The trade effects of antidumping duties: Evidence from the 2004 EU enlargement

The trade effects of antidumping duties: Evidence from the 2004 EU enlargement

Journal Pre-proof The trade effects of antidumping duties: Evidence from the 2004 EU enlargement Alexander-Nikolai Sandkamp PII: S0022-1996(20)30026...

3MB Sizes 1 Downloads 30 Views

Journal Pre-proof The trade effects of antidumping duties: Evidence from the 2004 EU enlargement

Alexander-Nikolai Sandkamp PII:

S0022-1996(20)30026-X

DOI:

https://doi.org/10.1016/j.jinteco.2020.103307

Reference:

INEC 103307

To appear in:

Journal of International Economics

Received date:

7 June 2018

Revised date:

9 January 2020

Accepted date:

27 January 2020

Please cite this article as: A.-N. Sandkamp, The trade effects of antidumping duties: Evidence from the 2004 EU enlargement, Journal of International Economics(2020), https://doi.org/10.1016/j.jinteco.2020.103307

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

© 2020 Published by Elsevier.

Journal Pre-proof

The Trade Effects of Antidumping Duties: Evidence from the 2004 EU Enlargement Alexander-Nikolai Sandkamp*

Draft

of

This version: January 2020

Abstract

ro

This paper exploits the 2004 EU enlargement as a natural experiment to estimate the effects of antidumping (AD) duties on import prices and quantities. The automatic extension of EU AD

-p

policy to new member states following their accession is exogenous to new members’ trade

re

shocks. Using this source of variation, the paper shows that, on average, AD duties raise producer prices of targeted exporters. However, import prices from countries with non-market economy

lP

status (NMES) - notably China - remain unchanged, while quantities fall more strongly. This is in line with expectations, as the EU applies special rules to NMES countries. The trade dampening

na

effects of AD duties persist over time. Moreover, the duties appear to induce exporters from

ur

non-targeted countries to increase their prices.

Keywords: Antidumping, Trade, European Union, Market Economy Status

Jo

JEL classification: F13, F14

*

University of Kiel (CAU), IfW Kiel & ifo Institute; Wilhelm-Seelig-Plat z 1, 24118 Kiel, Germany; Phone: +49

431 880 5631; E-mail: [email protected] I would like to thank two anonymous referees and the editor, Ralph Ossa, for excellent co mments and suggestions as well as Gabriel Felbermayr for his support throughout this project. I am also grateful to Andrea Ariu, Daniel Bau mgarten, Meredith Crowley, Carsten Eckel, Lisandra Flach, Jasmin Gröschl, Anna Gu mpert, Julian Hin z, Monika Schnitzer, Vincent Stamer as well as to participants of the Industrial Organization and Spatial Econo mics Conference 2018, the Aarhus-Kiel Workshop 2019, the ifo Center for International Economics Internal Seminar, the LM U IO and Trade Seminar, the IfW Research Seminar and the Trier University Research Colloquium for their extremely helpful remarks.

Journal Pre-proof

1

Introduction

Antidumping (AD) duties are by far the most frequently used trade defence instruments: in 2017, 1,600 such measures were in force worldwide. 1 However, estimating their effects on trade flows and prices is difficult. AD duties typically are imposed as a response to an increase in imports. Hence, endogeneity is inherent in these trade policy instruments. To tackle this problem, this paper exploits the EU enlargement of 2004 as a natural experiment. At the time of their accession to the EU, the new member states were required to adopt the existing EU AD policy. Under the assumption that the decision to join the EU was independent of

of

existing AD duties, the enlargement constitutes an exogenous treatment of new member states.

ro

Therefore, one can use a simple difference- in-differences regression to estimate the effect of AD duties on producer (i.e. pre-duty) import prices and quantities. The approach exploits the change

-p

over time in import prices and quantities of treated exporting country-product combinations

re

relative to non-treated ones.

In addition, the natural experiment can shed light on the impacts of a recent policy change.

lP

In December 2017, the EU has adjusted its AD regulation, officially abandoning the concept of Non-Market Economy Status (NMES). 2 This may have important implications because Market

na

Economy Status (MES, assigned to the exporter by the imposing country) determines the way AD duties are calculated. In a nutshell, exporters in countries with MES receive individual firm

ur

specific duties, whereas NMES exporters are subject to a - typically larger - country-wide duty that is the same for all exporting firms. For this reason, the effects of EU AD duties on import prices

1

Jo

and quantities should differ between exporting countries with MES and those with NMES. This Data on g lobal AD measures in force is taken fro m the WTO’s I-TIP database (WTO, 2018). Du mping is defined

as exporting a product at a price below its normal value (WTO, 1994), where normal value is typically the domestic price of the product in the exporting country (for a detailed d iscussion s ee for example Felbermayr et al. (2016)). It is a common phenomenon in international trade that can have many causes, such as international price discrimination (Viner, 1923), production under demand uncertainty (Ethier, 1982), reciprocal dumping with oligopolistic firms (Brander and Krugman, 1983), dynamic competition (Gruenspecht, 1988; Clarida, 1993), subsidies (Dixit, 1988; Blonigen and Wilson, 2010) or cyclical aspects (Staiger and Wolak, 1992). WTO rules allow member states to counteract dumping practices with antidumping duties. 2

Regulation (EU) 2017/2321 (Europeran Parliament, 2017). The EU legislation now refers to the concept of

significant distortions which permits the use of alternative methodologies to calculate dumping margins. The criteria to identify these distortions are similar in part to the old NMES methodology.

Journal Pre-proof paper provides a test for this prediction. The paper shows that AD duties raise pre-duty import prices on average by 25%. However, prices only increase for imports originating from countries with MES. Producer prices of imports from non- market economies remain unchanged. Import quantities on average fall by 74%. This effect is stronger for NMES exporters (on average 85% compared to 68% for MES exporters) which is due to the higher average duties they face. Price as well as quantity effects of AD duties persist over time, even after their revocation. Finally, producer prices of imports from countries not targeted by AD duties also increase. AD duties may thus have signalling effec ts, causing

of

non-targeted exporters to update their beliefs of becoming subject to AD investigations and raise

ro

prices in anticipation.

This paper relates to three strands of literature, namely the impact of AD duties on import

-p

prices, on quantities as well as on third countries. Regarding import prices, AD duties work through two channels. Like tariffs, they directly increase consumer prices (assuming positive

re

pass-through). In addition, and in contrast to ordinary tariffs, they should incentivise exporters to

lP

raise their prices. Having the official objective to protect the importer’s domestic market from “unfair” foreign competition (European Parliament, 2017; European Commission, 2016), AD duties can be adjusted if the exporter increases ex-factory prices (Feenstra, 2008).

na

Consequently, theory predicts pass-through rates larger than 100% as exporters increase prices to achieve a reduction of AD duties in subsequent periods (Blonigen and Haynes, 1999;

ur

Blonigen and Park, 2004). Over time, this results in a worsening of the importer’s terms of trade.

Jo

Hence, in addition to determining the effectiveness of this policy instrument, measuring price responses of exporters constitutes an important component when evaluating the effects of AD duties on welfare.

Empirical investigations of price effects of AD duties remain scarce. Blonigen and Haynes (2002) examine the impact of US AD duties on imports of iron and steel from Canada. They find that AD duties indeed lead to higher pre-duty import prices from the point of view of the AD imposer. Lu et al. (2013) use Chinese customs data to investigate the effect of US AD duties on Chinese exports to the US. The authors do not find positive price effects. Indirect evidence for price effects is provided by Gourlay and Reynolds (2012), who show that US duties against China (and other countries) fall following the first review. This indicates an increase in producer prices. Nita and Zanardi (2013) find that EU duties against China increase

Journal Pre-proof slightly after the first interim review (they fall for other countries), indicating further dumping. Whether the price effects of AD duties depend on the MES of the exporter has so far been completely ignored by existing studies. This paper adds to the literature by investigating the universe of EU imports. This enables a comparison of the effects of AD duties across different exporting countries. The findings indicate that the seemingly conflicting results of Blonigen and Haynes (2002) (increasing producer prices following AD duties against Canada) and Lu et al. (2013) (no producer price effects for exports from China) are driven by the MES of the exporters considered in the respective studies. While the

of

effect of AD duties on average export prices is positive, these price increases are driven by

ro

exporters from MES countries such as Canada. Import prices from NMES countries (including China) remain unchanged. Hence, this paper is the first to identify differential trade effects of AD

-p

duties by MES of the exporter.3

The second strand of literature to which this paper contributes concerns the effects of AD

re

duties on import values and quantities. 4 Prusa (1997, 2001) investigates the implementation of US

lP

AD duties, showing that they reduce US imports from targeted countries by up to 50%. In contrast to that, Egger and Nelson (2011) find much smaller effects. 5 For the European Union, Messerlin (1989), Lasagni (2000) and Konings et al. (2001) estimate treatment effects similar in magnitude

na

to those of Prusa (1997, 2001). 6 Vandenbussche and Zanardi (2010) look at several AD imposing countries, finding that AD duties imposed by the so called “new adopters” have trade chilling

ur

effects on bilateral trade flows. Following the availability of firm- level export data, a growing

3

Jo

literature is also starting to look at impacts of AD duties on exporting firms.7

Existing studies are either descriptive, co mparing levels of AD duties for MES and NM ES exporters (Det lof and

Fridh, 2006; Felbermayr et al., 2016; Felbermayr and Sandkamp, 2019) or look at the effect of MES on the number of AD investigations (Urdinez and Masiero, 2015). 4

For an overview see for example Bolnigen and Prusa (2003, 2016) and Nelson (2006).

5

Other studies include the investigation of individual stages of the AD process (Staiger and Wolak, 1994) as well

as particular sectors (Carter and Gunning-Trant, 2010). 6

The AD process itself also plays a role for the EU, with Bran (2015) finding that withdrawn or rejected cases only

have temporary effects, while trade effects of final duties are strong and lasting. 7

Besedeš and Prusa (2013) find US AD to induce firm exit. Lu et al. (2013) show that US AD duties against China

induce exit and reduce exports of surviving firms. Jabbour et al. (2019) demonstrate that Chinese exporters reduce exports to the EU following the imposition of EU AD duties, but also become larger and more productive. Felbermayr

Journal Pre-proof The above studies potentially suffer from endogeneity bias due to simultaneity of AD duties and imports. AD duties typically increase consumer prices and thus reduce import quantities of targeted products. However, being designed to protect domestic industry, they are more likely to be imposed on products with low prices and high import quantities. This simultaneity of imports and AD duties results in biased estimates of the treatment effect (Bown and Crowley, 2013). For quantity effects, the bias is likely to be positive, leading to an underestimation of the (negative) treatment effect. 8 For prices - assuming that AD duties do indeed raise prices - the bias is likely to be negative, as AD duties are more likely to be implemented on products for which dumping is

of

suspected, i.e. import prices are low.

ro

Measuring these effects correctly is, however, essential for inferring the consequences on welfare resulting from trade destruction and increased prices of intermediates. It is also required

-p

for the effective design of AD policies. This paper adds to the above strand of literature by exploiting the EU enlargement of 2004 as a natural experiment to tackle simultaneity (as well as

re

omitted variable bias) and obtain unbiased estimates of the effect of AD on imports. Estimated

lP

effects are larger than those found by previous studies, indicating that these may indeed suffer from simultaneity bias, which results in an underestimation of the treatment effect. Third, the paper relates to the literature on the effects of trade policies on untargeted

na

countries. Bown and Crowley (2007) and Baylis and Perloff (2010) find evidence for trade deflection (increased exports to third countries) following the imposition of US AD duties against

ur

Japanese and Mexican exports respectively. Similarly, Nguyen et al. (2016) show that EU duties

Jo

imposed on Vietnamese footwear increase Vietnamese exports to the US. Chandra (2016) finds evidence for trade deflection following the imposition of US temporar y trade barriers against China. Other studies do not find systematic evidence for trade deflection following the imposition of US AD duties (Lu et al., 2013) or EU and US import restrictions (Bown and Crowley, 2010) against China. Prusa (1997) shows that US AD duties divert imports from targeted to non-targeted countries. This paper finds that producer prices of imports from non-targeted countries increase and Sandkamp (2019) provide evidence that both EU and US AD duties decrease firm exports and induce exit, with small firms being affected most severely. 8

Felbermayr and Sandkamp (2019) show that not accounting for demand side effects that are correlated with the

decision to implement AD duties results in an underestimation of the true treatment effect.

Journal Pre-proof following the imposition of AD duties against another country. This result can be explained by the uncertainty surrounding the AD process. Dumping allegations for the same product are often investigated separately for different exporting countries. Given the uncertainty surrounding the AD investigation process (Blonigen and Park, 2004), the imposition of AD duties against one exporting country may induce producers of the same product in other exporting countries to update their beliefs regarding the likelihood of being investigated and becoming subject to duties. They thus raise prices in order to avoid future duties. In line with findings by Prusa (1997), the price increase is smaller than the one observed for targeted countries. 9

of

The remainder of this paper is structured as follows. Section 2 outlines the EU AD

ro

methodology and its implications for exporter behaviour. Section 3 presents the empirical strategy. This is followed by an overview of the data used, including some descriptive evidence in Section

-p

4. Section 5 presents the core results of the paper. Section 6 offers several extensions and

The EU AD Methodology

lP

2

re

robustness checks and Section 7 concludes.

In order to understand why one might expect different effects of AD duties on exports from MES

na

and NMES countries, this section starts by briefly outlining the EU’s AD methodology during the

Setup

Jo

2.1

ur

investigation period. It then discusses the theoretical implications on the price effects of AD.

WTO rules permit the imposition of AD duties if import prices (net of transport costs) are below a product’s so called normal value (WTO, 1994). For exporting firms in countries with MES, the EU’s AD regulation defines normal value as the price the firm charges in its own domestic market (European Union, 2009). If price data is not available, normal value may also be based on production costs. The dumping margin is then calculated as the difference between a firm’s own export price to the EU and normal value. 9

This finding also relates to the work on AD echoing by Tabakis and Zanardi (2016). The authors find that

different importing countries tend to impose AD duties on products from the same exporter. Th is paper finds evidence for non-targeted exporters echoing price responses of targeted exporters. The possibility of AD echoing would provide further incentives for exporting firms to raise prices.

Journal Pre-proof For exporting firms in countries with NMES, the dumping margin is calculated as the difference between average export prices across all firms exporting a particular product and normal value. The latter is, however, calculated very differently compared to the MES case. It is constructed using analogue production costs or prices in a third country with MES (Euorpean Union, 2009).10 The AD duty is the same for all exporters. According to the EU’s lesser duty rule, the AD duty either equals the dumping margin or the injury margin (a measure of the exporter’s price relative to prices of EU domestic firms), whichever is lower. This procedure is independent of whether or not the exporter is granted MES.

of

Duties can be reviewed and adjusted if the (pre-duty) export price changes.

ro

China is by far the largest NMES country, both in terms of exports and the number of AD duties it receives. Chinese exporters can obtain market economy treatment (MET) if they can

-p

successfully show that they are active in a market economy environment. In this case they are treated as if they were situated in an MES country (Felbermayr et al., 2016). A hybrid case

re

between MES and NMES is the so called individual treatment (IT). Here the EU uses firm specific

Implications for the Price Effects of AD

na

2.2

lP

export prices but still relies on analogue prices and costs to determine normal value.

The MES procedure has two implications for the effect of AD duties on import prices. First, given

ur

the same normal value (i.e. domestic prices or production costs), firms charging a lower price in the EU receive a larger duty. This may force low price exporters to exit the market, driving average

Jo

prices up. Second, exporting firms have an incentive to raise (pre-duty) prices in order to receive a lower duty following a review. Both channels imply an increase in average import prices. The implications of the NMES methodology are threefold. First, firms exporting the same product receive the same duty, independent of their own export price. Hence, there should be no systematic exit of low price exporters and thus no effect on average prices through this channel. Second, adjusting own export prices does not change the duty the exporting firm faces, so exporters have no incentive to raise prices. As the duty works like a normal tariff, firms may even reduce pre-duty prices (less than 100% pass-through).11 10

The difference in methodologies is driven by the assumption that domestic prices in NMES countries are

distorted. For a more detailed discussion see Detlof and Fridh (2006) or Felbermayr et al. (2016). 11

In the case of imperfect co mpetition, the direction of the price change fro m this channel is, however, not clear. If

Journal Pre-proof Third, as constructed normal value is usually larger than an exporter’s domestic price, AD duties calculated using the NMES methodology are typically larger than those calculated using the MES methodology (Detlof and Fridh, 2006; Felbermayr et al., 2016; Felbermayr and Sandkamp, 2019). In the sample used in this study, EU duties imposed against NMES exporters are on average 16 percentage points (48%) larger than those imposed against MES exporters for the same CN8 product. Assuming a constant price elasticity of demand, the imposition of an AD duty against a n NMES exporter should hence reduce import quantity by more than the imposition of a duty against an MES exporter.

of

Chinese exporters subject to MET or IT should have the same incentive to adjust prices as

ro

exporters from MES countries. Felbermayr and Sandkamp (2019) show that the total export market share of MET and IT firms exporting to the EU is less than 10% in most sectors, and almost

-p

never more than 20%. Assuming MES exporters indeed raise prices following AD while NMES exporters do not, the price coefficient for China - which incorporates these MET firms - would be

re

overestimated. There would be an increased risk of wrongly rejecting the null hypothesis of the

lP

true coefficient being zero. Since it will be shown that the NMES price coefficient remains insignificant despite this potential bias, this provides reassurance that export prices of NMES exporters do not change following the imposition of EU AD duties.

na

Even exporters in MES countries (or those receiving MET in China) may not adjust prices and apply for a review because of the high costs of the review process combined with

ur

administrative uncertainty (Gourlay and Reynolds, 2012) or strong competition in the destination

Jo

market (Felbermayr and Sandkamp, 2019). Under uncertainty, exporters may even set higher prices in expectation of a possible AD duty (Blonigen and Park, 2004). Together with the fact that NMES exporters have no incentive to increase prices and ask for a review of the duty, this may explain why only 36% of affirmative EU AD cases were ever reviewed (Nita and Zanardi, 2013).12 Overall, the above reasoning suggests an increase in (pre-duty) import prices following the imposition of AD duties against MES countries (as found for Canada by Blonigen and Haynes

an exporting firm has significant market power (through its own size or collusion) it can affect the average price and hence the duty it receives. The dominant firm(s) may therefore raise prices and apply for a reduction of the duty. Alternatively, they could reduce prices further to induce an increase in the duty (paid by all exporters) and hence drive out competitors. 12

A case refers to an investigation by the trade authority of the importing country.

Journal Pre-proof (2002)) but no such effect for imports originating from NMES countries (in line with the results for China in Lu et al. (2013)). On the other hand, import quantities should react more strongly to AD duties if imported from NMES countries. These hypotheses can be tested by comparing price and quantity effects of AD duties for exporters from countries with MES with those from NMES countries.13

3

Estimation Strategy

of

Identification of the treatment effect relies on a difference- in-differences estimation. It exploits the change over time (before and after the EU enlargement in 2004) in import prices and quantities of

ro

treated exporting country-product combinations relative to two control groups: First, the same

-p

product imported from untreated exporting countries (within product across country variat ion) and second, untreated products imported from the same exporting country (within country across

re

product variation).14

The baseline analysis considers aggregated imports of the ten accession states. 15 The years

lP

2003 and 2005 are chosen as pre- and post-treatment period respectively, as they constitute a symmetric time period around the accession of the ten new member states in May 2004. The panel

na

is balanced by dropping exporting country-product combinations that were only observed in one

which is given by

ur

year. The difference-in-differences specification is estimated with a first differences regression

(1)

Jo

 ln yih = ADih  i  h   ih .

The dependent variable  ln yih is the change in the natural logarithm of import price (quantity) of product h imported from exporting country i between 2003 and 2005. ADih is the treatment dummy that equals one if an exporting country-product combination becomes subject to AD duties 13

Testing whether the within firm or between firm effect dominates the results, however, requires the use of firm

level data. As both channels work in the same direction, the exact channels at work are not the primary concern of this paper. 14

Unit values are constructed by dividing import values by quantities. Import quantities provide a clearer picture

of changing trade flows than values because the latter incorporate price effects. 15

These are Cyprus, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Malta, Po land, Slovakia and Slovenia.

Bulgaria, Romania and Croatia, who joined the EU in 2007 and 2013, are dropped. Treating individual countries as separate entities does not offer any additional information as treatment takes place at the EU level.

Journal Pre-proof  tells how import prices (quantities) of treated

in May 2004 and zero otherwise. 16

country-product combinations (for which ADih = 1 ) change relative to untreated country-product combinations (for which ADih = 0 ) once the AD duty is implemented through accession to the EU.  i and  h are exporter and product fixed effects respectively.  ih is an error term.17 In order to test for differential effects of duties on imports by applied AD methodology, a second regression is run with the treatment dummy nested by AD regime. The treatment dummy

ADih is interacted once with a dummy that is equal to one if the exporter has MES and once with

of

a dummy identifying if the exporter has NMES. 18 The baseline regression only considers AD cases for which final duties were implemented

ro

by the end of 2003 (i.e. before the accession) and that were in force throughout 2005 (i.e. not

-p

revoked in 2005 or before). This yields a clear pre- and post-treatment period. All duties considered were in force in EU15 countries but not in accession states in 2003 (pre-treatment

re

period), entered into force at the same time in 2004 from the perspective of new member states and

16

The dummy

lP

still were in force in 2005 (post-treatment period).19

AD is zero for all ih in 2003 and changes to one in 2005 only for those ih that are subject to

When t=2, taking first differences yields the same results as the - perhaps more familiar - fixed effects

specification

ln yiht =  ( ADih postt )  ih  it  ht   iht . Here the dependent variable ln yiht is the natural

ur

17

na

EU AD duties.

ADih identifies the treatment

Jo

logarithm of import price (quantity) of product h imported from country i at time t.

group and is a time invariant dummy that is equal to one if imports from country i of product h are subject to EU AD duties and zero otherwise.

postt is a time dummy that equals zero before 2004 and one after 2004. ADih postt is

the treatment dummy that is an interaction of the AD dummy and the time dummy so that effect.

 iht

is an error term.

 ih ,  it

and

 ht



identifies the treatment

are exporter-product, exporter-time and product-time fixed effects

respectively. The three two dimensional fixed effects can be implemented using the “reghdfe” stata command by Correia (2016). 18

The resulting estimation equation becomes

 ln yih =  MES MES ADih   NMES NMES ADih  i  h   ih . 19

This is also the reason why the 2007 accession round is not considered. If 2008 was chosen as the post -treatment

period so as to include Roman ia and Bulgaria, all duties implemented or revoked between 2005 and 2008 would have to be removed from the sample. As several duties were revoked during this time period, this would have reduced the

Journal Pre-proof The advantage of the natural experiment is that the implementation of AD duties already in force in the EU is exogenous from the perspective of new member states. New member states were required to adopt the existing AD policy (treatment) because they joined the EU. Under the identifying assumption that accession states did not join the EU because of its AD policy (independence of decision to join the EU and existing EU AD regulation), the difference- in-differences strategy yields unbiased estimates of the treatment effect. Independence of the decision to join the EU and existing EU AD regulation also means that new member states’ imports did not determine whether AD duties were introduced by EU15

of

countries before 2004. Knowing that the new member states were about to join the EU in 2004, it is

ro

possible that EU AD decisions were adjusted even before 2004 in order to accommodate the need for protection of future member states. AD duties imposed before 2004 would thus not be

-p

exogenous from the perspective of the accession countries.

This claim can be rejected for three reasons. First, according to the EU AD legislation,

re

duties can only be imposed if there is proof of material injury of the domestic (i.e. EU15) industry.

lP

From a legal perspective, AD duties can therefore not be imposed before 2004 if only the domestic industry of EU accession states is affected by dumping practices. Second, only four out of the ten new member states imposed AD duties before joining the EU, indicating limited interest in the

na

instrument. 20 Finally, for almost all AD cases that were successfully imposed by the accession states, the EU imposed no case covering similar products and exporting countries, indicating that

ur

the EU did not adjust its AD policy before 2004. 21

Jo

The estimation set-up also rules out a number of further concerns that are typically problematic when evaluating trade policy. First, all time invariant unobserved country-product characteristics are eliminated by the first differences approach. This also includes subsidies that do size of the treatment group significantly. 20

These are the Czech Republic (one case), Latvia (one), Lithuania (seven) and Poland (nine). Slovenia started

one investigation which however was withdrawn (Bown, 2015). 21

One exception is the case of graphite electrodes from India that were investigated by Poland and the EU

simultaneously in 2003 and became subject to AD duties by both economies. On the other hand, pocket lighte rs exported by China, Taiwan, Indonesia and Vietnam that became subject to Polish AD duties in 2000 were investigated by the EU in 2002. No final duties were imposed by the EU. Similarly, styrene-butadiene rubber fro m Russia became subject to Polish AD duties in 2003 and was subsequently investigated by the EU in 2004 and 2005. Even though dumping was determined to take place, no evidence for injury was found so that no duties were imposed.

Journal Pre-proof not vary between 2003 and 2005. Second, including product fixed effects after taking first differences controls for unobserved demand side variables such as changes in tastes and preferences that may induce a correlation of imports of EU accession states with (potentially endogenous) imports of EU15 countries. They also capture average changes in MFN tariffs over time.22 Third, exporter country fixed effects control for unobserved time varying characteristics of the exporting country. These include supply side factors such as non-product specific market distortions and changes in the price index of intermediates in individ ual exporting countries as

of

well as time-varying multilateral resistance terms (Feenstra, 2008). Overall, the combination of

ro

first differences with country and product fixed effects controls for all unobserved variables that vary across the exporter-product, exporter-time or product-time dimension.

-p

Omitted supply side factors which vary across the exporter-product-time dimension and may cause omitted variable bias cannot be controlled for with fixed effects because this variation is

re

required to estimate the effect of AD duties. Another necessary assumption is therefore no

lP

systematic change in unobserved exporter-product characteristics between 2003 and 2005 that correlate with EU AD duties. The assumption - which is tested in Section 6 - would be violated by exporter-product specific subsidies that vary over time. This is very relevant as AD duties may be

na

imposed in response to such subsidies.

However, in the context of the natural experiment, only AD cases imposed by (and hence

ur

initiated before) 2003 are included in the sample. Their implementation in the past (including

Jo

possible correlations with export subsidies) should not be correlated with time varying country-product characteristics in 2003 and 2005. The fact that they are inherited by the new member states from 2004 onward does not imply a change in unobserved exporting country-product characteristics (such as subsidies) between 2003 and 2005. Nevertheless, subsidies could also be designed to increase continuously over time, i.e. even after the original AD duty was imposed. This problem is addressed in a robustness check by looking at the change in EU15 imports between 2003 and 2005 (for which AD duties remained unchanged). The paper provides evidence that there are no unobserved exporter-product-time specific variables (such as subsidies) that correlate with import prices and AD duties and that 22

Moore and Zanardi (2009, 2011) show a positive correlation between the use of AD and trade liberalisation

(reduction in MFN tariffs).

Journal Pre-proof would lead to an underestimation of the treatment effect.

4

Data

Data on EU trade is obtained from the Eurostat Comext Database (Eurostat, 2017). It supplies data on annual bilateral import values and quantities for all EU member states at the CN8 digit product level. This paper uses data for the years 1999 to 2009, with a focus on 2003 and 2005. 23 For 2003 and 2005 the dataset covers imports of 10,636 CN8 products from 223 countries.

of

Information on EU AD duties at the CN8 digit product level is taken from the World Bank’s Global Antidumping Database (Bown, 2015). The European AD process involves three

ro

stages: Initiation of a case, preliminary (temporary) duties and final duties. Once imposed, AD

-p

duties typically remain in force for at least five years (Euorpean Commission, 2013). Only cases for which final duties were imposed by the end of 2003 and that remained in force until at least

re

2005 (i.e. not revoked in 2005 or earlier) are considered. This leaves 87 AD cases covering 82 CN8 products from 21 exporting countries. 24 The persistence of AD duties implemented by 2003 is

lP

illustrated in Figure 1.

The datasets are merged by exporting country, CN8 product and year. Using import (rather

na

than export) data has the advantage that the importer’s product nomenclature is used, which coincides with the nomenclature reported in Bown (2015) who also relies on importers’

ur

declaration of AD duties. As HS codes are only comparable across countries up until the HS6 digit level (Lu et al., 2013; Bown and Crowley, 2016), studies using exporter data have to restrict their

Jo

analysis to this higher level of aggregation. AD duties are, however, often implemented at a more disaggregated level. Using aggregated data means that HS6 products which are assigned AD treatment incorporate trade flows that are in fact not subject to AD duties. Estimations relying on exporter data may hence suffer from attenuation bias, leading to an underestimation of the treatment effect.

23

1999 is the first year for which Eurostat provides trade data for EU member states that joined in 2004. Using

data until 2009 provides a symmetric five year window around the treatment year 2004. 24

Overall, 145 (115) cases were in force in 2003 (2005). Only those in force in both years are included in the

analysis. Each case can cover several products, while several cases may cover the same product, but for different exporting countries.

Journal Pre-proof Figure 1: The persistence of EU AD duties: Cases imposed by 2003 and remaining in force Note: Cases in force both in 2003 and onward (several products per case).

After the merge, the balanced baseline sample includes imports of 8,366 CN8 products from 149 countries. 55 products imported from 13 countries are subject to EU AD duties. 25 As can be seen from Table 1, only two exporters in the sample are assigned NMES. 26 China is by far the largest NMES country, both in terms of total exports as well as the number of AD cases. However, it will be shown that China and Belarus react remarkably similar to AD duties, indicating that

of

NMES is driving the results for both countries.

ro

Figure 2 presents descriptive evidence for the effect of AD duties on import prices and quantities in the ten EU accession states (aggregated to one entity). For the years 2001 to 2007, it

-p

shows the average price and quantity deviation of treated and untreated exporter-product combinations from their seven year average. The top left panel clearly indicates that prices of

re

treatment and control group followed the same trend up until 2003. Once the new member states

lP

joined the EU in May 2004 and EU AD po licy was implemented, prices of targeted exporter-product combinations increased relative to untargeted exporter-product combinations.27

na

Table 1: Number of AD cases and affected products in sample AD Cases

Affected CN8 Products

MES Exporters

Jo

ur

Country

25

Note that a product may be subject to AD duties if imported from one country, but not the other.

26

Information on NMES of exporters is taken from Detlof and Fridh (2006) and Felbermayr et al. (2016).

Countries that are assigned NM ES by the EU in the per iod of investigation are A lbania, Armen ia, Azerbaijan, Belarus, China, Georgia, Kazakhstan, Kyrgyzstan, Moldova, Mongolia, North Korea, Tajikistan, Turkmenistan, Uzbekistan and Vietnam. Out of these 15 countries, only five (Armenia, Belarus, China, Kazakhs tan and Vietnam) have ever become subject to EU AD duties. 27

Prices may also be affected by exchange rate fluctuations. This should however only be the case if the currency

of countries subject to AD duties reacted differently to the EU en largement than cu rrencies of countries not subject to AD duties. The difference-in-differences specification relies on variation within countries across products as an additional identification channel which is not affected by exchange rate fluctuations. Country fixed effe cts capture average exchange rate fluctuations by exporting country.

Journal Pre-proof 2

3

Hong Kong

1

1

India

3

3

Japan

3

10

Romania

2

6

Russia

7

18

South Korea

3

3

Taiwan

1

1

Turkey

2

6

USA

1

Ukraine

5

of

Croatia

1

ro

17

3

China

18

Total

51

re

Belarus

-p

NMES Exporters

4

25 55

lP

Note: Only cases in which final duties were imposed are included. The sum of CN8 products across all exporting countries does not equal the total since several countries exporting the same

na

product may be targeted by AD duties.

ur

Similarly, starting in the year of accession, import quantities fell for the treatment group while they even increased slightly in the control group (top right panel). The bottom two panels of

Jo

Figure 2 divide the treatment group into MES and NMES countries. Price effects are strong for MES exporters (bottom left panel), while prices of NMES exporters decrease in 2004 before increasing in 2005. The decrease in import quantity is clearly visible for both MES as well as for NMES exporters (bottom right panel). It is, however, not obvious whether the drop in imports of treated products stems from AD or is simply a consequence of the EU accession. As imports of untreated products include imports from EU countries, the graphs could simply show import diversion from non-EU exporters towards EU exporters. Figure A.1 in the Appendix hence excludes imports from EU exporters. The overall picture remains similar, indicating that results are not driven by import diversion following the accession.

Journal Pre-proof Figure 2: Average import prices and quantities by treatment status Note: The vertical line denotes the EU accession (beginning of treatment) in May 2004. Import price (left panels) or quantity (right panels) on vertical axis, year on horizontal axis. Prices and quantities of individual exporter-product combinations are demeaned by dividing by their seven year average before constructing average deviations by treatment group.

Further descriptive evidence is provided by Table B.1 in the Appendix. It shows results from regressing import prices and quantities on an AD dummy with product fixed effects before

of

and after the accession (treatment). Products that become subject to AD duties in 2005 are indeed

ro

cheaper than non-targeted products in 2003, indicating dumping (Column 1). This price discount decreases under AD in 2005 (indicating that AD duties raise producer prices). The change in

-p

relative prices only takes place for products exported by MES exporters (Column 2). Country-product combinations targeted by AD duties experience higher import quantities in 2003

re

(Column 3). However, the gap in import quantities between targeted and non-targeted products

lP

falls following the imposition of AD duties in 2005. This is true for both MES and NMES

na

exporters (Column 4).

Econometric Baseline Results

5.1

Effects on Prices

ur

5

Jo

Table 2 provides the baseline estimation results, with the change in the logarithm of import price, quantity and value as dependent variable. Column (1) of Table 2 shows the price effects of AD duties following the basic difference- in-differences estimation as given in Equation 1. The coefficient of the AD dummy (0.2206) is positive and statistically significant. It indicates that import prices (before tariffs and duties) increase by 25% following the imposition of AD duties. 28 From a welfare perspective, this implies a negative terms of trade effect for the importer. This is in stark contrast to the usually positive terms of trade effects following incomplete pass-through of a normal tariff (Feenstra, 1989, 1995). Considering that AD duties often involve intermediate products such as steel, the increase in prices constitutes a negative cost shock for

28

The marginal effect is calculated using the formula

100(e  1)% .

Journal Pre-proof domestic producers who rely on imported inputs. The results are in line with Blonigen and Haynes (2002), who also find pass-through rates of more than 100%, but not with Lu et al. (2013), who do not find any price effects for imports from China.

Table 2: The effects of AD duties on imports (1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Dep. var.

 ln price  ln price  ln quantity  ln quantity  ln value  ln value  ln value  ln value

AD

0.2206*

-1.3518**

*

*

(0.0879)

(0.2338)

AD*NM

*

*

(0.2048)

-0.8736*

-0.9057*

**

**

(0.2721)

(0.2274)

(0.2368)

0.1471

-1.8852**

-1.7381*

-1.7582*

*

**

**

(0.3435)

(0.3519)

(0.3446)

lP

(0.1527) 0.1223

0.1359

0.1359

0.1703

0.1704

0.1420

0.1420

na

0.1223

(0.2032)

(0.1148)

ES

R2

of

-1.1253**

**

ro

0.2518*

**

-p

S

-1.1384*

re

AD*ME

-1.1312*

Note: OLS regressions (first differences) with exporter and product fixed effects. Robust

ur

standard errors clustered by product in parenthesis. *** p<0.01, ** p<0.05, * p<0.1. (1) - (6): 144,998 observations. (7) - (8): Larger sample including trade flows for which no information on

Jo

quantity is available. 184,889 observations.

To check whether this difference stems from the specific AD procedure applied to non- market economies such as China, the AD dummy is additionally interacted with a dummy indicating whether the exporter has MES, and a (mutually exclusive) dummy indicating whether the exporter has NMES. The results are presented in Column (2). When comparing the estimated coefficients for MES and NMES countries, it is evident that aggregate results presented in Column (1) are driven by MES countries. The interaction coefficient of the treatment dummy and the MES dummy is positive and statistically significant, while the interaction coefficient of the treatment dummy and the NMES dummy is smaller and not statistically significant. Producer prices of products imported from MES countries increase following the imposition of AD duties. On the

Journal Pre-proof other hand, one cannot reject the hypothesis that producer prices of products imported from NMES countries do not change following the imposition of AD duties. The policy implication of this finding is that the MES methodology increases the likelihood that AD duties achieve an increase in import prices, which is the officially stated objective of the instrument. 29

5.2

Effects on Quantities

Columns (3) and (4) of Table 2 summarise the effects of AD duties on import quantities. Column (3) shows regression results for the basic difference- in-differences specification following

of

Equation 1. The coefficient of the AD dummy is negative and statistically significant at the 1%

ro

level, indicating that the imposition of AD duties reduces import quantities of EU accession states. In terms of magnitude the coefficient of -1.3518 in Column (3) indicates that imports fall by 74%

-p

following the imposition of AD duties. This estimate is at the high end of the existing literature,

re

suggesting that not sufficiently addressing endogeneity concerns indeed leads to an underestimation of the treatment effect. From a welfare perspective, this implies that trade

lP

destructing effects may be larger than previously thought. Column (4) presents the estimated effect of AD duties on import quantities separated by

na

MES and NMES. While both coefficients are highly statistically significant, the estimated treatment effect for NMES countries is larger in terms of magnitude than the one for MES

ur

countries. The difference is statistically significant at the 10% level. This result is to be expected given the higher average AD duties imposed against NMES exporters observed in the literature.

Jo

However, it also shows that AD duties remain an effective protectionist instrument if applied against MES exporters. Even if the EU treats China as a full MES exporter in the future (the new regulation allows for exceptions), AD duties would thus still reduce imports into the union.

5.3 29

Effects on Values As mentioned before, product-level data does not allo w the determination o f whether the price increase for M ES

exporters stems fro m exporting firms increasing their prices or fro m lo w p rice exporters receiv ing high duties and thus exiting the market, leaving only high price exporters behind. From the perspective of the importer, the result is the same. It may nevertheless have long term implications if the exporter composition is affected (e.g. inefficient exporters driven out of the market, leav ing only efficient ones behind. See for example Lu et al. (2013) and Jabbour et al. (2019) for a more detailed discussion).

Journal Pre-proof The baseline regression focuses on quantity effects to estimate the impact of AD duties on real trade flows. For completeness, value effects (in EUR) are also estimated. By construction, value = price  quantity so that  ln value =  ln price   ln quantity . This is also true for the

estimated coefficients which are reported in Columns (5) and (6) of Table 2. They are similar to quantity effects but smaller in magnitude. This is due to the positive price effects of AD duties which are incorporated in the value effects and reduce the magnitude of the (negative) coefficient. The difference between estimated coefficients for MES and NMES countries increases in significance (5%) relative to the quantity regression.

of

An advantage of using import values is the resulting increase in sample size, as information

ro

on import values is more frequently available than information on import quantity. Running the same regression with a larger sample (Columns 7 and 8) yields coefficients of similar magnitude,

5.4

re

-p

indicating that results are robust to a change in sample composition.

Effects over Time

lP

By comparing import flows in 2003 and 2005 for treated and untreated products, the baseline regression provides a snapshot of the trade effects of AD duties. In order to investigate whether the

na

effect of AD duties persists over time, the sample is extended, covering trade flows for the years 1999 to 2007. Instead of estimating one treatment effect, separate treatment effects are estimated

ur

for each year from 2001 to 2007. This is done by interacting the AD dummy (which varies across products and exporters) with year dummies. Each of the resulting dummies hence only switches

Jo

from zero to one in one year, identifying the effect o f AD duties on import prices and quantities in that specific year.30

The AD cases included in the sample are the same as in the baseline in order to avoid the need to switch treatment in a year other than 2004. Since a few of these cases were revoked in 2006 and 2007 (Figure 1 above), the resulting coefficients for these years may be biased towards zero. Estimates thus constitute a lower bound of the true treatment effect. The results are illustrated graphically in Figure 3. 30

The graph already provides

Effects over time are estimated using the fixed effects specification

ln yiht = T =2001 2007

31

31

yearT

( ADih yearT )  ih  it  ht   iht with yearT = 1 if t = T and zero otherwise.

Detailed coefficients for each year are provided in Columns (1) and (2) of Table B.2 in the Appendix.

Journal Pre-proof transformed effects, so that the point estimates show percentage changes in import prices and quantities of treated exporting country-product combinations for each year relative to non treated ones. Price and quantity effects are not statistically significant before the new member states joined the EU in 2004. Both coefficients become significant in 2004 (the new member states officially joined the Union in May 2004) and increase in magnitude in 2005. From 2005 onwards, effects remain stable. Since only a part of 2004 is treated, the smaller coefficient for this year is to be expected. The results imply that AD duties quickly unfold their full effect on trade. Small delays could be driven by contracts which fix prices and quantities in the short run. On the ot her

ro

research is needed to decompose these potential channels.

of

hand the results could be taken as evidence that exporters adjust their prices in steps. Further

-p

Figure 3: Effect of AD duties on import prices and quantities, by year Note: EU accession (beginning of treatment) in May 2004. Percentage change in import

re

prices and quantities of treated products on vertical axis, year on horizontal axis. The plot shows

lP

point estimates with 95% confidence intervals, indicating large and significant price and quantity effects of AD duties since their introduction in 2004.

na

Separate results for MES and NMES exporters are shown in Figure A.2 in the Appendix. Qualitatively, the results are in line with the baseline estimation. Price effects become positive and

ur

significant for MES exporters in 2004 and remain so until 2007, while they are insignificant for

Jo

NMES exporters over all years. Quantity effects are significantly negative for both MES and NMES exporters throughout the post-treatment period.

6

Extensions & Robustness Checks

6.1

China

China is not only the largest non-market economy, but also the major target of EU AD duties (Table 1). It is hence possible that estimated coefficients of the effect of AD duties against NMES countries are driven by China. Since China and Belarus are the only NMES exporters in the sample, the AD dummy is nested by MES, China and Belarus in a robustness check. Regression results are reported in Columns (1) and (2) of Table 3. The MES coefficient is almost identical to

Journal Pre-proof the baseline. The estimated coefficients for China and Belarus are both very similar to the baseline NMES coefficient reported in Table 2 and not significantly different from each other. Both countries therefore react very similarly to EU AD duties, but very different compared to the average MES exporter, indicating that NMES, not just China, is driving the baseline results. Given the sample limitations, it cannot of course be ruled out that other NMES exporters react differently to AD duties. As China is by far the largest non- market economy, its behaviour in the face of AD is, however, most relevant for EU policy makers. In addition, the results presented in this paper allow a prediction on how China’s reaction to AD may adjust following a change in

ro

of

the EU’s AD regulation.

Table 3: China and individual importing countries baseline

baseline

individual

 ln price

 ln quantity

AD

R2

individual

individual

individual

countries

countries

countries

countries

 ln price

 ln price

 ln quantity

 ln quantity

0.1680***

-0.9261***

(0.0644)

(0.1522) 0.2305***

-0.8041***

(0.1149)

(0.2725)

(0.0777)

(0.1752)

0.0418

-1.1727***

(0.0934)

(0.2605)

0.1507

na

-1.1261***

-1.8490***

Jo

AD*Belarus

(6)

0.2517**

AD*NMES

AD*China

(5)

ur

AD*MES

(4)

-p

(3)

re

Dep. var.

(2)

lP

Sample

(1)

(0.1766)

(0.3639)

0.1252

-2.1095**

(0.1058)

(0.9895)

0.1223

0.1359

0.2031

0.2031

0.1995

0.1995

Note: OLS regression with first differences. *** p<0.01, ** p<0.05, * p<0.1. Columns (1) to (2): Exporter and product fixed effects. Robust standard errors clustered by product in parenthesis. 144,998 observations. Columns (3) to (6): Importer-product and exporter-importer fixed effects (controlling for exchange rates). Robust standard errors clustered by importer-product in parenthesis. 337,822 observations.

Journal Pre-proof

6.2

Individual Importing Countries

Since all ten EU accession states are subject to the same treatment, the AD dummy does not vary across EU importers. Import values and quantities are aggregated to one single importing entity because the disaggregated importer dimension does not offer any additional variation to identify a treatment effect. 32 As an additional robustness check, the baseline regression is rerun on a sample with ten individual importing countries. The estimation is adjusted by expanding the fixed effects by the importer dimension. 33 This also ensures that bilateral exchange rate fluctuations are controlled for.

of

The results are reported in Columns (3) to (6) of Table 3. Coefficients become slightly

ro

smaller but overall remain similar in magnitude to those of the baseline estimation shown in Table 2. The price coefficients (except for NMES exporters) increase in statistical significance. The

-p

difference in price coefficients between MES and NMES exporters becomes statistically

Import Diversion and Spillover Effects

lP

6.3

re

significant at the 10% level.

Treatment effects are estimated by exploiting within exporter across product variation as well as

na

within product across exporter variation. The baseline sample includes imports by accession states from other EU member countries (both EU15 as well as the ten accession states). The second

ur

identification channel could therefore be affected by import diversion. In particular, increased imports from EU15 countries as a consequence of the accession may result in an overestimation of

Jo

the effect of AD duties on import quantities. To exclude this possible channel, the baseline regression is performed on a sample that excludes imports from EU member states. Hence, imports from targeted exporting countries are only compared to imports from non-targeted non-EU exporting countries. The results are presented in Columns (1) to (4) of Table 4. Estimated coefficients for the effect of AD on import quantities (Columns 3 and 4) remain stable and even increase in magnitude, indicating that import diversion following the accession does not drive the results. 32

In addition, many country pairs do not trade every CN8 product every year, so that aggregating over all

accession states increases the number of imported products observed in both periods. 33

The regression hence includes exporter-importer and importer-product fixed effects. Standard errors are

clustered by importer-product.

Journal Pre-proof Estimated price coefficients (Columns 1 and 2) remain positive but turn insignificant. This indicates that prices of targeted products increase relative to imports from EU countries (baseline), but not relative to imports from non-targeted non-EU countries. That observation can be explained by spillover effects. Allegations of dumping concerning the same product are typically split by exporting country and investigated in separate cases either simultaneously or sequentially. If one country is found guilty of dumping a particular product, then other exporters of the same product may expect to become the subject of investigations in the future and raise prices in anticipation. Such spillover effects would mean that non-targeted countries raise prices following the

of

imposition of antidumping duties against another country. This would violate the stable unit

ro

treatment assumption and cause underestimation of the treatment effect. As EU exporters are

-p

never subject to EU AD duties, no anticipation spillovers are to be expected for them.

(2)

(3)

no EU

no EU

no EU

exporte

exporte

exporters

rs

rs

(4)

(5)

(6)

(7)

(8)

no EU

all

all

all

all

exporters

exporte

exporte

exporters

exporters

rs

rs

lP

Sample

(1)

re

Table 4: Excluding EU exporters and spillovers

 ln price  ln price ln quantity ln quantity ln price ln price ln quantity ln quantity

AD

0.1374

0.2532*

-1.3111**

*

**

*

(0.2779)

(0.0936

(0.2424)

ur

-1.5504**

Jo

(0.1006

na

Dep. var.

) AD*MES

0.1296

(0.1348

) -1.3146**

0.3547*

-1.1606**

*

**

*

(0.3250)

(0.1225

(0.2875)

) AD*NMES

0.1557

) -2.1050**

0.0121

-1.8597**

* (0.1656

*

(0.3626)

(0.1597

) Non-targete

(0.3950)

) 0.1207*

0.1508

Journal Pre-proof d

(0.0678

(0.1731)

) Non-targ.*

0.2358*

MES

0.0577

** (0.0744

(0.2010)

) Non-targ.*N

-0.1284

0.1772

MES

(0.1065

(0.2811)

) 0.1980

0.2001

0.2002

0.1223

0.1224

of

0.1980

R2

0.1359

0.1359

ro

Note: OLS regression with first differences. All regressions include exporter and product fixed effects. Robust standard errors clustered by product in parenthesis. *** p<0.01, ** p<0.05, *

re

-p

p<0.1. (1) -(4): 51,962 observations. (5) - (8): 144,998 observations.

The hypothesis of price spillovers can be tested by investigating the effect of AD duties on

lP

imports of targeted products from non-targeted countries. 34 The estimated coefficient for the effect of AD duties on import prices from non-targeted countries is indeed positive and statistically

na

significant (Column 5 of Table 4). This indicates that prices of imports from non-targeted countries do increase following the imposition of AD duties against other countries.

ur

The magnitude of the price change is roughly ha lf of the effect for treated countries, indicating significant spillover effects. This reaction can only be observed for AD duties imposed

Jo

against MES countries (Column 6). Given the different ways AD duties are calculated under MES and NMES, untargeted exporters may find it more likely to be targeted by an AD duty in the future if it is imposed against another MES exporter. Overall, the results provide evidence that non-targeted exporting countries react to AD duties imposed against other exporters. AD duties seem to have a signalling effect, as they induce non-targeted exporters to raise prices. This constitutes an (unintended) negative welfare effect for 34

To do this, a dummy “Non-targeted” is created and assigned the value of one for imports of targeted products

imported from non-targeted non-EU countries. Imports of targeted products from EU countries are assigned a zero. The treatment effect is hence identified by using variation in imp orts from non-targeted non-EU countries (where spillovers due to anticipation o f further A D cases may be expected) relative to imports fro m non -targeted EU countries (which will never be subject to EU AD duties).

Journal Pre-proof consumers in the importing country. The estimated effects of AD duties on imported quantities from non-targeted countries are insignificant (Columns 7 and 8).

6.4

The Persistence of AD Duties

Figure 3 has illustrated that the trade dampening effects of AD duties persist over several years. The estimated coefficients show the average effect over time of AD duties that were in force 2005. However, they may underestimate the treatment effect for the years 2006 and 2007 because the baseline sample only includes AD cases that were in force until at least 2005. Cases revoked in

of

2006 or 2007 are still treated as being subject to AD duties in the baseline sample, even though

ro

they are not in force anymore.

A robustness test hence performs the same regression, estimating treatment effects by year,

-p

but only including cases in force until at least 2007. Estimated quantity and price coefficients

re

increase for all post-treatment years but become insignificant in 2004 (Columns (3) and (4) of Table B.2 in the Appendix). This is not surprising given the smaller number of cases used to

lP

identify the treatment effect and since some products that were treated in 2004 and 2005 are not assigned treatment anymore.

na

In order to investigate an even longer time horizon, treatment effects are also estimated for the years 2000 - 2008 (using the baseline sample) as well as for 2000 - 2009. Results are reported

ur

in Columns (5) to (8) of Table B.2 and graphically in Figure A.3 in the Appendix. Both price and quantity effects remain significant until the end of the sample period, despite half of the cases

Jo

being revoked before (Figure 1). The magnitude of the price coefficient remains almost constant between 2005 and 2008, indicating that the effect of AD duties on prices persists beyond their revokement. This finding should be considered in the design of AD policy. Even though the estimated quantity coefficient falls over time, it does by no means halve, as would be expected if AD duties only affected trade as long as they are in force. The removal of AD duties constitutes a source of variation that has so far not been used to estimate their effect on trade. An extension therefore investigates the removal of those duties that were in force in the new member states before they joined the EU in 2004. As mentioned in Section 3 above, four of the accession countries had AD duties in place before the y joined the EU (18 cases in total). These were abandoned when they adapted the EU’s common trade policy. The Global Antidumping Database (Bown, 2015) does not report product codes for these

Journal Pre-proof products. This paper hence uses concordance tables to manually assign CN8 codes to product names. A new variable “AD Before” is created that equals 1 if an AD duty was in force before the accession and 0 afterwards. As the removed AD cases are importer specific, imports of each of the new member states are treated separately, rather than being aggregated to one entity. The regression results are reported in Table B.3 in the Appendix. Both aggregate coefficients as well as those nested by MES are qualitatively similar to the baseline AD coefficient. However, they are not statistically significant, indicating no change in import prices and quantities following revokement of AD duties.

of

One possible explanation for this observation is that large AD duties completely drive out

ro

some exporting firms (Lu et al., 2013; Jabbour et al., 2019). Once an exporter is eliminated, it is impossible for her to re-enter the market quickly following the elimination of the duty. This could

-p

be due to market entry costs or a strengthening of the domestic industry during the protection period. Alternatively, surviving exporters may fear a new AD investigation if they increase exports

re

and lower prices following the revokement of the duty. The trade destructing effect of AD duties

lP

thus persists beyond their duration.

The sample used in the baseline regression does not include AD cases that were revoked before 2006 to make sure only cases in force in 2005 are included in the post-treatment period.

na

When including cases that were revoked in 2004 and 2005 (results reported in Columns (1) to (4) of Table B.4 in the Appendix) estimated coefficients remain significant but become smaller in

ur

magnitude. This is not surprising as wrongly assigning treatment to products that are not treated

Jo

(anymore), leads to an underestimation of the treatment effect. Nevertheless, if there is a persistent effect, these cases might still affect trade flows, even if only to a lesser extent than duties that are still in force. In another robustness test, the baseline regression is therefore performed on a sub-sample of AD cases that were revoked in 2004, i.e. in force in 2003 but not anymore in 2005. The insignificant price and quantity coefficients, reported in Columns (1) and (2) of Table B.5 in the Appendix, suggest no lasting impact of these duties. 35 The sub-sample includes cases revoked before the accession of the new member states in May 2004 as well as cases revoked after they joined the EU. In the next step, the sample is divided further into cases revoked before May 2004 (which were never in force in the new member states) and cases revoked between June and December 2004. Cases revoked before May 2004 did not 35

These cases do not target imports from NMES countries, so that separate effects cannot be estimated.

Journal Pre-proof impact import prices or quantities of the new member states (Columns 3 and 4). For AD cases that were briefly in force in 2004, the price coefficient is positive and significant (Column 5), providing some evidence for a persistent effect of those cases that were only briefly in force.

6.5

Further Sources of Endogeneity

One channel that constitutes a threat to identification of the treatment effect and may lead to biased coefficients is the potential of AD duties to cause a reversal of trade deflection. If EU AD duties imposed before 2004 result in increased exports of targeted economies to third countries -

of

including the new member states - then imports of new member states would be larger in the

ro

pre-treatment period than what they would have been without the EU AD duty. The imposition of AD duties in new member states following accession to the EU could thus have two effects, first

-p

the standard trade destruction effect and second the reversal of previous trade deflection.

re

Under the assumption of linear demand, the larger pre-treatment import quantity would be associated with more inelastic demand in new member states. The quantity response to an

lP

ad-valorem duty would be smaller than in the absence of trade deflection. For external validity purposes, this would constitute an underestimation of the treatment effect. In the case of a specific

na

duty, the duty would, however, be larger as a percentage of the pre-duty price, which may result in an overestimation of the treatment effect.

ur

Anticipation might also constitute a threat to identification. The accession of the ten member states and its consequences for their AD policy were known by importers and exporters

Jo

years before 2004. If the change in AD regulation was anticipated it is hence possible that firms exporting to the new member states may have increased their prices before 2004 in order to avoid the imposition of AD duties once the EU AD rules are in force. Only looking at post-treatment price effects would hence underestimate the treatment effect. Alternatively, it is also possible that exporters engaged in excessive dumping before 2004 to sell as many dumped products as possible before the regulation enters into force. This would result in an overestimation. Figure 3 has already shown that treated products did not react differently to untreated products before 2004, indicating that neither trade deflection nor anticipation took place before the accession. As an additional robustness check, import quantities and prices of EU accession states are regressed on AD duties imposed by the EU in the pre-accession period 2000 to 2003. The treatment dummy ADt switches from zero to one in the year in which final AD duties are

Journal Pre-proof imposed and remains equal to one until the end of the sample period. AD cases revoked between 2000 and 2003 are excluded from the sample. The results of the fixed effects estimation are reported in Columns (1) and (2) of Table 5. The coefficient of the time varying AD dummy in Column (1) is insignificant, indicating no effect of EU15 duties on import prices of new member states in the period before the accession. The estimated coefficient for quantity effects is negative and statistically significant at the 10% level (Column 2). Both coefficients provide evidence for the absence of trade deflection of EU imports

of

towards the new member states.

(2)

(3)

(4)

EU Accession

EU Accession

EU 25

EU 25

Sample

pre 2004

pre 2004

post 2004

post 2004

Dep. var.

ln price

ln quantity

ln price

ln quantity

ADt

0.0489

-0.2213*

(0.0448)

(0.1327)

0.0659**

-0.8647***

(0.0257)

(0.0910)

0.0611*

-0.5956***

(0.0354)

(0.1041)

0.8890

0.9072

re

lP

na

ADtEU 15

-p

(1)

Importer

ur

ADtAccession

0.8976

R2

ro

Table 5: Trade deflection and post accession effects

0.8897

Jo

Note: OLS regression (fixed effects). Regressor: AD Dummy (time variant). *** p<0.01, ** p<0.05, * p<0.1. Regressions (1) and (2) include exporter-product, exporter-year and product- year fixed effects. Robust standard errors clustered by exporter-product in parenthesis. 931,883 observations. Regressions (3) and (4) include exporter- importer-product, exporter- importer- year and importer-product- year fixed effects. 2,467,857 observations. Robust standard errors clustered by exporter-importer-product in parenthesis. Sample period pre 2004: Import data 1999 - 2003, AD Duties imposed 2000 - 2003, Sample period post 2004: Import data 2005 - 2009, AD Duties imposed 2006 - 2009.

They also indicate that there were no anticipation effects for AD duties imposed in the years before the accession. If exporters had increased prices before the accession to avoid the

Journal Pre-proof implementation of AD duties by EU accession states after 2004, one would observe positive price effects. Similarly, if exporters increased exports to new member states before 2004 to sell as many products as possible before the imposition of duties, this would have resulted in a positive coefficient in Column (2). Even though the data does not allow to make a statement on duties implemented before 2000, it does permit the conclusion that EU AD duties imposed between 2000 and 2003 did not cause trade deflection to new member states. The significantly negative quantity coefficient in Column (2) is much smaller in magnitude than the coefficient of -1.3518 in the baseline regression (Table 2). It hence should not be

of

interpreted as evidence that the new member states already adopted the EU AD policy before their

ro

accession in 2004. If this was the case, the coefficient would be larger in magnitude. In addition, this would have resulted in significant coefficients for the pre-treatment years in Figure 3.

-p

The finding is in line with Bown and Crowley (2010), who also find weak evidence for trade chilling effects of exports of targeted countries to third countries. The authors interpret this

re

finding as a political chilling effect. Regarding the European market, an alternative explanation

lP

would be that EU15 countries constitute the primary market for some exporting firms. When EU15 AD duties drive them out of this market, they cease production and also stop exporting to other

spillover effect.

na

countries, including the new member states. The small negative effect could constitute such a

As discussed in Section 3, identification relies on the assumption of no change in

ur

unobserved exporter-product characteristics (e.g. continuously increasing subsidies) between

Jo

2003 and 2005. In a further robustness check, the experiment is thus carried out for EU15 countries, with the years 2003 and 2005 as pre- and post-treatment period respectively. Under the assumption that a continuously increasing subsidy would reduce import prices for all EU countries (i.e. EU15 and new member states), one would expect a significantly negative price coefficient and a positive quantity coefficient (AD duties remain constant for EU15 countries between 2003 and 2005). The results are summarised in Columns (1) to (4) of Table B.6 in the Appendix. Both price and quantity coefficients are not significantly different from zero. This shows that EU15 import prices and quantities of products targeted by AD in 2003 and 2005 did not change between the two periods. It provides evidence that there are no unobserved time varying exporter-product specific variables (such as subsidies) that correlate with import prices and AD duties and that would lead to

Journal Pre-proof an underestimation of the treatment effect. One exception is the MES quantity coefficient, which is negative and statistically significant at the 10% level (Column 4). However, it is much smaller in magnitude than the baseline coefficient of -1.13 provided in Column (4) of Table 2. It might capture the observation made in Figure 3 that AD duties take some time to unveil their full effect (recall that a few duties were only imposed in 2003). The small R 2 (about half the size of the R 2 in the baseline regression) is a further indication that duties do not explain any variation in import prices and quantities within the post-treatment period.

of

Similarly, the experiment is carried out for accession states, but with an assumed accession

ro

year of 2002, using 2001 and 2003 as pre- and post-treatment periods respectively. Results are provided in Columns (5) to (8) of Table B.6. As expected, all coefficients are not significantly

re

6.6

-p

different from zero.

External Validity

lP

The effect of AD duties may vary by imposing country (Felbermayr and Sandkamp, 2019). This raises the question of external validity, more precisely whether the results can be transferred to

na

EU15 countries or whether they are specific to EU accession states. To test if the new member countries react differently to the imposition of AD duties than EU15 states, both EU15 and EU

ur

accession states’ import quantities and prices are regressed on AD duties (nested by EU15 and EU accession states) imposed after the EU enlargement in 2004. The sample period consists of the

Jo

years 2005 - 2009.36

The results are reported in Columns (3) and (4) of Table 5. Comparing price and quantity effects for EU15 and EU accession state importers reveals coefficients of very similar size, indicating no systematic difference between the two entities. The price coefficients (Column 3) are not statistically significantly different from each other. Both are positive and significantly different from zero, while quantity effects in Column (4) are negative and significant. The results are thus in line with the baseline regression, although smaller in magnitude. This provides further evidence for endogeneity leading to underestimation of the treatment effect when not relying on the natural

36

Only AD duties imposed from 2006 - 2009 are considered to have a pre-treatment period. The sample includes

two importers, namely EU15 countries and accession states.

Journal Pre-proof experiment.

6.7

Alternative Estimation Strategies

By using dummies to identify the treatment, the baseline regressions estimate average changes in import prices and quantities following the imposition of AD duties. These effects depend on the average size of the duty as well as the implied elasticity. To investigate how import prices and quantities react to a change in the size of AD duties, an extension uses information on average product specific duty rates to estimate elasticities. 37 Columns (1) to (4) of Table B.7 in the

of

Appendix provide estimates for semi-elasticities. The estimated coefficient in Column (1) shows

ro

that a one percentage point increase in AD duties leads to an increase in (producer) import prices of 0.34%. The results are again driven by MES countries (Column 2). There is no evidence that

-p

import prices from NMES countries react to AD duties.

re

Looking at the impact of AD duties on import quantities (Column 3), the coefficient of -0.02 means that import quantities fall on average by 2% for each percentage point increase in AD

lP

duties. Coefficients for MES and NMES countries are not significantly different from each other (Column 4), indicating that the difference observed in the baseline regression is indeed driven by

na

differences in average AD duty rates. Given the same estimated elasticity, imports from NMES countries on average fall by more following the imposition of AD due to the higher average duty

ur

rates they face. Elasticity estimates provided in Columns (5) to (8) show similar results. To show that results are not driven by the regression method used, the baseline regressions

Jo

are also carried out using the poisson-pseudo-maximum- likelihood (PPML) estimator. 38 Results are provided in Columns (1) to (4) of Table B.8 in the Appendix. Price effects lose significance but increase in magnitude by a factor of 3.5 (Column 1). Results continue to be driven by the MES price coefficient, which remains significantly positive (Column 2). Quantity effects remain stable in magnitude and significance (Columns 3 and 4). 37

As duties are often firm specific, the duties used in the regression are the maximu m duties imposed as provided

by Bown (2015). Nine country-product combinations are targeted by price undertakings or specific duties (e.g. EUR per ton). Since ad-valorem duties are difficult to calculate in these cases, they are dropped from the sample. 38

Since PPML does not permit negative dependent variables, a fixed effects estimation is employed instead of a

first differences estimation. Estimations are performed using the stata commands “poi2hdfe ” by Guimarães and Portugal (2010) and Figueiredo et al. (2015) and “ppmlhdfe” by Correia et al. (2019a,b).

Journal Pre-proof In the baseline analysis, zero trade flows are omitted as they are not reported in the trade statistic. If a country-product combination is only observed in one year, it is dropped in order to balance the panel as pre- and post-treatment observations are needed to estimate a treatment effect. However, these non-observed zero trade flows potentially contain information, because AD duties are expected to reduce imports. If duties are prohibitively high, eliminating trade flows entirely, the observation drops out of the sample, leading to an underestimation of the treatment effect. Rather than balancing the panel by dropping country-product combinations that are only observed once, the sample is expanded by filling up the missing years with zero trade flows. Since

of

the natural logarithm of zero is not defined and OLS yields unreliable results when zero trade

ro

flows are included (Santos Silva and Tenreyro, 2006), quantity effects are estimated using poisson-pseudo-maximum- likelihood (PPML). The results are reported in Columns (5) and (6) of

-p

Table B.8 in the Appendix. The estimated coefficients are almost identical to those obtained from the baseline ppml regression excluding zero trade flows (Columns 3 and 4).

re

The baseline estimation includes product and exporter fixed effects. Not controlling for

lP

these does not significantly alter the results. As shown in Columns (5) - (8) of Table B.4 in the Appendix, estimated coefficients are robust to excluding product and exporting country fixed effects. This indicates that the quasi-experimental setup addresses omitted demand and supply side

6.8

ur

na

variables that typically have to be controlled for using fixed effects.

Further Sources of Heterogeneity

Jo

Using disaggregated imports and nesting AD duties by importing country enables an investigation of potentially heterogeneous effects. Estimated coefficients vary considerably across importers. Price effects are positive and significant for Cyprus, the Czech Republic, Malta and Slovenia (Column 1 of Table B.9 in the Appendix). They are positively correlated with average prices charged before the accession to the EU. 39 Higher prices in a destination country may indicate a degree of market power of the exporter, enabling her to increase prices without losing market share. Quantity effects are significantly negative for the Czech Republic, Hungary, Malta, Poland, Slovenia and Slovakia (Column 2). 39

Malta, the Czech Republic, Cyprus and Slovenia are the four countries with highest average import unit values.

Considering the average price deviat ion (import price of a part icular product divided by the average import price of the same product across all importing countries) also puts Cyprus, Slovenia and Malta on top.

Journal Pre-proof In another regression, AD duties are divided into three groups according to duty size (low, medium and high duties). Prices and quantities are regressed on three different dummies that correspond to each of the groups. As shown in Column (3) of Table B.9, both low and high duties lead to higher prices. The three coefficients are not significantly different from each other. These findings indicate that differences in price effects for MES and NMES countries are not driven by differences in duty rates. Quantity coefficients (Column 4) are all significantly negative and not significantly different from each other. Finally, the AD dummy is nested by sector. 40 Column (5) of Table B.9 reveals positive

of

price coefficients for chemicals, metals and miscellaneous products. Estimated quantity

ro

coefficients (Column 6) are significantly negative for the three aforementioned sectors as well as for mineral products and textiles. The estimated price (quantity) coefficient for the transport sector

-p

is significantly negative (positive). However, this result only relies on one single treated

Conclusion

lP

7

re

country-product combination.

This paper exploits the EU enlargement of 2004 as a natural experiment to estimate effects of AD

na

duties on import prices and quantities. Following their accession to the European Union, the new member states inherited the EU’s AD duties. Under the plausible assumptions that the accession

ur

countries did not join the EU because of its AD policy and that the EU did not act on behalf of the new member states before their entry, this implementation can be seen as exogenous. The resulting

Jo

estimation consequently does not suffer from endogeneity bias due to simultaneity (larger import values increasing the likelihood and size of AD duties) that has not sufficiently been addressed in the existing AD literature. Omitted variable bias by means of unobserved changes in preferences or subsidies is also addressed. Using this new estimation strategy, the paper provides evidence that AD duties do increase producer prices and reduce import quantities. Estimated effects are larger than suggested by existing studies, implying that trade destructing effects may be larger than previously thought. Price effects of AD duties are only present when implemented against countries with MES. This 40

Except for the sectors chemicals, metals and machinery, the results should be interpreted with caution since

estimations are based on extremely small treatment groups so that coefficients may be driven by outliers.

Journal Pre-proof result aligns seemingly contradicting findings of previous studies by showing that differing estimates of price effects are driven by MES of the exporter investigated in the respective sample. Imports from non-market economies fall by more following the imposition of AD duties, which can be explained by the larger average AD duties they receive. Data limitations (only two NMES countries are targeted by AD in the sample) leave some doubt on whether the findings can be extended to all NMES exporters. However, it is clear that China - the largest NMES exporter and most frequent target of AD duties - behaves differently from the average MES exporter. Against the background of a recent change in European AD

of

legislation, this paper shows that both AD methodologies significantly reduce import quantities.

ro

The EU’s move away from the NMES methodology therefore does not render its trade policy toothless.

-p

Furthermore, the paper shows that import prices of products from non-targeted countries also increase. This has strategic implications for the use of AD policy, as the imposition of AD

re

duties may have spillover effects across many exporters. Subject to the usual caution, this paper’s

lP

findings may be applicable also to other AD imposing economies, in particular EU15 countries. Finally, evidence is presented that trade dampening effects of AD duties tend to persist over several years and even beyond their revokement, indicating that their impact cannot easily be

References

ur

na

undone.

Jo

Baran, J., 2015. The impact of Antidumping on EU Trade. IBS Policy Paper 12. Baylis, K., Perloff, J.M., 2010. Trade diversion from tomato suspension agreements. Canadian Journal of Economics 43, 127–151. Besedeš, T., Prusa, T.J., 2013. Antidumping and the Death of Trade. NBER Working Paper 19555. Blonigen, B.A., Haynes, S.E., 1999. Antidumping Investigations and the pass-through of exchange rates and antidumping duties. NBER Working Paper. Blonigen, B.A., Haynes, S.E., 2002. Antidumping Investigations and the Pass-Through of Antidumping Duties and Exchange Rates. American Economic Review 92, 1044–1061. Blonigen, B.A., Park, J.H., 2004. Dynamic pricing in the presence of antidumping policy: Theory and evidence. American Economic Review 94, 134–154.

Journal Pre-proof Blonigen, B.A., Prusa, T.J., 2003. Antidumping, in: Choi, T., Harrigan, J. (Eds.), Handbook of International Trade. Blackwell Publishing Ltd. Blonigen, B.A., Prusa, T.J., 2016. Dumping and Antidumping Duties, in: The Handbook of Commercial Policy. 1 ed.. Elsevier B.V. Blonigen, B.A., Wilson, W.W., 2010. Foreign subsidization and excess capacity. Journal of International Economics 80, 200–211. Bown, C.P., 2015. Global Antidumping Database. The World Bank June. Bown, C.P., Crowley, M.A., 2007. Trade deflection and trade depression. Journal of International

of

Economics 71, 176–201.

ro

Bown, C.P., Crowley, M.A., 2010. China’s export growth and the China safeguard : threats to the world trading system ? Canadian Journal of Economics 43, 1353–1388.

-p

Bown, C.P., Crowley, M.A., 2013. Self-enforcing trade agreements: Evidence from time-varying trade policy. American Economic Review 103, 1071–1090.

re

Bown, C.P., Crowley, M.A., 2016. The Empirical Landscape of Trade Policy, in: The Handbook

lP

of Commercial Policy.

Brander, J., Krugman, P., 1983. A ’reciprocal dumping’ model of international trade. Journal of International Economics 15, 313–321.

na

Carter, C.A., Gunning- Trant, C., 2010. U.S. trade remedy law and agriculture: Trade diversion and investigation effects. Canadian Journal of Economics 43, 97–126.

ur

Chandra, P., 2016. Impact of temporary trade barriers: Evidence from China. China Economic

Jo

Review 38, 24–48.

Clarida, R.H., 1993. Entry, Dumping, and Shakeout. The American Economic Review 83, 180–202.

Correia, S., 2016. A Feasible Estimator for Linear Models with Multi-Way Fixed Effects. Working Paper. Correia, S., Guimarães, P., Zylkin, T., 2019a. PPMLHDFE: Fast Poisson Estimation with High-Dimensional Fixed Effects. Mimeo. Correia, S., Guimarães, P., Zylkin, T., 2019b. Verifying the existence of maximum likelihood estimates for generalized linear models. Mimeo. Detlof, H., Fridh, H., 2006. The EU Treatment of Non- Market Economy Countries in Antidumping Proceedings. Swedish National Board of Trade.

Journal Pre-proof Dixit, A., 1988. Anti-dumping and countervailing duties under oligopoly. European Economic Review 32, 55–68. Egger, P., Nelson, D., 2011. How bad is antidumping? Evidence from panel data. Review of Economics and Statistics 93, 1374–1390. Ethier, W.J., 1982. Dumping. Journal of Political Economy 90, 487–506. European Commission, 2013. Reviews URL: http://trade.ec.europa.eu/doclib/docs/2013/april/tradoc{_}151019.pdf.

Council. Official Journal of the European Union.

of

European Commission, 2016. Regulation (EU) 2016/1036 of the European Parliament and the

ro

European Parliament, 2017. Regulation (EU) 2017/2321 of the European Parliament and the Council of 12 December 2017. Official Journal of the European Union.

-p

European Union, 2009. Council regulation (EC) No 1225/2009 on protection against dumped imports from countries not members of the Europan Community. Official Journal of the

re

European Union.

lP

Eurostat, 2017. Eurostat COMEXT Database.

Feenstra, R.C., 1989. Symmetric pass-through of tariffs and exchange rates under imperfect competition: An empirical test. Journal of International Economics 27, 25–45.

na

Feenstra, R.C., 1995. Estimating the effects of trade policy, in: Handbook of International Economics. volume 111. chapter 30, pp. 1554–1590.

Jo

Press.

ur

Feenstra, R.C., 2008. Advanced international Trade: theory and evidence. Princeton University

Felbermayr, G., Sandkamp, A., 2019. The trade effects of anti-dumping duties: Firm-level evidence from China. European Economic Review. Felbermayr, G., Sandkamp, A., Yalcin, E., 2016. New trade rules for China? Opportunities and threats for the EU. European Parliamentary Research Service Figueiredo, O., Guimarães, P., Woodward, D., 2015. Industry localization, distance decay, and knowledge spillovers: Following the patent paper trail. Journal of Urban Economics 89, 21–31. Gourlay, S., Reynolds, K.M., 2012. Political Economy of Antidumping Reviews: The Impact of Discretion at the International Trade Administration. American University Department of Economics Working Paper Series.

Journal Pre-proof Gruenspecht, H.K., 1988. Dumping and dynamic competition. Journal of International Economics 25, 225–248. Guimarães, P., Portugal, P., 2010. A Simple Feasible Alternative Procedure to Estimate Models with High-Dimensional Fixed Effects. The Stata Journal 10, 628 – 649. Jabbour, L., Tao, Z., Vanino, E., Zhang, Y., 2019. The good, the bad and the ugly: Chinese imports, European Union anti-dumping measures and firm performance. Journal of International Economics 117, 1–20. Konings, J., Vandenbussche, H., Springael, L., 2001. Import diversion under European

of

Antidumping Policy. Journal of Industry, Competition and Trade 1, 283–299.

ro

Lasagni, A., 2000. Does Country-targeted Anti-dumping Policy by the EU Create Trade Diversion? Journal of World Trade 34, 137–159.

-p

Lu, Y., Tao, Z., Zhang, Y., 2013. How do exporters respond to antidumping investigations? Journal of International Economics 91, 290.

re

Messerlin, P.A., 1989. The ec antidumping regulations: A first economic appraisal, 1980-85.

lP

Weltwirtschaftliches Archiv 125, 563–587.

Moore, M.O., Zanardi, M., 2009. Does Antidumping Use Contribute to Trade Liberalization in d‘Economique 42.

na

Developing Countries?*. Canadian Journal of Economics/Revue Canadienne

Moore, M.O., Zanardi, M., 2011. Trade Liberalization and Antidumping: Is There a Substitution

ur

Effect? Review of Development Economics 15, 601–619.

Jo

Nelson, D., 2006. The political economy of antidumping: A survey. European Journal of Political Economy 22, 554–590. Nguyen, T.H., Nguyen, T.T., Pham, H.V., 2016. Trade Diversion as Firm Adjustment to Trade Policy : Evidence from EU Antidumping Duties on Vietnamese Footwear. The World Economy. Nita, A.C., Zanardi, M., 2013. The first review of european union antidumping reviews. World Economy 36, 1455–1477. Prusa, T.J., 1997. The Trade Effects of U.S. Antidumping Actions, in: The Effects of U.S. Trade Protection and Promotion Policies. volume 44, pp. 191–214. Prusa, T.J., 2001. On the spread and impact of anti-dumping. Canadian Journal of Economics 34, 591–611.

Journal Pre-proof Santos Silva, J., Tenreyro, S., 2006. The log of gravity. The Review of Economics and Statistics 88(4), 641–658. Staiger, R.W., Wolak, F.A., 1992. The effect of domestic antidumping law in the presence of foreign monopoly. Journal of International Economics 32, 265–287. Staiger, R.W., Wolak, F.A., 1994. Measuring Industry Specific Protection: Antidumping in the United States. Brookings Papers on Economic Activity: Microeconomics 1, 51–118. Tabakis, C., Zanardi, M., 2016. Antidumping Echoing. Economic Inquiry. Urdinez, F., Masiero, G., 2015. China and the WTO: Will the Market Economy Status Make Any

of

Difference after 2016? The Chinese Economy 48, 155–172.

European Economic Review 54, 760–777.

ro

Vandenbussche, H., Zanardi, M., 2010. The chilling trade effects of antidumping proliferation.

-p

Viner, J., 1923. Dumping: A Problem in International Trade,. Chicago University Press. WTO, 1994. Agreement on the Implementation of Article VI of the General Agreement on Tariffs

re

and Trade 1994 , 147–171.

Additional Figures

ur

A

na

Appendix

lP

WTO, 2018. I-TIP Database. URL: http://i-tip.wto.org.

Jo

Figure A.1: Average import prices and quantities by treatment status, excluding EU exporters Note: The vertical line denotes the EU accession (beginning of treatment) in May 2004. Import price (left panel) or quantity (right panel) on vertical axis, year on horizontal axis. Prices and quantities of individual exporter-product combinations were demeaned by dividing by their seven year average before constructing average deviations by treatment group.

Figure A.2: Effect of AD duties on import prices and quantities, by year and MES Note: EU accession (beginning of treatment) in May 2004. Percentage change in import prices and quantities of treated products on vertical axis, year on horizontal axis. The plot shows point estimates with 95% confidence intervals. Detailed regression results are available on request.

Journal Pre-proof Figure A.3: Effect of AD duties on import prices and quantities, 2000 - 2009 Note: EU accession (beginning of treatment) in May 2004. Percentage change in import prices and quantities of treated products on vertical axis, year on horizontal axis. The plot shows point estimates with 95% confidence intervals, indicating large and significant price and quantity effects of AD duties since their introduction in 2004.

Additional Regressions

of

B

Table B.1: Characteristics of products subject to AD duties in new member states (2)

ln price

ln price

-0.4716***

AD (MES,

na

2003) AD (MES,

AD (NMES,

Jo

2003)

ur

2005)

AD (NMES, 2005)

ln quantity

0.9756*** (0.2874)

lP

(0.0964)

ln quantity

(0.2771)

re

-0.2658***

(4)

2.0112***

(0.0957) AD (2005)

(3)

-p

AD (2003)

(1)

ro

Dep. var.

-0.4251***

1.8357***

(0.1134)

(0.3014)

-0.1615

0.9526***

(0.1350)

(0.3256)

-0.5802***

2.4266***

(0.1477)

(0.4967)

-0.5123***

1.0278*

(0.1118)

(0.5582)

Note: OLS regression with product fixed effects. Robust standard errors clustered by product in parenthesis. *** p<0.01, ** p<0.05, * p<0.1. The AD dummy identifies the same products in 2003 (pre-treatment) and 2005 (post-treatment). 144,998 observations per year.

Table B.2: Effects over time Sampl e

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

1999 -

1999 -

1999 -

1999 -

1999 -

1999 -

1999 -

1999 -

2007

2007

2007

2007

2008

2008

2009

2009

Journal Pre-proof AD

baseline

baseline

if 2007

if 2007

baseline

baseline

baseline

baseline

Cases Dep.

ln price

ln quantity ln price ln quantity ln price

ln quantity ln price

ln quantity

var. 2000

2006

2007

(0.1475)

(0.0587)

(0.1480)

0.0098

0.0228

0.0116

(0.0567)

(0.1461)

(0.0811)

(0.1990)

(0.0626)

(0.1737)

(0.0625)

(0.1736)

-0.0088

0.0303

-0.0237

0.0898

0.0127

0.0953

0.0108

0.0948

(0.0563)

(0.1533)

(0.0782)

(0.2146)

(0.0620)

(0.1847)

(0.0619)

(0.1855)

0.0492

-0.0322

-0.0583

0.1285

0.0630

0.0425

0.0598

0.0406

(0.0574)

(0.1584)

(0.0743)

(0.2132)

(0.0646)

(0.1805)

(0.0649)

(0.1813)

0.1545*

-0.4459**

0.0585

-0.2762

0.1731*

-0.3729*

0.1743*

-0.3838**

*

*

(0.0655)

(0.1725)

(0.0825)

(0.2308)

(0.0683)

(0.1907)

(0.0683)

(0.1917)

0.3035*

-1.2927**

0.3147*

-1.3430**

0.3240*

-1.1848**

0.3192*

-1.1665**

**

*

*

*

**

*

**

*

(0.0955)

(0.2379)

(0.1232)

(0.3055)

(0.0976)

(0.2496)

(0.0977)

(0.2497)

0.3224*

-1.4998**

0.3435*

-1.7155**

0.3344*

-1.3881**

0.3313*

-1.3779**

**

*

*

*

**

*

**

*

(0.1229)

(0.3326)

(0.1735)

(0.4732)

(0.1253)

(0.3459)

(0.1254)

(0.3450)

0.3367*

-1.5096**

0.4105*

-2.0314**

0.3531*

-1.3744**

0.3471*

-1.3535**

*

*

*

*

*

*

*

*

(0.4007)

(0.2010)

(0.6034)

(0.1464)

(0.4116)

(0.1465)

(0.4094)

0.3187*

-0.8147**

0.3161*

-0.8170**

(0.1441) 2008

-p

ro

of

0.0226

*

re

2005

(0.0586) -0.0991

lP

2004

0.1705

0.0669

na

2003

0.0351

-0.0631

ur

2002

0.1714

-0.0001

Jo

2001

0.0346

*

** (0.1077)

** (0.3562)

2009

Obs.

1,716,48

1,716,485

5

R2

0.8537

1,716,48

1,716,485

5 0.8471

0.8537

1,930,78

1,930,787

7 0.8471

0.8498

(0.1065)

(0.3503)

0.2157*

-1.1313**

*

*

(0.1064)

(0.3945)

2,127,80

2,127,801

1 0.8435

0.8456

0.8388

Journal Pre-proof Note: OLS regressions including exporter-product, exporter-year and product-year fixed effects. Robust standard errors clustered by exporter-product in parenthesis. *** p<0.01, ** p<0.05, * p<0.1.

Table B.3: Revoked AD cases for individual counties (2)

(3)

(4)

 ln price

 ln price

 ln quantity

 ln quantity

AD

0.1682**

-0.9281***

(0.0712)

(0.1866)

0.0743

-0.5877

(0.1684)

(0.4856)

0.2305**

-0.8040***

-p

AD*MES

ro

AD Before

of

(1) Dep. var.

Before*MES

na

AD

0.2031

-1.0343

(0.2356)

(0.9010)

0.0417

-1.1731***

(0.1046)

(0.2598)

-0.0098

-0.0428

(0.2372)

(0.5361)

0.2031

0.1995

0.1995

ur

R2

0.1265

lP

AD*NMES

Before*NMES

(0.2204)

re

AD

(0.0920)

Note: OLS regressions (first differences) with importer-exporter and importer-product fixed

Jo

effects. Robust standard errors clustered by importer-product in parenthesis. *** p<0.01, ** p<0.05, * p<0.1. 337,822 observations.

Table B.4: Including revoked cases and excluding fixed effects

Sample

Dep.

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

incl.

incl.

incl. rev.

incl. rev.

baseline

baseline

baseline

baseline

rev.

rev.

cases

cases

cases

cases

 ln price  ln price  ln quantity ln quantity ln price  ln price  ln quantity ln quantity

var. AD

0.1563*

-0.9112**

0.2130*

-1.3071**

Journal Pre-proof **

*

*

*

(0.0595)

(0.1726)

(0.0831)

(0.2048)

AD*ME

0.1496*

-0.7232**

0.2709*

-1.1545**

*

*

*

*

(0.0694)

(0.1887)

(0.1065)

(0.2436)

0.1852

-1.7249**

0.0752

-1.6702**

S

AD*NM ES

* (0.1361) 0.1223

(0.3164)

0.1223

0.1358

(0.1517)

0.1359

0.0000

0.0000

(0.3083) 0.0003

0.0003

of

R2

*

Note: OLS regressions with first differences. 144,998 observations. Robust standard errors

ro

clustered by product in parenthesis. *** p<0.01, ** p<0.05, * p<0.1. (1) - (4): Sample includes cases revoked in 2004 and 2005. Including exporter and product fixed effects. (5) - (8): The low

-p

R 2 can be explained by the relatively small treatment group. While AD duties may explain a lot of

re

variation for the treated group, they do not explain any variation in the vast majority of

lP

observations that are not subject to AD duties.

Table B.5: AD cases revoked in 2004 (3)

(4)

(5)

(6)

Revoked in

Revoked in

Revoked in

Revoked in

Revoked in

Revoked in

2004

2004

2004

2004 before

2004 after

2004 after

before May

May

May

May

ur

AD cases

(2)

na

(1)

 ln quantity

 ln price

 ln quantity

 ln price

 ln quantity

0.1083

-0.1731

0.0824

-0.1989

0.4314***

-0.0056

(0.1360)

(0.6495)

(0.1537)

(0.7317)

(0.0252)

(0.0477)

Observations

143,359

143,359

143,317

143,317

143,045

143,045

R2

0.1225

0.1360

0.1225

0.1360

0.1227

0.1362

Jo

 ln price

Dep. var. AD

Note: OLS regressions with first differences, including exporter and product fixed effects. Robust standard errors clustered by product in parenthesis. *** p<0.01, ** p<0.05, * p<0.1.

Table B.6: Placebo tests Treatme nt

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

2004

2004

2004

2004

2002

2002

2002

2002

Journal Pre-proof EU15

EU15

EU15

EU15

 ln price  ln price  ln quantity ln quantity ln price  ln price  ln quantity ln quantity

Dep. var. AD

0.0082

-0.2810

0.0608

-0.1013

(0.0547)

(0.1757)

(0.0600)

(0.1443)

0.0162

-0.3921*

0.0792

-0.1467

S

(0.0654)

(0.2287)

(0.0627)

(0.1422)

AD*NM

-0.0136

0.0204

-0.0165

0.0893

ES

(0.0791)

(0.2027)

(0.1155)

(0.3450)

R2

0.0604

0.0604

0.0664

of

AD*ME

0.0664

0.0783

0.0783

0.0931

0.0931

ro

Note: OLS regression with first differences. All regressions include exporter and product fixed effects. Robust standard errors clustered by product in parenthesis. *** p<0.01, ** p<0.05, *

re

-p

p<0.1. (1) - (4): 267,578 observations. (5) - (8): 151,943 observations.

Table B.7: Elasticities and Semi-elasticities

Treat

(2)

(3)

(5)

(6)

(7)

(8)

 ln price  ln price  ln quantity ln quantity ln price  ln price  ln quantity ln quantity Duty

Duty

Duty

. var. 0.0034*

Duty

Duty Duty Duty ln(1  Duty 100 ) ln(1  100 ) ln(1  100 ) ln(1  100 )

-0.0210**

0.4858*

-2.8798**

*

*

*

(0.0040)

(0.1960)

(0.5368)

Jo

* (0.0014)

Duty

ur

Duty

(4)

lP

var.

(1)

na

Dep.

0.0038*

-0.0199**

0.5506*

-2.6949**

*

*

*

*

MES

(0.0017)

(0.0049)

(0.2368)

(0.6724)

Duty

0.0023

-0.0243**

0.3029

-3.4018**

*

*

*

NM

(0.0032)

*

(0.0079)

(0.4360)

(0.9897)

ES

R2

0.1223

0.1223

0.1359

0.1359

0.1223

0.1223

0.1359

0.1359

Note: OLS regressions with first differences, including exporter and product fixed effects. Robust standard errors clustered by product in parenthesis. *** p<0.01, ** p<0.05, * p<0.1. 144,989

Journal Pre-proof observations.

Table B.8: PPML regressions

AD

(3)

(4)

(5)

(6)

baseline

baseline

baseline

baseline

zero trade

zero trade

sample price

sample price

sample

sample

flows

flows

quantity

quantity

quantity

quantity

0.7668

-1.4427***

-1.4421***

(0.4752)

(0.3815)

(0.3719)

AD*MES

AD*NMES

of

var.

(2)

0.8301*

-1.4329***

-1.4327***

(0.4891)

(0.4377)

(0.4275)

-1.4641***

-1.4626***

(0.3054)

(0.2968)

ro

Sample Dep.

(1)

-0.2786

289,996

289,996

289,996

289,996

430,030

430,030

re

Observations

-p

(0.3061)

Note: PPML regressions with exporter-product, exporter-time and product-time fixed effects.

lP

Robust standard errors clustered by product in parenthesis. *** p<0.01, ** p<0.05, * p<0.1.

(1)

(3)

(4)

(5)

(6)

 ln quantity

 ln price

 ln quantity

 ln price

 ln quantity

ur

Cyprus

 ln price

(2)

0.3059*

-0.6957

Jo

Dep. var.

na

Table B.9: Sources of Heterogeneity

(0.1644)

(0.6169)

Czech

0.3340*

-1.2919***

Republic

(0.2026)

(0.4484)

Estonia

0.4413

-0.0857

(0.2815)

(0.4212)

0.0380

-0.8614*

(0.1545)

(0.4568)

-0.1666

-0.4538

(0.1824)

(0.4521)

0.1487

0.0636

(0.1298)

(0.3677)

Hungary

Lithuania

Latvia

Journal Pre-proof

Slovenia

Slovakia

-1.5584***

(0.2834)

(0.3475)

0.0896

-1.4888***

(0.1334)

(0.3247)

0.6691***

-1.7819***

(0.2039)

(0.4342)

-0.1282

-0.5914*

(0.1267)

(0.3368)

Low duty

Medium duty

0.2998*

-1.3921***

(0.1771)

(0.5203)

0.1029

-1.1235***

(0.1084) 0.2523*

(0.1471)

(0.3660)

lP

Prducts Chemicals

na

Plastics and rubber

Jo

ur

Textiles

Machinery

-1.5366***

re

Mineral

Metals

(0.3330)

-p

High duty

of

Poland

2.4740***

ro

Malta

Electrical Transportatio n

-1.9479***

(0.3208)

(0.1243)

0.5251**

-2.3045***

(0.2379)

(0.6984)

-0.0499

0.2461

(0.4519)

(1.6500)

-0.6669

-1.7252***

(0.5614)

(0.5648)

0.1812*

-1.2143***

(0.0928)

(0.2407)

0.1817

-0.5708

(0.2309)

(0.5622)

-0.1708**

1.3501***

*

Miscellaneous

R2

-0.0648

0.2031

0.1995

0.1223

0.1359

(0.0253)

(0.0427)

0.4457***

-1.9503***

(0.0146)

(0.0275)

0.1223

0.1360

Note: OLS regression with first differences. *** p<0.01, ** p<0.05, * p<0.1. (1) - (2):

Journal Pre-proof Exporter- importer and importer-product fixed effects. Robust standard errors clustered by importer-product in parenthesis. 337,822 observations. (3) - (6): Exporter and product fixed

Jo

ur

na

lP

re

-p

ro

of

effects. Robust standard errors clustered by product in parenthesis. 144,998 observations.

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6