The Effects of Antidumping Policy on Trade Diversion: A Theoretical Approach

The E?ects of Antidumping Policy on
Trade Diversion: A Theoretical Approach
Arastou KHATIBI1
February 2007
Abstract
The purpose of this paper is to contribute theoretically to the lit-
erature on the e?ects of antidumping policy on trade diversion. Trade
diversion refers to a shift in trade ?ows away from ?rms whose imports
are under scrutiny for dumping (named ?rms) to ?rms that import
the same product but are not faced by any investigations (non-named
?rms). Previous empirical studies show that import diversion in Eu-
rope - compared to the United States - is limited, suggesting that
EU’s antidumping policy is more e?ective in keeping imports out. The
explanations formulated to account for the heterogeneity in trade di-
version are the lower duty levels and the greater extent of uncertainty
surrounding the EU antidumping policy. This paper develops a model
to explain the empirical evidence and formulates new explanations on
the e?ects of antidumping policy on trade diversion.
Keywords and Phrases: Trade Diversion, Antidumping, Cournot
Competition, Signal Extraction.
1PhD student, Department of Economics, Universite Catholique de Louvain (UCL).
I wish to deeply thank my supervisor Hylke Vandenbussche for guidance and numerous
comments. I am also grateful for comments by Marcelo Phe-Funchal, Tanguy Isaac and
participants at the Doctoral Workshop January 2007 at CORE, UCL.

---

1
Introduction
Consecutive multilateral negotiations at the level of the WTO have resulted
in the reduction of import tari?s, but not much initiative has been directed
toward nontari? barriers. Moreover, recent trends in the usage of unfair
trade laws indicate that they are being used more frequently, by more
countries, and against more products (Stiglitz (1997) and Prusa (2005)).
Advocates of unfair trade measures claim that it deters unfair trade and
guarantees a more healthy competitive environment for the industries who
petition the law suits. Opponents, on the other hand, claim that the abuse
of these laws represents one of the most ominous threats to international
trade (Stiglitz (1997)). Nonetheless, there is a growing consensus among
advocates and opponents of the need to better understand the e?ects of
nontari? barriers on both domestic and foreign ?rms.
Antidumping (henceforth, AD) is the most popular trade defense instru-
ment, used intensively by the United States and the European Union to pro-
tect their producers from, supposedly, unfair foreign competition. However,
unlike traditional forms of protection, current AD measures are selective
and less transparent than tari?s (Ethier and Fischer, 1990), giving rise to
less desirable externalities. One of these externalities is trade diversion, that
is the shift in trade ?ows away from the alleged or named countries to the
bene?t of non-named countries that import the same good.
Previous studies have quantitatively measured the amount of trade di-
version for the United States (Prusa, 1997), the European Union (Konings,
Vandenbussche and Springael, 2002) and Mexico (Niels 2003). An inter-
esting result drawn from these studies is that although AD protection in-
duces substantial trade diversion from named to non-named countries in
the United States (Prusa 1997), it does so to a less signi?cant degree in the
European Union (Konings et al 2002) and Mexico (Niels 2003), implying
that AD measures are more e?ective in reducing foreign imports in either
Europe or Mexico. Three main explanations are advanced to explain the
lower amount of trade diversion found for Europe (Konings et al). The ?rst
is that the lower duty levels in the EU limits the bene?ts of protection for
the non-named countries. The second is the lack of transparency regarding
the antidumping procedure in Europe that could explain a more prudent
reaction by exporters from non-named countries to the EU. And ?nally, the
higher degree of market fragmentation in Europe could explain the lower
degree of import diversion. To date, these studies have mainly focused on
1

---

whether there is substantial trade diversion and have not provided a theo-
retical explanation.
This paper endeavours to provide a theoretical explanation for the empir-
ical evidence of country level heterogeneity in trade diversion as a response to
AD policy. In particular, it addresses two of the three explanations advanced
by the empirical literature, namely the duty level and the transparency of
the AD policy. As for the ?rst, the result is straightforward and intuitive,
it reaches the same conclusion as the one formulated by the empirical liter-
ature. It concludes that the lower level of duty leads to less trade diversion
because it limits the bene?ts for the non-named countries to increase their
exports to the EU. As for the second, regarding the opacity of AD policy
leading to lower trade diversion, the theoretical model shows that under
some conditions trade diversion can also be low under a transparent AD
policy. This is an interesting result because it provides a new explanation
to the literature on the e?ects of AD policy on trade diversion.
The underlying argument in this paper is conceptually simple. While
named ?rms directly face an AD tari?, non-named ?rms face the uncer-
tainty of being future AD targets. To understand the consequences of this
uncertainty on trade diversion it is helpful to know how ?rms forecast the
future. It is argued in the model that ?rms receive some public information
- through signaling, that reveals the future intentions of the AD author-
ity, and the more transparent the AD policy, the more accurate the signal.
Transparency in this paper implies that the ?rms face less uncertainty when
forming their expectations. For example, if the public signal reveals the in-
formation that the government is anxious to protect the domestic industry,
then under a perfectly transparent AD policy, non-named ?rms would be
responsive and reduce their exports. However, under opacity, ?rms will not
be as responsive for they do not know how the government would react in
the future. Although non-named ?rms reduce their trade ?ows, the model
shows that they do to a less signi?cant degree than under transparency.
This result clearly has relevance for European AD policy, because up to
now the literature on the AD e?ects on trade diversion has seemed to con-
clude that the low amount of import diversion in Europe suggests that the
AD policy is more e?cient in protecting the domestic industry from foreign
imports (Konings and al 2002). And since the AD policy in the EU is not
as transparent as the one the the US, it is doing a better at protecting the
domestic market. The theoretical result in this paper, however, points out
2

---

that it is debatable.
The theoretical model developed in this paper formalizes the idea that
AD policy does not only a?ect the imports of targeted countries but also
the imports of non-named countries, due to information asymmetries sur-
rounding the decision making process of the investigations. The analysis is
conducted using a model of Cournot competition, with information trans-
mission that is revealed a form of a noisy signal, involving three ?rms and a
government administrative authority in a two-stage game. In the ?rst stage,
the government imposes an optimal duty on the named ?rm by maximizing
its objective function described by a weighted average of the domestic ?rm’s
pro?t, consumer surplus and the AD tari? revenu . In the second stage, ?rms
set their output quantities in the presence of a publicly available stochastic
signal S, that informs the ?rms on the future use of AD policy.
Earlier papers have analyzed theoretically information transmission in
Cournot and Bertrand competition where ?rms observe a signal (Gal-Or,
1986; Cooper and Riezman, 1989), and recent papers have applied the model
to study trade policy design (Caglayan, 2000; Caglayan and Usman (2004)).
The present paper contributes to the line of theoretical literature that fo-
cuses on AD policy in imperfectly competitive markets.
The rest of the paper is structured as follows: Section 2 presents the
assumptions and the information structure of the model. Section 3 analyzes
the model by using the concept of backward induction and compares the
output of the ?rms when they receive some information about the future
AD policy in the form of a noisy signal and a benchmark case, in which
trade diversion is total. Section 4, analyzes the ?rst stage and the govern-
ment’s objective function. Section 5, aims to explain the di?erences in trade
diversion between the EU and the US. Section 6, provides an extension to
the model by explaining how the named ?rm gets targeted in the ?rst place.
Finally, Section 4 concludes.
2
Setup of the Model
Consider three asymmetric risk-neutral Cournot ?rms, a home (h) and two
foreign ?rms, one named (n?) and one non-named (nn?). All ?rms produce
a homogeneous good that is sold on the domestic market. The point of de-
parture of the game is as follows. First, the AD administration names one
3

---

of the foreign ?rms by imposing an AD duty on the ?rm. Second, all ?rms
set their output quantities in the presence of a publicly available stochastic
signal S which informs them on the intentions of the AD administration in
using future AD policy on the non-named ?rm. The information content
of the signal depends on the transparency of the AD policy. The present
analysis focuses mainly on how the exports of the non-named ?rm to the
domestic country is a?ected given the information content of the signal.
The domestic ?rm has constant marginal cost of production ch and its
output for the home market is Qh. Among the foreign ?rms, one has a low
constant marginal cost denoted by c? and the other a high marginal cost
L
c? such that: c? > c? . In particular, it is assumed that marginal cost of
H
H
L
the foreign ?rms include transport costs per unit. Note that, in order for
the government to use AD policy, a foreign ?rm has to be causing injury to
the home market, i.e., it must be that the marginal cost including transport
cost has to be lower than the domestic cost of production. Consequently,
it is assumed that both foreign ?rms have a cost advantage over the home
?rm, such that: ch > c? > c? .
H
L
For simplicity, it is assumed at ?rst that the government chooses to name
the low cost foreign ?rm in an AD investigation. This simplifying assump-
tion will be relaxed in section 6 of the paper, where the naming procedure
of the ?rms is endogenised. Accordingly, the low cost foreign ?rm is referred
to as the named ?rm and the high cost ?rm as the non-named. Therefore,
the exports of the named and non-named foreign ?rm to the home market
are respectively denoted by Qn? and Qnn? .
The inverse market demand function for the product is assumed to be
linear and is given by
P (Q) = ? ? ?Q
where P (Q) denotes the market price and Q the total demand. The demand
parameters are non-stochastic and positive: ?, ? > 0 and ? > ch. At equi-
librium, market demand equals the total supply of the individual outputs
such that
Q = Qh + Qn? + Qnn?
If t denotes unit duty tari?, then the marginal cost of serving the relevant
market for the named ?rm is c? + t. The home ?rm and the foreign named
L
?rm’s pro?t functions are respectively
?h = [P (Q) ? ch]Qh
4

---

?n? = [P (Q) ? (c?L + t)]Qn?
While the named ?rm is directly faced with a tari?, the non-named ?rm
faces the uncertainty of being targeted in the future by an AD measure. In
particular, it is assumed the marginal cost of the non-named ?rm is c? + ? ,
H
where ? is a stochastic variable drawn from some distribution with a strictly
positive mean E(? ) = ¯
? > 0 and variance V ar(? ) = ?2? whenever t > 0.
It should be stressed that when t = 0 then ? = 0. Indeed, under such a
setting since the government does not impose any AD duty on the named
?rm, the non-named ?rm does not face any uncertainty as well. In addition,
let ?H = c ? c? represent the cost di?erence between the home and the
H
non-named ?rm. It is assumed that ¯
? = ¯
? (?H ) and that ? ¯
? /??H > 0,
implying the higher the cost advantage of the non-named ?rm, the higher
the risk of causing injury and consequently facing a higher duty. Thus the
non-named ?rm’s pro?t functions is
?nn? = [P (Q) ? (c?H + ? )]Qnn?
Before proceeding, it is important at this point to formalize the infor-
mation structure of the model. Asymmetric information is modeled by a
random variable ? , which indicates the future uncertainty surrounding the
non-named ?rm being targeted with an AD duty. Ex post, when uncertainty
is realized, ? > 0 means that the non-named ?rm will face an AD duty. Ex
ante however, before uncertainty is realized, ?rms have two sources of in-
formation. The ?rst is their general knowledge about the future outcome,
that is they know that they could be future targets. Mathematically, this
translates in having prior knowledge on the distribution of ? . The second,is
the additional information they gain (at the beginning of the second stage)
in the form of a noisy signal
S = ? +
where
is a white noise with mean E( ) = 0 and V ar( ) = ?2 ? 0. The
extent to which the ?rms take this information into consideration depends
on how clearly the signal informs them about the future AD policy.
The de?nition of transparency in this paper focuses on the quality of
information the signal contains. It does not refer to having perfect informa-
tion about the future. Transparency refers to how clearly the ?rms, after
having received a publicly available information, can extract useful infor-
mation about the future state of the world - which still remains uncertain.
For example the more transparent the AD policy, the more clear the signal
5

---

- translated by a low ?2, consequently leading ?rms to anticipate more pre-
cisely the future AD policy. Under opacity, i.e. a high ?2, the ?rms cannot
extract clearly from the signal how the government intends to use the AD
mechanism in the future.
Some further consideration are needed to make the computation simple.
It is assumed that
and ? are independently distributed, which implies that
Cov(?, ) = 0. Under the above assumptions, and using the result from
Ericson (1969), the conditional expectation of ? given a signal S is
E(? |S)
=
E(? ) + ?(S ? E(? ))
=
¯
? + ?(S ? ¯
? )
(1)
where ? = ?2?/(?2? + ?2) is a measure of the quality of information and
therefore transparency. Expression (1), is a Bayesian updated expectation
that describes how information is used to update prior expectations. It
corresponds to a weighted average of the prior belief E(? ) and the new in-
formation obtained from the signal.1
To summarize, in this two stage game of incomplete information, the
time and information structure of the ?rms results as:
Stage I.
The AD authority imposes an optimal tari? on the named
?rm by maximizing the weighted sum of tari? revenues, consumer surplus,
and producer’s surplus.
Stage II.
The ?rms set output quantities Qh, Qn? and Qnn? in the
presence of a publicly available stochastic signal S.
1For example, if the signal is equal to zero (S = 0), the non-named ?rm receives the
information that it will not be subject to a future AD duty. Moreover, if the AD procedure
is fully transparent so that ?2 = 0, then the expected future AD duty given the signal is
?2
E(? |S) = ¯
? +
?
(0 ? ¯
? ) = 0
?2? + 0
In this case trade diversion is maximum because the ?rms trust completely the information
they receive.
6

---

3
Equilibrium Strategies
As explained in the previous section, AD policy will a?ect the equilibrium
volumes in the market for all ?rms. The present section characterizes the
(Bayesian) equilibrium strategies, while comparing them to benchmark (de-
noted by subscript B)where the non-named ?rm does not face any threat
as to the use of future AD actions. It corresponds to the case where trade
diversion is total. In terms of the present model, the non-named ?rm re-
ceives perfect information that it will not be subject to an AD duty. More
precisely, the benchmark scenario allows one to compare how information
uncertainties can e?ect the equilibrium outputs. In the course the analysis,
these equilibria are derived by the standard backward induction method.
3.1
Total Trade Diversion as The Benchmark
It is straightforward to show that total trade diversion - the case where
the intended e?ect of antidumping policy in reducing ?erce competition for
the domestic producer is totally o?set by the increase in competition from
non-targeted ?rm - corresponds to the situation where the non-named ?rm
knows with certainty that it will not be targeted by an AD action. Here,
each ?rm maximizes its pro?t by setting its output to solve
max ?i, i ? {h, n?, nn?}
Qi?0
If optimal production resulting from the Cournot maximization problem for
the home, named and non-named ?rms are positive, the output of each ?rm
in the industry resulting from an imposition of a duty tari?, obtained from
the ?rst order conditions for the home, named and non-named ?rm are:
? + (c? + t) + c? ? 3ch
Qh
L
H
B
=
(2)
4?
? + ch + c? ? 3(c? + t)
Qn?
H
L
B
=
(3)
4?
? + ch + (c? + t) ? 3c?
Qnn?
L
H
B
=
(4)
4?
Having derived the optimum quantities, one can now consider what happens
to the total supply and the price in the domestic market in presence of AD
protection. It is easy to verify that in absence of AD duty, the total output
is higher and the price of the good lower than in the presence of an AD
7

---

measure.
The result obtained in the benchmark case is standard. The imposition
of a duty on the named ?rm can be seen as an upward shift of the home and
non-named ?rm’s best response function. In other words, the antidumping
measure, decreases the volume of imports for the named ?rm and increases
the output of both the domestic and non-named ?rm.
The Benchmark
scenario puts to doubt the e?ectiveness of AD policy as a tool for protection.
3.2
Equilibrium Strategies with Uncertainty
In this section, it is assumed, that the antidumping authority in the relevant
market, imposes a duty t on the named ?rm and at the same time the
?rms receive a signal about the future intentions of the government. In the
following scenario, the ?rms determine their output level conditional on the
signal. They choose their quantities by maximizing the conditional expected
payo?
max E(?i|S), i ? {h, n?, nn?}
Q??0
i
By using equation (1) and since the signal is common to all ?rms, the output
E(Qi|S) is denoted by Qi . The equilibrium strategies for the home, named
S
and non-named ?rms are
1
Qh
S
=
Qh
B +
¯
? + ?(S ? ¯
? )
(5)
4?
1
Qn?
S
=
Qn?
B +
¯
? + ?(S ? ¯
? )
(6)
4?
3
Qnn?
?
S
=
Qnn?
B
¯
? + ?(S ? ¯
? )
(7)
4?
Notice that the optimal outputs are composed of a the benchmark quan-
tities and a stochastic part. Consequently, two e?ects are at play. First, the
duty e?ect decreases the output of the named ?rm and increases the quan-
tity of the home and non-named ?rm. Second, the signaling e?ect decreases
the output of the non-named ?rm but at the same time increases the output
of the home and named-?rm. Indeed, once the non-named ?rm faces uncer-
tainty, it becomes more prudent in increasing its import, giving opportunity
to the home ?rm to increase its output and to the named ?rm to regain a
part of its market share. Bare in mind that the second e?ect can be equal
to zero, it corresponds to the extreme case of a perfectly transparent AD
policy is (? = 1) and ?rms receiving information that the non-named ?rm
8

---

will not be subject to a future AD duty (S = 0). The interesting conclusion
is that these two e?ects both work out to the advantage of the home ?rm,
and this leads to the following proposition:
Remark 1 When ?rms update their belief with a signal that reveals infor-
mation about the future intention of the AD administration in protecting the
home country:
1. the imports from the non-named ?rm are lower than (or equal to) the
benchmark quantity
2. the imports of the named ?rm are higher than (or equal to) the bench-
mark quantity
3. the production of the domestic industry is higher than (or equal to) the
benchmark quantity
Again it is interesting to consider what happens to the total supply and
the price in the domestic market with respect to the benchmark scenario. It
is easy to verify that in the presence of the signaling e?ect, the total output
is lower (or equal) and the price of the good higher than (or equal to) the
benchmark.
4
The Governments Objective Function
This section analyzes the ?rst stage of the game. It is assumed that the
government maximizes its objective function described by a weighted aver-
age of the domestic ?rm’s pro?t, the consumer surplus and the antidumping
tari? revenue:
G = ? CS + ?h + tQn?
(8)
where ? ? [0, 1] is the weight attached to the consumer surplus. The
reason for choosing a weight on the consumer surplus rather than the pro-
ducer surplus is to be able to compare the institutional di?erence between
the United States and the European Union. Indeed, in the United states
since there is no community interest clause in the AD law, this is interpreted
by ? = 0, that is the government imposes a tari? without taking into ac-
count the consumer surplus. On contrary, in Europe, the AD law includes a
community interest clause and this is hence interpreted by ? = 1. In reality,
? could be somewhere in between zero and one, but for analytical purposes
9

---