Multiproduct firms and antidumping duties: Evidence from India

ABSTRACT

With the growing availability of high-quality higher-dimension data in international trade, many new stylized facts have also emerged. One such stylized fact is that multiproduct firms play a significant role in international trade. In this paper, we investigate the effect of US antidumping duties on the exports of Indian multiproduct firms. In particular, we study whether US antidumping duties lead the Indian exporter to alter their product-scope to third country markets (aka to trade partners other than the US). Using a unique transaction-level data from India, we find that firms affected by US antidumping duties increased the number of products exported to other destinations by about 0.7 products, on average. This translates to a substantial 40% increase in the product-scope of these firms because a typical Indian exporting firm exported an average of 1.8 products to a given destination in our sample. We also find that firms whose products spanned multiple sectors drove most of this increase. However, we do not find any difference in the product-scope response of firms producing differentiated vs. those producing homogenous products. We find our results to be robust across various specification and sample size changes.

KEYWORDS:

• Antidumping duties
• India
• multiproduct firms
• product scope

JEL CLASSIFICATIONS:

• F12
• F13
• F15
• L25

Acknowledgment

I thank Sweta Ghosh, Kara Reynolds, the editor, two anonymous reviewers, and the participants at the Southern Economic Association conference at New Orleans in November 2015 for many helpful comments and suggestions. All remaining errors and omissions are my own.

Disclosure statement

No potential conflict of interest was reported by the author.

ORCID

Piyush Chandra http://orcid.org/0000-0001-6182-866X

Notes

1 The idea is that each firm will increase the product variety until the cost of introducing marginal variety is just sufficient to cover the variety introduction fee. Hence, with the rising variety fee the most productive firms will also be the ones with the largest product scope. Since globalization would tend to increase the profit of the marginal variety of the most productive firms, these firms would increase their product scope.

2 In addition to the conflicting theoretical predictions, empirical evidence on the effect of trade liberalization on product scope is mixed. For instance, Baldwin and Gu (2009) study the impact of CAFTA and find that trade liberalization reduced the product scope but only for smaller Canadian firms. Similarly, Iacovone and Javorcik (2010) provide evidence of product scope reduction for Mexican firms following NAFTA. Berthou and Fontagné (2013), on the other hand, find evidence of an increase in the product scope for French firms following a reduction in trade costs spurred by the introduction of the Euro.

3 Bown and Crowley (2007) identify four distinct effects of the imposition of a trade barrier that is not common across all trade partners. In particular, imposition of a trade barrier by Country A against country B, but not against country C, could (a) cause country B to increase its exports to country C (trade deflection); (b) cause country B to import less from country C, as country B exporters would sell part of the lost sales to country A domestically (trade depression); (c) increase country C’s exports to A (trade diversion); or, (d) reduce the exports of B to A (trade destruction).

4 Using firm level data from China between 2000 and 2006, Chandra and Long (2013) find that those firms that were specifically targeted in the US antidumping investigation faced up to 10% decrease in their productivity.

5 In a recent paper, Vandenbussche and Viegelahn (2018) also study the impact of antidumping duties on multiproduct firms in India. However, they investigate the impact of antidumping duties on the firms’ choice of inputs. They find that an increase in import protection on some inputs causes firms to switch to the unprotected inputs, which in turn also leads to product switching in the output market. Note that, this is different from the current paper in two respects. First, we are interested in studying the impact of antidumping duties on firms’ final output as opposed to its input. Second, and more importantly, the antidumping duty is imposed by India’s trade partner in this paper, whereas, they are interested in investigating the impact of India’s own antidumping duty actions.

6 If our dependent variable were export of a given product from India to the US, regressing it on whether or not there is a US antidumping duty on that product, would lead to inconsistent estimates due to the presence of reverse causality.

7 Once the antidumping duty is imposed on a given product, the size of the antidumping duty can itself vary across firms (Chandra 2016; Chandra and Long 2013). However, in this paper, we do not exploit across firm differences in the intensity of duties. Instead, we classify the firm as facing an antidumping duty as long it exported any product on which the US had imposed an antidumping duty. This will actually bias against us finding any effects.

8 Previous studies have shown that US macroeconomic environment might affect the likelihood of imposition of antidumping duty (Knetter and Prusa 2003).

9 We use the reghdfe module by Correia (2017) to implement the multiway clustering in STATA.

10 Eleven Indian ports in the database are Calcutta (air, sea), Chennai (air, sea), Cochin (air, sea), Haldia (sea), Mangalore (sea), Mumbai (air, sea) and Visakhapatnam (sea).

11 We tried using a user-written STATA program reclink2 by Wasi and Flaaen (2015) as well as writing our own matching program based on a similar fuzzy algorithm. However, none of them produced satisfactory results. We supplemented these algorithms by manually matching the firm names.

12 A complete list of the countries in our sample is provided in the appendix in Table A1.

13 Croatia was not officially a member when we started the project and hence is not included in the list of EU countries. However, this affects a minuscule number of observations in our sample and is unlikely to change any of our results.

14 A complete list of antidumping investigations and its outcome included in our sample is provided in the appendix in Table A2.

15 Anderson et al. (2018) show that even within a single eight-digit code there could be multiple products as verified by their product names and unit prices etc.

16 All estimates based on (2) use HS six-digit classification to classify products as HS classifications are common across countries only up to six-digits. Everywhere else in the paper where we are only interested in whether that firm has a US antidumping duty imposed, we use HS eight-digit nomenclature to classify a product.

17 One such as demand side variable that affects all firms is import tariffs. Note that, while import tariffs are not strictly time-invariant, they did not change much during our sample. An alternative specification, where we include import tariffs directly into the regressions instead of ${\alpha }_{hy}$ fixed-effect does not change our results.

18 (exp(−0.17)−1).

19 i.e. $TotalExport{s}_{fht}=\sum _{k\ne h}Export{s}_{fkt}$.

20 See, for instance, Blonigen and Prusa (2008), Chandra and Long (2013).

21 Another potential size measure would be to control for firms’ total sales or firms’ productivity. Unfortunately, our data comes from customs and does not have firm level information. However, as is widely known, larger and more productive firms are more likely to be exporters, have larger exports, as well as export to more destinations. Thus, it is likely that our results would not change even after the inclusion of those additional variables. Unfortunately, in the absence of a measure of productivity, we do not have a way to directly test this assertion.

22 Note that, our measure of firm-specific tariff faced by the Indian firm in a given time period is slightly different from Qiu and Zhou (2013). While they were interested in a firm-specific trade cost measure, our variable varies by both firm and destination. In Tables 12 and 13, we show that our results remain unchanged if we ignore the destination dimension and create a firm-specific tariff as in Qiu and Zhou (2013).

23 To clarify, let us write Equation (1) using the usual difference-in-difference estimation structure: (1a) $NProduct{s}_{fct}={\alpha }_{fc}+{\alpha }_{ct}+{\beta }_{1}Treatmen{t}_f+{\beta }_2{\tau }_{USA,ft}+{\beta }_3{\tau }_{USA,ft}\ast Treatmen{t}_f+\gamma Z_{fct}+{\epsilon }_{fct},$(1a) where NProducts_fct is the number of products exported by firm f to country c in time t, and ${\alpha }_{fc}$ and ${\alpha }_{ct}$ are the firm*country and country*time fixed effects, as above. Similarly, ${\tau }_{USA,ft}$ is an indicator variable which shows whether firm f faces a US antidumping duty in period t, and $Z_{fct}$ is a vector of control variables. The main difference between Equations (1) and (4) is the inclusion of $Treatmen{t}_f$, which is defined as an indicator that takes the value one for all firms that had an antidumping duty imposed against them by the US during our sample period. Whereas, $Treatmen{t}_f$ is zero for those products where US started an antidumping investigation but where no antidumping duty was eventually imposed. The coefficient on the interaction term, ${\beta }_3$ then gives us the difference-in-difference estimate of the US antidumping duty. The advantage of the panel data structure is that since $Treatmen{t}_f$ is time-invariant, it drops out of the estimating equation. Similarly, the interaction between ${\tau }_{USA,ft}\ast Treatmen{t}_f$ is also functionally equivalent to ${\tau }_{USA,ft}$. Thus, our estimating equation reduces to (1b) $NProduct{s}_{fct}={\alpha }_{fc}+{\alpha }_{ct}+{\beta }_3{\tau }_{USA,ft}+\gamma Z_{fct}+{\epsilon }_{fct},$(1b) which is identical to Equation (1). Here ${\beta }_3$ can be interpreted as the differential impact of the treatment on the treated under the usual common trend assumption (Wooldridge 2002; Bertrand, Duflo, and Mullainathan 2004).

24 Note that in addition to a negative decision, the petitioner also has the right to withdraw the petition at any point, which would also terminate the investigation.

25 For a complete list of the US antidumping investigations included in our sample period as well as their outcome, please see appendix Table A2.

26 See also, Dhingra (2013), who emphasizes competing effects of trade liberalization through increased economies of scale and through higher competition from other firms. According to her framework, a firm facing tougher competition might be induced to lower product lines to reduce within-brand cannibalization. However, lowering product lines also exposes the firms to reduced brand visibility, and hence, to tougher between-brand competitions.

27 Rauch (1999) uses SITC rev. 2 nomenclature in its classifications. We used the publicly available concordance from WITS to concord these products to HS six-digit classification before matching. Also, we used the liberal definition of product differentiation available from Rauch (1999) but changing it to a more conservative definition does not alter our results.

28 An alternative would be to use the share of products belonging to the differentiated products category or the corresponding weighted shares coupled with a threshold. However, we have not tried these alternative definitions.

29 Note that, there is some controversy as to whether the fixed effects negative binomial is the appropriate method in these scenarios. For instance, Allison and Waterman (2002) argue that the fixed effects negative binomial regressions proposed by Hausman, Hall, and Griliches (1984) are not true fixed effects. In any case, one needs to note while interpreting the results from Poisson and the negative binomial regression in the case of panel data, that the negative binomial fixed effects is not the fixed effect in the same sense as the fixed effects in Poisson. In the case of negative binomial, fixed effects refer to the dispersion parameter.

30 We used STATA module by Guimaraes and Portugal (2010) for Poisson estimates. Also, while the standard errors are clustered at firm-country level for Poisson estimates, no robust or clustered standard errors are available for the negative binomial regressions.

31 Developed countries refer to those classified as high-income countries by the World Bank.

32 The list of countries is given in Table A1. We have a total of 39 countries in Table A1 including the US. All of our main results except in Tables 3 and 4 exclude the US.

33 These results are available from the author on demand.