\^/Ps-2    s
POLICY RESEARCH WORKING PAPER           2955
The Perversity of Preferences
The Generalized System of Preferences
and Developing Country Trade Policies, 1976-2000
(4aglar Ozden
Eric Reinhardt
The World Bank
Development Research Group
Trade
January 2003



POLICY RESEARCH WORKING PAPER 2955
Abstract
Industrial countries maintain special tariff preferences,    country's trade policy choice, using a dataset
namely the Generalized System of Preferences (GSP), for      of 154 developing countries from 1976 through 2000.
imports from developing countries. Critics have              They find that countries removed from the GSP adopt
highlighted the underachieving nature of such                more liberal trade policies than those remaining eligible.
preferences, but developing countries continue to place      The results, corrected for endogeneity and robust to
the GSP at the heart of their agenda in multilateral         numerous alternative measures of trade policy, suggest
negotiations. What effect do such preferences have on a      that developing countries may be best served by full
recipient's own trade policies? Ozden and Reinhardt          integration into the reciprocity-based world trade regime
develop and test a simple theoretical model of a small      rather than continued GSP-style special preferences.
This paper-a product of Trade, Development Research Group-is part of a larger effort in the group to study global trade
regimes. Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please
contact Paulina Flewitt, room MC3-333, telephone 202-473-2724, fax 202-522-1159, email address
pflewitt@worldbank.org. Policy Research Working Papers are also posted on the Web at http:/Iecon.worldbank.org. The
authors may be contacted at cozden@worldbank.org or erein@emory.edu. January 2003. (33 pages)
The Policy Research Workmig Paper Series disseminates the findings of wvork in progress to encourage the exchange of tdeas about
development issues. An objective of the series is to get the findings out qutickly, even if the presentations are less than fully polished. The
papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this
paper are entirely those of the authors They do not necessarily represent the view of the World Bank, its Executive Directors, or the
countries they represent.
Produced by the Research Advisory Staff



The Perversity of Preferences: GSP and Developing
Country Trade Policies, 1976-2000*
Qaglar Ozdentand Eric Reinhardtt
JEL Classification: F13, F14, D72, D78, 019.
'We would like to thank Caroline Freund, our discussant at the AEA meetings, plus Jagdish Bhagwati,
Owen Beelders, Pravin Krishna, Rod Ludema, Joy Mazumdar, Jerry Thursby, Marie Thursby, Christopher
Zorn and seminar participants at the AEA Meetings in Atlanta, University of Maryland-College Park, the
World Bank, International Trade Corunission and the IMF for comments and suggestions.  Francisco
Parodi provided research assistance. This research was partially funded by Institute for Comparative and
International Studies and University Research Council of Emory University
tCorresponding author. Dept of Economics, Emory University and World Bank. Mailing address: Inter-
national Trade Unit, World Bank Mailstop: MC3-303. 1818 H Street NW Washington, DC 20433. E-mail:
cozden(worldbank.org. Phone: 202-473-5549.
tDept of Political Science, Emory University, Atlanta, GA 30322. Email: erein@emory.edu Web:
http://userwww.service.emory.edu/-erein/.






1 Introduction
The United Nations [2001] declaration that "increased trade is essential" for the world's poor
countries "to reap the potential benefits of globalization" is a commonly shared view. This
principle is enshrined in the postures industrialized countries and international institutions
adopt in dealing with developing countries. Yet there is a lively political and theoretical
debate on how best to accomplish this end. The prevailing approach, known as "special and
differential treatment", grants certain products from developing countries preferential access
to industrialized countries' markets without reciprocal liberalization in turn. The General-
ized System of Preferences (GSP) is the leading instrument for such nonreciprocal trade
preferences. Advocates contend that GSP should "increase the export earnings, ... promote
the industrialization, and ... accelerate the economic growth" (GATT [1972]) of recipient
countries. Even now, 31 years after General Agreement on Tariffs and Trade (GAIT) mem-
bers first authorized GSP as a "temporary" measure, it remains "highly popular" among
developing country beneficiaries (Jackson [1997]). Strengthening preferential market access
without reciprocal obligations is the centerpiece of the developing world's agenda in the
current "Development Round" of multilateral trade negotiations (WTO [1999, 2000], Pana-
gariya [2002a]).
For critics, however, nonreciprocal preferences like GSP are a "Faustian bargain" (World
Bank [1987]). As numerous empirical studies have demonstrated, protectionist forces in
advanced industrial states severely restrict GSP benefits. For instance, such preferences
are typically revoked when they actually start to boost the recipients' exports, as shown in
Ozden and Reinhardt [2002]. The problem is that nonreciprocal preferences like GSP lie
outside the purview of the binding GATT legal system, so they can be unilaterally modified
or cancelled by donor countries at any time. Hence, critics advocate turning developing
countries into regular participants who swap trade concessions for reciprocal market access
(Panagariya [2002b]; Whalley [1990]; Krueger [1995]).
The theoretical literature does not provide answers to this policy debate. Starting with
the pioneering work of Johnson [1953-4], the literature recognizes the efficiency-increasing
properties of trade negotiations, especially in the presence of terms-of-trade externalities.
Bagwell and Staiger [1999] show that rectprocal liberalization enables trading partners to
internalize these externalities and reach Pareto superior outcomes compared to protectionist
policies. Krishna and Mitra [2000] and Coates and Ludema [2001] construct stylized political-
economy models that show untlateral liberalization by a large country may induce small
developing country trade partners to liberalize under certain conditions. Thus, although
trade liberalization is one of the most strongly advocated development policies, there is no
agreement on whether reciprocal or unilateral liberalization by developed countries would
best induce such behavior among developing countries.
This paper asks how nonreciprocal preferences such as GSP affect the trade policies
of beneficiary countries. The answer is important for both the policy and the theoretical
debate. In particular, we argue that nonreciprocal preferences have the perverse effect of
2



delaying trade liberalization by recipients. That is, developing countries withdrawn from
GSP subsequently lower their trade barriers more than countries remaining eligible for such
preferences. The reason is simple. Governments set trade barriers while balancing political
support from import-competing and export sectors. GSP shifts the balance to the import-
competing sector. As Hudec [1987] observed, "the non-reciprocity doctrine tends to remove
the major incentive that [developing country] export industries have ... for opposing protec-
tionist trade policies at home ... instead of trying to enlist the support of the export sector
for liberal trade policy." Furthermore, if there is a threat of removal from the GSP program
when its exports increase significantly, then the recipient has the perverse incentive to im-
plement even more protectionist policies to limit its exports and avoid such outcomes. On
the other hand, if GSP is withdrawn and access to the export market becomes conditional
on its own trade policy through the reciprocity rule, the recipient government's optimal level
of protectionism declines.
We develop this argument using a simple two-sector general equilibrium model of a small
country where the optimal trade policy may be influenced by political economy motivations.
This paper's chief contribution is, however, empirical. We provide the first extensive evidence
on the effects of the GSP program on beneficiaries' own trade policies. Specifically, we
test the effects of GSP removal on beneficiaries' trade policies using a dataset of annual
observations of 154 developing countries since the United States started the program in
1976. We demonstrate that countries dropped from GSP subsequently adopt lower trade
barriers than those remaining eligible for nonreciprocal (as opposed to reciprocal) preferences.
These findings control for income, market size, geography, growth, and other factors; they
are robust to five different measures of trade barriers; and they hold when correcting for
possible endogeneity of the US GSP program.
Writing before they were implemented, Harry Johnson argued that nonreciprocal pref-
erences would fail if developing countries did not cease import substitution practices that
create "cost disadvantages ... frequently ... far greater than the competitive advantage that
could be conferred by preferences from the developed countries" (Johnson [1967]). We go one
step further, in claiming that nonreciprocal preferences actually delay removal of the very
policies Johnson said would make GSP ineffective. To the extent that GSP's perverse effects
are magnified by donors' protectionism, our findings support critics who argue against unilat-
eral modification of GSP commitments. If increasing trade is indeed a key to development
and growth, then developing countries would be better served by abandoning reliance on
nonreciprocal preferences altogether, becoming members with full responsibilities and thus
rights in the world trading regime. We accordingly conclude, with Roessler [1998], that "the
cause of development was manifestly not served by releasing developing countries from their
GATT obligations."
Our results have farther-reaching implications, especially for the future of the multilateral
trading regime. Recent decades have witnessed the proliferation of discriminatory and non-
reciprocal practices that push both the letter and the spirit of the GATT/WTO rules to their
limits. There can be little doubt that GATT/WTO's twin principles of non-discrimination
3



and reciprocity contributed significantly to the success of the multilateral negotiations for
trade liberalization (Jackson [1997]). The place of programs like GSP in the regime must be
rationalized if the WTO is to continue as the basis of future liberalization efforts, especially
those, like the Doha agenda, dealing principally with developing countries. As our paper
shows, these programs damage trade liberalization efforts in two distinct ways. First, they
prolong protectionism among recipient countries. Second, and more importantly, develop-
ing countries have no avenues to protect their rights since these policies are outside WTO
jurisdiction, permitting donor countries to change these programs as they wish.
The next section outlines the main features of the GSP program and surveys some relevant
literature. Then we provide a brief model that leads to the.conjectures tested in the empirical
section, followed by sections describing the data, the econometric model, and the empirical
results along with extensions. Conclusions follow.
2 GSP Basics
The Generalized System of Preferences is composed of trade preferences granted by the de-
veloped world to developing countries en masse on a nonreciprocal basis, i.e., without market
access concessions in return. It was first proposed by the United Nations Conference on Trade
and Development (UNCTAD) during the Kennedy Round to encourage the participation of
developing countries in GATT.V The main principle of GSP is to provide better-than-MFN
("most-favored nation") treatment to imports from qualifying countries. For instance, the
United States' GSP program sets a zero tariff on 6,409 articles from beneficiary states (out
of total 15,467 tariff lines) while other countries face higher MFN tariffs.
GSP clashes with two critical principles of the multilateral trade regime (Hudec [1987]).
The first is reciprocity, the strategic mechanism through which states can escape Prisoner's
Dilernma problems in trade negotiations. The second is nondiscrimination, the main pillar of
the regime as encoded in Article I of the 1947 GATT text. Recognizing these inconsistencies,
GATT members approved special waivers for GSP, temporarily in 1971, and permanently in
1979 through the "Enabling Clause" of the Tokyo Round agreements. However, the Enabling
Clause lacks effective guidelines on GSP implementation, allowing donors to design their
schemes "as they see fit" (GAO [1994]; Hudec [1987]; Jackson [1997]). In sharp contrast to
reciprocal GATT concessions, "he who giveth may taketh away" freely in the context of
GSP-conferred market access.
The United States was the last country to enact such preference schemes, taking effect
in January 1976. The 1974 Trade Act allows the President to confer GSP eligibility on
'The US also administers the Caribbean Basin Initiative, the Andean Community preference, and a
even more preferential system in favor of "least developed countries," a subset of the poorest GSP-eligible
states; but GSP in 2000 covered about 75 percent of all such US preferential imports. The European Union
maintains the Lome Convention, a better-than-GSP scheme for former colonies in Africa, the Caribbean,
and the Pacific.
4



any country except those which (a) harbor international terrorists, (b) nationalize American
property without compensation, (c) are members of a commodity export cartel causing "se-
rious disruption to the world economy," or are (d) communists (except those with permanent
normal trading status). The law stipulates other criteria that may be used in eligibility deci-
sions, such as (a) level of economic development, (b) protection of workers' and human rights
and (c) whether the country receives preferences from other countries. Also, specific prod-
ucts may be removed from eligibility if the exports exceed a certain "competitive need limit,"
currently $100 million per tariff line, per year, per country. The President has discretion
over when and how to apply these criteria. In practice, an Assistant US Trade Representa-
tive chairs an interagency committee which makes decisions after reviewing petitions from
interested parties (the country in question, import competing domestic firms, relevant US
labor unions, intellectual property groups, human rights/environmental NGOs). GSP eligi-
bility can be removed at the country, product, or country-product level. Out of 154 eligible
countries, 36 have "graduated" from the GSP program since 19762 (e.g., Singapore, Hong
Kong, Taiwan, Korea, Malaysia, Mexico, Botswana). The list of major countries remaining
eligible includes Brazil, India, Russia, Indonesia, lurkey, South Africa, and Thailand.3 Be-
cause GSP decisions are legally unilateral, import-competing lobby groups have effectively
made GSP the last bastion of truly unregulated protectionism in the United States (Hudec
[1987]).
Existing studies of GSP focus on recipients' export performance, highlighting several
common points. First, GSP programs - and not just that of the United States - fail to
cover some products in which beneficiary states have the greatest comparative advantage,
such as textiles.4 In 2000, for instance, just 47 percent of the $175.6 billion of US imports
from GSP beneficiaries were in tariff lines listed under the program (USITC [2002]). Second,
by invocation and even more by anticipation, the export ceilings on GSP product eligibility
are often binding.5 Third, complexity of the system (especially its rules of origin paperwork)
and technical incapacity of developing country exporters inhibit full use of GSP preferences
even when eligibility is not a problem (UNCTAD [1999, 2001]). Fourth, as Clark and Zarrilli
[1992] demonstrate empirically, GSP donors disproportionately substitute non-tariff for tariff
protection on sensitive GSP-eligible products. Not surprisingly, the universal conclusion of
numerous empirical studies is that GSP has underperformed, yielding at best a "modest"
increase in imports from beneficiary states, with some of those gains due merely to trade
diversion.6 To quote Whalley [1990], "available quantitative studies ... seem to point to the
conclusion that special and differential treatment has had only a marginal effect on country
economic performance, especially through GSP."
Surprisingly, however, no studies have examined how nonreciprocal trade preferences af-
2Being dropped from the program is often called "graduation," because in many (but not all) cases the
stated rationale is a sufficiently high per capita income.
3Some dropped countries were subsequently reinstated.
4WTO [2000], UNCTAD [1999, 2001], Ray [1987], Devault [1996].
-5MacPhee and Rosenbaurn [1989], Hoekman and Kostecki [2001], Finger and Winters [1998].
6'MacPhee and Oguledo [1991]; Brown [1987, 1989]; Grossman [1982]; Sapir and Lundberg [1984].
5



fect a recipient's own trade policy. So far this subject has been a matter of mere conjecture.
For example, Hudec [1987] argues that, by removing "external legal constraints," GSP un-
dercuts the ability of a developing country leader to tie his hands when facing protectionist
pressures, even "making things worse by giving greater legitimacy to claims for trade pro-
tection." Roessler [19981 somewhat more cynically agrees, claiming that GSP "provided
developing countries with a justification for refusing to make market-access conunitments."
Finger and Winters [1998] likewise observe that "the wish to maintain preferences could
undermine developing countries' willingness to invest in rounds of negotiated trade liberal-
ization." 7
Recent theoretical on the role of reciprocity recognizes its efficiency increasing proper-
ties. Specifically, Bagwell and Staiger [1999] argue that reciprocal liberalization enables large
trading partners to internalize the terms-of-trade externalities inherent in multilateral trade
negotiations, enabling them to reach Pareto superior outcomes.8 A second set of papers asks
whether unilateral liberalization by a large country can cause a small country to liberalize
as well. Krishna and Mitra [2000] and Coates and Ludema [2001], in particular, emphasize
political economy mechanisms through which the large country's unilateral liberalization
modifies the lobbying game inside the smaller country and induces liberalization. Hence,
in terms of the large country's policy choice, the contrast for Bagwell and Staiger [1999] is
between protectionism and reciprocal liberalization, while for the latter papers it is between
protectionism and unilateral liberalization. To evaluate the effects of GSP in a world oth-
erwise dominated by GATT-based reciprocity, however, the appropriate contrast is between
unilateral and reciprocal liberalization, from the standpoint of a developed state vis-a-vis
developing country imports. We develop a simple model to this end in the next section.
3 Model
We consider an import and an export sector in a small country whose trade policies can only
influence domestic prices but not world prices. All variables related to the import (export)
sector are identified by the superscript m (x). The domestic price of the import good is
denoted by pm and its foreign market price by pT. Similar variables (pZ and px,) are defined
for the export good. Variable p = pm/pX (P, = pT/pw) is the relative price ratio faced by
consumers and producers in the domestic (world) market. Furthermore, p,,. is the inverse of
the terms of trade for the small country.
We assume that the only policy tool available to the government is the tariff (or the sub-
sidy) on the import good. This might seem restrictive but, in general equilibrium settings,
7See also Nowzad [1978]; Michalopoulos [1999] .
8McLaren [1997] provides an interesting model highlighting the importance of commitment to trade
liberalization in the presence of sunk costs. One of the benefits of reciprocal liberalization under GAIT is
this commitment value.
6



the Lerner symmetry theorem allows us to focus only on the trade policy in one sector. Im-
port policies are certainly more common in practice. We express the trade policy parameter
as t, which is one plus the non-prohibitive ad valorem tariff rate in the domestic market.
This implies p = tpw, since pZ = pZ. We further have that dp/dt > 0 and Op.OIt = 0, since
this is a small country.
Production of the import and the export goods in this country is characterized by a
production possibilities frontier. The domestic price ratio p determines the marginal rate of
transformation and the production levels (QX (p) and Qm (p)) of the two goods.9 The demand
functions Dm (p, T) and D' (p, T) are also determined by the domestic price ratio p and the
overall tariff revenue T, which is distributed to the consumer. Tariff revenue is a function of
domestic and world prices, which makes domestic demand a function of these as well. Next
we impose a balanced trade constraint: Pw [Din (Ppp) - Qrn (p)] = [Qm (p) - Dm (p, p.)]-
Thus, the exogenous world price Pw and the tariff rate t determine the domestic price level
p, and this, in turn, determines the demand and supply levels of both goods as well as the
tariff revenue.lU
Government preferences are represented through a very general objective function that
depends on endogenous domestic and exogenous world prices as parameters, following Bag-
well and Staiger [19991: G (p,pw) = G (p (t,pW) ,Pw) . This is somewhat different than the
norm in most papers, where the objective is a simple function of the tariff rate, but there
is no loss of generality. We impose the restriction that the improved terms of trade lead to
higher welfare for the government when the domestic prices are fixed: 9G (p, p,,) /lPw < 0. 11
This structure is quite general: most models of trade policy formation'2 can be represented
this way.
The optimal trade policy tP is given by the first order condition:
dGdp
dp dt
We assume that the second order condition is satisfied: Gpp < 0. We should note that
the optimal policy is free trade (t = 1) in most models regardless of world prices if the
government were to maximize social welfare. The presence of positive tariffs (t > 1) implies
that political economy motivations are in place. The optimal trade policy might also be a
function of exogenous world prices pw*
9The production possibilities frontier can be derived from a specific factors model in which each sector
employs two factors. One of these factors (capital) is sector-specific and is supplied in fixed quantities while
the other (labor) is mobile. An increase in the price of the import good (and thus an increase in p) increases
the demand for labor employed in that sector and the wage which, in turn, decreases the profit level in the
export sector. This structure is quite different than the ones found in some papers (such as Grossman and
Helpman [1994]) where labor is assumed to be in excess supply and to have a constant price so that are there
are no production externalities between sectors. This model is presented in more detail in an earlier version
of the paper. It is available upon request.
lf)For a more detailed version of this setting, see Bagwell and Staiger [19991.
We should note that Pw is the inverse of the terms of trade given the way we defined our variables.
12Grossman and Helpman [1994] and see Rodrik [1995] for an extensive survey.
7



3.1 Granting GSP
What happens to t* when the small country receives nonreciprocal preferences? The GSP
status granted by the United States and other developed countries allows a developing coun-
try to export its goods to these markets without the regular tariffs applied to other suppliers.
This special treatment increases the price exporters from GSP-eligible countries receive for
their products, because the GSP donor markets are big enough to affect the world prices
through their trade policies. In other words, being included in the GSP program is equivalent
to an increase in the exogenous variable pX or a decline in Pw in our model."3 If we totally
differentiate the first order condition (1), we obtain dt*ldp  = - (Gppt + GpPj) /Gppp  or
sign (d   ) = stgn (Gppt + Gpp.)
We know that Gpp < 0 but the sign of Gppw is ambiguous and, actually, in most models, it
is positive. Thus the impact of the initial conferral of GSP (a decline in Pm) on the small
country's optimal trade policy (t*) is also ambiguous; it depends on the details of the model
at hand. Note that the baseline for comparison here is a unilaterally protectionist regime
by the large, donor state.
A decline in pw, (or increase in p,) is equivalent to unilateral liberalization by a large
country, and this is analyzed in the models by Krishna and Mitra [2001] and Coates and
Ludema [2001]. They show that, under certain assumptions about the underlying political
economy framework, this causes the small country to liberalize. Our simple scenario, on the
other hand, states that it is not possible to derive unambiguous results about the impact
of a movement from a unilaterally protectionist regime to a unilaterally liberal regime by
large countries. That is, our generic model highlights the sensitivity of their results to
assumptions about the domestic political economy setting. However, that is not this paper's
main concern; rather, we aim to compare unilateral lzberalhzat%on (via GSP) to reczprocal
lzberalizatzon (as in GATT/WTO). And, in this respect, the results are more clear.
3.2 Imposing Reciprocity
Reciprocity has always been one of the most important norms of GATT negotiations. The
general principle of reciprocity requires countries to lower their tariff rates and other trade
barriers together, at least roughly balancing the market access concessions offered by other
members. Thus Pareto efficient outcomes can be achieved compared to the inefficiencies of
unilateral tariff choices. The methods of implementing reciprocity and assessing the balance
'3Note that p_ is actually the terms of trade faced by exporters and importers trading with the large
country. In the absence of GSP this is identical to the actual world price. However, under GSP the terms
of trade differ from the actual world price, as these firms do not have to pay tariffs due to their preferential
status.
8



of concessions have changed from one GATT round to the next, but the general principle
has prevailed since the first days of GATT (Hoekman and Kostecki (2001]).
Once a country "graduates" from GSP, it is no longer eligible for the (positive) discrim-
inatory treatment and duty-free access to the US market. Furthermore, if the graduating
country wants to receive the MFN tariffs granted by the United States to its other WTO
partners, it may be compelled to reciprocate by lowering its own barriers to imports, espe-
cially for those coming from the US. In other words, the principle of reciprocity imposes a
conditionality on the export prices of the developing country in terms of the domestic prices
of its imports. We represent this conditionality in very general terms as pX = f (pm) where
f/ < 0.
We must emphasize, however, that we are not offering a positive model of the donor
state's decision to impose such reciprocity. Indeed, the incentives for the US to demand
such reciprocity may decline with the market size of a GSP beneficiary. On the other hand,
there are many other economic or institutional reasons to the demand for reciprocity by the
US as well. Our claim here thus addresses the consequences of a policy change by the GSP
donor state.
Returning to the model, we see that the reciprocity rule implies that a higher domestic
tariff rate t on the import goods leads to lower export prices pX and worsened terms of trade
p. Under reciprocity, the objective function of the government becomes
G (P, PW) =G (PM)     P. ))
where pm = tpm. The tariff rate is again the only decision variable for the government.
However, unlike the previous section, the tariff rate chosen by the government influences pX
and, therefore, the domestic price ratio p and the terms of trade p1. The first order condition
of the objective function under the reciprocity rule is
[1 - fp] Gp - f'p.Gp. = 0
We know that Gp,, < 0 and f' < 0, which imply Gp = lf'P-,GpW > °.
The critical question is whether the withdrawal of GSP (and the attendant shift to
reciprocity) leads to a lower equilibrium tariff rate t. To answer this question, let PGsp denote
the export prices under the GSP program, tRBC (t*sp) be the equilibrium tariff rate under
reciprocity (GSP), and pT be identical under the GSP regime and the reciprocity rule. Then,
suppose the tariff rate chosen under the reciprocity rule leads to the same export prices as
the GSP regime. In other words, the developing country government chooses tariffs so that
it receives the same tariffs on its exports as it did under GSP: f (tRECPT) = PREC = PGSP-
Therefore, the export prices px (and hence the terms of trade p,w,) are the same under the
two regimes, but the import prices pm (and therefore domestic prices p) might be different
due to differences between the equilibrium tariff rates tREC and t*,. But we have Gp > 0
9



in equilibrium under the reciprocity rule, while Gp = 0 under GSP. This implies that PREC
and PREC are lower than PGsp and PGSP, respectively, since G is a concave function of p,
and pZ is the same under the two regimes. The next statement summarizes this result.
Conjecture 1 Suppose the reczprocity rule leads to the same world pmces for the export
good x (i.e., yielding the same tamffs from the United States) as the GSP regzme. Then the
equilibrium tamff implemented under the reczprocsty rule zs lower than the one implemented
under GSP.
This conjecture states that removing a developing country's eligibility for GSP, and thus
giving market access only in exchange for reciprocal tariff reductions, indeed induces the
developing country to reduce its import barriers. The reciprocity rule compounds the costs
- by lowering export prices - of the developing country's decision to maintain higher
tariffs. This increase in the marginal cost of tariffs lowers their equilibrium level. In other
words, if a developed country liberalizes its trade policies reciprocally, within the scope of
the GATT/WTO regime, it will induce more liberalization by developing countries than if
it had liberalized unilaterally, as under GSP. Removing a beneficiary from GSP should thus
cause it to adopt more liberal trade policies.
4 Data and Variables
This section tests the hypothesis that countries dropped from the GSP program adopt more
liberal import policies than those remaining on GSP. We construct a dataset with one
observation per year per independent developing country, from its first year as a United States
GSP beneficiary (minimum 1976, the start of the program) through 2000. Our definition
of "independent" follows the official US GSP manuals (e.g., USTR [19991).14 This yields a
total of 154 countries and 3210 country-years in the panel.
We previously noted that most developed countries maintain their own versions of GSP.
If our hypothesis is correct, all of these programs should have an impact on developing coun-
try trade policies, weighted perhaps by the significance of the industrialized nation's market
for the exports of the country in question. However, we have data on only the United States'
GSP program. This is no great loss, because the overall margin of preference conferred by
the US GSP program is by most accounts fundamentally comparable to the preferences of
the European Union and other developed countries (GAO [2001], UNCTAD [1999]). Fur-
thermore, the US GSP-authorizing law stipulates that the President may base GSP eligibility
1'Following the US GSP program guidelines cited earlier, most OPEC members and members of the
communist bloc were never granted GSP eligibility (until they experienced a regime change), with some
exceptions, namely, Angola, Ethiopia, Grenada, Laos, Madagascar, Mozambique, Nicaragua, Romania, and
Yugoslavia (communist regimes) and Indonesia, Nigeria, and Venezuela (OPEC members). The others are
thus not in our sample, which only includes a state from its first year on the GSP program.
10



decisions on "the extent to which other major developed countries are undertaking a compa-
rable effort," i.e., extending GSP preferences of their own to the country in question (USTR
[1999]). Thus, different GSP programs are correlated"5 and the measurement of GSP status
here, while technically representing the US program only, offers a reasonable approximation
of all developed country preferences for developing countries.
4.1 Key Variables
GSP Status. The dummy variable GSP,,t is 1 if country i was a United States GSP benefi-
ciary in year t; 0 otherwise"6. This information was collected from USTR [1991] and Federal
Register [various]. The GSP dummy admittedly glosses over important distinctions between
countries with widely-varying amounts of actual US GSP benefits. In order to capture that
variation, we also construct the variable GSP Fraction,,t, defined as country i's GSP exports
to the US (USITC [2002]) as a percent of its total exports in year t (World Bank [2001b]).
Unfortunately, this variable is available only for the last decade,17 so we use GSP Practzon
for sensitivity testing only.
Trade Polzcy Measures. Available data on aggregate national trade protection are no-
toriously imperfect.  Our strategy is redundancy: if we obtain similar findings with each
of the several most frequently used measures of trade policy, our inferences will be more
robust. The most common method of measuring protectionism is to examine trade flows,
not policy, in the form of trade volume divided by GDP, mainly because of data availability.
Accordingly, our variable Closure,,t is country i's total imports as a percent of GDP in year t.
This is subtracted from zero to make higher values represent greater import restrictiveness,
yielding a sample mean of -46. It has the virtue of capturing the effect of non-tariff as
well as tariff barriers. However, as many have argued (e.g., Dollar and Kraay [2001] and
Rodrfguez and Rodrik [2001]), trade openness reflects not just polzcy but also geography
and market size (though we control on the right hand side for these factors). Among the
measures which reflect policy more directly, we adopt Dutiesq,t (duties as a percent of total
trade, from World Bank [2001b]) and Tariff,t (the unweighted average nominal tariff, from
World Bank [2001a]). These variables average 9.4 and 20.6, respectively, in our dataset.
Rodrfguez and Rodrik [2001] make a strong case for Duties as the best of an imperfect lot,
since Tariff weights policies on small-volume products like golf putters the same as tariffs
on automobiles. However, because highly protected industries have fewer imports (and thus
fewer duties), Duttes understates protection levels. Below we estimate models using these
I5For instance, the US announced the withdrawal of Malaysia from its GSP program in October 1996, and
the European Union followed suit in 1997 by slashing eligibility for all of Malaysia's chief exports (palm oil,
plastics, rubber, wood, clothing, consumer electronics, and all agricultural products), if not cutting it from
the program formally.
16The GSP eligibility of Pakistan and Argentina was halved in the late 1990s; we code this as withdrawal
of GSP.
17Earlier data on GSP imports by country are disaggregated by 10-digit product code, by customs entry
point, by month, and available only in microfiche form.
11



three dependent variables separately. Because data on all three are missing for a sizable
portion of the country-years in the overall dataset (13, 52, and 73 percent missing, respec-
tively), the dependent variable is the greatest constraint on the size of the samples used in
these regressions.18
Bzvanate Associattons. If the hypothesis is correct, then countries on GSP should be
more protectionist (with higher Dutzes, Tariff, and Closure) than countries dropped from
GSP. In Figure 1, which displays averages across all country-years in the sample, on and
off GSP, we see this is indeed true. The average ex-GSP recipient, compared to the average
beneficiary, has less than half the nominal tariff and duties, and imports nearly 9 percent
more of its GDP per year. This association is evident for the continuous measure of GSP
benefits as well. For example, the correlation between GSP Fraction,,t and Tarnff,,t is 0.15
(p < 0.001, n = 572). Of course, these simple summaries are by no means definitive, since
they fall to control for alternative explanations of trade policy, but they nonetheless indicate
the association of developing country protectionism with GSP.
4.2 Control Variables
Trend. Virtually all developing countries have liberalized their trade policies significantly
over the course of the past decade or two. Since 36 countries have been dropped from the US
GSP program over time, a positive correlation between GSP withdrawal and liberalization
could result from this secular trend alone. To isolate GSP's impact from the secular trend,
we include the variable TRendt, i.e., the mean of the (relevant) dependent variable across all
countries in the dataset in year t.19
Condstionality. Many developing countries in this time period have experienced acute
financial crises, and, in most cases, sought assistance from the International Monetary Fund
(IMF), whose conditionality terms typically include trade liberalization. Compliance with
IMF conditionality is seriously lacking on average (e.g., IMF [1998] and Goldstein [2000]).
Rodrik [1994] similarly argues that trade liberalization in the developing world has not been
driven by pressure from international financial institutions. Nonetheless, to control for this
potential cause of freer trade policies, we include the variable IMF,,t, a dummy which is 1
if an ongoing IMF program is underway in country i in year t (Vreeland [2002]),20 and 0
otherwise. IMF programs were underway in about 38 percent of the country-years in the
dataset.
Market Size. Countries with larger markets may be more diversified and thus have a
smaller proportion of GDP involved in trade. Hence we introduce a control for (the natural
18Testifying to the importance of redundancy for this analysis, the three measures are not highly correlated
For Dutzes,t and Tariff,,t, r = 0.37 (n = 542); Dut%es,,t and Closure,,t, r = 0.13 (n = 1540); Tanff,,t and
Closure,,t, r = 0.42 (n = 864).
"9We thank Jagdish Bhagwati for suggesting the inclusion of this variable.
2('Thanks to Jim Vreeland for the most current update of his dataset of IMF program participation.
12



log of) country i's market size in year t, Log GDPi,t (in constant 1995 US dollars), which
should affect Duties, Tariff, and Closure in line with conventional expectations.
Income. Income is one of the most important determinants of trade flows, if not trade
policies2". High income is also a statistically significant predictor (though not the only one)
of removal from US GSP eligibility (Ozden and Reinhardt [2002]). In order to limit the
potential for spurious inference about the effects of GSP, we control for Log Income,,t (using
per capita GDP in constant 1995 US dollars).22
Geography. A country's distance from important suppliers and markets affects transport
costs and thus trade levels, as gravity model studies have repeatedly shown (Leamer and
Levinsohn [1995]). In addition, such geography-induced transport costs may serve as a sub-
stitute for policy-created import barriers. In contrast to the usual pairwise gravity model,
however, we control for location using a country-specific geographical indicator. Specifically,
Gallup and Sachs [1999] give the proportion of a country's population that lives more than
within 100 kilometers of a coast or navigable river (here, Landlocked,) - though their data
are not time-varying. The proportion of "landlocked" population in the dataset's average
country is 0.39.
Growth. The business cycle is often said to affect trade policy. Protectionist pressures
are often greatest during economic downturns (Dornbusch and Frankel [1987] and Rodrik
[1995]). Yet severe recessions also may catalyze reform, while prolonged growth may retard
it (Rodrik [1994]). For that reason, we control for real GDP growth (Growth1,t) in our
analyses (World Bank [2001b]).
5 Econometric Models
We estimate several sets of regressions, comparing countries still eligible to those withdrawn
from the US GSP program. The first set (models 1, 2, and 3) concerns GSP status as
a dummy variable, thereby permitting use of the entire 1976-2000 period.  We lag all
explanatory variables by one year (except Landlocked) as a first-cut method of addressing
concerns about endogeneity.23 For robustness, we estimate the same model separately for
each of the three primary measures of the dependent variable. Hence, if Yi,t is either Duties,t,
Tarnff,,t, or Closure,,t for country i in year t then the estimated equation is
Y.Nt =  Ce + 51 GSPI,-1 + ,62 Rhendt.I + 833IMF,,t-1 + 14 (Log GDP1,t-1)  (2)
+135 (Log Income,,t-) + /B6Landlocked, + /7Growth,,t-j + e,,t.
21According to Baier and Bergstrand [2001], it is the most important deterninant.
22The source for GDP and per capita GDP figures is World Bank [2001b].
23We set GSP,,t,1 = 1 for each country's first year in our dataset. (Recall that the sample begins the
first year a country is eligible for GSP.)
13



Due to the cross-sectional time-series nature of the data, heteroscedasticity and autocor-
relation are likely. A common estimation approach in this setting is feasible generalized
least squares (GLS) with contemporaneous error correlation, groupwise heteroscedasticity,
and first-order unit-specific autocorrelation (e.g., Greene [2000]). However, GLS assumes
we know the true error covariance matrix, the basis of the FGLS transformation. Because
there are N x (N + 1)/2 contemporaneous covariances (where N is the number of countries
in the panel), and because unit-specific serial correlation draws on just one country's string
of observations, good estimates of the error covariance matrix demand many more years per
country than countries themselves. Using Monte Carlo simulation, Beck and Katz [1995]
demonstrate, in precisely the situation our analysis faces,24 FGLS standard errors dramati-
cally understate true variability, whereas OLS with appropriately corrected standard errors
is less efficient but consistent (see also Greene [2000]). Hence we estimate equation 2 with
OLS, reporting Newey-West standard errors robust to heteroscedasticity and first-order au-
tocorrelation (e.g., Greene [2000]).25 Recall that 31 should be positive in each version of
the equation, using Dutzes, Tamff, and Closure. Table 1 displays descriptive statistics for
the regression samples.
5.1 Estimation Results
The results strongly support our hypothesis that removing GSP induces liberalization. In
every model, GSP is a statistically significant determinant of protection, with a powerful
substantive impact as well. These findings are robust to a wide variety of sensitivity analyses.
Table 2 shows the full-sample OLS estimates of the Duties, Tariff, and Closure equa-
tions (models 1, 2, and 3). All models fit the data well, as evidenced by the tables' model
test statistics. Adjusted R2s in the OLS models lie between 0.31 and 0.41. Diagnostics
indicate heteroscedasticity and first-order serial correlation.26  Hence the more conserva-
24We have more than a hundred cross-sections but, after missing data, at best no more than 19 years per
country
25We do not include error components. The random effects assumption that country-specific random
errors are uncorrelated with the regressors is clearly not appropriate, given the concerns about endogeneity.
Moreover, most variation in the dependent variables and in GSP is cross-sectional rather than longitudinal
- variation that is at the heart of our analysis but which would be obscured by country-specific dummies
In model 2's Tariff regression sample, for instance, 55 of the 111 countries have 5 or fewer observations
each. Rather than subsume such cross-national variation with dummies, we control for the most important
theoretically-justified cross-sectional factors, like market size, income, and geography, adding appropriately
conservative standard errors. The first-difference technique is often suggested as an alternative to error
components in cases like this, but here it is inappropriate because GSP is a dummy which very rarely
changes (36 times out of our 3210 observations) and whose greatest effects are felt over a long stretch of
subsequent years, not just one year following removal (e.g., Wooldridge [20021). One solution, however, is
a before-and-after test, looking solely at those countries ultimately dropped; we report the results of this
robustness test below.
26Specifically, for OLS models 1, 2, 3, White's general test rejects homoscedasticity with p < 0.01 in each
case (Greene [2000]). Regression of OLS residuals -,,t on E,,t-l yield estimates of first-order serial correlation
14



tive, heteroscedastic- and autocorrelation-consistent Newey-West standard errors are indeed
appropriate.  Correlations among the independent variables are very low in most cases,
moderate in just a few,27 and never higher than 0.5; multicollinearity is not a problem. Be-
cause the conventional GLS approach is more efficient but, as noted above, becomes biased
in wide but short panels like ours (Beck and Katz [1995]), we conduct Hausman tests of
the appropriateness of the consistent but less efficient estimator, OLS. Such tests reject the
equivalence of OLS and GLS coefficients with p < 0.01 (for models 1, 2, 3, and 7), validating
the use of OLS despite its lower efficiency. Of course, inefficiency simply raises the bar for
our hypothesis tests.
Consider the OLS estimates in Table 2. Just as expected, the coefficient of GSP is
positive and significant for all three dependent variables (models 1, 2, and 3). When other
factors are held constant, a GSP recipient has average duties 1.66 points higher as percent of
trade (model 1) and average tariffs higher by 3.83 percent (model 2). Furthermore, its total
imports are 8.5 points lower as percent of GDP (model 3). These numbers imply that GSP
recipients are roughly 15-20 percent more protectionist than countries no longer eligible,
because these differences are about one-sixth to one-fifth of the means of these trade policy
measures (Table 1). Had Costa Rica hypothetically been dropped from GSP in 2000, model
3 predicts it would have imported $8.6 billion instead of $7.3 billion that year.
Of course, the comparison here is between countries removed from GSP, on one hand, and
the same countries before removal, plus all countries never dropped from GSP, on the other.
If the countries ultimately dropped are systematically different from those never dropped, in
some unobservable way, our inference about GSP's effect might be tainted. To confirm that
our results are not affected by such bias, we re-estimated models 1, 2, and 3 using only the
36 countries ultimately dropped from GSP. In every case, the coefficient for GSP was still
positive and significant. Even those countries ultimately dropped were more protectionist
when they were eligible for GSP, controlling for other factors as well. Similarly, the GSP
findings in models 1, 2, and 3 do not change if we add dummies for World Bank-defined
regions.28
We should note that, among the control variables, income level is significant in all models.
The trend in liberalization across the developing world is significant in the Dutzes and Tarqff
models while total GDP is significant in the Tariff and Closure models, all with the expected
signs. On the other hand, IMF and Landlock are not significant in any of the models.
of about p = 0.8, p < 0.01, for all four OLS models as well.
27Depending on the model and sample in question, Log Income,,ti, exhibits a correlation with IMF,,t-l
of about -0.25, with Log GDP,,t-l of about 0.35, and with Landlocked, of roughly -0.50.
2"Results available upon request.
15



5.2 Endogeneity of GSP
The second set of regressions addresses the potential for endogeneity of GSP status.29 This is
a critical issue, though indeterminate as to direction. Granting and removing GSP eligibility
is, after all, a political decision. In particular, endogeneity may arise if a recipient's import
policy affects the United States removal decision. On one hand, if US exporters (rather than
primarily US import-competing producers) influence the decision, then beneficiary countries
with high trade barriers of their own would be prime targets for removal. This suggests
GSP recipients may be more, not less, open to imports than ex-beneficiaries. If so, the
estimates from the first set of regressions may be conservative. On the other hand, the US
may be most likely to maintain GSP eligibility for precisely those countries that have failed
to overcome the legacy of import substitution, because they have prospered the least and are
in greatest apparent need of foreign aid. This implies GSP should be associated with worse
policies (e.g., high trade barriers), just as we have proposed - but for a different reason.
If true, the results from our OLS regressions may overstate GSP's effect. Either way, we
must explicitly address potential endogeneity bias in order to have greater confidence in our
findings. However, we emphasize that this paper makes no claim about the direction of the
reverse causality. Rather, the point is simply to see whether GSP still independently affects
import policies even in a model that acknowledges its endogeneity.
We use an instrumental variables (IV) approach to isolate the "treatment effect" of GSP
on trade policy when the treatment itself is endogenous. That is, imagine a prior equation,
GSPt =    'x%t-1 + Uz,t                         (3)
where GSP,,t = 1 if GSP*t > 0 but 0 otherwise, and x is a vector of instruments thought
to be uncorrelated with E in equation (2). Because of the discrete nature of GSP, the usual
2SLS instrumental variables method may overstate the coefficient estimates. Instead, we
use the well-known "treatment effects" instrumental variables (IV) approach, much like a
Heckman selection model but with observed trade policy outcomes even for those countries
not receiving the "treatment" (GSP). This model assumes E,,t in (2) and u,,t in (3) are
correlated, and we estimate it using maximum likelihood (Greene [2000, 933-934]). We thus
use values of GSPt,t (not GSP,,t-., since we are correcting for simultaneous causation here)
fitted from a first-stage probit regression as the optimal instrument for GSP,,t in equation (2).
Models 4, 5, and 6 separately apply this method to Dut%es, Tarff, and Closure, paralleling
models 1, 2, and 3, reporting traditional heteroscedastic-consistent standard errors.
The decisive issue is exogeneity of the instruments. Fortunately, as Ozden and Reinhardt
[2002] demonstrate, there are a number of good predictors of US GSP eligibility; many of
these have no conceivable direct link to the developing country's own trade policy. These
include (a) a dummy denoting the presence of other foreign policy sanctions imposed by the
US on country 230; (b) capitol-to-capitol distance from country i to the United States, in
29We thank Caroline Freund for suggesting these additional tests.
30Institute for International Economics [2000]; Office of Foreign Assets Control [2002])
16



miles; (c) a dummny indicating country i maintains a formal alliance with the United States31;
(d) the percent of all once-eligible developing countries still on the US GSP program; and (e)
the US GDP growth rate (World Bank [2001b]); (f) a sufficiently lagged value of GSP status
itself. All are measured for year t - 1, except the last, which is for year t - 3. The US is
likely to remove GSP status when it also employs national security sanctions against a target
country, a criterion even written into the GSP-authorizing statute. Yet such sanctions are
not directly related to the country's own import policy. Likewise, distance and alliance
status affect a country's trade with the United States in particular but not its import policy
overall. US macroeconomic conditions affect the pressure from domestic import-competing
industries to remove many significant exporting nations from GSP but have little or no effect
on the target country's own import barriers. We avoid concerns about endogeneity of the
lagged GSP instrument by setting the lag sufficiently deep. While all results we report below
are valid without the use of the GSPi,t.3 instrument, it increases the explanatory power of
the first-stage IV regression, and the use of stronger instruments has been shown to reduce
IV bias (e.g., Staiger and Stock [1997]). We include these instruments, in addition to the
right-hand-side variables in 2, in the x vector in equation 3. We verify that our instruments
are properly exogenous (i.e., not correlated with E in 2): tests of overidentifying restrictions
in IV models 4, 5, 6 (and 8) all fail to reject the null of exogeneity (Davidson and MacKinnon
[1993]).32
Table 3 presents the IV estimates of the Duties, Tariff and Closure equations (models 4, 5
and 6). GSP's coefficient is highly statistically significant and in the hypothesized (positive)
direction in each column.33  Indeed, the coefficients from the IV estimation are slightly
higher than the respective values in the OLS estimation. (What this implies regarding
the direction of the reverse causal effect, however, is irrelevant for our claim about the
independent second-stage impact of GSP.) Hence, the findings in models 1, 2, and 3 are
not an artifact of endogeneity, since the association between GSP status and protectionism
shows up even when controlling for endogeneity bias in the IV models.
5.3    Other Measures of GSIP
The third set of regressions examines the impact of GSP Fraction,,t instead of the GSP
dummy. This is a better measure of the GSP program as it measures GSP exports of country
i to the United States (USITC [2002]) as a percent of its total exports in year t. Because the
time series for Dutzes and Tariff are frequently interrupted and GSP Fraction itself is only
31Distance and alliance data are from the Correlates of War Project as distributed by Bennett and Stam
[2000], updated by the authors to 2000.
32These statistics yield p = 0.548, p = 0.446, p = 0.607 for models 4, 5, 6, respectively.
33Fit in the first-stage probit (of GSP on the instruments) is quite good in each case, all with large x2 values
and pseudo-R2s around 0.5 or higher. Thus, by Staiger and Stock's [1997] rule of thumb, our instruments
have more than adequate explanatory power.
17



available for the last decade (48 percent of observations), we restrict our attention to the
dependent variable with the most nonmissing cells, Closure. We substitute GSP Fraction
for GSP and use OLS as above. Because GSP Fractson is a continuous variable, our IV
method is now standard two-stage least squares instead of the treatment effects approach
used in the previous section. In all of these regressions we expect GSP Fraction's coefficient
to be positive, increasing trade barriers. All use the same set of controls as in equation (2),
again lagged by one year (except Landlocked).
Models 7 and 8 in Table 4 show OLS and IV estimates for the regression of Closure
on GSP Fraction. The results from the previous sections stand even if we adopt a more
nuanced measure of GSP benefits. The more a country's total exports benefit from (and
depend on) United States GSP duty-free status, the more closed to imports it should be.
Just as expected, the coefficient of GSP Fractson is positive and significant in both columns.
Dropping the average beneficiary from the GSP program increases its imports by 1.7 percent
of its GDP in the OLS estimation. This is a more modest and conservative impact than
the previous results (models 3 and 6), because GSP does not cover a country's entire export
portfolio and because beneficiaries do not rely wholly on the United States alone as an export
market. In other words, the impact of GSP removal would be greatest for countries like the
Dominican Republic, for whom 75 percent of its 1990 global exports went to the US, with
13 percent of these qualifying for GSP duty-free treatment.
We should note that the Irend variable (measuring global trade policy patterns) is no
longer significant in these regressions, while all the other variables maintain the same signs
as in the earlier regressions.
5.4 Non-Tariff Barriers
Two additional, but subjective, measures of trade barriers offer a useful comparison. Both
assess the overall restrictiveness of trade policies, including non-tariff as well as tariff barriers,
grouping countries into ordered categories, annually but onily for the last few years. The
first is the International Monetary Fund's (IMF) "Trade Restrictiveness Index" (TRIX,t). It
weights NTBs more than tariffs (IMF [1998]), using public and private information at the
IMF's disposal.34 With 10 representing the most protectionist policy (e.g., India) and 1 the
least (e.g., Hong Kong), the average for our dataset is 4.6. The second is the trade policy
component (O'Driscoll, Holmes, and O'Grady [20021) of the Heritage Foundation's "Index of
Economic Freedom" (HF,,t). This draws purely on published sources but factors in customs
corruption as well as formal tariffs and NTBs. It goes from the most open score of 1 (e.g.,
Hong Kong) to 5 (e.g., Bangladesh), averaging 3.8.  In the fourth set of regressions, we test
how our hypothesis fares with these subjective measures of trade barriers.
34We thank Anne McGuirk and Dustin Smith at the IMF's Trade Policy Division for providing these data.
35'These two measures are only moderately correlated (r = 0.47, n = 463). Both are likewise correlated
with Duties and Tariff only moderately (r = about 0.6), with Closure even less (r = about 0.25).
18



Models 9 and 10 in Table 5 give estimates (with heteroscedastic-consistent standard
errors) of ordered probit regressions of TRI and HF on GSP Fraction for over 100 developing
countries in several years in the late 1990s. GSP Fraction again has a statistically significant
effect, increasing protectionism (higher values of TRI and HF). Holding other variables at
their sample means, we find that moving from Brazil's 1998 export dependence on US GSP
benefits (about 4 percent) to that of Hong Kong (zero) increases the predicted Prob( TRI,,t <
5) by a quarter, going from 0.39 to 0.49. This accords with the observed TRI figures
for Hong Kong and Brazil in 1998, which are 1 and 5, respectively. The same change,
holding other variables at their sample means in model 10, yields an increase in the predicted
Prob(HF,,t < 4) by about half, going from 0.29 to 0.43 - where Hong Kong's and Brazil's
HF scores that year are 1 and 4.
(B Examples
In case after case, we can see trade policy following the path of GSP eligibility and withdrawal
we have mapped above. Consider (South) Korea. Prior to being dropped in 1989, Korea
had pursued a long, slow program of import liberalization, having cut average (unweighted)
nominal tariffs a little from the 1980s average of 21 percent. In the five year period preceding
graduation from GSP, import duties as a percent of imports slid down 1.2 points as a result.
In the four years after being dropped, however, Korea cut the average nominal tariff by
6 points (down to 8.9 in 1993), so import duties as a percent of imports decreased much
more rapidly, by 3.5 more points, reaching a low of 4.4 in 1994. Note that this occurred in
the absence of any additional multilateral tariff commitments by Korea, since the Uruguay
Round was not implemented until 1995.
Samoa offers another interesting illustration.  RFom 1993 to 1996, when Samoa was
ultimately dropped from the US GSP program, Samoa's imports as percent of GDP actually
fell 16 points, from 75 to 59. After GSP withdrawal, however, Samoa increased its imports,
up to 82 percent of GDP by 2000. Similarly, after years with a steady average nominal
tariff of 9 percent, Samoa's government announced a wholesale trade liberalization program
in early 1998, designed to cut tariffs to zero by 2010 (O'Driscoll, Holmes, and Kirkpatrick
[2001]).
Mexico, too, is revealing. The US graduated Mexico from GSP in 1994, upon Mexico's
accession to the North American FRee Trade Agreement (NAFTA). This would seem an ex-
ceptional case, especially because Mexico in the early 1990s had, according to most accounts,
already begun a significant program of unilateral trade liberalization. However, Mexico's
imports as a percent of GDP had not increased from 1990 to 1993, and in just two years
after leaving GSP it rose by over a third, from 21.7 to 30.1 percent. And this was during the
'tequila crisis,' which witnessed a severe devaluation of the peso and thus downward pres-
sure on imports. Mexico's average nominal tariff, which had in fact increased a little from
1989 to 1994, dropped 3.4 points from 1994 to 1999, and its duties as a percent of imports
19



dropped to one-third of its previous level in this period as well. Granted, the catalyst for
this major import liberalization was the desire for export revenue in the US market; GSP
removal followed rather than started the policy change. But this is an exception that proves
the rule, since Mexico had not liberalized as significantly during the GSP period. By the
mnid-1990s, Mexico had finally learned the lesson that you have to 'pay to play' - a lesson
GSP status had obscured for years.
The experience of Chile comes close to a smoking gun in support of our clairn. Chile was
suspended from the GSP at the beginning of 1988 for violating internationally recognized
worker rights as the Pinochet regime stalled the last stage of democratic transition (Adams
[1989]). Its GSP status was later restored in 1991, after democratic elections had occurred.
What happened to its trade policy in the interval? Just days after the GSP withdrawal,
Chile's Finance Minister Hernan Buchi announced that the formerly "sacred" 20 percent
nominal tariff would be cut to 15. He stated explicitly that the cut aimed to lower the
burdensome imported input costs for Chilean exporters, to compensate for the loss of GSP
duty-free status in the US market (Adams [1989]; Frasca [1988]). Chile cut its tariff another
4 points, down to 11 percent, by the year its GSP eligibility was restored, but made no
cuts for the rest of the decade (once GSP was restored).  Imports as a percent of GDP
went up from the 1986-87 average of 26.7 to the 1989-90 score of 31.3 (a level Chile has not
since reached even now), and then, upon GSP restoration, down again to 29.6 for 1992-93.
The decision to drop Chile was essentially exogenous for our purposes, yet the removal of
GSP increased and determined the timing of Chile's trade liberalization, and its restoration
stalled that liberalization subsequently.
7 Conclusion
The aim of this paper was to contribute to the ongoing debate on the best ways to integrate
the developing countries into the world trading system and to lower their trade barriers.
One of the dominant views favors the "special and differential treatment of LDCs" through
programs such as the GSP. However, this paper provides strong empirical evidence against
such programs and our results thus indicate quite clearly that (a) remaining eligible for
GSP makes a country less likely to liberalize its own trade policy and (b) the GSP "dose"
matters as well, i.e., greater export dependence on US GSP preferences boosts a country's
resistance to liberalization. These findings are extremely robust. First, the estimates of
GSP's impact on protection are similar in statistical and substantive significance regardless
of the differences among our five trade policy measures. Second, the results hold across
two different measures of the independent variable, GSP status: contrasts both (1) between
countries on and off GSP and (2) among countries receiving various degrees of GSP benefits
equally support the hypothesis. Third, the GSP treatment effect holds up if we correct
for the potential endogeneity of US GSP eligibility decisions. There is little option but to
conclude that the evidence is strongly consistent with our hypothesis.
20



These results are quite surprising. After all, the US is not the largest trading partner
for many of these developing countries, and most exports to the US by GSP beneficiaries do
not qualify for GSP duty-free status. Other factors, such as IMF conditionality, the overall
secular trend toward liberalization in the developing world, macroeconomic performance,
or exogenous characteristics like geography and market size, are all given greater weight in
conventional explanations of developing countries' trade policies. Our results indicate there
is merit in some of these explanations, but the GSP program also has significant influence.
What is particularly striking is that our results obtain despite the presence of several pariah
states on the list of countries removed from the US GSP program - countries like Ethiopia in
the 1980s, as well as Myanmar and Mauritania in the 1990s, whose import regimes verged on
autarky. These countries drag our estimates down, and the use of an inefficient estimation
method leaves our estimates relatively imprecise. Given these considerations, we have set
up a hard test for the model. Yet, as we have found, GSP actually matters in trade policy
decisions of beneficiary states.
Is endogeneity bias responsible for the apparent association between GSP and protection?
We address this concern directly, and the answer is unambiguously no. First, recall that we
lag the GSP variable by a year in the OLS models, and, because the US GSP review process
actually takes place in the 12 months prior to the announcement, there is an effective lag
of nearly two years. Second, as Ozden and Reinhardt [2002] demonstrate using a discrete
duration model of time till GSP withdrawal for all US GSP beneficiaries from 1976-2001,
neither the state's overall trade openness, its openness to US exports in particular (as a
percent of GDP), nor its duties as a percent of trade have any significant effect on GSP
withdrawal decisions, when controlling for other important factors. (Rather, what matters is
the volume of its exports to the US and its per capita income, plus US-specific macroeconomic
conditions.) After all, the GSP as approved within the General Agreement on Tariffs and
Trade legal system is not supposed to be reciprocal in nature. To be sure, the US links other
issues to GSP status reviews (e.g., worker rights), but GSP is not conditioned on reciprocal
market access in practice. As one observer has noted, GSP status decisions are "controlled
essentially by importers" (Wilson [1992]). Third, in any case, our hypothesis passes the
truest test of this objection, in IV models controlling for endogeneity bias.
Several factors might account for the negative effect of the US GSP program on the
trade policies of developing countries. One of them might be the lack of lobbying by export
groups for liberalization, since they face low tariffs on their exports in any case (under GSP);
hence policy is determined by protectionist import-competing groups. Another might be the
way the GSP eligibility is determined. The political process leading to GSP decisions is
heavily influenced by protectionist domestic interest groups in the US. Since 'he who giveth
may taketh away,' the non-guaranteed nature of GSP preferences prevents the recipients
from fully focusing on their export sectors. Developing countries in GATT/WTO have often
levied the critique that sector- and country-specific 'graduation' criteria are "contrary to
principles of non-discrimination and non-reciprocity that underpin the GSP and therefore
21



alien to the original intentions underlying the GSP concept" (WTO [2000])." Thus the
negative effects of GSP should not be blamed just on developing countries, but also on the
way in which GSP is administered in US, highly exposed as it is to unregulated traditional
protectionist pressures. One of the main implications of our results is that participating in
reciprocal trade agreements through formal institutional frameworks such as the GATT is
more likely to lead to liberalization by developing countries.
As part of the buildup to what many have called the "Development Round," i.e., WTO's
next multilateral trade negotiation, many developing countries have continued to lobby fellow
WTO members for continued and even increased nonreciprocal trade preferences (WTO
[2001a]), in some cases explicitly pleading cessation of pressure for any further liberalization
of their own (e.g., WTO [1999]). Based on the evidence in this paper, we argue that the
preferred scenario is one in which developing countries give up GSP in favor of the reciprocity-
driven trade regime embodied in GATT/WTO relationships among developed states. The
normative criteria here is the extent of trade liberalization achieved by developing countries,
of course. A world in which GSP is administered outside of the GATT/WTO legal structure,
exposing it to protectionist political economy dynamics within developed countries, is the
least ideal.
Hopefully, our results lead to more questions than they answered. Some of these questions
are about the actual political economy mechanisms that cause the effects we identified.
Second, the links between GSP status and export performance need to be analyzed. Our
model has implications for how GSP affects the export level of a country, but at the same
time GSP eligibility has been shown to be negatively affected by export volume (Ozden and
Reinhardt [2002]). This endogeneity issue is crucial for correct identification of the true
linkages between exports (if not import openness, the issue here) and GSP status. We hope
to address these questions in subsequent papers.
References
[1] Adams, Paul H. [19891, "Suspension of Generalized System of Preferences from Chile
- The Proper Use of a Trade Provision?", George Washzngton Journal of International
Law and Economzcs 23: 501-530.
[2] Bagwell, K. and R. Staiger [1999], "An Economic Theory of GATT," Amerncan Eco-
nomic Revzew, vol. 89, p. 215-248.
[3] Baier, Scott and Jeffrey Bergstrand [2001], "The Growth of World Trade: Tariffs, Trans-
port Costs, and Income Similarity," Journal of International Economics, 53 (February):
1-29
36This critique is ironic in the sense that GSP is itself discriminatory.
22



[4] Beck, Nathaniel, and Katz, Jonathan N. [1995], "What to Do (and What Not to Do)
With Time-Series Cross-Section Data," Amerncan Politscal Sczence Revzew 89 (Septem-
ber): 634-647.
[5] Brown, Drusilla K. [1987], "General Equilibrium Effects of the US Generalized System
of Preferences," Southern Economic Journal 54: 27-47.
[6] Brown, Drusilla K. [1989], "Trade and Welfare Effects of the European Schemes of the
Generalized System of Preferences," Economic Development and Cultural Change 37
(July): 757-776.
[7] Clark, Don P., and Zarrilli, Simonetta [1992], "Non-Tariff Measures and Industrial Na-
tion Imports of GSP-Covered Products," Southern Economic Journal 59 (October):
284-293.
[8] Coates, Daniel E., and Ludema, Rodney D. [2001], "A Theory of Trade Policy Leader-
ship," Journal of Development Economics 65: 1-29.
[9] Davidson, R. and J. G. MacKinnon [1993], Estimation and Inference rn Econometrzcs,
Oxford University Press.
[10] Devault, James M. [1996], "Political Pressure and the US Generalized System of Pref-
erences," Eastern Economwc Journal 22 (Winter): 35-46.
[11] Dollar, David, and Aart Kraay [2001], "Trade, Growth, and Poverty," Working Paper
2615, World Bank.
[12] Dornbusch, Rudiger, and FRankel, Jeffrey A. [1987], "Macroeconomics and Protection,"
in Robert M. Stem (ed.) US Trade Polictes mn a Changing World Economy, MIT Press,
77-130.
[13] Federal  Register.  Various.  Weekly   Compilatzon  of   Preszdential  Docu-
ments.  Washington,   DC:  National  Archives  and   Records  Administration.
http://www.access.gpo.gov/nara/nara003.html??
[14] Finger, J. Michael, and Winters, L. Alan [1998], "What Can the WTO Do for Developing
Countries?", in Anne 0. Krueger (ed.) The WTO as an Internattonal Organzzatzon,
University of Chicago Press, 365-392.
[15] Frasca, Tim [1988], "Chile Cuts Tariffs and Raises Tempers, Trying to Offset US Sanc-
tions," Christian Science Monitor (January 26): 10.
[16] Gallup, J. L. and J. D. Sachs [1999], "Geography and Economic Development," Center
for International Development Working Paper 1, Harvard University.
[17] General Accounting Office [2001], Internatzonal Trade: Comparison of US and European
Union Preference Programs, GAO-01-647, GAO.
23



[18] General Agreement on Tariffs and Trade [1972], Basic Instruments and Selected Docu-
ments, 18th supplement, Geneva.
[19] General Agreement on Tariffs and Trade [1988], "Meeting on 8 March 1988," MTN.SB/5
(March 29), Geneva.
[20] Goldstein, M. [2000], "IMF Structural Programs," NBER Conference on Economic and
Financial Crises in Emerging Market Economies, October 19-21.
(21] Greene, William H. [2000], Econometrnc Analysis, 4th ed., Prentice-Hall.
[22] Grossman, Gene M. [1982], "Import Competition from Developed and Developing Coun-
tries," Review of Economwcs and Statzstzcs 64: 271-281.
[23] Grossman, G. M. and E. Helpman [1994], "Protection for Sale," Amertcan Economzc
Revzew, vol. 84, p.833-850.
[24] Hoekman, Bernard M., and Kostecki, Michel M. [2001], The Polztwcal Economy of the
World Tradzng System, 2nd ed., Oxford University Press.
[25] Hudec, Robert E. [1987], Developing Countries zn the GATT Legal System. Gower.
[26] Institute for International Economics. [2000], "Chronological Summary of Economic
Sanctions for Foreign Policy Goals, 1914-99." Washington, DC: Institute for Interna-
tional Economics. http://www.iie.com/topics/sanctions/sanctions-overview.htm
[27] International Monetary Fund [1998], Trade Liberaltzatton in IMF-Supported Programs.
IMF.
[28] Jackson, J. H. [1997], The World Trading System: Law and Polzcy of Internatzonal
Economzc Relat%ons, 2nd ed. MIT Press.
[29] Johnson, Harry G. [1967], Economzc Policzes Toward Less Developed Countries. Wash-
ington, DC: Brookings.
[30] Krishna, Pravin and Devashish Mitra [2000], "Reciprocated Unilateralism: A Political
Economy Approach," mzmeo
[31] Krueger, A. 0. [1995], Trade Polzczes and Developing Nations, Brookings.
[32] MacPhee, Craig R., and Oguledo, Victor Iwuagwu [1991], "The Trade Effects of the US
Generalized System of Preferences," Atlantic Economic Journal 19 (December): 19-26.
[33] MacPhee, Craig R., and Rosenbaum, David I. [1989], "The Asymmetric Effects of Re-
versible Tariff Changes under the United States GSP," Southern Economzc Journal 56
(July): 105-125.
24



[34] McLaren, John [1997], "Size, Sunk Costs and Judge Bowker's Objection to Free Trade,"
Amerncan Economic Remew, 87, (june): 400-20
[35] Michalopoulos, Constantine, [1999], "Trade Policy and Market Access Issues for Devel-
oping Countries: Implications for the Millenium Round," Working Paper 2214, World
Bank.
[36] Nowzad, Bahram [1978], "Differential Trade Treatment for LDCs," Ftnance & Devel-
opment 15 (March): 20-23.
[37] O'Driscoll, Gerald P., Jr., Holmes, Kim R., and Kirkpatrick, Melanie [2001], 2001 Index
of Economic Freedom. Heritage Foundation.
[38] O'Driscoll, Gerald P., Jr., Holmes, Kim R., and O'Grady, Mary Anastasia [2002], 2002
Index of Economic Freedom, Heritage Foundation.
[39] Office  of  Foreign  Assets  Control  [2002],  "Sanctions  Program   Sum-
maries."  Washington,  DC:   United  States  Department  of the   Treasury.
http://www.treas.gov/offices/enforcement/ofac/sanctions/index.html
[40] Oyejide, T. Ademola [2000], "Interests and Options of Developing and Least-Developed
Countries in a New Round of Multilateral Trade Negotiations," G-24 Discussion Paper
Series No. 2, UNCTAD, Geneva.
[41] Ozden, C. and E. Reinhardt [2002], "The Political Economy of US Trade Preferences
for Developing Countries, 1976-2001," manuscript, Emory University.
[42] Panagariya, Arvind [2002a], "Developing Countries at Doha: A Political Economy
Analysis," forthcoming World Economy
[43] Panagariya, Arvind [2002b], "EU Preferential TRade Policies and Developing Countries,"
mimeo.
[44] Pomfret, R. [1986], "MFN Tariff Reductions and Developing Country Trade Benefits
Under the GSP: A Comment," Economic Journal, vol. 96, p. 534-536.
[45] Ray, Edward John [1987], "The Impact of Special Interests on Preferential Tariff Con-
cessions by the United States," Review of Economics and Statzstics 69 (May): 187-193.
[46] Rodrik, D. [1994], "The Rush to Free Trade in the Developing World: Why So Late?
Why Now? Will It Last?", in S. Haggard and S. B. Webb (eds.), Voting for Reform:
Democracy, Political Ltberalzzatton, and Economwc Adjustment, Oxford.
[47] Rodrik, D. [1995], "Political Economy of Trade Policy," in G. M. Grossman and K.
Rogoff (eds.), Handbook of International Economwcs, vol. III, Elsevier, 1457-1494.
25



[48] Rodriguez, Francisco, and Rodrik, Dani [2001], "Trade Policy and Economic Growth:
A Skeptic's Guide to the Cross-National Evidence," in Ben S. Bernanke and Kenneth
Rogoff (eds.) NBER Macroeconomtcs Annual 2000, MIT Press, 261-325.
[49] Roessler, F. [1998], "Domestic Policy Objectives and the Multilateral Trade Order:
Lessons from the Past," in A. 0. Krueger (ed.), The WTO as an International Organz-
zation, University of Chicago Press, p. 213-230.
[50] Sapir, A. and L. Lundberg [1984], "The U.S. Generalized System of Preferences and Its
Impacts," in A.O.Krueger and R.E.Baldwin (eds.) The Structure and Evolution of US
Trade Policy, NBER.
[51] Staiger, Douglas, and Stock, James H. [1997] "Instrumental Variables Regression with
Weak Instruments," Econometrica 65 (May): 557-586.
[52] United Nations [2001], "Third United Nations Conference on the Least Developed Coun-
tries, Brussels, Belgium, 14-20 May 2001 - Declaration," A/CONF.191/L.20. New
York: UN.
[53] United Nations Conference on Trade and Development [1999], "Quantifying the Ben-
efits Obtained by Developing Countries from the Generalized System of Preferences,"
UNCTAD/ITCD/TSB/Misc.52 (October), Geneva.
[54] United States International Trade Commission [2002], ITC Trade Dataweb. Washington,
DC: USITC. http://www.dataweb.usitc.gov??
[55] United States Trade Representative. 1999. US Generalized System of Preferences Guzde-
book. Washington, DC: USTR.
[56] Vreeland, James Raymond [2002], "The Effect of IMF Programs on Labor," World
Development, vol. 30, p. 121-139.
[57] Whalley, John [1990], "Non-Discriminatory Discrimination: Special and Differential
Treatment under the GATT for Developing Countries," Economic Journal, vol. 100,
p. 1318-1328.
[58] Wooldridge, Jeffrey M. [2002], Econometrnc Analyszs of Cross Sectzon and Panel Data.
MIT Press.
[59] World Bank [1987], World Development Report 1987. New York: Oxford University
Press.
[60] World  Bank  [2001a],   "Trends  in  Average   Tariff  Rates  for  Develop-
ing  and  Industrial  Countries  1980-99."  Washington,  DC:  World   Bank.
http://wwwl.worldbank.org/wbiep/trade/Data/tradedata.htm??
26



[61] World Bank [2001b], World Development Indicators 2001 CD-ROM. Washington, DC:
World Bank.
[62] World Trade Organization [1999], "Proposals for Addressing Concerns on Marginaliza-
tion of Certain Small Economies," WT/GC//W/361, Geneva, October 12.
[63] World Trade Organization [2000], "Implementation of Special and Differential Treat-
ment Provisions in WVTO Agreements and Decisions: Note by the Secretariat,"
WT/COMTD/W/77 (October 25), Geneva.
27



8 Tables and Figures
Table 1. Descriptive Statistics
Dependent:          Duties,,t     Tariff,t      Closuret,t      Closure,,,
Models 1 & 4  Models 2 & 5   Models 3 & 6    Models 7 & 8
8.52-6.22    20.18-15.13   -44.59-28.95    -46.28-26.81
Dependent           [0, 55.82]   [0,102.21     [-446.3, -1.2]  [-174.9, -1.2]
1987.3-6.5    1991.9-5.1     1989.0-6.9      1995.9-2.5
Yeart             [1976,2000]   [1981,2000]     [1976,2000]    [1992,2000]
0.95-0.21     0.91-0.28      0.94-0.24
[0,1]         [0,11          [0, 1]
GSP Fraction,,t-l     -             -               -           0.90-1.52
[0,14.79]
9.83-3.03     21.45-5.91    -45.42-3.23    -46.69-0.84
Trendt..1         [3.92,16.87]  [11.37,30.69]  [-53.4, -38.4]  [-47.7, -45.0]
0.41-0.49     0.51-0.50      0.41-0.49       0.48-0.50
IMIFt,t1             [0,1]         [0,1]          [0, 1]          [0,1]
22.79-1.94    23.48-1 84     22.31-2.03      22.59-2.07
Log GDP1,t_1       [17.9,27.3]   [18.6,27.3]    [17.1,27.3]    [17.4,27 3]
7.23-1.18     7.37-1.25      7.08-1.21       7.16-1.28
Log Jncomet,t.i   [4.51,10.17]  [4.59,10.15]    [4.44,10.17]   [4.44,10.17]
0.40-0.37     0.41-0.36      0.41-0.38       0 41-0.38
Land1ockedt          [0,1]         [0, 1]         [0, 1]          [0,1]
3.83-5.68     3.59-4.64      3.43-6.45       3.59-6.35
Growth1, ..i      [-26.5,27.2]  [-22.9,19.0]   [-50.2,86.9]    [-50.2,86.9]
Countries             112           111            140             135
Observations         1503           821            2640            993
Mean-SD, [Min, Max]
28



Table 2. OLS Estimates of Trade Policy Models
Dependent Variable:  Duties1,t  Tariff1,t  Closure1,t
Model:                  1          2           3
Method:               OLS        OLS         OLS
Years:              1976-2000  1981-2000   1976-2000
Constant       ~~27.118**  -16.417°   -127.215**
Constant             (3.036)    (7.256)     (11.876)
GSPI't-,             1.663**    3.827**      8.458*
(0.481)    (1.189)      (3.825)
0.2768*    0.933t*      0.356
Trendt_..i          (0.059)    (0.082)     (0.203)
-0.521     -0.716       0.537
(0.367)    (1.038)      (1.028)
-0.148     2.806**     7.424**
Log GDP1,~.._1       (0.136)    (0.401)      (0.320)
-2.6798*   -7.014**    -10.564**
Log Income1,,...i    (0.226)    (0.813)      (0.746)
-0.342     -3.192       2.155
Landlockedj          (0.546)    (1.959)      (1.660)
0.052      0.161       -0.328*
Growth1,t..1        (0.028)    (0.084)      (0.129)
Countries              112        111         140
Observations          1503        821         2640
FReq(GSPi,t-1=0)       68         71          157
R                     0.312      0.414       0.359
F                    75.03**    39.38**     149.96**
* denotes two-tailed p < 0.05; t, p < 0.01. Newey-West robust SEs in parentheses.
29



Table 3. IV Estimates of Trade Policy Models
Dependent Variable:  Duties1,t   Tariff,,t   Closure1,t
Model:                  4           5            6
Method:                IV          IV           IV
Years:              1976-2000   1981-2000    1976-2000
26.166**   -23.971**    -132.908**
Constant             (2.438)     (6.498)      (10.812)
GSPIt                2.298**     8.448**      11.846**
(0.766)     (1.482)     (3.961)
0.267**     0.885**      0.350*
Trendt..i            (0.047)     (0.068)      (0.174)
-0.581      -0.948        0.328
(0.302)     (0.894)     (0.802)
-0.130      2.893**      7.483**
Log GDP1,t-.1        (0.102)     (0.340)      (0.250)
-2.668**    -6.683**    -10.416**
Log Incomeq-,t-      (0.169)     (0.669)      (0.572)
-0.346      -3.134        2.225
Landlocked1           (0.418)    (1.650)      (1.220)
0.051*      0.181*      -0.328**
Growth1,t-1          (0.025)     (0.078)      (0.119)
Countries              112         111          140
Observations           1503        821         2640
Freq(GSP,,t=0)          68          71          157
Model x2             844.12**    365.74**    1791.68**
* denotes two-tailed p < 0.05; **, p < 0.01. White robust SEs in parentheses.
30



Table 4. Effect of GSP Benefits on Import Penetration
Dependent Variable:           Closure1,t
Model:                   7              8
Method:                 OLS            IV
Years:               1992-2000      1992-2000
-132.188**     -145.326**
Constant              (42.106)       (43.130)
GSP Fractioni,t-i      (0.68786)
GSP Fraction,,t         -            (3.19293)
-0.390         -0.726
TEendt1l              (0.882)         (0.922)
-2.812         -2.656
(1.701)        (1.705)
Log GDP,,t-1          6.334**         6.081**
(0.556)        (0.577)
-10.453**      -10.184**
Log Income1,t...1    (1.068)         (1.047)
Landlocked,            3.808          4.366
(2.603)        (2.658)
Growthi,t     ~       -0.620¢*       -0.649**
(0.191)        (0.204)
Countries               135            135
Observations            993            993
Freq(GSP=0)          113 (t - 1)      120 (t)
-2                     0.353          0.345
F                      57.17**        56.064*
* denotes two-tailed p < 0.05; -, p < 0.01. Newey-West robust SEs in parentheses.
31



Table 5. Effect of GSP Benefits on Trade Barriers
Dependent Variable:     TRLi,t         HFI,t
Model:                    9             10
Method:             Ordered Probit Ordered Probit
Years:                1998-2000      1996-2000
GSP Fraction1,t-1       0.066*         0.102*
(0.033)        (0.040)
TRendt-I                0.445          0.550
(0.293)        (0.441)
-0.935**       -0.300**
(0.135)        (0.102)
Log GDP1, t-            0.064*        -0.059
(0.030)        (0.033)
-0.319**       -0.448**
Log Income2,t...1     (0.063)        (0.058)
Lanidlocked, ~          0.111          0.004
(0.167)        (0.147)
Growth,,t-I            -0.004          0.015
(0.009)        (0.010)
Countries                138            120
Observations             386            546
Freq(GSP,,t-1=0)         58             72
LL                     -767.24        -716.29
Model x2               62.72**        159.04**
Pseudo-R2               0.049          0.095
* denotes two-tailed p < 0.05; **, p < 0.01. Heteroscedastic-consistent robust SEs in
parentheses. Ordered probit intercepts omitted for brevity.
32



Duties as % Trade                  [l Average Nominal Tariff, %
Vf1 Imports as % GDP
50 -
D-Cd!- .
40- ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ t  2 - -   _ _ _ _ _ _ _ __,_ _ _
40 -1J3   s t _ 11 _ '_s3 >* J _
30 -
~D 21 "> 
-41~~~~~~~~~~~~~~
20 -                       *;   f3_3vi'<_       ~                  s  5  ]
Figure~~~ 1: Trd  olic by GS  Satu, ConryYa Avergs 1976200
tJ3~~~~ 1       3",,         s    __               (3< _ N
Dropped from GSP                       On GSP
Figure 1: Trade Policy by GSP Status, Country-Year Averages, 1976-2000
The figure includes 141 (29) countries with at least one year on GSP (dropped from
GSP), for a total of 2466 GSP country-years and 176 ex-GSP country-years. T-tests (with
unequal variances) of equal means across GSP and ex-GSP country-years for Duties, Taiff,
and Closure, with two-tailed p = 0.00, p = 0.00, and p = 0.01, respectively, reject the null
hypothesis in each case.
33









Policy Research Working Paper Series
Contact
Title                           Author                  Date             for paper
WPS2938 Recurrent Expenditure Requirements Ron Hood              December 2002     M Galatis
of Capital Projects Estimation for  David Husband                         31177
Budget Purposes                 Fei Yu
WPS2939 School Attendance and Child Labor  Gladys L6pez-Acevedo  December 2002     M Geller
in Ecuador                                                                85155
WPS2940 The Potential Demand for an HIV/  Hillegonda Maria Dutilh  December 2002   H Sladovich
AIDS Vaccine in Brazil           Novaes                                   37698
Expedito J A. Luna
Mois6s Goldbaum
Samuel Kilsztajn
Anaclaudia Rossbach
Jose de la Rocha Carvalheiro
WPS2941 Income Convergence during the    Branko Milanovic        January 2003      P Sader
Disintegration of the World                                               33902
Economy, 1919-39
WPS2942 Why is Son Preference so Persistent Monica Das Gupta    January 2003       M Das Gupta
in East and South Asia9 A Cross-  Jiang Zhenghua                          31983
Country Study of China, India, and the Li Bohua
Republic of Korea               Xie Zhenming
Woojin Chung
Bae Hwa-Ok
WPS2943 Capital Flows, Country Risk,     Norbert Fiess           January 2003      R lzquierdo
and Contagion                                                             84161
WPS2944 Regulation, Productivity, and    Giuseppe Nicoletti     January 2003       Social Protection
Growth OECD Evidence            Stefano Scarpetta                         Advisory Service
85267
WPS2945 Micro-Finance and Poverty Evidence Shahidur R Khandker   January 2003      D Afzal
Using Panel Data from Bangladesh                                          36335
WPS2946 Rapid Labor Reallocation with a  Jan Rutkowski           January 2003      J Rutkowski
Stagnant Unemployment Pool. The                                           84569
Puzzle of the Labor Market in Lithuania
WPS2947 Tax Systems in Transition        Pradeep Mitra          January 2003       S Tassew
Nicholas Stern                            88212
WPS2948 The Impact of Contractual Savings  Gregorio Impavido     January 2003      P Braxton
Institutions on Securities Markets  Alberto R Musalem                     32720
Thierry Tressel
WPS2949 Intersectoral Migration in Southeast  Rita Butzer        January 2003      P Kokila
Asia Evidence from Indonesia,   Yair Mundlak                              33716
Thailand, and the Philippines   Donald F Larson
WPS2950 Is the Emerging Nonfarm Market   Dominique van de Walle  January 2003      H Sladovich
Economy the Route Out of Poverty  Dorothyjean Cratty                      37698
in Vietnam9
WPS2951 Land Allocation in Vietnam's     Martin Ravallion        January 2003      H Sladovich
Agrarian Transition             Dominique van de Walle                    37698



Policy Research Working Paper Series
Contact
Title                           Author                  Date             for paper
WPS2952 The Effects of a Fee-Waiver Program  Nazmul Chaudhury   January 2003       N. Chaudhury
on Health Care Utilization among the  Jeffrey Hammer                      84230
Poor: Evidence from Armenia     Edmundo Murrugarra
WPS2953 Health Facility Surveys. An      Magnus Lindel6w         January 2003      H Sladovich
Introduction                    Adam Wagstaff                             37698
WPS2954 Never Too Late to Get Together   Bartlomiej Kaminski    January 2003       P. Flewitt
Again Turning the Czech and Slovak Beata Smarzynska                       32724
Customs Union into a Stepping Stone
to EU Integration