positive sign for its coefficient.

As a final consideration, we note that the Indian government's use of these particular import-

restricting polices also, in principle, requires legal justification in the form of petitioning industries

providing evidence that they have faced dumped imports and are injured (antidumping) or are at least

injured (safeguards). To address the potential concern of omitted variables bias, in our preferred

specifications we also include 4-digit ISIC industry fixed effects to control for changing market

conditions at the industry level that may be associated with evidence of dumping and injury. One

22

implication of including such effects in the probit analysis is that in such specifications we focus only on

the cross-product variation within antidumping-using industries.

The summary statistics for each of the variables used in the probit analysis, as well as the

expected sign of their impact on the antidumping initiation outcome variable, is illustrated in table 3b.

4.3 Estimates from the probit model

Table 5 reports estimated marginal effects of the probit model. In addition to the determinants already

discussed, in all specifications we also include exporting country fixed effects to control for the concern

that exporting countries such as China are more likely to be targeted across products (Bown, 2007b).

Specification (5) is our baseline specification. While this is not our preferred specification because it does

not also include industry-level fixed effects, the coefficient estimates from the model provide evidence

that is nevertheless generally consistent with predictions of the theory.

Consider first the coefficient on the primary variables of interest. In specification 5, the

coefficient on the tariff change variable is negative and significant, indicating that larger tariff reductions

during the 1990s trade liberalization period increase the probability of initiating an antidumping

investigation in 2000/2001. This reinforces the results from the structural Grossman and Helpman model

that we obtained in the previous section. Second, the retaliation variable has a positive and significant

coefficient estimate. This variable indicates that, on average, India's imports of products from countries

that have recently targeted Indian exporters in the same 4-digit industry with antidumping are themselves

more likely to be the target of Indian antidumping.

The coefficients on the remaining control variables are mostly consistent with theory. Import

value has the expected sign and is statistically significant, while the applied tariff in 2000/2001 has a

negative coefficient, although insignificant. 33 The coefficients on the industry level variables for output,

33 This differs from the positive (and not statistically significant) coefficient obtained in our benchmark estimation of the Grossman and Helpman model. Note that we have also replaced the import value with specifications using import growth as well as the import penetration ratio, and the qualitative pattern of results were unaffected.

23

the number of employees and the number of establishments are all statistically significant and have the

predicted sign. The indicator variable of a previous AD or SG investigation has a positive coefficient,

though it is not statistically different from zero. 34 Finally, the one variable whose estimated effect runs

counter to the theory is the elasticity of import demand, which has a coefficient estimate that is negative.

Column 6 of table 5 presents our preferred specification in which we also control for unobserved

industry-level heterogeneity through 4-digit ISIC industry fixed effects. The first item to note is that this

reduces our sample size by two thirds, as our use of a binary dependent variable and the probit model

implies we are now only able to exploit the cross-product variation within those industries that used

antidumping against at least one of its 6-digit HS products in 2000/2001. Nevertheless, the qualitative

pattern of results is essentially unchanged from specification 5. 35 With respect to the variables of interest,

there is a negative and significant relationship between the size of the 1990-1997 tariff change and the

2000/2001 resort to antidumping, as well as a positive estimated impact of the retaliation variable.36

4.4 Additional sensitivity analysis

Columns 7 through 9 present a number of additional robustness checks to the estimation. In column 7, we

redefine the dependent variable to be a binary indicator taking on a value of one if there is an AD or SG

initiation facing a given product-exporter pair in 2000/2001.37 In specification 8, we redefine the tariff

34 We also tried replacing this indicator with an indicator of whether there was an AD measure against that exporter-product pair still in force in 2000 (from an initiation before 2000), and the results are very similar. 35 Once we add the 4-digit ISIC fixed effects, all of the variables defined at the 3-digit ISIC level (e.g., output, employment, establishments, elasticity) are dropped from the estimation. 36 The results are robust to shortening the period used to define the retaliation indicator to include initiations against India since 1998 or 1999 only. 37 Even though the SG is supposed to be applied on an MFN basis, as noted above, many exporting countries are frequently exempted from the policy for a number of reasons (Bown and McCulloch, 2003). In the case of India's application of SG during this time period, for example, it exempted a number of de minimus developing country exporters from the SG. To reflect this feature of the policy, we thus treat these particular exporters of the product also as if they did not face the SG investigation either. Note that in this specification the AD previous-initiation indicator is replaced with one based on previous AD or SG initiations. The sample size increases because we were able to include some 4-digit ISIC industries that were users of a SG - but that were not users of AD – in 2000/2001.

24

change variable. Instead of using the absolute difference in tariff levels in 1997 and 1990, in this

specification we measure it as the difference between those tariffs (each scaled by 100) divided by one

plus the average of the tariffs.38 Finally, in column (9), we report estimates from a linear probability

model instead of the binomial probit. This specification is designed to address the econometric concern

that the use of fixed effects in non-linear models has the potential problem that the estimators may be

inconsistent.39 Nevertheless, as the table indicates, the key results are unchanged under each of these

different sensitivity checks.

In addition to the sensitivity analysis illustrated in table 5, we have undertaken a number of other

robustness checks. First, we included as an explanatory variable the difference between the applied and

the bound tariff rate in 2000, defined as a percentage of the applied rate. We expect that this variable

(frequently referred to as "tariff overhang") would have a negative coefficient, i.e., that a smaller

difference indicates less flexibility for India to increase its applied tariff in 2000 while remaining

consistent with its WTO obligations and thus increasing the probability of AD or SG protection. While

the coefficient on this variable was negative and statistically significant, the result was not robust to the

way in which the variable was defined (e.g., level differences versus percentage differences, etc.). More

importantly, including this variable also did not significantly affect the estimates of the primary variables

of interest either.

Finally, to address the concern that the applied tariff in 2000/2001 could be endogenous to the

existence of an AD or SG initiation during that period, we also estimated our benchmark specifications in

the following ways: i) by excluding the 2000/2001 applied tariff variable, and ii) by replacing the

2000/2001 applied tariff variable with the tariff from an earlier year (e.g., 1997 or 1992). These

alternative specifications also did not result in any underlying changes on the coefficients of interest.

38 Specifically, we redefine the tariff change variable to be ( ) ( )[ ]2/100/100/1100/100/ 1990199719901997 tttt ++− . 39 We should also point out that Greene (2004) finds that this problem is reduced significantly as the number of observations per group increases.

25

4.5. Economic significance of the estimated effects

While the estimated effects on the variables of interest in table 5 are statistically significant, are they

economically important determinants of antidumping use? First note that, using our preferred

specification 6 from table 5, the predicted probability of an antidumping initiation is 0.0043 when the

estimated coefficients are evaluated at the mean value of each explanatory variable. In terms of the size of

the estimated marginal effects estimates, a 1 percentage point increase in the tariff reduction between

1990 and 1997 increases the probability of initiating an investigation by 0.000050 (or approximately 1

percent of the predicted probability value). Given the large tariff reductions that actually took place in

India during that period - e.g., the mean in the sample is 50 percentage points and the standard deviation

is 40 percentage points – these estimates are economically significant. A one standard deviation increase

in the tariff reduction away from its mean implies a predicted increase in the probability of a 2000/2001

investigation by 0.0022, i.e., a 51 percent increase in the predicted probability of an investigation.

Furthermore, the estimated marginal effect of the retaliation variable is also economically sizable.

A product is 1.4 percentage points more likely to initiate an antidumping investigation if it is an India

import from a trading partner that has hit an Indian export in the same industry with its own antidumping.

The size of this effect is also quite large, given that the predicted probability of an antidumping initiation

is 0.0043 when the estimated coefficients are evaluated at the means.

4.6 Industry-level analysis

In table 5 we presented results that, when averaged across industries, there is an expected negative

relationship between the size of India's 1990-1997 tariff cut and its 2000/2001 antidumping use, and there

is a positive relationship between an Indian product being the target of foreign antidumping and its

industry's own subsequent use of antidumping. Table 6 presents model estimates in which we interact the

key explanatory variables of interest with 3-digit industry-level indicator variables in order to identify

which industries’ products are driving these results.

26

Consider first specification (10) of table 6, in which we interact the tariff change variable with

industry-level indicators.40 The coefficient on the tariff change is negative for three industries: textiles

(ISIC 321), paper and products (ISIC 341), and iron and steel (ISIC 371). While the estimated marginal

effect is only statistically different from zero for two of these industries – paper and products and iron and

steel – they are major Indian users of antidumping and safeguards during 2001/2002. The implication is

that, of the sectors that are driving the results on the link between tariff liberalization and antidumping

use, they are important AD and SG users, accounting for 32 percent of the total number of products with

at least one initiation in 2000/2001, and they are significant importing industries, combined accounting

for 6.2% of India's manufactured imports during 1992-2004 (table 2).

Nevertheless, another noteworthy result is that the effect is neither negative nor statistically

different from zero for India's dominant user of antidumping and safeguards during 2000/2001 (as well as

the full 1992-2004 period) – i.e., the industrial chemicals sector (ISIC 351). What might explain cross-

product use of antidumping within industrial chemicals since it is not the size of the product-level tariff

cut between 1990-1997? To investigate one potential explanation, in specification (11) of table 6 we

interact the retaliation variable with industry level indicators. We find statistically (and economically)

significant evidence that products within industrial chemicals that are imported from countries which

have themselves targeted antidumping against Indian exports of industrial chemicals products will be

more likely to face an Indian antidumping investigation. This effect is also positive, although not

statistically significant, for products in the steel industry. Nevertheless, this is consistent with the theory

that antidumping can also be used to enforce collusive, market segmenting arrangements – as has

frequently been alleged in reasonably concentrated global industries such as chemicals and steel.

40 We proceed sequentially in this section because many fewer Indian industries are targeted by foreign antidumping use, which cause them to drop out of specification (11).

27

4.7 Duration of antidumping measures and trade liberalization

As a final exercise, we examine an additional margin along which we expect to observe a relationship

between the size of India's 1990s tariff cuts and antidumping protection – i.e., the duration of time that

antidumping measures remain in place providing protection to the domestic industry. While we have

illustrated evidence that, on average, India was more likely to use antidumping in the early 2000s in

products that had larger tariff cuts over 1990-1997, is the Indian government also providing a longer spell

of import protection to products that suffered larger tariff cuts?

We can provide some preliminary evidence consistent with this effect. Consider the set of all

products in India that received antidumping protection prior to 2001 – i.e., products for which sufficient

time elapsed for us to have data regarding whether their antidumping protection was removed prior to the

WTO-mandated 5 year period under its "Sunset Review" provisions. Calculate the mean percentage tariff

reduction from 1990-1997 for those 6-digit HS products that had their AD measures removed within five

years versus products that had the measures extended beyond five years.41 We find that the average 1990-

1997 tariff reduction was 64 percent for products with measures revoked within the five year period,

while it was 76 percent for products which had AD measures extended beyond five years. 42 We conclude

that "temporary" antidumping protection may be more likely to become "quasi-permanent" protection, the

larger was the product's original tariff cut as well.

4.8 Summary and implications of results

The result of an empirical link between the size of Indian tariff cuts in the 1990s and subsequent resort to

antidumping and safeguards is potentially important as it indicates that at least part of its trade

liberalization was reversed by the reapplication of new forms of import protection via exceptions

41 We calculate this as the difference in the tariffs from 1997 and 1990 divided by their average. 42 This differential is statistically significant at the 10 percent level for all sectors combined. We test the null hypothesis that the means are equal against the alternative that the average tariff cut is larger for products with AD measures lasting more than 5 years. For the industrial chemicals sector alone, for which we found that retaliation may have played a predominant role, there is also a difference though it is not statistically significant.

28

permitted under the WTO. This evidence is consistent with the results of Vandenbussche and Zanardi

(2006), who estimate a gravity equation for a group of countries to quantify the effects of the adoption of

AD laws on trade flows. For the case of India, they find the effects of AD measures have offset most of

the gains from trade liberalization, providing further support for the results identified here.

Finally, our results also suggest a caveat for the emerging literature that uses India’s 1990s tariff

reductions alone to study the impact of trade liberalization. In particular, our result of a relationship

between the size of the tariff reduction and subsequent use of antidumping and safeguards in a number of

economically sizable sectors indicates less dispersion in the actual reduction of protection across products

than in the tariff-only data that many prior studies have used.43 While much of this research examines data

from the period prior to India's run in up in antidumping and safeguard use, in the least, our results

identify a caveat for future research seeking to extend this approach to more recent time periods.

5 Conclusion

This paper uses India's exogenously-induced tariff reform in the 1990s to test the theory of a relationship

between trade liberalization and the imposition of new import protection via WTO-permitted policy

exceptions such as safeguards and antidumping. We exploit cross-product variation and provide evidence

that India’s resort to antidumping and safeguard protection in 2000/2001 is related to the size of its tariff

reform in 1990-1997 – i.e., the larger the tariff cut, the more likely was the product to subsequently seek

and receive new import protection under these policy exceptions. These results derive from structural

estimates of a modified Grossman and Helpman (1994) model, and they are confirmed by reduced-form

specifications able to exploit larger samples and additional margins of the underlying data. Our results

have important implications for understanding India’s import market access reforms in the 1990s, as tariff

changes are increasingly used to study the impact of trade liberalization on development in areas such as

productivity, poverty, and labor demand.

43 See again the literature in footnote 8. Note that we also found that 91 percent of the AD/SG initiations from 1992-2004 were in industries in which the standard deviation of the tariff cut was larger than the median for all industries.

29

Our evidence is consistent with theories of trade agreements as a commitment device (e.g.,

Staiger and Tabellini, 1987; Maggi and Rodriguez-Clare 1998, 2007) and the concern that policy

exceptions such as safeguards and antidumping reduce the credibility of trade liberalization

announcements. Nevertheless, despite the Indian setting providing an excellent natural experiment

allowing investigation of a product-level link between the size of the Indian tariff cuts and its subsequent

resort to such safeguard exceptions, there are important limits to how to interpret our results. Neither our

approach nor our results allow us to make the bold claim that India's knowledge of ex post access to

“exceptions” via antidumping or safeguards served to facilitate its trade liberalizing tariff cuts ex ante.

Furthermore, our product-level focus does not allow us to rule out other theories of such policy

exceptions (e.g., Bagwell and Staiger, 1990). For example, we do not test whether India's selective,

product-level use of antidumping and safeguards was an escape valve preventing a spillover of new

protectionism to other products thus maintaining a broader program of trade liberalization.

Finally, while we find economically and statistically significant effects of the tariff cuts on

average, we identify substantial heterogeneity of the size of this result across sectors. In particular, we

find tariff cuts are an important determinant of antidumping and safeguard use for products within the

steel, iron and paper industries. While these industries are frequent users of these policies and large

importers, nevertheless, we do not find evidence of such an effect within industrial chemicals, which is

India's dominant user of antidumping and safeguards during this period. However, we provide strong

evidence that retaliatory or collusive effects intended to discipline foreign competitors may drive cross-

product use within industrial chemicals. This result, in particular, speaks to the flexibility that access to

antidumping and safeguard policies provide policymakers and industries, as well as the lack of discipline

that the WTO rules on these polices had on constraining their use in India during this time period.

30

References

Bagwell, Kyle and Robert W. Staiger (1990) “A Theory of Managed Trade,” American Economic Review

80(4): 779-795.

Bagwell, Kyle and Robert W. Staiger (2005) “Enforcement, Private Political Pressure and the

GATT/WTO Escape Clause,” Journal of Legal Studies 34(2): 471-513.

Blonigen, Bruce A. and Chad P. Bown (2003) "Antidumping and Retaliation Threats," Journal of

International Economics 60(2): 249-273.

Blonigen, Bruce A. and Thomas J. Prusa (2003) “Antidumping,” in E. Kwan Choi and James Harrigan

(eds.) Handbook of International Trade Oxford, UK: Blackwell Publishers.

Bown, Chad P. (2004) “How Different are Safeguards from Antidumping? Evidence from U.S. Trade

Policies toward Steel,” Brandeis University manuscript, July.

Bown, Chad P. (2007a) “Global Antidumping Database,” (Version 3.0), Brandeis University working

paper, June [available on line at http://people.brandeis.edu/~cbown/global_ad/ .]

Bown, Chad P. (2007b) “China’s WTO Entry: Antidumping, Safeguards, and Dispute Settlement,”

NBER Working Paper No. 13349, August.

Bown, Chad P. and Meredith A. Crowley (2005) "Safeguards," in Patrick F.J. Macrory, Arthur E.

Appleton, and Michael G. Plummer (eds.), The World Trade Organization: Legal, Economic and

Political Analysis. Springer: New York.

Bown, Chad P. and Rachel McCulloch, (2003) "Nondiscrimination and the WTO Agreement on

Safeguards," World Trade Review 2(3): 327-348.

Cadot, Olivier, Jean-Marie Grether, and Marcelo Olarreaga (2003) “India’s Trade Policy for Sale: How

Much? Who Buys?” CEPR Discussion Paper no. 4168, December.

Cerra, Valerie and Sweta C. Saxena (2002) “What Caused the 1991 Currency Crisis in India?” IMF Staff

Papers 49(3), 395-425.

Crowley, Meredith A. "Why are Safeguards Needed in a Trade Agreement?" Contingent Protection in the

WTO: a Law and Economic Analysis. Kyle Bagwell, George Berman, and Petros C. Mavroidis

(eds.) Cambridge University Press: Cambridge, UK, forthcoming.

Edmonds, Eric V., Nina Pavcnik and Petia Topalova (2007) “Trade Adjustment and Human Capital

Investments: Evidence from Indian Tariff Reform,” NBER working paper no. 12884, February.

Feinberg, Robert M. and Kara M. Reynolds (2006) "The Spread of Antidumping Regimes and the Role of

Retaliation in Filings," Southern Economic Journal 72(4): 877-890.

Feinberg, Robert M. and Kara M. Reynolds (2007) “Tariff Liberalization and Increased Administered

Protection: Is there a Quid Pro Quo?” The World Economy 30(6): 948-961.

31

Fischer, Ronald D. and Thomas J. Prusa (2003) “WTO Exceptions as Insurance,” Review of International

Economics 11(5): 745-57.

Gawande, Kishore and Usree Bandyopadhyay (2000) “Is Protection for Sale? Evidence on the Grossman-

Helpman Theory of Endogenous Protection,” Review of Economics and Statistics 82(1): 139-152.

Gawande, Kishore, Pravin Krishna and Michael Robbins (2006) “Foreign Lobbies and U.S. Trade

Policy,” Review of Economics and Statistics 88(3): 563-571.

Goldberg, Pinelopi K. and Giovanni Maggi (1999) “Protection for Sale: An Empirical Investigation,”

American Economic Review 89(5): 1135-55.

Greene, William (2004) “The Behavior of the Fixed Effects Estimator in Nonlinear Models” The

Econometrics Journal 7 (1): 98-119.

Grossman, Gene M. and Elhanan Helpman (1994) “Protection for Sale,” American Economic Review

84(4): 833-50.

Hasan, Rana, Devashish Mitra and K.V. Ramaswamy (2007) “Trade Reforms, Labor Regulations and

Labor Demand Elasticities: Empirical Evidence from India,” Review of Economics and Statistics

89(3): 466-481

Hoekman, Bernard M. and Petros C. Mavroidis (1996) “Dumping, Antidumping and Antitrust,” Journal

of World Trade 30(1): 27-52.

Hoekman, Bernard and Michael M. Kostecki (2001) The Political Economy of the World Trading System:

The WTO and Beyond (2nd ed.) New York: Oxford University Press.

Kelejian, Harry (1971) “Two Stage Least Squares and Econometric Systems Linear in Parameters but

Nonlinear in the Endogenous Variables,” Journal of the American Statistical Association 66: 373-

374.

Krishna, Pravin and Devashish Mitra (1998) “Trade Liberalization, Market Discipline and Productivity

Growth: New Evidence from India,” Journal of Development Economics 56(2): 447-462.

Maggi, Giovanni and Andres Rodriguez-Clare (1998) “The Value of Trade Agreements in the Presence of

Political Pressures,” Journal of Political Economy 106(3): 574-601.

Maggi, Giovanni and Andres Rodriguez-Clare (2007) “A Political-Economy Theory of Trade

Agreements,” American Economic Review 97(4): 1374-1406.

Narayanan, Prakash (2006) “Anti-dumping in India—Present State and Future Prospects,” Journal of

World Trade 40(6): 1081-1097.

Nicita, Alessandro and Marcelo Olarreaga (2007) “Trade and Production: 1976-2004,” World Bank

Economic Review 21(1): 165-171.

Prusa, Thomas J. (1992) "Why Are So Many Antidumping Petitions Withdrawn?" Journal of

International Economics 33, 1-20.

32

Prusa, Thomas J. and Susan Skeath (2002) "The Economic and Strategic Motives for Antidumping

Filings," Weltwirtschaftliches Archiv 138: 389-413.

Pursell, Garry, Nalin Kishor, and Kanupriya Gupta (2007) " Manufacturing Protection In India Since

Independence," ASARC Working Paper 2007/07, Australian National University.

Smith, Richard J. and Richard W. Blundell (1986) “An Exogeneity Test for a Simultaneous Equation

Tobit Model with an Application to Labor Supply,” Econometrica 54(3): 679-686.

Srinivasan, T.N. (2001) “India’s Reform of External Sector Policies and Future Multilateral Trade

Negotiations,” Economic Growth Center Discussion Paper no. 830, Yale University.

Staiger, Robert W. and Guido Tabellini (1987) “Discretionary Trade Policy and Excessive Protection,”

American Economic Review 77(5): 823-837.

Staiger, Robert W. and Guido Tabellini (1999) "Do GATT Rules Help Governments Make Domestic

Commitments?" Economics and Politics 11(2): 109-144.

Staiger, Robert W. and Frank A. Wolak (1994) “Measuring Industry-Specific Protection: Antidumping in

the United States,” Brookings Papers on Economic Activity: Microeconomics 51-118.

Topalova, Petia (2004) “Trade Liberalization and Firm Productivity: The Case of India”, IMF working

paper no. 04/28.

Topalova, Petia (2006) “Trade Liberalization, Poverty and Inequality: Evidence from Indian Districts,” in

Ann Harrison (ed.) Globalization and Poverty. Chicago: University of Chicago Press.

Vandenbussche, Hylke and Maurizio Zanardi (2006) “The Chilling Effects of Antidumping Law

Proliferation,” CEPR Discussion Paper no. 5597, April.

Vandenbussche, Hylke and Maurizio Zanardi (2008) "What Explains the Proliferation of Antidumping

Laws?" Economic Policy 23: 98-103.

WTO (2007a) “Anti-dumping initiations: by exporting country” and “Anti-dumping measures: by

exporting country”, available online at http://www.wto.org/english/tratop_e/adp_e/adp_e.htm.

WTO (2007b) “Safeguard initiations: by reporting Member” and “Safeguard measures: by reporting

Member”, available online at http://www.wto.org/english/tratop_e/safeg_e/safeg_e.htm.

WTO (2008) "India and the WTO: Dispute cases involving India," available online at

http://www.wto.org/english/thewto_e/countries_e/india_e.htm .

33

Figure 1. Average Tariffs by Year for Indian Imported Products with and without AD/SG Initiations in 2000/2001 (Selected Industries)

Source: Authors’ calculations using data from Bown (2007a).

Iron and Steel (ISIC 371)

0

50

100

150

200

250

300

1990 1992 1996 1997

Paper and Products (ISIC 341)

0

20

40

60

80

100

120

1990 1992 1996 1997

Industrial Chemicals (ISIC 351)

0

20

40

60

80

100

1990 1992 1996 1997

AD/SG products No AD/SG products

34

Table 1. India’s Antidumping (AD) and Safeguard (SG) Initiations and Outcomes

Number of AD Number of Initiations Number of SG Number of Initiations Initiations with Final AD Measure* Initiations with Final SG Measure

1992 5 5 0 01993 0 0 0 01994 6 6 0 01995 6 5 0 01996 21 21 0 01997 13 13 1 11998 28 18 5 31999 63 49 3 22000 40 33 2 12001 67 60 0 02002 79 56 3 12003 32 20 1 02004 14 2 1 0

Total 374 288 16 8

* Excludes cases with only price undertakings. There was only one (in 2002).

Year

Source: Authors’ calculations using data from Bown (2007a). Note that India's antidumping statute was established in 1985, although its first investigation did not take place until 1992. India's safeguard statute was established in 1997, and its first safeguard investigation took place in that year.

35

Table 2. India’s Antidumping and Safeguard Initiations by Industry: 1992-2004

Industry (3-digit ISIC) Number of

AD Initiations

Number of Final AD Measures

Number of SG

Initiations

Number of Final SG Measures

Percentage of Manufacturing

Imports

311- Food Products 1 0 1 0 4.7 313- Beverages 0 0 0 0 0.0 314- Tobacco 0 0 0 0 0.0 321- Textiles 9 9 0 0 1.9 322- Wearing apparel except footwear 0 0 0 0 0.1 323- Leather products 0 0 0 0 0.4 324- Footwear except rubber or plastic 0 0 0 0 0.0 331- Wood products except furniture 0 0 0 0 0.1 332- Furniture except metal 0 0 0 0 0.1 341- Paper and products 9 6 1 0 2.3 342- Printing and publishing 0 0 0 0 0.5 351- Industrial chemicals 214 173 9 6 15.5 352- Other chemicals 18 17 2 2 2.7 353- Petroleum refineries 3 3 0 0 9.9 354- Misc. petroleum and coal products 2 1 0 0 0.8 355- Rubber products 2 2 0 0 0.5 356- Plastic products 0 0 0 0 0.4 361- Pottery china earthenware 3 2 0 0 0.1 362- Glass and products 2 2 0 0 0.4 369- Other non-metallic mineral products 9 4 0 0 0.3 371- Iron and Steel 36 25 0 0 3.9 372- Non-ferrous metals 8 1 0 0 13.6 381- Fabricated metal products 2 2 0 0 1.3 382- Machinery except electrical 17 7 1 0 12.6 383- Machinery electric 14 13 0 0 7.4 384- Transport equipment 0 0 0 0 5.0 385- Professional and scientific equipment 4 2 0 0 2.6 390- Other manufactured products 0 0 0 0 12.7

Source: Authors’ calculations using data from Bown (2007a).

36

Table 3. Summary Statistics

a. Data used to estimate the Grossman and Helpman Model

Variable Mean Standard

Deviation Minimum Maximum

Dependent variable: AD/SG measures coverage ratio 0.005 0.052 0 0.99 AD/SG initiations coverage ratio 0.006 0.059 0 1

Explanatory Variables:

I x z/e a 0.043 0.605 0 19.0

z/e a 0.181 1.862 0† 84.8

Applied tariff in 2000/2001 33.027 11.638 0 210

Tariff change (1997 – 1990) -47.314 38.392 -265 35

Notes: a indicates that the variable was scaled by 10,000,000. †Exact value is 0.542/10,000,000. Number of observations is 3772.

b. Data used to estimate the Binomial Probit Model

Variable Expected

Sign Mean Standard

Deviation Minimum Maximum

Dependent variable:

AD initiation 0.004 0.060 0 1 Explanatory Variables:

Tariff change (1997 – 1990) b [-]

-0.005 0.004 -0.027 0.004 % Tariff change (1997 – 1990) [-]

-0.281 0.182 -0.991 0.298 Within industry retaliation indicator [+] 0.148 0.355 0 1 Import value a [+]

0.0001 0.0014 0.000 0.049 Applied tariff in 2000/2001 b [-]

0.003 0.001 0 0.021 Output a [+]

0.118 0.078 0.004 0.320 Number of employees a [+]

0.004 0.003 0.0002 0.013 Number of establishments b [-]

0.605 0.459 0.040 2.283 Elasticity of import demand c [-]

-0.015 0.024 -0.387 -0.005 Indicator for AD initiation before 2000 ?

0.006 0.074 0 1 Indicator for AD/SG initiation before 2000 ?

0.008 0.089 0 1

Notes: a, b, and c indicate that the variable was scaled by 10,000,000, 10,000 and 100, respectively. Number of observations is 35471.

37

Table 4. Estimation of Grossman and Helpman Model's Determinants of

Indian Use of Antidumping and Safeguards

Dependent variable is 1/

Explanatory Variables AD/SG measures

coverage ratio AD/SG initiations

coverage ratio AD/SG measures

coverage ratio AD measures coverage ratio

(1) (2) (3) (4)

I x z/e 0.000017* 0.000022* 0.000740*** 0.000015 (0.000010) (0.000013) (0.000126) (0.000009) z/e -0.000028*** -0.000035*** -0.000548*** -0.000024*** (0.000010) (0.000012) (0.000110) (0.000009) Applied tariff in 2000/2001 0.0066 0.0096 -0.0085 0.0060 (0.0049) (0.0064) (0.0108) (0.0045) Tariff change (1997 – 1990) -0.0029*** -0.0043*** -0.0023** -0.0027***

(0.0010) (0.0013) (0.0011) (0.0009) Constant -2.0435*** -2.5472*** -0.8480** -1.8742***

(0.3049) (0.3955) (0.3657) (0.2836)

Method of determining industry organization variable

Cadot et al Cadot et al Industry initiated

any AD investigations

Cadot et al

Log likelihood -260.20 -256.32 -159.51 -247.55 Observations2/ 3772 3772 3772 3772

Notes: Standard errors of the tobit model's estimates are in parentheses with *, **, and *** indicating statistically significant at 10%, 5% and 1% levels, respectively. 1/ Calculated as the share of imports for which there was a measure/initiation in either 2000 or 2001. 2/ Observations are cross section of 6-digit HS products averaged over 2000/2001.

38

Table 5. Marginal Effects Estimates of the Probit Model of Antidumping and Safeguard Initiations

Binary dependent variable =1 if there is AD initiation 1/

Baseline Add industry fixed effects

Change dependent

variable to AD/SG initiation

Redefine tariff change

variable

Linear probability

model Explanatory Variables (5) (6) (7) (8) (9)

-0.063*** -0.504*** -0.470*** -- -1.191** Tariff change (1997 – 1990) b

(0.022) (0.139) (0.154) (0.530) % Tariff change (1997 – 1990) -- -- -- -0.011*** -- (0.004)

0.003** 0.014*** 0.013*** 0.013*** 0.010** Within industry retaliation indicator (0.001) (0.005) (0.005) (0.005) (0.005) Import value a

0.029* 1.660*** 1.332** 1.357** 6.288** (0.017) (0.597) (0.555) (0.558) (2.675) Applied tariff in 2000/2001 b

-0.155 -1.774 -1.204 -0.980 -1.745 (0.097) (1.144) (1.114) (1.138) (2.035) Output a

0.009*** -- -- -- -- (0.003) Number of employees a

0.122* -- -- -- -- (0.067) Number of establishments b

-0.0019*** -- -- -- -- (0.0006) Elasticity of import demand c

-0.006** -- -- -- -- (0.003)

0.003 0.017* -- -- -- Indicator for AD initiation before 2000 (0.002) (0.010)

-- -- 0.015* 0.015* 0.027* (0.008) (0.008) (0.014)

Indicator for AD/SG initiation before 2000

Exporter fixed effects Yes Yes Yes Yes Yes 4-digit ISIC fixed effects No Yes Yes Yes Yes

Log likelihood -674.03 -592.74 -758.61 -762.47 -- Pseudo R2 or R2 0.20 0.15 0.12 0.12 0.02 Observations 2/ 35471 11400 14632 14632 14632

Notes: Standard errors adjusted for clustering at the 6-digit HS product level in parentheses where *, **, and *** indicate statistically significant at 10%, 5% and 1% levels, respectively. a, b, and c indicate that the variable was scaled by 10,000,000, 10,000 and 100, respectively. 1/ Variable equals 1 if there is at least one AD/SG initiation in 2000 or 2001. 2/ Observations are cross section of 6-digit HS product-exporter combinations.

39

Table 6. Marginal Effects Estimates of the Probit Model of Antidumping and Safeguard Initiations: Industry Effects

Binary dependent variable = 1 if

there is AD/SG initiation 1/ Explanatory Variables (10) (11) Tariff change (1997 – 1990) b -- -0.695*** (0.210) Textiles x Tariff change (1997 – 1990) b -0.126 -- (0.112) Paper products x Tariff change (1997 – 1990) b -1.812* -- (0.956) Industrial chemicals x Tariff change (1997 – 1990) b 0.434 -- (0.338) Iron and steel x Tariff change (1997 – 1990) b -0.739*** -- (0.243)

Within industry retaliation indicator 0.012*** -- (0.004) Industrial chemicals x Within industry retaliation indicator -- 0.015** (0.006) Iron and steel x Within industry retaliation indicator -- 0.014 (0.009) Indicator for AD/SG initiation before 2000 0.014* 0.025* (0.007) (0.013) Import value a 1.175** 2.081*** (0.511) (0.744) Applied tariff in 2000/2001 b -1.331 -1.470 (1.030) (1.697) Exporter fixed effects Yes Yes 4-digit ISIC fixed effects Yes Yes

Log likelihood -742.67 -656.71 Pseudo R2 0.14 0.09 Observations 2/ 14632 11783

Notes: Standard errors adjusted for clustering at the 6-digit HS product level in parentheses where *, **, and *** indicate statistically significant at 10%, 5% and 1% levels, respectively. a and b indicate that the variable was scaled by 10,000,000 and 10,000, respectively. 1/ Variable equals 1 if there is at least one AD/SG initiation in 2000 or 2001. 2/ Observations are cross section of 6-digit HS product-exporter combinations.

40

Appendix: Antidumping Initiations in 1997-1999:

We redid our estimation of equation (4) and the Grossman and Helpman model replacing the 2000/2001

antidumping and safeguard initiations with initiations for the 1997-1999 period, and thus using the tariff

change from 1990 to 1996. The results are shown in table 7. Specification (12) presents the estimates

when the dependent variable is the antidumping and safeguard measure coverage ratio. The statistically

significant coefficients of ( )iii zI ε× and ( )iiz ε are positive and negative, respectively (as predicted by

the model). Nevertheless, their corresponding first-stage residuals were statistically significant. The

applied tariff is again positive but not significant. 44

When estimating the model on this sample of data, the estimated impact of the 1990-1996 tariff

change on AD/SG is now positive and significant, a counter-intuitive result. This indicates that products

that received smaller tariff reductions from 1990 to 1996 also received higher AD/SG protection during

1997-1999. However, closer inspection of the data reveals a potentially intuitive political economy

explanation for this phenomenon – during the 1997-1999 period, the applied tariffs on many products

were actually raised (see again the discussion above in section 2.1), and there is a negative correlation in

the raw data between the size of the 1990-1996 tariff cut and the 1997-1999 tariff increase.45 The sectors

that engaged in less tariff cutting during 1990-1996 (perhaps because they were politically powerful) were

able to raise their tariffs in 1997-1999.

One interpretation of the positive sign of the impact of the tariff cut variable for the 1997-1999

period is that there is a complementary nature between trade policy instruments during this time period.

The same industries that were receiving increases in their applied tariffs were also more likely to pursue

protection via initiating AD/SG petitions, perhaps as an insurance plan in case their requests for applied

tariff increases were denied.

44 We use the applied tariff from 1997, as we do not have tariff data for 1998 or 1999. We also tried using the average of the tariffs from 1997 and 2000 and the results did not change much. 45 For example, during this period the percentage of products with tariff increases was actually higher than the percentage of products with tariff reductions (see also Topalova, 2006).

41

Table 7 shows this result for the 1997-1999 period is robust to the same specification changes that

we reported in table 4 when estimating the model on AD/SG initiations data from the 2000/2001 period

instead. The exception is when we use our classification of organized and unorganized industries (column

14) the tariff change becomes not significant and, in contrast to what we found for the 2000/2001 period,

when we use only AD measures as the dependent variable (column 15), the coefficient on ( )iii zI ε× is

significant at the 1 percent level.

Note that in unreported results, we also estimated similar specifications of our reduced form

probit model on data from the 1997 to 1999 period. The results regarding the tariff change are similar to

those obtained in our estimation of the Grossman and Helpman model for the same period. Again, our

interpretation of this result is that during 1997-1999 period applied tariffs were changing and actually

increasing for several products, thus signaling a period of a complementary relationship between

increased tariff protection and increased antidumping and safeguards protection.

Because applied tariffs themselves were changing in 1997-1999 in India, we conclude that it was

not a particularly "clean" period to examine our question of interest regarding the link between trade

liberalization and subsequent resort to antidumping and safeguards. Hence the body of the paper focuses

our model estimates on the 2000/2001 period.

42

Table 7. Grossman and Helpman Model Estimates of India's use of Antidumping and Safeguards in

1997-1999

Dependent variable1/

Explanatory Variables AD/SG measures coverage ratio

AD/SG initiations coverage ratio

AD/SG measures coverage ratio

AD measures coverage ratio

(12) (13) (14) (15)

I x z/e 0.00012*** 0.00011*** 0.00338*** 0.00011***

(0.00004) (0.00004) (0.00076) (0.00004)

z/e -0.00015*** -0.00013*** -0.00335*** -0.00014***

(0.00004) (0.00004) (0.00076) (0.00004)

Applied tariff in 1997 0.0012 -0.0083 0.0010 0.0003

(0.0101) (0.0123) (0.0124) (0.0100) Tariff change (1996 – 1990) 0.0057** 0.0047** 0.0006 0.0058**

(0.0026) (0.0023) (0.0016) (0.0026)

Constant -1.9098*** -1.6980*** -0.6995 -1.8149***

(0.4420) (0.4754) (0.4636) (0.4344)

Method of determining industry organization variable

Cadot et al Cadot et al Industry initiated

any AD investigations

Cadot et al

Log likelihood -261.40 -305.67 -133.79 -248.17 Observations2/ 3983 3983 3983 3983

Notes: Standard errors of the tobit model's estimates are in parentheses with *, **, and *** indicating statistically significant at 10%, 5% and 1% levels, respectively. 1/ Calculated as the share of imports for which there was a measure/initiation in any year from 1997-1999. 2/ Observations are cross section of 6-digit HS products.