Economic Policy July 2005 Printed in Great Britain
© CEPR, CES, MSH, 2005.
Blackwell Publishing, Ltd.Oxford, UK
ECOPEconomic Policy0266-4658© CEPR, CES, MSH, 2005.43 Original ArticleRULES OF ORIGIN PATRICIA AUGIER, MICHAEL GASIOREK and CHARLES LAI-TONGRules of origin
A great deal of post-war trade liberalization resulted from regional, preferential
trade agreements. Preferential trade agreements cut tariffs on goods originating only
in those nations that have signed the agreement. Therefore, they need ‘rules of origin’
to determine which goods benefit from the tariff cut. Rules of origin have long been
ignored for two good reasons: they are dauntingly complex and at first sight appear
mind-numbingly dull. The third standard reason for ignoring them – the assertion
that they do not matter much – turns out to be wrong. We show that rules of origin
are important barriers to trade. Moreover, such rules are emerging as an important
trade issue for three additional reasons. First, preferential trade deals are prolifer-
ating worldwide. Second, the global fragmentation of production implies complex
international supply chains which are particularly constrained and distorted by rules
of origin. Third, the extent to which regionalism challenges the WTO-based trading
system depends in part on incompatibilities and rigidities built into rules of origin.
Our empirical results exploit a ‘natural experiment’ that was created by technical
changes to Europe’s lattice of rules of origin (ROOs) in 1997. This change, known as
‘diagonal cumulation’, relaxed the restrictiveness of rules of origin on trade among the
EU’s free trade agreement (FTA) partners without changing the degree of tariff preference.
Our analysis allows us to establish a lower-bound and upper-bound estimate of trade
impact of ROOs reduced trade among the EU’s trade partners. The lower bound we
find is something like 10% while the upper bound is around 70%. The second part
of the paper draws the policy lessons that arise from considering the implications of
our empirical findings. The most direct lessons are for FTA-writers. We argue that Europe’s
implementation of ‘cumulation’ is a good way of reducing the welfare-reducing impact
of overlapping rules of origin without gutting their fraud-fighting ability. We also suggest
a three-part procedure for establishing a more multilateral framework for rules of origin
which would be more transparent, flexible, administratively feasible and negotiable.
— Patricia Augier, Michael Gasiorek and Charles Lai-Tong
Rules of origin
RULES OF ORIGIN 569
Economic Policy July 2005 pp. 567–624 Printed in Great Britain
© CEPR, CES, MSH, 2005.
The impact of rules of origin
on trade flows
Patricia Augier, Michael Gasiorek and Charles Lai Tong
CEFI FRE 2778, CNRS – Université de la Méditerranée; Sussex University and
GREQAM; CEFI FRE 2778, CNRS
Trade liberalization has been an important driver of growth and economic integration
worldwide. Much of this liberalization has taken place in the multilateral WTO/GATT
context, but a large fraction of the tariff-cutting resulted from regional, preferential
trade agreements. Preferential trade agreements, however, only cut tariffs on goods
originating in nations that have signed the agreement. To establish which goods get the
tariff preference (thus preventing tariff fraud) these agreements need ‘rules of origin’.
Rules of origin are usually ignored for two good reasons: they are dauntingly
complex and at first sight appear mind-numbingly dull. The third standard reason
for ignoring them – the assertion that they do not matter much – turns out to be
wrong. Evidence presented in this paper together with other recent work demonstrates
The authors would like to thank the referees of
for comments from David Evans, Peter Holmes, Sherman Robinson and Alan Winters, as well as discussants and participants
at a joint CGD-GDN workshop, and a joint IADB-DELTA-INRA-CEPR conference on earlier drafts of this work. We are also
very grateful to Barry Reilly for his advice, and finally, we would like to thank the
extremely helpful guidance on this paper and for his detailed suggestions and comments. Any errors or omissions of course
remain our responsibility.
for their helpful and constructive comments. We are also grateful
Managing Editor for his
570PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
that rules of origin are important barriers to trade. Trade barriers are always a matter
of concern for economic policy makers, but rules of origin in particular are a trade
issue of growing importance for three very clear reasons.
The first and most obvious reason is the proliferation of preferential trade
deals worldwide – more than 150 such agreements have been officially launched
since the GATT was formed five decades ago; 100 of those were launched in the
last 10 years. Well-functioning preferential trade arrangements in Europe and
North America now cover about 40% of world trade, and East Asia appears
set to implement its own lattice of regional deals, so this figure could soon rise
The second more subtle reason involves the global fragmentation of production.
Staying competitive in today’s world frequently requires firms to set up a complex
international supply chain. But since such a large share of world exports are to
nations with preferential tariffs, much of this global production optimization is
constrained and distorted by rules of origin.
The third reason concerns the challenges that regionalism poses for the world
WTO-based trading system. The growth of overlapping and intersecting preference
trade deals, each with its own and differing rules of origin, has highlighted the
possible distortions created by those rules and of the incompatibilities among them.
Jagdish Bhagwati has referred to this as the ‘spaghetti bowl’ problem, with ‘preferences
like noodles criss-crossing all over the place’ (Bhagwati and Mayer 2003, p. 5). The
greater the incompatibilities and rigidities built into the rules of origin the more those
criss-crossing noodles serve to fragment and distort world trade, and the more likely
it is that preferential trading arrangements may make multilateral liberalization more
difficult to achieve.
1.1. If you can’t measure it, it doesn’t exist
While the potential for rules of origin to distort trade is simple to identify in theory
and frequently bemoaned by exporting firms, empirical evidence is scant. It is easy
to understand why. On the one hand, the rules’ technical opaqueness makes it
difficult to quantify the severity of particular rules of origin. For example, where the
main text of a typical Association Agreement between the EU and a Barcelona
process country is between 20–30 pages long, the annex covering the rules of origin
for thousands of individually mentioned products is close to 100 pages long; NAFTA’s
rules-of-origin annex is close to 200 pages. Moreover, one often requires detailed
information about the industry to understand the true impact of a particular rule.
On the other hand, even if one could accurately measure the severity of the rules,
intrinsic empirical difficulties make it hard to isolate the impact of rules of origin
(ROOs); ROOs are formulated and come into force concurrently with the preferen-
tial trading agreements, so it is exceedingly difficult to separate the impact of ROOs
separately from the impact of the trade deal itself.
RULES OF ORIGIN 571
1.2. What the paper does
In the first part of this paper, we provide formal evidence of the impact of rules of
origin by exploiting a ‘natural experiment’ that arose due to a technical change in
Europe’s rules of origin – a change known as ‘diagonal cumulation’.
this change are complex, and the complexity is important for our empirics, but as an
introduction it is useful to think of cumulation as reducing the restrictiveness of the rules
of origin without changing preferences. This allows us to avoid both of the standard
empirical problems. Since the 1997 introduction of cumulation affects the restrictive-
ness of rules of origin without affecting tariffs, and it only applies to certain bilateral
trade flows, we can gauge ROOs’ effect by comparing the change in affected trade
flows with the change in unaffected trade flow, controlling, of course, for other factors.
Our empirical findings suggest that rules of origin do restrict trade in an important
way. Our indirect methodology establishes a lower bound estimate of 8–22% and an
upper bound estimate of 25–70%. These findings can be viewed as a contribution to
two distinct literatures. First, it provides a new line of evidence in the growing body
of work that demonstrates that rules of origin do indeed affect trade flows. Second,
in the context of the EU’s trading relations, our results strongly suggest that rules of
origin serve to restrict trade between the EU’s partners and those partners’ trade with
The second part of the paper draws the policy lessons that arise from considering
the implications of our empirical findings. There are two sets of lessons.
Rules of origins are a necessary evil – every preferential trade deal must have them
to prevent tariff fraud; the first policy issue concerns the design of systems that
minimize the evil. We argue that Europe’s implementation of ‘cumulation’ suggests
a good way of reducing the welfare-reducing impact of overlapping rules of origin
without gutting their fraud-fighting ability. The second set of lessons concerns more
global issues. Taken in their entirety, preferential trading agreements may help or
hinder multilateral trade liberalization – there are serious proponents of both views.
But in either case, the design of rules of origin can influence the extent to which the
proliferation of regional trade agreements helps or hinders multilateral liberalization.
We discuss a number of principles that would reduce the chances that rules of origin
will prove to be stumbling blocks to world trade integration.
The details of
1.3. Plan of the paper
The structure of our paper is as follows. Section 2 provides a primer on rules of origin
– what they are and the likely impact from a theoretical perspective. Section 3
introduces the necessary details concerning the ‘natural experiment’, namely the
Technically there are different types of cumulation which can be distinguished. These are bilateral, diagonal and full. The
differences between these are explained in more detail in Appendix 3. Much of this paper is concerned with diagonal cumula-
tion, and unless otherwise indicated, use of the term ‘cumulation’ in this paper is taken to mean diagonal cumulation.
572 PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
Pan-European Cumulation System. Section 4 of the paper covers our formal empir-
ical work and Section 5 discusses the policy implications. The final section presents
Before turning to the body of the paper, we put our contribution in the context of
the existing literature.
1.4. Existing empirical evidence
There is only limited work on the empirical impact of rules of origin. Until very
recently, most studies either gave anecdotal evidence or cited Herin (1986) who
showed that a great deal of trade between the European Free Trade Association
(EFTA) and the European Union (EU) paid the non-preferential tariff despite the
EFTA-EC free trade agreements which would have, in principle, allowed them to
claim duty-free status. The assertion was that this indicated that the administrative
costs of the ROOs were greater than the tariff, so firms opted to pay the tariff.
Specifically, he found that non-preferential tariffs were paid on 21.5% of EFTA’s
imports from the EC, and 27.6% of EC imports from EFTA.
More recent work in the same vein by Brenton and Machin (2003) shows that the
Baltic States’ exports to the EU paid the non-preferential tariff on a substantial
proportion of supposedly tariff-free trade. They argue that a significant part of the
explanation for this derives from the restrictive rules of origin applied by the EU.
Inama (2004) provides similar evidence on utilization by poor nations that have, in
principle, the right to export duty-free to rich nations under the so-called Generalized
System of Preference (GSP) scheme. Using the examples of Canada, the EU, Japan
and the USA he calculates the rate of GSP utilization as falling from 55.1% in 1995
to 38.9% in 2001. This low level of utilization suggests that even where there are GSP
preferences, developing countries appear to have difficulties in actually realizing
tariff-free access to developed country markets, and that a key explanatory factor lies
with ROOs. Flatters and Kirk (2004) focus on South Africa, and argue that the South
African Development Community (SADC) rules of origin were heavily influenced by
highly protectionist domestic industries. They then provide a number of detailed
examples which illustrate the point.
More recently, studies by Estevadeordal (2000), and Estevadeordal & Suominen (2004),
. (2002), Augier
Carrère and De Melo (2004) have attempted to estimate the impact of ROOs on
In the context of the NAFTA agreement, Estevadeordal (2000) shows that ROOs
tend to be more restrictive the greater the difference between US and Mexican tariffs;
and that there is a strong correlation between restrictive ROOs and those sectors
with long phase-out periods for tariff liberalization. The conclusion therefore is that
restrictive ROOs tend to be more prevalent in those industries which also seek greater
RULES OF ORIGIN 573
origin have a negative impact on the volume of preferential trade. Mattoo
assessed the African Growth and Opportunity Acts and suggest that the benefits to
Africa would have been approximately five times greater without the restrictive rules
of origin that were in place (in particular with regard to yarn).
(2004) use a sectoral cross-section model to focus on the impact of
rules of origin for textiles. Their results indicate that lack of cumulation of rules of
origin in textiles may reduce trade between non-cumulating countries by up to 73%
in 1995 and 81% in 1999.
Estevadeordal and Suominen (2004) employ a synthetic ROO restrictiveness index
compiled on the basis of the underlying features of ROO across a range of preferen-
tial trade agreements (PTAs). This index is then used in an augmented gravity model.
Their results suggest that rules of origin do serve to restrict trade, and that measures
that allow for a relaxation of their restrictiveness (such as diagonal cumulation, or
clauses) encourage trade. Unlike the preceding analyses Carrère and De
Melo (2004) address the issue of the costs of ROO compliance, also in the context of
NAFTA. They find that these tend to be lowest where underlying ROO refers to a
change in tariff classification, and tend to be highest where the ROO specified given
production processes. They also assess the consistency of their results with the synthetic
index derived by Estevadeordal and Suominen.
(2004), also in the context of US–Mexico trade, show that rules of
2. RULES OF ORIGIN: A PRIMER
This section provides readers with the background they need to understand how ROOs
can distort trade flows and how and why they may be used as protectionist devices.
2.1. The formulation of rules of origin
As mentioned in the introduction, rules of origin are an unavoidable aspect of free
trade agreements because the introduction of multiple tariff rates opens the door
to fiscal fraud. The EU, for example, charges a zero tariff on optical fibres imported
from Switzerland but a 2.9% tariff on those imported from Japan. To prevent tariff-
evading trans-shipment of Japanese optical fibres via Switzerland, EU customs
authorities need rules for determining whether the goods crossing the Swiss–EU
border are actually made in Switzerland. This is the job of rules of origin.
Globalization has made it much harder to determine a good’s country of origin. Most
exported goods contain a large measure of imported parts and materials and thus
may be viewed as originating from many different nations. The standard convention
is that a good is considered as having been made in the last country in which it
underwent a ‘substantial transformation’. There are three standard tests for substantial
transformation. Continuing with the EU–Swiss example for the sake of concreteness,
574 PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
Change in tariff classification
tariff classification than do spectacles so if this rule is applied, eyeglasses made
in Switzerland from optical glass imported from Japan are considered as ‘Made
in Switzerland’ as far as tariffs are concerned (because the imported input, glass,
falls under a different tariff classification than does the exported good, spectacles).
Value added content.
Under this rule, if at least
Switzerland, the good is considered as Swiss. The exact percentage varies across
products and preferential agreements; typical values range from 40% to 60%.
Specific production process rule
. This is where the rules can get really idiosyncratic.
For example, the EU considers ‘wood sawn lengthwise’ imported from Switzerland
to be Swiss if any ‘planing, sanding or end-jointing’ was done to it in Switzerland.
. For example, optical glass falls under a different EU
% of a good’s value is added in
These three tests can be employed together. To give the reader a taste for the
complexity and specificity involved, here is the EU’s rule of origin for ‘glassware of a
kind used for table, kitchen, toilet, office, indoor decoration or similar purposes’. The
glassware is viewed as originating in the exporting nation if it was:
Manufactured from materials of any tariff heading, except that of the product, or
Cut, provided that the total value of the uncut glassware used does not exceed
50% of the ex-works price of the product, or
Hand decorated (except silk-screen printing) of hand-blown glassware, provided
that the total value of the hand-blown glassware used does not exceed 50% of
the ex-works price of the product.
2.1.1. All-or-nothing and complexity.
nothing status. One can imagine that this facet of ROOs was adopted for simplicity’s
sake – it means that firms only have to track one feature of each imported component (the
component’s origin) rather than, say, tracking the source of all inputs into each of its
imported component. However, this all-or-nothing feature can have important distortionary
effects. Marginal changes in the source of even a minor component may cause a firm
to lose duty-free status for the entire value of its exports. One feature that greatly
contributes to the daunting complexity of ROOs – the ‘spaghetti bowl’ problem in
Bhagwati’s memorable words – is that each preferential trade agreement has its own set
of ROOs. (Box 1 summarizes the application of ROOs in different PTA arrangements.)
Importantly, ‘origin’ is treated as an all-or-
Box 1. An overview of rules of origin in regional trading
agreements (RTAs) European Union
The EU has in recent years tried to move towards a common set of ROOs
across each of its preferential trading arrangements. That common set of rules
is known as the Pan-European system and was introduced in 1997. In practice
(minor) differences across the EU’s regional arrangements still remain depending
RULES OF ORIGIN575
on when the original agreements were signed. Hence, for example, the ROOs
applicable to the Association Agreements with Egypt and Jordan are virtually
identical to the Pan-European rules, whereas the agreements with Morocco
and Tunisia are slightly different for certain product categories. The EU ROOs
use all of the above criteria depending on the product being considered with the
change of tariff classification and the minimum value being most frequently
(2004) show that in the Pan-EU system the change of tariff
classification rule applies to 33% of products listed, the value added rule applies
to 11% of products, 13% of products are covered by both of these criteria,
and 18% of products by the combination of either of these with the specific
production processes rule.
United States and the Americas
In the Western Hemisphere, there is generally more variation in the usage of the
different criteria than in Europe. The rules in the NAFTA agreement (and
then also in many other related agreements such as US–Chile, Mexico–Bolivia,
Chile–Canada etc.) are complex, requiring either a change of tariff chapter,
heading, or item depending on the product, and these are often combined with
one of the other criteria. In contrast in the Latin American Integration
Agreement there is a general rule involving a change in tariff classification
or a regional value added of at least 50% of the f.o.b. (free on board) value of
exports. Similar rules apply in CARICOM and in the Andean Community.
Where the EU is attempting to harmonize its ROO regimes across different
agreements, this is not the case for the US, where the regimes vary widely
between different agreements.
The use of simpler rules is often the case in many other regional integration
schemes. Some only use the change in tariff classification rule, e.g. Canada–
Israel, or CACM. Others focus primarily on the minimum value-content rule
(e.g. ANZCERTA, SPARTECA, ECOWAS, COMESA, MERCOSUR) which
can range widely. Hence this can vary from, for example, 25% domestic
content in the Namibia–Zimbabwe FTA, to 35% in COMESA, to 60%
domestic content for MERCOSUR. Other regimes, however, do introduce more
sectoral selectivity such as SADC, or the Japan–Singapore Economic Partnership
This box draws heavily upon WTO (2002a), and Estevadeordal and Suominen (2004).
These give a comprehensive discussion of the incidence of the different criteria across a wide
range of PTAs, as well as an analysis of some of the key differences in ROO regimes across
different regional groupings.
576 PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
2.2. How might ROOs distort trade patterns?
Rules of origin may act as trade barriers via two main channels.
They impose administrative costs on exporters. In this way they act as transac-
tions costs that tend to offset the bilateral trade creation;
They may induce firms to switch suppliers in order to meet the rules of origin.
In this way, they tend to exaggerate the classic trade diverting effect of preferential
liberalization. The ROOs-as-transaction-costs effect is the easiest to understand,
so that is where we begin.
2.2.1. ROOs as transaction-costs: Impact on trade between FTA partners.
When a free trade agreement (FTA) is signed, two things affect the bilateral trade
between the partners. The bilateral tariff drops to zero, but firms need to prove
‘origin’ in order to get duty-free treatment. This leads to two simple points. First, the
administrative burden of ROOs tends to offset the impact of the bilateral tariff
liberalization. Second, the burden of the ROOs-as-transaction-costs channel is
limited by the level of the non-preferential tariff. After all, firms can avoid all the
ROO administrative costs by paying the non-preferential tariff. As a matter of fact,
a great deal of this goes on in Europe since non-preferential tariffs are fairly low (see
Table 1). This limits the potential for ROOs to distort trade between the partners.
This insight led to the first wave of empirical research on ROOs in the 1980s. As
was mentioned earlier, several researchers have shown that a great deal of bilateral
trade between FTA partners actually pays the non-preferential rate (the MFN, or
most favoured nation tariff, in the jargon). Here we present evidence that this is also
true of many of the FTAs signed by the EU.
Table 1 gives information on what is known as ‘utilization rates’ for a range of
southern Mediterranean countries (Meds), and three Central and Eastern European
countries (CEECs). Utilization rates are the share of exports from the listed countries
to the EU that are actually granted the preferential tariff rate, i.e. the proportion for
which ‘origin’ status has been established. Tens of thousands of goods are traded, so
it is necessary to summarize the information. What we do is group products at a fairly
disaggregated level, namely the two-digit level in the so-called Harmonized System
(HS) of goods classification (there are 96 two-digit groups), and then calculate the
proportion of two-digit sectors with utilization rates less than 25%, 50% and 75%
(respectively) for each country and year. Hence, if we look at the first row of the table
for the Czech Republic, it can be seen that in 1998 out of all the 97 two-digit sectors
12.63% of the sectors had utilization rates of less than 25%, 24.21% had utilization
rates of less than 50%, and 52.63% had utilization rates less than 75%.
There are several interesting features, which emerge from this table. First, the
utilization rates are typically very low, overall. ‘Origin’ was established for more than
75% of the exports in only about half of the product categories. In 20–40% of the
product categories, less than half the exports were granted ‘origin’ status (and thus
RULES OF ORIGIN577
eligible for the preferential tariff). One interpretation of this is that there is a great
deal of frustrated trans-shipment going on, i.e. goods made in, say, the US or Japan,
are being trans-shipped to the EU via the Czech Republic, but the rules of origin are
frustrating the fiscal fraud. More reasonably, we can view the Table 1 numbers as an
indication that firms in these countries find that the transaction costs involved in
establishing origin to be greater than the cost of paying the EU’s MFN tariff. In other
words, the low utilization rates indicate that the ROOs are offsetting a great deal of
the bilateral liberalization that would have otherwise happened when bilateral tariffs
were lowered under the various preferential trade agreements (FTAs in the case of
the CEECs, specifically Europe Agreements, and unilateral GSP preferences in the
case of the southern Mediterranean nations).
Utilization rates tend to be higher for the CEECs than for the Meds, and the
proportion failing to meet the three cut-off criteria for all of the southern Mediterra-
nean countries except perhaps Morocco tends to rise over time, whereas for the
CEECs (except the Czech Republic) it tends to fall. Thus, for example for Egypt,
whereas in 1998 15.96% of industries had utilization rates less than 25%, by 2001
Table 1. Utilization rates
Percentage of sectors with utilization rates
below a certain threshold
Czech Republic < 25
Source: Own calculations on data supplied by the European Commission.
578 PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
this appears to have risen to 27.08%. The corresponding figures for Poland are 16.49%
and 14.43%. Since the EU’s MFN tariff was stable or falling during this period (some
Uruguay Round cuts were still being implemented), this pattern suggests that the
administrative burden of ROOs was rising for the Meds, but falling for the CEECs.
There could be several reasons for this, and it is also well known that the data on this
is imperfect. Nevertheless, the simple fact that the rates are so low suggests that the
first distortionary aspect of ROOs – namely the extent to which they offset the
bilateral liberalizing effects of preferential arrangements – is important.
While the first channel through which rules of origin act as trade barriers concerns
the trade between the FTA partners – the EU and CEECs and Meds, for example –
the second channel affects trade between the EU’s FTA partners and third nations.
2.2.2. Supply-switching effects of ROOs: impact on third nations.
two countries sign a free trade agreement, they lower tariffs on bilateral trade while
simultaneously establishing rules of origin. As per the classic trade-creation-trade-
diversion analysis, the FTA will lead to some supply switching from third nation
suppliers to FTA suppliers; preferential tariffs change the relative price of imports
from partner nations versus third nations and so naturally induce supply-switching in
favour of partner-nation firms. The classic analysis only considers the impact of the
preferential tariff. As noted above, however, rules of origin can also induce FTA-based
firms to reduce their purchases of inputs from third nations.
Hub and spoke
empirical work, it is useful to present some facts concerning the tangle of bilateral
and plurilateral trade arrangements surrounding the EU. During the 1990s, the
period of our data, the EU had 15 members and economically speaking, was the
hegemon of Europe. The EU15’s GDP was about $10 000 billion while the next
largest non-EU economy (Switzerland) was only $277 billion; other non-EU economies
were significantly smaller. Since trade arrangements are all about market access, the
EU’s trade arrangements with its neighbours can be described as hub-and-spoke
bilateralism. (The large economy, the EU, is called the ‘hub’ and the smaller economies
are called the ‘spokes’, since a schematic representation of the pattern of bilateral free
trade agreements resembles an old-fashioned wagon wheel, see Figure 1.) Of course,
the hub-and-spoke picture is a radical simplification of the tangle of trade arrange-
ments in Europe and the Mediterranean. A hint of this complex can be seen in the
diagram. Some of the ‘spokes’ – i.e. nations that have bilateral free trade deals with
the EU – have free trade agreements among themselves, for example Iceland and
Switzerland are members of the European Free Trade Association (EFTA) and
Poland and Hungary are members of the Central European Free Trade Agreement
(CEFTA). And some of the spokes have unilateral preferential access to the EU; for
example, the EU15 granted substantial tariff preferences to Moroccan goods without
demanding that EU goods receive the same preference in Morocco.
To illustrate this point while simultaneously preparing the ground for our
RULES OF ORIGIN 579
Supply-switching effects of ROOs
suppliers. The figure shows a small part of the hub-and-spoke system of bilateral
FTAs that surrounds the EU. Specifically, it shows two spoke economies, Country B
and Country C, which have signed bilateral FTAs with the EU. The export flows
between the spokes and between a typical spoke and third nations (RoW, short for
‘rest of world’) are shown with arrows. As the labels on the arrows indicate, the
bilateral FTAs tend to reduce trade between spoke-nation B and spoke-nation C for
two distinct reasons: (1) the preferential tariffs that each ‘spoke’ grants firms located
in the EU but not those located in the other spoke, and (2) rules of origin that lead
Figure 2 illustrates how ROOs may lead firms to switch
Figure 1. European hub-and-spoke bilateralism, c. 1996
Figure 2. Trade impact of rules of origin
580PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
firms in the spoke economies to prefer sourcing their inputs domestically or from the
EU for purely origin-establishing reasons.
This outlines the fundamental supply-switching effect of ROOs. Of course, the extent
to which such supply switching occurs will then depend on a number of further factors,
such as the nature of the underlying market structure (e.g., Vousden 1987; Krishna and
Krueger 1995), or on how ‘sufficient working or processing is defined’ (Krishna and
Itoh 1988), and of course on the costs of not being able to fulfil the originating
requirement, and in particular the height of the importers’ tariff (Hoekman 1993;
2002). Clearly then the rules may serve to protect certain sectors from
the degree of liberalization that might otherwise be implied by the free trade agreement.
These issues are also discussed in Brenton and Machin (2002), Falvey and Reed
(2002), Burfisher, Robinson and Thierfelder (2004), and Hoekman (1993).
As far as spoke-RoW trade is concerned, both the ROOs and the preferential
tariffs tend to depress exports from RoW to the spoke economies, but there is no
first-order effect on the exports of the spoke economies to the rest of the world. For
simplicity, we omit the impact on spoke-EU trade and RoW-EU trade. (Our empirics
focus on the spoke-spoke and spoke-RoW trade flows.)
The identification problem
flows in the same direction. The core difficulty in estimating the impact of ROOs on
trade flows lies in separating the classic effects of tariff preferences (trade creation
between partners and trade diversion with third nations) and the impact of the ROOs
that always accompany tariff preferences (reduced trade creation between partners
and exaggerated trade diversion with third nations). The impact of preferences on
trade diversion, however, only applies where the spokes offer preferential access
to the EU and not to the other spokes. If countries B and C also have a free trade
agreement between themselves, then the issue of trade diversion should not apply.
Similarly, if a spoke has an asymmetric trade agreement with the EU, where it is the
EU which is offering tariff free access to its market but not vice versa, then again
trade diversion should not be an issue (the spoke levies the same tariff on imports from
all sources, so no trade diversion arises with respect to its trade with third nations).
Another aspect of ROOs that suggests a way to separate effects is their extra-large
impact on intermediate goods. While the trade diversion arising from the tariff pref-
erence will affect final and intermediate goods equally, the ROOs tend to have a
much greater impact on intermediates than they do on final goods. Finally also,
ROOs may have a bigger impact on nations that rely more heavily on imported
intermediates, e.g. small nations.
As the figure shows, preferences and ROOs tend to affect trade
As far as our empirical work is concerned, the main points to retain are that rules of
origin are likely to:
RULES OF ORIGIN581
Reduce trade between ‘spokes’, i.e. nations with which the EU has a bilateral
FTA, while encouraging trade between the EU and each spoke nation;
Affect trade in intermediate goods more than final goods;
Affect small countries more than large since they are likely to depend more on
imported intermediate goods;
Affect trade more for those countries where vertical specialization is more
3. THE CUMULATION OF RULES OF ORIGIN: A NATURAL EXPERIMENT
By the mid-1990s, a network of trade agreements had almost completely eliminated
tariffs and quotas on European trade in industrial goods. This tangle of arrangements,
however, was not the same as continent-wide free trade due in a large part to rules
This ‘spaghetti bowl’ problem posed problems for European manufacturers. In
Europe, as elsewhere, staying competitive requires firms to scour the world for the
cheapest inputs. Frequently, this involved setting up a complex supply chain in which
components were shipped among many nations. In the mid-1990s, there were some-
thing like 60 bilateral FTAs in Europe, each with its own complex set of origin rules.
Such complexity made it difficult to optimize manufacturing structures, especially
given the all-or-nothing nature of ‘origin’. Some final goods have hundreds of
intermediate inputs, some of which pass through several nations during their pro-
duction. Since any given part can lose its origin-status each time it crosses a border
– and the origin of the final good depends in a complex way on the origin of each
component – the problem of multiple, complex ROOs became a nightmare for
European businesses. It could be very difficult if not impossible for a firm to be
absolutely sure how the outsourcing of one of its intermediate goods would affect the
origin status of its final-good exports.
It is exactly this complexity that led EU businesses to lobby for diagonal cumula-
tion in the form of the Pan-European Cumulation System. Pushed by their manufac-
turers, the EU15, the EFTA4 (Iceland, Liechtenstein, Norway and Switzerland), and
10 then applicant-nations in Central Europe decided to amend their various FTAs
by substituting a common set of rules of origin for those they originally contained.
Value could thus be cumulated between different European countries without
prejudicing the duty-free status of end products. The result was the Pan-European
Cumulation System (PECS), introduced in 1997 and extended to Turkey in 1999.
The PECS is not a minor event in world trade. PECS nations account for about
two-fifths of world trade and have an aggregate population of 554 million.
This bit of commercial history is critical to our empirical strategy. Note that what
was being liberalized with the cumulation system was the trade-distorting impact of
ROOs; there was no change in the preferential tariff rates. Given the focused nature
of the liberalization inherent in cumulation, changes between pre- and post-cumulation
582PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
trade flows can give us a lower bound measure on how distortionary the ROOs
where in the first place – lower bound since cumulation reduces the restrictiveness of
ROOs but does not fully eliminate it.
3.1. Expected trade impact of cumulation
Since the impact of cumulation is at the heart of our empirical contribution, and the effects
are somewhat complex, it is useful to illustrate the expected effects with a diagram.
Figure 3 shows the impact of adding cumulation to bilateral FTAs considered in Figure 2.
Recall that each bilateral FTA tended to reduce trade in intermediates between
the spoke economies (Countries B and C) since such inputs did not help firms based
in, for example, Country B with the ROO requirements for exporting to the EU
(typically their main market). Cumulation changes this.
Under the Pan-European Cumulation System, it is easier for inputs sourced from
C or B or the EU to count as ‘domestic’ as far as eligibility for duty-free treatment
is concerned. We should, therefore, expect a rise in spoke-spoke trade after the
implementation of cumulation, i.e. after the PECS started operating in 1997. Of
course, even with cumulation and harmonized rules, rules of origin continue to distort
trade flows in Europe. Cumulation reduces, but does not eliminate the administrative
costs involved in demonstrating origin. Moreover, the ‘specific production process’
ROOs continue to act as very high non-tariff barriers for certain inputs, especially as
regards imports from non-PECS nations.
3.1.1. Not for third nations.
sourced from third nations. We should not, therefore, as a first order impact, expect
Importantly, the cumulation does not extend to inputs
Figure 3. Trade impact of cumulation on trade flows
RULES OF ORIGIN583
RoW exports to the spoke economies to rise. However, it should be noted that by
allowing greater use of PECS intermediates and hence increasing their proportion,
this may in turn allow for an increased use of RoW intermediates without violating
the rules of origin. Cumulation may therefore have a second order impact on third
nations, such that the net effect on RoW-spoke trade is ambiguous. In contrast we
would not expect any impact on either intra-RoW trade, nor intra-EU (intra-hub)
trade. These differences are the crux of our identification strategy. To summarize:
By diminishing the trade-inhibiting impact of ROOs on spoke-spoke exports, but
not on RW-spoke exports, implementation of the PECS constitutes a natural
experiment that allows us to establish a lower bound estimate on how restrictive
ROOs were in the first place. As usual, the differential impact should be greater
for intermediate goods.
This analysis clearly suggests the use of a difference-in-differences estimating strategy,
a suggestion we follow in the next section.
Before turning to the formal statistical analysis, however, we present some
evidence that the PECS did indeed impact spoke-spoke exports in a manner that
was distinct from its impact on RoW-spoke exports.
evidence for cumulation’s impact
In order to see whether there is a
consider Figures 4 and 5. The diagrams give total imports relative to 1997 imports
from three sources: the EU, the non-EU countries that joined the PECS (denoted
PEU), and the rest of the world (denoted RoW). These indices were calculated for
each country separately and then averaged. In Figure 4 we show these flows for the
CEFTA and Baltic countries, and in Figure 5 for the EFTA countries.
case for PECS impacting on trade flows,
Figure 4. CEFTA and Baltic imports by source
584PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
From the figures it can be immediately seen that membership in PECS seems to
have made a difference. For both the CEFTA countries, and the EFTA countries, the
evolution of trade prior to 1997 was similar across all sources. After 1997, the CEFTA
countries increased their imports from non-EU PECS members by more than their
increase in imports from either the EU or the rest of the world. This difference is
even more marked for the EFTA countries. While clearly there are many other factors
also having an impact on trade flows, it is perhaps significant that there appears to
be a differential impact after 1997, the year in which PECS was introduced.
4. ESTIMATING THE TRADE IMPACT OF ROOS
We begin our formal statistical analysis by examining whether the introduction of
cumulation via the PECS had a positive impact on spoke-spoke trade controlling
for other factors. The idea behind this empirical strategy is simple, albeit somewhat
indirect. Our theory suggests that cumulation reduces the restrictiveness of ROOs
without changing tariff preferences. Thus, if we find that cumulation boosts spoke-
spoke trade, then it must have been that ROOs were depressing this trade before
cumulation. Of course, cumulation does not fully eliminate the distortionary effects
of ROOs, so cumulation’s impact will be a lower bound estimate of the initial restric-
tiveness of the ROOs. Plainly, Figures 4 and 5 suggested spoke-spoke trade did rise
at the same time as cumulation was introduced. The crucial addition of this section
is to formally control for other factors that might have accounted for the post-1997
surge in spoke-spoke trade.
To control for other factors, we need a model of what trade flows would have been
without the PECS and here we use the gravity model since it is now the standard solution.
See Anderson (1979), Bergstrand (1985, 1989), Helpman and Krugman (1985), Deardorff (1998), Frankel (1997) and Anderson
and Wincoop (2003) for theoretical justifications.
Figure 5. EFTA imports by source
RULES OF ORIGIN 585
In its simplest, ‘barebones’ form, the gravity model postulates bilateral trade as
increasing with the economic ‘mass’ of the exporting and importing nations (usually
their GDPs), and decreasing with the bilateral distance between them. (The resem-
blance to the formula for the gravitational force between two masses gives the model
This ‘barebones’ version is typically augmented with additional controls. The most
common are the nations’ populations and various dummies to capture pro-trade
factors, e.g. a common language, a common border, and preferential trade arrange-
ments between the two countries (see Appendix 1 for a more formal discussion). In
its latest evolution, several authors suggest that the bilateral trade is also affected
by third nation variables since sales from nation A to nation B depend in part on
the price of A’s goods relative to the price of goods from third nations. This line
of reasoning suggests inclusion of an importer-specific price-index-like term – often
called a ‘remoteness’ or ‘multilateral trade resistance’ measure. Specification of
this measure is complex, but Anderson and Wincoop (2003) show that an equivalent
methodology is to use country-specific fixed effects. These types of fixed effects are
also recommended in Matyàs (1997) and have been used by many authors, including
Hummels (1999), and Redding and Venables (2001).
We also control for the presence of trade agreements among nations in our sample.
For our analysis this is important because trade agreements can distort trade and
lead, for example, to trade diversion, as discussed earlier. We include controls for
multi-country agreements such as the EU, EFTA, CEFTA and BAFTA. We also
control for bilateral agreements between our spokes. These are divided into two
categories of agreement – those between PECS countries, and those between PECS
and non-PECS spokes. We also control for agreements between the EU and its
The introduction of cumulation could, theoretically, also affect hub-spoke exports
and RoW-spoke in both directions. We therefore include dummy variables for hub-
spoke, and RoW-spoke trade, where we distinguish the direction of trade. Lastly, there
were also some changes in the preferential trade agreements between countries in our
sample. We have taken account of each of these changes by including extra dummy
variables that switch when the new free trade agreements take effect. These include
the changes in CEFTA membership, agreements between PECS spokes (in Table 2
below these are entitled
), and agreements between PECS and non-PECS
In terms of the standard cross-section estimating equation the advantage of the Anderson and Wincoop approach is a more
correct treatment of third-country competition; however, as the fixed effects are perfectly collinear with the key activity variables
(the GDP levels of the exporting and importing countries) the Anderson and Wincoop approach effectively implies unitary
restrictions on the GDP coefficients which may or may not be justified. In the empirical work reported here for the cross-section
estimations we employ both approaches.
586PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
4.1. Measuring cumulation’s impact: a difference-in-differences approach
The precise statistical analysis used to establish a lower bound on the impact of
ROOs is a technique called difference-in-difference analysis. This compares the
behaviour of two groups of bilateral trade flows. The ‘treatment’ group includes all the
bilateral trade flows that should have been boosted by the PECS. The ‘control’ group
is made up of the bilateral trade flows that should not have been affected by the PECS.
In essence, the procedure is to compare how much treatment-group trade flows rose as a
result of cumulation (this is the first difference) and compare this with the change in flows
for the control group (the second difference) – hence the term difference-in-differences.
Of course, the introduction of cumulation was not the only thing that changed
between the pre-1997 and post-1997 periods, but we use the gravity model to control
for other factors. Additionally, we control for all sorts of unobservable pair-specific
factors (e.g. historical ties, business networks, etc.) by employing a statistical technique
called fixed effects at the country-pair level. We also hope that this goes some way
to correcting for the issue of reverse-causality (namely, the idea that membership in
PECS may have been more likely for nations with high spoke-spoke trade flows, so
trade is influencing PECS membership rather than vice versa).
4.1.1. The ROO and ROO+TD dummies.
cumulation on affected spoke-spoke trade flows should show up in the coefficient
on a dummy that switches from zero to 1 when PECS is introduced. Specification of
the dummy, however, is subject to some complication since the PECS ‘natural exper-
iment’ was not perfectly designed from the econometrician’s perspective.
As discussed in Section 2.2.2, the main difficulty in estimating the impact of ROOs
lies in distinguishing between classic trade-diversion effects that arise from tariff pref-
erences and the extra supply switching that arises from the imposition of the ROOs
that come with the preferences. This applies principally when considering the impact
of the introduction of a preferential trading arrangement – where one would expect
both effects to be present – and therefore also when comparing trade between cumu-
lating and non-cumulating partners at any given point in time. The conflation of the
ROO impact with trade diversion is less of an issue when considering the impact of
the introduction of the PECS system as this occurred (largely) in the presence of
existing trade agreements. Nevertheless, from this perspective, the natural experiment
was purer for some spoke-spoke trade flows than it was for others. For example,
Hungary and Poland had a free trade agreement between themselves and each had
a free trade agreement with the EU. Thus, the PECS made rules of origin less
restrictive in a setting of zero tariffs. For other spoke-spoke flows directly affected by
PECS, the experiment was rather less pure. Turkey and Estonia, for instance, did not
have a bilateral FTA, but each had free trade with the EU. In this case, it is possible
that Turkey-Estonia trade could be affected both because of ROO supply-switching
considerations and because of changes in the degree of trade diversion. However, as
In practical terms, the impact of
RULES OF ORIGIN 587
there is little direct reason to suppose any change in the degree of trade diversion, in
reality this is unlikely to be significant.4
Nevertheless, as the possibility does exist, in order to distinguish these two cases,
we specify two ROO dummies. ROOIMPACT is the dummy for PECS-affected spoke-
spoke trade flows where trade preferences are not an issue. ROOIMPACT+TD is the
dummy for PECS-affected spoke-spoke trade flows where trade diversion may be an
issue (see Appendix 2 for the precise classification of all bilateral trade flows).
4.1.2. Control groups.
that were unaffected by the PECS, but this group should also be as large as possible
to boost statistical precision. Since everything affects everything in general, selection
of the control group is subject to a trade-off between group size and the extent to
which we are sure that the PECS had no influence. We use three different groupings.5
The first group comprises all bilateral trade flows in our sample that are not in the
treatment group. Note that this includes exports by the rest of the world (RoW) to
the spokes, as well as trade between PECS and non-PECS spokes (e.g. between
Morocco and Poland). As discussed earlier in the context of the impact on RoW-
spoke trade, the net effect on these flows of improved cumulation arrangements is
ambiguous due to secondary effects; it is possible that these flows are indirectly
affected by the PECS and so should not be viewed as proper controls.
To deal with this, we set up a second, more narrowly defined control group by
taking out these bilateral trade flows. This second control group almost certainly
captures the impact of cumulation more accurately. Finally, it is also possible that
cumulation may have had an impact on sales from the EU (the hub) to the spokes,
again due to secondary effects. To address this possibility, we created a third, even
narrower control group that excludes all hub-spoke flows as well as all RoW-spokes
flows. Thus it includes only intra-EU flows, intra-RoW flows and flows between the
EU and the rest of the world. As with the second control group, this is more likely to
correctly capture the impact of cumulation on intra-spoke trade – which is precisely
where the theory predicts the most unambiguous results. The three sets of regressions
are respectively labelled Control 1, Control 2, and Control 3 in Table 2 below.
The control group should consist of bilateral trade flows
4.1.3. Data sample.
– all of the EU countries, three EFTA countries (Iceland, Norway and Switzerland),
the CEFTA countries, the Baltic States, six countries taking part in the Barcelona
process (Turkey, Jordan, Israel, Egypt, Tunisia, Morocco), as well as the United States,
Canada, China, Japan and Australia. The regressions were carried out using total
Our estimations are based on trade flows between 38 countries
4 Likewise, when the EU-spoke preferential trade agreement is asymmetric as it is with developing nations (the EU offers tariff
reductions without receiving reciprocal tariff preferences), then there would be no spoke-spoke trade diversion, and thus no
possibility of conflation of trade diversion and ROOs effects.
5 Tables A1–A3 in Appendix 2 provide more details of the treatment and control groups used.
PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
Table 2. Panel regression with country-pair fixed effects, 1995–1999
Control 1 Control 2Control 3
TotalManuf. Int.TotalManuf.Int. TotalManuf.Int.
No. of obs
Notes: The estimated equation is xijt = α0ROOijt + α1yit + α2yjt + α3nit + α4njt + α5PTAijt + αij + γt + eijt, where xijt is the real bilateral export from i to j in period t, ROO is the
cumulation dummy between i and j in year t, and the y’s and n’s are GDPs and populations respectively, αij is the pair-specific fixed effect, γ is the year effect, and e is the error.
See text for definition of control groups. Standard errors are in parentheses and *, **, and *** denote statistical significance at the 10%, 5% and 1% levels respectively. Since this
includes country-pair dummies all time-invariant variables are absorbed by the fixed-effects. Estimation is by ordinary least squares.
RULES OF ORIGIN 589
trade, manufacturing trade, and intermediate goods trade for the years 1995–99. Full
details concerning the data can be found in Appendix 2.
and intermediate goods trade with all three control groups. The clearest theoretical
predictions on cumulation consist of the impact on spoke-spoke trade – the effects
picked up by ROOIMPACT and also by ROOIMPACT+TD.
Table 2 reports on the results for total trade, manufacturing trade,
The pure treatment effect.
group constitutes all other countries in the sample. Here the results for ROOIMPACT and
ROOIMPACT+TD in the fifth and sixth rows indicate that trade between the spokes was
higher after 1997, but that for all categories of trade the results are not statistically
significant. Remember, however, that as discussed above there are trade flows
included in the control group here where the anticipated net effect on trade is ambig-
uous. These are then excluded in the middle and right hand panels entitled Control
2 and Control 3.
If we take both these panels we see that the ROOIMPACT coefficient is statistically
significant in all cases except one. The percentage equivalent of these dummies can
be found by taking [exp(dummy) − 1] * 100. Applying this suggests that cumulation
served to increase trade by 9.5% with respect to Control 2, and between 5.6% and
11.5% with respect to Control 3. With respect to both the second and third control
groups, the smallest impact on trade is for manufactured goods, and the biggest
impact on intermediate trade and total trade. As discussed above, this provides direct
evidence that ROOs had an independent, negative impact on spoke-spoke trade.
Since cumulation did not eliminate the restrictiveness of ROOs – rather it merely
relaxed it – this is likely to be a lower bound estimate of the supply-switching impact
The coefficients on hub-spoke trade are positive and significant across all categories
of trade, and suggest that exports from the hub to the spokes rose in the order of
11.9–14.7% for manufacturing trade and between 19.9–24.4% for intermediate
goods trade after 1997. Note that this is a period where there was on-going symmetric
trade liberalization between the EU and the CEFTA and Baltic spokes, and therefore
much of this increase is likely to be explained by this, rather than by cumulation per
se. There is also some evidence of an increase in spoke-hub trade. Finally, both RoW-
spoke and spoke-RoW trade increased after 1997 when looking at the first control
group, which is consistent with on-going trade liberalization between the rest of the
world and the spokes. The evidence on RoW-spoke trade is also interesting as the
theory suggested possible ambiguity regarding the impact on trade. On the one hand
there could be trade diversion arising from the introduction of the PECS system, on
the other hand by allowing a greater proportion of (PECS) originating goods, this
could result in more goods being imported from the rest of the world. Our results do
not suggest that there is any evidence of the former. If we turn to the regional
Turning first to the left panel of the table, where the control
590PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
integration dummies we see that the CEFTA dummy and PECS dummy, which is
picking up on bilateral agreements between PECS countries (e.g. Norway-Poland),
are both positive and highly statistically significant. Not surprisingly they suggest that
signing a regional integration agreement serves to positively increase trade flows
between member countries.
4.1.5. Robustness checks.
undertaken some sensitivity analysis with regard to the preceding regressions. Here
we conduct two forms of sensitivity analysis. First, we have excluded from our sample
extreme outliers, i.e. those trade flows which are unusually large or unusually small
given all the factors that we are controlling for.6 Secondly, we have excluded from our
sample all bilateral trade flows with a very small value – where we define small as
where the bilateral trade is less than $1 million for total trade in 1999. The reason
Table 3 presents some further results where we have
6 Formally this is defined by analysing the residuals from the panel regression for total trade (Control 1). All observations for a
given country-pair are removed if any of its estimated residuals in this regression exceed in absolute value 3.5 times the standard
error of the residuals.
Table 3. Comparing ROOIMPACT across different control groups
Total Manuf.Int.Total Manuf. Int.
Small flows excluded
Small flows excluded
Small flows excluded
Notes: *, **, and *** denote statistical significance at the 10%, 5% and 1% levels respectively. Estimations are
with the same variables as in Table 2, i.e. they include country-pair dummies, spoke dummies, RTA dummies,
as well as year dummies.
RULES OF ORIGIN 591
for excluding small value trade flows is that these may be more subject to fluctuation
and possibly to measurement error.7 Of course, choosing a cut-off point is to some
extent arbitrary, but the point of the exercise is to explore the sensitivity of the results
to such changes in the underlying sample.
The results corroborate those in Table 2. If we take our preferred control groups
(2 and 3), the results with no outliers for ROOIMPACT suggest that cumulation increased
trade between 7.4% and 16.3%, and with the small flows excluded by between
10.8% and 15.5%. When looking at the ROOIMPACT+TD coefficient for our preferred
control groups the results suggest that the increase in trade as a result from cumula-
tion ranged between 7.5–11.2% with no outliers, and between 15.5–22.1% with the
small trade flows excluded. It is, of course, important to underline that in principle
here there may be both trade diversion effects and supply switching effects that are
being captured. However, as there is little reason to suppose much change in trade
diversion over this time period, both ROOIMPACT and ROOIMPACT+TD are likely to be
largely capturing the effect of increased cumulation possibilities.
4.1.6. Summary of difference-in-difference findings.
approach allows us to exploit the ‘natural experiment’ aspect of the Pan-European
Cumulation System introduced in 1997. This system relaxed the restrictiveness of
ROOs on trade among the EU’s FTA partners without changing the degree of tariff
preference. This allows us to establish a lower-bound estimate of the extent to which
the ROOs reduced spoke-spoke trade in the first place. The size of the effect we find
is large – but not unbelievably so. Specifically, we find that with our preferred control
groups the impact ranges from 7.4% to 22.1%.
4.2. Digging deeper
The question so far has been: Did the PECS change in cumulation arrangements
affect trade? The answer so far is ‘yes’ – there is clear evidence that the PECS system
increased trade between the non-EU nations that joined the PECS. Now we turn to
another aspect of our natural experiment. Instead of focusing on the change, we focus
on the level of trade flows. In particular, we ask whether the level of trade between
non-cumulating spokes is lower. If ROOs affect trade in the ways the theory suggests,
we would expect lower trade flows between countries that do not enjoy cumulation,
controlling as usual for other factors. It is this issue to which we now turn, and we
do so by running the gravity model first as a pure cross-section for each of our years,
and then as a panel but with Anderson–Wincoop-style fixed effects. Note that just as
7 For example, the raw data for the Baltic countries suggested both a different pattern with regard to their exports than for the
other PECS countries and a greater variation especially in the latter period. There could be several reasons for this – the signing
of a free trade agreement with EFTA, the economic crisis in Russia, or even the fact that these countries were pursuing quite
different (substantially more fixed) exchange rate regimes to many of the other, e.g. CEFTA, countries.
592PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
the difference-in-difference approach could be viewed as establishing a lower bound
of the trade impact, this cross-section approach can be viewed as an upper bound;
the ROO dummies absorb all the differences between the spokes with and without
cumulation. In particular, it is likely that the trade flows among the non-EU PECS
members were higher than other non-EU flows due to their long history of integra-
tion. Indeed, as mentioned above, there is the possibility that a high trade flow was
implicitly a condition for PECS membership.
Given the different aims of this section, we alter the ROO variable definition so
that it identifies trade reduction rather than trade promotion. That is, we specify the
main variable of interest – the ROO dummy – to capture the trade impact of not having
cumulation. Hence, if the lack of cumulation does indeed have a negative impact on
trade we would expect the coefficient to be negative. Hence, the ROO coefficient is
now picking up on the levels of trade between those spokes, which are not part of the
PECS system. The spokes are those countries which have a bilateral trade agreement
with the EU – the hub. Hence, for 1995 and 1996, which is prior to the introduction
of the PECS, the ROO coefficient is focusing on the levels of trade between all the
spokes. For the period of 1997–99 the coefficient is then picking up on trade between
the non-PECS spokes. As discussed earlier, however, we need to control carefully for
free trade agreements by distinguishing between trade flows that may be lower
because of lack of cumulation only, and those that may be lower due to an absence
of cumulation and because of trade diversion. Analogously to above, we segment the
two sets of trade flows by defining two ROO-related dummies, ROO and ROO+TD.
Table 4 presents the full set of results from the gravity modelling cross-section
estimation for 1995 and 1999 for total trade, manufacturing trade and intermediate
goods trade. For this first set of estimations we have amended the standard model
with the addition of the two ROO variables, and the results presented are without the
Anderson and Wincoop fixed effects. As with the panel regressions we also include
control variables for the (implemented) preferential trade agreements between the
countries in our sample. These include those for the principal regional trading agree-
ments such as the EU, EFTA, CEFTA and BAFTA, the agreement between the
United States and Israel, as well as any further trading arrangements between PECS
spokes (entitled PECSijt in the table), agreements between PECS-NonPECS spokes
(PECS-NonPECSijt), symmetric agreements between the EU and the Spokes (EU-Spoke 1),
and asymmetric agreements between the EU and the spokes (EU-Spoke 2). Finally, we
also include a control variable for trade between Israel and the Arabic Mediterranean
partners. For historical and political reasons it is well known that these bilateral trade
flows are very low, and thus it is important to control for this.
Table 5 then gives the results solely for the rules of origin dummy variables, ROO
and ROO+TD, for each category of trade and for each of the years 1995–99.8 We do
8 A full set of results is available on request.
RULES OF ORIGIN 593
this both for the standard gravity model given above (columns entitled ‘Base’), but also
where we allow for separate exporter and importer fixed effects (columns entitled ‘Fixed
Effects’) in the spirit of Anderson and Wincoop (2003). Essentially, the idea behind the
Anderson and Wincoop fixed effects is that there may be factors specific to either the
exporting or the importing country, which may have an impact on bilateral trade flows.
Table 4. Trade and cumulation
Variables TotalManufacturing Intermediates
Notes: Standard errors are in parentheses and *, **, and *** denote statistical significance at the 10%, 5% and
1% levels respectively. The equation estimated is given by Equation (A4) in Appendix 1 with the exclusion of
the tariff variable.
594 PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
Turning first to Table 4, almost all the coefficients are statistically significant. The
coefficients on GDP and on distance have the expected sign and show that bilateral
trade flows increase with GDP and decrease with distance. Looking at the GDP
variables it is interesting to note that these results suggest that in most cases these are
statistically different from unity. The variables on the regional trading blocs indicate
that trade between the CEFTA economies, between the Baltic States (BAFTA), as well
as between the United States and Israel is significantly higher across all categories
and years. The coefficients for the EU and EFTA are generally insignificant, and for
the EU slightly negative for 1995. The negative variable on the EU is consistent with
results obtained by previous studies (such as Winters and Soloaga 2001), though it
does appear counter-intuitive. As expected also, the variable on trade between Israel
and the Mediterranean partner countries is negative and highly statistically signifi-
cant. The border variable is perhaps surprisingly negative, though usually statistically
insignificant. Similarly the language variable is not statistically significant.
Most interestingly from our point of view is the negative and statistically significant
variable on the ROO and the ROO+TD dummies in almost all the estimations. The
size of the ROO coefficient falls between 1995 and 1999 for each category of trade,
Table 5. Impact of rules of origin by year and category
TotalManuf. Int.Total Manuf.Int.
Notes: Standard errors are in parentheses and *, **, and *** denote statistical significance at the 10%, 5% and
1% levels respectively. The equation estimated is given by Equation (A4) in Appendix 1 with the addition of
importer and exporter dummies for the fixed effects estimations. For these, then, the GDP and population
variables are absorbed by the fixed-effects. The adjusted R2 for the OLS Base models range from 0.78 to 0.86
while those for the fixed effects are between 0.83 and 0.90.
RULES OF ORIGIN595
but is not significant for intermediate goods trade. Taking the significant variables
these coefficients suggest that the lack of cumulation is associated with bilateral trade
flows that are approximately 24–42% lower. In contrast to the ROO variable, the size
of the ROO+TD dummy rises as we move between 1995 and 1999, and as one moves
from total trade through to intermediate goods trade. Hence the largest negative
impact on trade is for intermediates in 1999, where the coefficient suggests trade is
over 70% lower where there is a combination of both trade diversion and supply
switching from lack of cumulation.
These results are explored more fully in Table 5 where we give the results for each
of the five years and where we also run the regressions with the Anderson and
Wincoop fixed effects.9 The results in Table 5 show the same pattern as that reported
above and are interesting in several regards. The ROO coefficient is consistently
negative and highly statistically significant for total trade and manufactured goods
across all the regressions. The results with the inclusion of the country fixed effects
are fairly similar for total trade and manufacturing trade as the classical gravity
regression results. However, the inclusion of the fixed effects tends to increase the size
and significance of the coefficient on intermediate goods trade. Hence, in the base
regressions the coefficient on intermediates was always lower than for total trade and
manufactured trade; this is no longer the case in the fixed effect model, where the
difference across the different categories of trade is much smaller. For 1997 and 1998
there is some evidence that lack of cumulation mattered more for intermediate goods
trade than total trade or manufactured trade.
When looking at the ROO+TD coefficient in the base model, it appears that the
combination of lack of cumulation and trade diversion rose over time for all categories
of trade, and was particularly pronounced with regard to intermediates. With the fixed
effects model the size of the coefficient is markedly smaller and is significant for the
years 1997, 1998 and 1999. As before, the biggest combined impact of cumulation
and trade diversion is on intermediate goods trade for each of these years.
4.2.1. Caveats and confounding factors.
results it is important to note that the composition of the countries and therefore
trade flows changes across the years, and in particular after 1997 with the introduc-
tion of the PECS for the EFTA, CEFTA and Baltic countries, and in 1999 with the
inclusion of Turkey into the PECS. Each of these did not have cumulation in 1995,
but they did by 1999. This then is an important part of the explanation for the
variation in the coefficients over time. Hence if we look at the coefficients there is
typically a decline in the coefficient after 1998. That decline in the coefficient sug-
gests that the impact of the lack of cumulation was more significant for Turkey, than
for the remaining non-PECS spokes. Similarly the fall in the coefficient after 1996
In terms of interpretation of these
9 The regressions were run using both OLS and Tobit, and the results were highly comparable. The results reported here are
based on the Tobit estimates.
596PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
suggests that the lack of cumulation would appear to matter less with respect to trade
between the CEFTA/Baltic countries and the Mediterranean, between the EFTA
countries and the Mediterranean, and with respect to intra-Mediterranean trade.
As earlier the percentage equivalent of these dummies can be found by taking
[exp(dummy) – 1] * 100. With the fixed-effect model this suggests that trade between
non-cumulating countries was approximately 39–46% lower in 1998, and slightly less
in other years.
In the panel estimations reported earlier in Section 4.2 we controlled for country-
pair fixed effects. In other words we are controlling for any reason, other than those
included in the regression equation, why trade flows between any given pair of coun-
tries may be consistently different over our time period. This is a standard procedure
in this form of panel estimation. However, while there may be effects that are specific
to any given bilateral pairing (as dealt with above), there may also be effects that
are specific either to given exporting countries, and/or effects that are specific to
given importing countries. In the next set of regressions we report on below, therefore,
we control separately for both importing and exporting country effects. By controlling
for country specific effects we are thus examining whether there are effects specific
to either imports or exports such as exchange rate effects, administrative barriers,
export subsidies, level of institutional capacity / corruption etc. which may have an
impact on trade flows.
The country-specific fixed-effects model is thus in the spirit of Anderson and
Wincoop, as discussed earlier, but in a panel framework. Table 6 reports on the results
of the panel estimation with the country-specific effects.10 As before, we present the
results for our principal coefficients of interest, the rules of origin dummy variables –
ROO and ROO+TD. We include estimations carried out over the entire time period,
10 We have also run regressions with either exporting or importing country effects; as well as testing the model for random as
opposed to fixed effects. As the results are qualitatively similar to those given in the table and given that the diagnostic tests
supported fixed over random effects these results are not reported here.
Table 6. Panel regression with country fixed effects
TotalManuf. Int. TotalManuf.Int.
Whole period, 1995–1999
−0.336*** −0.290*** −0.321*** −0.098
Pre-cumulation, 1995–1996 −0.504*** −0.451*** −0.421**
Post-cumulation, 1997–1999 −0.408*** −0.394*** −0.460*** −0.364*** −0.322*** −0.449***
(0.122) (0.088) (0.091)(0.107)
Notes: Standard errors are in parentheses and *, **, and *** denote statistical significance at the 10%, 5% and
1% levels respectively. The equation estimated is given by Equation (A5) in Appendix 1, where the estimations
include importer and exporter dummies as well as year dummies. All adjusted R2 are between 0.83 and 0.89.
RULES OF ORIGIN597
as well as over two sub-periods – pre- and post-1997. The full set of results for the
1995–99 regressions are available in Table A5 of the Appendix. Here we are con-
cerned not only with the relative magnitude of this ROO coefficient but also with how
that magnitude may vary over time, and also between different categories of trade.
As before those categories of trade are total, manufacturing and intermediate goods.
The first aspect to note is that the magnitudes of the coefficients are overall similar
to those reported in Table 5. Secondly there is some evidence in the latter period for
the lack of cumulation being more significant for intermediate trade than for either
manufacturing or total trade. This applies both to the pre-cumulation coefficient, ROO,
as well as to ROO+TD where we may also be picking up on some trade diversion.
Thirdly, there is some evidence to suggest that lack of cumulation appeared to impact
significantly more on total trade and manufacturing trade between the spokes over
1995–96, than over 1997–99. Over the entire period, for the ROO variable, the results
indicate that trade between non-cumulating spokes was between 25% and 29% lower.
The largest impact of the lack of cumulation on intermediates is over 1997–99, where
trade was between 37% lower. Similar results can be seen when looking at the combined
impact of lack of cumulation and trade diversion, though the results are not signifi-
cant for all periods. If we take the latter period, the coefficients suggest that trade was
lower by between 28% and 36% with the largest impact on intermediate goods trade.
4.2.2. Bottom line.
regression analysis. They suggest that the lack of cumulation can indeed significantly
impact on trade between spokes and potentially by a substantial amount. The size of
the coefficients suggests a potentially substantial impact of lack of cumulation and
hence of rules of origin.
It is worth underlining that the ROO dummy captures the extent to which trade
between the flows included is significantly less than the remaining flows, where we
have controlled for all other reasons that might reasonably be expected to have an
impact on trade flows. Of those other reasons notably we have included all other
preferential trading agreements. As these tend to increase trade between partner
countries, the inclusion of these dummies serves to increase the size of the negative
coefficient on the ROO dummy. The coefficient can therefore be interpreted as saying
that, relative to those spokes, which have preferential trading agreements between
themselves, the trade of the remaining spokes is much lower. A key reason for this is
likely to be rules of origin, and the lack of cumulation of rules of origin. As our
variables are dummies, it is of course possible that there may be other explanations
for this, which we have not been able to capture, and it may also be that the results
suffer from the problem of endogeneity. Hence, rather than the lack of cumulation
resulting in lower levels of trade between the spokes, it could be that where countries
trade little with each other there is little reason to offer cumulation.11
The results in Tables 4–6 strongly support the previous panel
11 See also Cadot et al. (2004), for a discussion and treatment of the issue of endogeneity in the context of NAFTA.
598PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
There are several reasons, however, why we would argue that this is unlikely to be
a major issue in our analysis. The first is that the EU’s choice of bilateral trading
partners was not driven by the nature of spoke-spoke relationships. Hence with
regard to spoke-spoke trade, there is little a priori reason to suppose that these are
inherently lower than other bilateral trade flows. In a similar vein, the choice of
PECS countries was not selective. The choice of PECS members was driven by the
desire to increase trade between the current EU member states, the future EU member
states, the EFTA countries, and Turkey after 1999. Thirdly, the variation in the ROO
coefficient over time is consistent with the changing composition of the non-cumulating
countries. Fourthly, there is some evidence that lack of cumulation matters more for
intermediate goods trade, as would be expected from the theory. Finally, one would
expect that the impact of rules of origin would depend on the existing set of tariff
preferences, and we turn to this issue in the next sub-section.
We therefore view these results as providing clear evidence of the impact of lack of
cumulation on trade flows, but that the coefficients almost certainly provide an upper
bound on the effect.
4.3. Rules of origin and tariffs
Theory suggests that rules of origin may dampen trade between spokes while favouring
trade between the spokes and the hub; predictions that our empirical evidence tends
to confirm. The theory, however, can be used to generate more specific, empirically
testable predictions. As discussed in Section 2, tariff levels are like to have an impor-
tant impact. The point is that firms can always avoid the rules of origin by paying
the non-preferential tariff rate. This suggests two hypotheses that we can test. First,
if tariffs between the spokes are high, then bilateral trade is likely to be low in the
first instance, and hence the trade diverting impact of any ROO is likely to be lower.
Second, the level of tariffs may interact with cumulation; the negative impact of rules
of origin should be lower where tariffs are high – all other things equal. We have
information on tariffs so we can explore such interactions.
In Table 7 we report the coefficients of interest, namely those on our two ROO
dummies, ROO and ROO+TD, and the bilateral tariff variable. This gives the results
where we have added the tariff variable to the fixed-effect version of the model, and
we do this for total, manufacturing and intermediate goods trade for the years 1995
and 1999. The inclusion of the tariff variable impacts slightly on the ROO and
ROO+TD variables though not always in the same direction. The absolute magnitude
of the ROO coefficient is now larger in 1995, but smaller in 1999. It should be noted,
however, that the tariff variable is rarely statistically significant, and this was also
found when looking at other years. This probably reflects the difficulty of working
with average tariff variables in a gravity modelling framework.
Table 8 investigates the interaction between the height of the tariff and the distort-
ing impact of rules of origin. To test our second hypothesis, we divide each of our
RULES OF ORIGIN 599
ROO and ROO+TD dummy variables into two separate groups – ROOHIGH and
ROOLOW, and ROO+TDHIGH and ROO+TDLOW. ROOHIGH then includes all the cases
where tariffs are equal to or above a certain level (threshold), and ROOLOW all the cases
where tariffs are below that same level, and analogously for ROO+TDHIGH and
ROO+TDLOW. The objective is then to see if the negative impact of ROOs is lower
when tariffs are low by comparing the coefficients on ROOHIGH and ROOLOW. However,
as we do not know the appropriate level of the threshold we proceed by setting the
threshold tariff at two different levels, 5% and 10%, for 1999 and then compare the
Turning first to the results with a tariff threshold of 5%, we see that where tariffs
are low, the impact of the lack of cumulation between countries (ROOLOW) is higher
Table 7. Trade, cumulation and tariffs
Notes: Standard errors are in parentheses and *, **, and *** denote statistical significance at the 10%, 5% and
1% levels respectively. Estimation based on Equation (A4) in Appendix 1 with the inclusion of importer and
exporter dummies. The GDP and population variables are thus absorbed by these fixed-effects.
Table 8. The interaction between tariffs and cumulation – 1999
Threshold = 5%Threshold = 10%
Notes: Standard errors are in parentheses and *, **, and *** denote statistical significance at the 10%, 5% and
1% levels respectively. Estimation based on Equation (A4) in Appendix 1 with the inclusion of importer and
exporter dummies. The GDP and population variables are thus absorbed by these fixed-effects.
600 PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
than where tariffs are higher than the threshold (ROOHIGH). However, the coefficients
are not statistically significant. With regard to the ROO+TD, the differences are more
marked and consistent with what one would expect. All the coefficients are now
statistically significant. Looking at the results with a threshold of 10, we again see for
the ROO variable that the impact of the lack of cumulation appears higher for
intermediates – but once again the results are insignificant. For the ROO+TD variable,
the coefficients are statistically significant and once again where tariffs are low the
impact of the lack of cumulation is much greater.
These results suggest that the height of the tariff may indeed affect the impact
of the lack of cumulation – and that the higher the tariff, the smaller the impact.
However, it should be stressed that the evidence on this is mixed. This is no doubt
driven by the difficulties of working with average tariff rates, but also by the fact that
we are likely to be picking up on cross-country driven variations in tariff levels, which
makes it harder to disentangle the relationship between tariffs and lack of cumulation.
5. POLICY IMPLICATIONS
From a policy perspective there are two key issues that arise in considering the
implications of the empirical results in this paper.
Regional issues. Rules of origins are a necessary evil – every preferential trade
deal must have them to prevent tariff fraud – but how does one minimize the
Global issues. Taken in their entirety, preferential trading agreements may help
or hinder multilateral trade liberalization – there are serious proponents of both
views. But in either case, the design of rules of origin can influence the extent to
which the proliferation of regional trade agreements helps or hinders multilateral
5.1. Regional policy issues
Theory and empirical evidence tell us that ROOs distort trade. Why does this matter?
If a firm switches its suppliers to satisfy origin requirements, we can be confident that
this switch means using a higher-priced input – otherwise, the firm would have
already switched even before the ROO. Substituting high-cost inputs for low-cost
inputs lowers welfare by reducing efficiency.12
Moreover, analysts as august as Anne Krueger have suggested that the trade-
inhibiting impact of ROOs is not an accident, i.e. that these rules have been utilized
12 A good summary of the welfare effects of rules of origin under different assumptions can be found in Krishna (2005). There
is also some work which examines the circumstances under which restrictive rules of origin may be welfare increasing (e.g. Mussa
1984; Falvey and Reed 1998; Panagariya and Krishna 2002), the interaction between the welfare effects and the political
viability of a given FTA (Duttagupta and Panagariya 2002), as well as the impact on firm behaviour (Ju and Krishna 2005).
RULES OF ORIGIN601
for protectionist purposes. These points are identified in the EU’s summary report on
the future of rules of origin published in August 2004. The report concludes that
current rules of origin do not ‘fit economic reality’ because ‘they do not correspond
to the global production model of the market; they reflect past defensive policy aims,
they do not correspond to the new manufacturing and processing operations which
are currently taking place . . .’ (European Commission, 2004).
Our findings suggest that cumulation is one way of reducing the welfare costs of
rules of origin without diminishing their ability to defeat tariff fraud. Our empirical
analysis suggested that the introduction of cumulation of the PECS type raised trade
among the spoke economies by between 7.4% and 22.1%, and that the absence of
cumulation reduces trade substantially, with the lower and upper bounds of our
estimates being in the neighbourhood of 25% and 70% respectively. Calculating the
exact welfare effects of this new trade is not possible without extremely detailed
information on costs, technology and market structures for thousands of products and
dozens of nations, but we can be sure that the welfare gains are positive from revealed
preference arguments. If firms located in the spoke economies reacted to the cumu-
lation-induced relaxation of ROOs by buying more from other spoke economies, it
must be because doing so lowered their costs. This, in turn, must have led to some
combination of lower prices to customers and higher profits.
In theory, there could be an offsetting factor. The cumulation-induced favouring of
trade among spokes that belong to PECS could have come at the expense of these
nations’ purchases of lower priced inputs from nations outside the PECS. Here our
empirical results suggest that this was not the case. We identify an increase in trade
between these spokes and the rest of world. Admittedly, this part of our results is
somewhat muddied by the on-going trade liberalization process between Central
European nations and the rest of the world, but at least we can say that there is no
clear evidence that cumulation induced trade diversion with respect to non-PECS
Our first policy conclusion is thus that cumulation should be more widely applied.
From the perspective of the nations that have been involved in a lattice of trade deals,
cumulation is a good way of reducing the necessary evil inherent in rules of origin.
More specifically, expanding PECS membership to include the Euro-Mediterranean
countries would be good idea.
Cumulation, however, is not an easy policy option. All participating countries must
sign identical free trade agreements, and all parties must have identical rules of
origin. If this is not the case, something that might be called rules-of-origin fraud
13 The multilateral tariff cuts agreed in the WTO’s 1994 Uruguay Round were being phased in during our sample period, and
many of the Central European countries joined the WTO in the 1990s. As mentioned above, cumulation may have allowed
firms inside the PECS to increase their sourcing from RoW nations. The point is that since cumulation allowed them to count
more of their existing purchases within the PECS toward establishing origin, they may have been freer to expand inputs from
outside the PECS while still guarding the origin status of their final good.
602 PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
could arise. For example, if Country B confers originating status on imported inter-
mediates from Country C using laxer rules of origin than those that exist between
Country C and the EU, goods may be shipped from Country C to the EU via Country
B, thus circumventing the ROOs between Country C and the EU.14
Unifying rules of origin could be a very difficult task, given their level of detail and
their opacity. In the European case, however, it was relatively simple. In most trade
negotiation, where improved market access is the goal, the nation with the biggest
market has the largest negotiating power. Given this principle and the enormously
unbalanced commercial importance of the EU market versus the markets of the other
13 PECS members, it was clear that the EU15’s rules of origin would be the template,
even if some adjustments were made. Moreover, by the mid-1990s it was clear that
10 of the 13 would be members of the EU within a decade in any case, so for them
the PECS simply meant adopting the EU ROOs a few years early.
In other regions, unifying ROOs may be much harder, so cumulation cannot be
viewed as a panacea. Below we outline an alternative set of policies, which we argue
would be more flexible and more effective in minimizing the distorting impact of
rules of origin. We also argue that the alternative set of policies are more likely to
enable regional agreements to be compatible with multilateral liberalization.
5.2. Global policy issues
This issue of the impact of differential ROO regimes for the multilateral trading
environment is becoming particularly important given the proliferation of regional
trading agreements, and given the increasing fragmentation / vertical specialization
in production. The importance and policy relevance of ROOs is thus inevitably grow-
ing. Indeed the EU’s Green Paper referred to earlier suggests that rules of origin are
‘useful only in so far as they reflect the actual conditions of production and trade and
the needs that the preferential arrangements are designed to satisfy. However, the
present framework for determining, managing and verifying preferential origin no
longer seems wholly adapted either to these needs or to quantitative and qualitative
trends in the international economy’ (European Commission, 2003, p. 4).
The ‘present framework’ referred to above is currently not very substantial. Under
the WTO, discussions have tended to focus on non-preferential rules of origin, i.e. those
to be applied in determining origin for non-preferential reasons. WTO members
agreed in principle to harmonize these non-preferential rules by 1998 but in reality
the process is still ongoing. There is only a very brief treatment of preferential rules
of origin and there is no agreement on cumulation provisions, although in 2002 the
possible implications of rules of origin were identified within the WTO as one of the
14 Moreover it enables easier access to the EU than that granted to other third countries despite the fact that Country C is not
part of the EU–B Preferential trade agreement (PTA). It is this form of non-compliance with WTO provisions that has
concerned some WTO members.
RULES OF ORIGIN603
key ‘systemic’ issues that needs consideration. A key issue here is the consistency of
rules of origin regimes with WTO provisions. Under Article 24 any preferential trade
agreement (PTA) should not raise ‘duties and other regulations of commerce’ against
non-PTA members. Clearly rules of origin are capable of so doing, and our results
suggest that this may well de facto be the case. Nevertheless, discussions continue in
the WTO as to whether ROOs should be considered as an ‘other regulation of
commerce’ (ORC) (WTO, 2002b). Our view is that quite clearly ROOs can be used
to impact upon trade flows, and thus strictly speaking should be considered as an
ORC. Constraining rules of origin would then not be WTO compliant. However,
except in extreme cases, this is of course extremely difficult to prove.
5.3. ROOs: A way forward?
So the question that arises concerns the possibility and viability of drawing up a
framework for rules of origin, which would minimize any further distortionary impact,
and prevent rules of origin regimes from reinforcing PTAs as ‘stumbling blocks’ as
opposed to ‘stepping stones’ to multilateral liberalization. Are there effective ways
of preventing the spaghetti bowl? In order to achieve this we suggest that there are
certain key principles that ought to be borne in mind. These are: transparency,
administrative simplicity, flexibility and negotiability.
First, any system should be as transparent as possible. If ROOs can be distortionary
then it is desirable for the extent of that distortion to be as clearly understood and
evident as possible. The greater the transparency, the easier it is to identify whether
the rules are constraining. ROOs, which involve combinations of criteria, are thus
more complex. Transparency is also clearly important for the administration of origin
regimes. Indeed the second key principle is concerned with administrative simplicity. By
their very nature ROOs are relatively complex to administer and hence the costs are
also likely to be much higher. Complex administrative arrangements, together with
lack of transparency, merely raise those costs further as well as increasing the incen-
tives for fraud and circumvention of those rules. Thirdly, the rules need to be flexible
and this is important for three reasons. Flexibility with regard to differential ROOs
is important in a world of increasingly overlapping PTAs, as agreement, under the
auspices of the WTO on a harmonized set of preferential ROOs is unrealistic. Flex-
ibility is also important as rigid rules, such as requiring a specific production process,
which do not easily allow for changing economic circumstances and conditions are
likely to become more distortionary over time. Finally, it is important to have flexibility
in order to enable rules to be ex-post negotiable. Negotiability is the last and possibly
most important. If the objective is to minimize barriers to trade then introducing
systems which make it difficult to negotiate reductions in those barriers are surely to
be avoided. Unlike tariffs, rules of origin once introduced do not easily form part of
the future trade liberalization agenda. There are several possible reasons for this, but
if there is little flexibility in the rules than inevitably negotiation is difficult. For
604 PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
example, if the underlying rule concerns a change in the tariff classification then
any relaxation of this rule implies either no rule, or switching to a product-specific or
value-added type rule.
Of course these are heuristic principles applicable in many other contexts. The
point of detailing them, however, is twofold. First, to emphasize how far away from
these principles the current framework for rules of origin is. Currently that framework
is opaque, extremely complex, inflexible, and difficult to negotiate over. A partial
exception to this is diagonal cumulation. It is negotiable because there may be some
flexibility with respect to which non-partner countries can be considered as part of
the cumulation arrangements. However, strictly speaking that flexibility can only
arise where the countries have identical FTA tariff and administrative provisions.
In practice, therefore, diagonal cumulation is hard to achieve across overlapping PTA
arrangements. The second reason for detailing them is to highlight how an alterna-
tive framework could meet those principles and thus be desirable. That alternative
framework is based on three stages:
The widespread application of the value content rule, as opposed to the change
of tariff classification rule or the product process rule. Any rule is to a large degree
arbitrary. However, the value content rule is much more transparent, more flexible,
and consequently more negotiable.
The use of a value-added tariff rule in determining tariffs, if any, to be levied.
This is a proposal first made by Lloyd (1993). The principle is that the tariff is
levied in proportion to the amount of non-originating inputs. For example, sup-
pose the EU signs a PTA with Country B, where Country B used non-originating
intermediates which comprised 60% of the final price of the good. The good
would thus be subject to the export tariff (on the final good) weighted by the 60%
share of non-originating intermediates. Hence if the tariff were 10%, the tariff
levied would be 6%. In the original Lloyd formulation tariffs would be paid on
any portion of the non-originating intermediate inputs. However, this rule could
easily be combined with a minimum value-added rule, which confers originating
status. Clearly such a rule could also have an impact on trade flows as producers
still need to weigh up the costs of importing intermediates upon which a tariff
could be levied, and sourcing the intermediate from within the free trade area.
However, failure to meet the minimum originating requirement no longer results
in such a binary penalty system, thus giving producers greater incentive to source
their intermediates from the cheapest suppliers.
The introduction of full as opposed to diagonal cumulation (see Appendix 3 for
a more detailed discussion of the differences between diagonal and full cumula-
tion). With diagonal cumulation countries are required to have identical rules of
origin and identical PTAs. This does not apply to full cumulation with the appli-
cation of a value-added tariff. It is entirely possible for Countries B and C to have
a different minimum value content rule, which confers originating status, to that
RULES OF ORIGIN 605
between either B or C and the EU. Ultimately, whether a tariff or not is levied
on export to the EU will depend on the relevant proportions of value added from
the different suppliers.
The system outlined here is much more transparent, flexible and inherently nego-
tiable. Importantly it is likely to both minimize the distortionary impact of ROOs as
well as deal with the multilateral problems arising from the increasingly overlapping
nature of regional trade agreements, and thus meet the objectives outlined earlier.
Inevitably it is not without its drawbacks. Possible problems associated with the value-
added rule concern fluctuations in import prices arising from, e.g., exchange rate
fluctuations. A value content rule may also discourage local final goods producers
from reducing their costs as this then raises the proportion of intermediate inputs.
These are valid concerns. However, first, they are of much more concern in a frame-
work where failure to meet the origin requirement results in the full tariff being
imposed. This would not be the case here. Secondly, it is possible to introduce some
flexibility into the rules of origin analogous to the current system of tolerance clauses.
There is also the question of administrative simplicity. However, the level of docu-
mentation is in principle identical to that already required. In order to be granted
originating status producers already have to show the proportions of value added
derived from domestic production and that from imported intermediates.
The key elements of the above are first switching to a value-content rule, secondly
agreeing to the application of a value-content tariff rule, and thirdly allowing for full
cumulation. To the extent that constraining rules of origin may not be WTO-
compliant this would equally apply with any equivalent value-content rule. Hence the
first element on its own may not impact on this. Equally, employing the first element
alone would not, per se, prevent PTAs as stumbling blocks as opposed to stepping
stones. It is the application of the second and third elements respectively, which would
go a long way to meeting those concerns. This, however, could not be achieved
without the first element.
It is clear from the literature that there is no consensus as to which of the origin
rules – change of tariff classification, specific production processes, value-content – is
superior. All commentators appear to agree that each of these has merits and demerits
respectively. It is important to note, however, that the merits and demerits are invar-
iably considered in the context of a given PTA, as opposed to in the context of a
system of overlapping PTAs. Given the clear advantages of the value-content rule in
the latter context, it is our view that countries should endeavour to utilize the value-
content rule as widely as possible. This should apply to new agreements but also to
amending existing agreements. Indeed, were Article 24 to be amended such that this
were a requirement then this would be an important step towards making regional
trading arrangements greater stepping stones towards multilateral liberalization. Ideally,
too, the second of the above elements should also be incorporated, as it would then make
it much less likely that rules of origin would raise barriers against third countries.
606 PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
6. CONCLUDING REMARKS
The empirical results in this paper can be viewed as a contribution to two distinct
literatures. First, it provides a new line of evidence in the growing body of work that
demonstrates that rules of origin do indeed affect trade flows. For some time there
has been considerable, though admittedly casual, empirical evidence which suggests
that policy makers and firms see existing rules of origin as very substantial constraints
on their choice of intermediate inputs. The earlier empirical literature relied on
hand-compiled, qualitative indices measuring the restrictiveness of rules of origin.
This direct approach has many merits – it gives the researcher a direct measure of
the protectionist content of ROOs and thereby allows direct estimation of the trade-
distorting effects of ROOs. However, the qualitative indices are unavoidably arbitrary,
and since the compliers cannot possibly know all the technical and commercial realities
in each product covered by a particular ROO, these indices may over- or under-
represent the true restrictiveness. Our approach is indirect and does not allow us to
capture the full trade-distorting effect of ROOs, but since it is based on a natural
experiment, it does not involved constructed data.
Second, in the context of the EU’s trading relations our results strongly suggest that
rules of origin serve to restrict trade between the EU’s partners and those partners’
trade with third countries. In particular there are three features of both the panel
data and cross-section regression results which lend strong support to the conclusion
that it is the lack of cumulation, and hence the restrictiveness of the rules of origin
themselves, which the coefficients are identifying. First, the panel results clearly indicate
that trade between newly cumulating spokes rose after the introduction of cumulation.
This dynamic impact, which we have identified, is central to our results. Secondly,
with regard to the cross-section and panel country fixed-effect results the change in
the coefficient over time is consistent with the change in cumulation arrangements
between countries across the years. Thirdly we have shown that there is some weak
evidence of interaction between tariffs and cumulation, which suggests that rules of
origin may be de facto less constraining in the presence of high tariffs.
There are several key conclusions emerging from the empirical work presented
in this paper. First, that the introduction of the PECS system of cumulation appeared
to directly increase trade between the EU’s spokes, and that increase in trade was of
the order of 22%. Secondly, that the lack of cumulation of ROOs, between the EU’s
spokes, could serve to impede trade by between 25% and 70%, depending on the time
period and trade flows being considered. Thirdly, and consistently with the theory,
there is some evidence to suggest that the lack of cumulation is more important with
regard to intermediate trade than manufacturing trade. Fourthly, the results suggest
that there is some evidence that the higher the tariffs the smaller the impact of
cumulation. Again, this is consistent with the underlying theory.
Of course, there are a number of limitations with this analysis. One key limitation
concerns the aggregate nature of the analysis. To the extent that rules of origin and
RULES OF ORIGIN607
their cumulation matter then this is likely to be at the individual industry or product
level and is likely to affect countries differentially. This is clearly part of the future
research agenda. We view this paper as an important step forward in understanding
the impact of rules of origin but clearly there is a need for considerably more work
in this area. There is a need for a much clearer understanding of which types of ROO
regimes are more likely to be trade distorting, and the circumstances under which
rules of origin are likely to be constraining. There is also a need for more disaggregated
analyses at both the level of industry and country aggregation.
The last section of this paper addressed the policy implications raised by the
empirical analysis. The key issues raised here concern the optimality of rules of origin
both with respect to any given PTA, and with respect to the multilateral trading
system. Our empirical evidence suggests that rules of origin are indeed distortionary
and thus contribute to the spaghetti bowl of overlapping trade agreements. Building
on this we have suggested a specific three-part procedure for establishing a multilateral
framework for rules of origin which would be more transparent, flexible, administra-
tively feasible, and negotiable. In so doing the rules would be far more compatible
with the multilateral world trading system than is currently the case.
University of Paris 1, Pantheon-Sorbonne, PSE and CEPR 20
This paper’s main message is that rules of origins are not simply a technical nuisance
that only trade lawyers and trade negotiators should study. All economists interested
in international trade issues should become aware of the crucial role of rules of
origins in a world where regionalism and vertical specialization are becoming
more important. From this point of view, I believe this is a very successful paper.
The authors find indeed a large and negative effect of rules of origins on interna-
tional trade. By the authors’ estimates, EU partners to whom the least generous
rules of origins system applies see their bilateral trade reduced by between 10% and
Compared to the nascent empirical literature on rules of origins, this paper inno-
vates by using a very specific characteristic of the system of cumulation of rules of
origin. This differs from the approach by Estevadeordal and Suominen (2004) who
construct indices of the restrictiveness of rules of origins and regress bilateral trade in
a gravity-type model on such indices. In this study, the approach is similar in the sense
that a gravity model is estimated to quantify the specific effect of rules of origin. Both
approaches have their advantages and drawbacks. The present one is less ad-hoc and more
transparent than one based on a constructed index. On the other hand, the impact
of the rules of origins is analysed only through a single (even though important)
608 PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
aspect of the rules of origin. Rules of origin and cumulation principles are complex,
and the authors do a good job explaining them. The authors exploit a crucial (even
though technical) difference between what is called the bilateral and diagonal cumulation
systems. Bilateral cumulation implies that a country which has signed a preferential
trade agreement (PTA) with the EU cannot cumulate its own value added with
intermediate inputs imported from another PTA EU partner, to determine origin
status for exports of goods to the EU. In the case of the diagonal cumulation system
it can. The implication for bilateral trade between the two EU partners (the spokes)
is that they will prefer to import intermediate goods from the EU (the hub). It may
also be that they will prefer to import those intermediate goods from the rest of the
world although the effect on trade between the EU PTA partners and the rest of the
world are not of an obvious sign.
The impact of rules of origins is of growing importance because it affects primarily
trade in intermediate goods which is the most dynamic part of global trade in the
recent past. It also means that they impact negatively the process of vertical special-
ization and the creation of global production networks. Hummels et al. (2001) show
that one-third of world export growth was due to vertical specialization between
1970 and 1990. Hence, the negative effect may not be only on bilateral trade
between two EU partners but may spill over to other countries if the consequence is
that it reduces the incentive to put in place global production networks with stages
of production in different countries. An important question that is not directly
addressed in the paper is whether these rules of origin are used as export subsidies
for EU firms producing intermediate goods. A recent paper by Cadot et al. (2004)
suggests that this is the case.
Another consequence of the cumulation system for rules of origin is that by reduc-
ing market access of the spokes to the hub, they impact on the decision to locate by
firms. Non-diagonal cumulation will favour location in the hub (the EU) rather than
the spokes. This will be inefficient for all as production will not take place in the least
costly location and it will be unfair to the spokes. Finally, we know that there is
hysteresis in industrial location especially when vertical linkages exist. This implies
that temporary policies on market access can have long lasting consequences for these
countries which may not be reversed once the policy is reversed.
An econometric issue that is not very clear in the paper is what the rules of origin
dummy exactly capture in the cross-section gravity equations. The dummy is unity if
the importing country has a PTA with the EU without diagonal cumulation with an
exporting country. It should be made clear that this dummy captures the effect of the
system of rules of origin on trade between the spokes of the EU PTA system but also
between those spokes individually and the rest of the world. Whereas the theoretical
effect of the rules of origins system is clearly negative on bilateral spokes trade, it is
ambiguous in the case of trade between the spokes and the rest of the world. Hence,
because the negative coefficient on the rules of origin dummy is actually an average
of the effect of rules of origin on trade with other spokes and with the rest of the
RULES OF ORIGIN 609
world, the interpretation is not fully obvious. It would have been interesting to
distinguish the effect on bilateral trade between the spokes and the hub (the EU), on
bilateral trade between the spokes and on bilateral trade between spokes and the rest
of the world. For this, interaction terms in the gravity equation would be required.
The negative impact of rules of origin on trade is very large in the cross-section
regressions (around −50%) especially because this is the effect of a single character-
istic of the rules of origin. The ‘hard to believe’ reaction to this number is reminiscent
of other coefficients in gravity equations, in particular the effect of monetary unions
There are several possible explanations for this large estimated effect. The first one
is that the effect of rules of origin is non-linear and that the dummy captures the
negative effect of the cumulation system not only on bilateral trade but also on trade
in groups of countries facing the same cumulation system and linked by potential
vertical linkages. One way to think of this non-linear effect is to reinterpret Yi’s model
(2003) of vertical specialization with different stages of production in different coun-
tries. In his model, Yi shows that the effect of trade barriers are magnified when they
apply at each stage of production because when they are too high, it becomes unprof-
itable to locate each stage of production in different countries. In this case, a trade
barrier between Country A and B has a negative effect not only on trade between A
and B but also on trade between B and C which produces another stage of produc-
tion. As rules of origin can clearly be interpreted as trade barriers for intermediate
goods, Yi’s model can apply here. If this magnification explains the large effect found
in the paper, however, most of the negative impact should be found in later stages of
production, where the effect should be at its maximum. The fact that the elasticity
rises over time may also be due to growth in vertical specialization during the same
period. Given that global production networks are progressively put in place during
the period, the elasticity of trade flows to the cumulation system may have risen over
An alternative interpretation of the large size of the negative effect is less generous
to the authors. Rules of origin are not exogenous. Reverse causality is possible and
the authors are aware of the problem. It is possible that sectors will not lobby hard
for diagonal cumulation with some countries if local producers import little from
The addition of pair-specific fixed-effects regressions eliminates the problem of
unobserved variables that correlate with both trade and the choice of cumulation
system. The effect is a bit smaller in this specification but still very large. The addition
of pair-specific effect does not, however, fully eliminate the problem of reverse causality.
When the authors apply difference-in-difference methods, estimates are much smaller
and not always significant. This suggests that reverse causality is indeed present and
that the negative effect may not be as large as 50%. It remains that the paper is very
convincing in showing that the cumulation system is a sensitive issue in a world where
vertical specialization is growing.
610 PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
The London School of Economics
This stimulating paper in the field of international trade offers three original contri-
butions: (1) It explains to a non-specialist audience how rules of origin work and what
forms they can take; (2) It provides an empirical evaluation of the differential effect
on trade flows of various forms of rules of origin (ROOs); (3) It discusses policy
implications and suggests a strategy for improving existing rules.
Part (1) is well-executed – the authors manage to clarify the present byzantine
system of ROOs – so I am going to focus my comments on Parts (2) and (3).
Part (1) and (2) together constitute an interesting contribution to the policy debate
on trade. As I will argue, Part (2) is vulnerable to a strong methodological critique.
I do not think that the authors can do much about it, except acknowledging this
weakness and warning readers. Despite this issue, the empirical part is extremely
useful in that it lays out some strong stylized facts. Part (3) is in my opinion highly
Part (2) is potentially of great practical interest. The paper uses both cross-
sectional data and panel data. The cross-sectional results indicate a very large
effect of differential ROOs. However, as the authors acknowledge, there is a difficult
causality issue similar to the one studied by Persson (2001) in his paper on currency
unions. There is no question that ROOs (as well as preferential trade agreements)
are endogenous. Countries that strongly desire favourable trade treatment from
the EU have several ways of influencing decision-making. We may therefore
suspect that countries with a strong trade potential manage to wring more favour-
able ROOs from the EU. If that is the case, the cross-sectional results are
biased upwards. As an outsider, common sense makes me think that the large
coefficients obtained here may be mostly a consequence of this problem (this is
confirmed by the fact that the panel data paint a very different picture – to be
There is no obvious way of fixing the problem. Using controls is useful, but I doubt
that the authors can convince readers that the existing controls take care of unob-
served differences across countries. The authors could look for instruments but those
are hard to find for cross-country data. Thus the cross-sectional results should be
interpreted as a stylized correlation only.
Panel evidence does not solve the reverse causality problem. It just shifts the
emphasis to differences across countries in terms of dynamic behaviour. For instance,
countries that are moving toward a market economy may be both more eager to
receive favourable ROOs and more likely to increase their trade level. However, in
this case I suppose most people would find panel evidence more persuasive than
RULES OF ORIGIN611
I have serious doubts about the normative analysis of ROOs presented here. ROOs
exist in order to make sure that countries outside a trade agreement do not benefit from
the trade agreement. I really do not see how we can discuss the optimality of ROOs
without knowing why the EU wants to exclude certain countries from free trade.
Let me approach this from another angle. The EU has a free trade agreement with
Tunisia but not one with Chad. We impose ROOs on Tunisia to make sure that Chad
is left out. Why exactly do we want Chad to be left out? This is not an idle question
but it determines what kind of ROOs between Chad and Tunisia are optimal. Perhaps
we want to ‘punish’ Chad for some political reasons, in which case a horribly strict
and mechanical rule is optimal. Perhaps we want to protect a particular European
industry, in which case we could tailor the ROO to block only a certain kind of good.
The authors propose four common-sense criteria to design optimal ROOs. However,
as the example of Tunisia and Chad shows, it is impossible to map these criteria to
more general objectives of the EU trade policy. As long as we take the criteria as
general heuristic principles of good government, we learn very little (because they
apply to cheese production as much as ROOs). As soon as we try to translate them
into practical recommendations, we run into the problem that we have absolutely no
idea how these recommendations actually fit with the more general objectives.
The authors could have addressed this in two ways. They could have taken the bull
by the horns and built a full model of the EU trade policy. ROOs would then be
determined together with the other trade policy variables. This is a worthy goal (and
probably a major contribution in trade theory), but it is probably too much to ask for
the present publication. Alternatively, the authors could have dropped the normative
section as such. They could have a couple of pages of conclusions in which they
provide heuristic arguments as to why the practical guidelines they suggest may be
an improvement compared to the present situation. Perhaps, it could be presented as
a topic for policy debate rather than a list of recipes.
Jonathan Haskel wonders about the economic significance of the results and in
particular the difference between the OLS cross-section and difference-in-differences
results, with the former showing a much larger effect. He suspects that the reported
results (for instance 40% in Table 4) are too large and that it might be the result of
an endogeneity problem. The authors respond that 40% is not out of line with what
industry people expect.
Jaume Ventura thinks that the large increases in trade when the cumulation agree-
ment is in place might be due to one spoke country taking advantage of another’s
better developed distribution network to the large economy. So that it is not about
trade in the good as much as trade of service to other spoke countries.
612 PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
Jaume Ventura questions whether the presence of cumulation in Eastern Europe
and lack of it in the southern Mediterranean countries reflects different characteris-
tics of each group of countries. The authors disagreed, arguing that the two groups
were so different, and noting that their Finger-Kreinin indices measuring export
similarity were similar and that there was lack of cumulation not just between but
across countries; moreover, they control directly for this in the difference-in-difference
regressions and still find a positive, although smaller, effect.
Mary Amiti recognizes the importance of intermediate goods in the results and
asks how they are defined, as there is conflict in the literature regarding the appro-
priate definition. The author acknowledges the difficulty in appropriately classifying
goods between intermediate and finished products, and responds that he used the
BEC classifications. She asks whether the authors had tested whether the difference
between the coefficients for manufacturing and intermediate goods was significant.
She suggests putting them both in a single regression with an interactive term.
Tullio Jappelli thinks that the OLS estimates are not needed, that only the panel
data results should be reported and that other coefficients (not just those that vary
over time) should be reported in the paper table. He asks what the value added is
from running the panel with and without fixed effects. He wonders that if he writes
the equations out for the panel and difference-in-differences estimations if the second
is not just a more restrictive version of the first.
Paul Seabright endorses Andrea Prat’s call for more attention to the political
economy of the rules of origin but disagrees with her characterization of exclusion of
particular countries being important. He thinks that free trade agreements are gen-
erally negotiated to benefit the countries involved, rather than to exclude particular
countries; the excluded countries are merely residuals from the process. Hans-Werner
Sinn agrees on the importance of the normative discussion and asks for the authors
to be clearer regarding the motives for having free trade agreements in the first place
in light of incentives to restrict trade from a national welfare perspective and pressure
from domestic industry lobbies. The authors responded to the call for more normative
discussion by saying that the information just isn’t there. In discussions with trade
policy makers they have yet to find anyone who knows the origin of diagonal cumu-
lation. They do know that within the EU, it was the EU textile industry which pushed
for diagonal cumulation with the Mediterranean countries, indicating that they may
have originated from industry lobbying. However, industries were very influential in
drafting the NAFTA, yet it contains no provisions for diagonal cumulation.
APPENDIX 1: THE GRAVITY MODEL
A typical equation derived from the gravity literature takes the form:
RULES OF ORIGIN 613
where Xij represents the value of exports by i to j (in 1000s of dollars); Yi, Yj and Yw
levels of production in countries i and j, and the world; and Φ(·) is a term capturing
trade costs between countries. The estimating equation describes bilateral aggregate
trade flows between two countries, as a function of their respective levels of GDP,
and the distance and/or trade costs between them. Typically, the model is then aug-
mented with the respective populations of each country as well as a range of dummy
variables, e.g. to capture common language, or membership of a PTA. Hence imports
by country i from country j, are typically expressed as:
Ln(Xij) = α0 + α1Ln(GDPi) + α2Ln(Popi) + α3Ln(GDPj) +
α4Ln(Popj) + α5Ln(Distij) + α6Z + eij
where Xij is the value of exports by country j to country i; GDPk is the GDP of country
k, (k = i,j); Popk is the population of country k (k = i,j); Distij is the distance between
the economic centres of gravity; Z is the set of dummy variables; and eij is the error
term and where the standard assumptions apply.
Following Anderson and Wincoop (2003) more recent work has included country-
specific fixed effects (e.g., Matyàs 1997; Hummels 1999; Redding and Venables 2001).
These are designed to capture what Anderson and Wincoop term multilateral trade
resistance. In a cross-section framework the introduction of importing and exporting
country dummies results in collinearity with the GDP and population variables. In
practice, therefore, researchers tend to rewrite Equation (A1), by taking the activity
variables to the left hand side resulting in:
In terms of Equation (A2), the logarithmic version of Equation (A3) implies unitary
restrictions on the parameters α1 and α2.
In our work we have added to the standard gravity model in order to evaluate the
potential impact of the cumulation of rules of origin as well as allowing for the
inclusion of a tariff term. Hence the extended version of the gravity model equation
Ln(Xij) = α0 + α1Ln(GDPi) + α2Ln(Popi) + α3Ln(GDPj) +
α4Ln(Popj) + α5Ln(Distij) + α6PTAij + α7Borderij +
α8Languageij + α9Tariffij + α10ROOij + eij
where the following are the relevant dummy variables: PTAij represents the relevant
free trade agreements (EU, CEFTA and EFTA); Borderij assesses the potential role
of a common border between countries; Languageij assesses the potential role of a
common language between countries; Tariffij gives the bilateral MFN or preferential
average tariffs between countries; ROOij gives the rules of origin dummy variable as
The panel model correspondingly takes the following form:
614PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
Ln(Xijt) = CountryFixedEffects + γt + α1Ln(GDPit) + α2Ln(Popit) +
α3Ln(GDPjt) + α4Ln(Popjt) + α5Ln(Distij) + α6PTAijt +
α7Borderij + α8Languageij + α9ROOijt + eijt
where Xijt is the real bilateral export from i to j in period t, and GDPit and GDPjt are
the real GDPs of i and j and γt are year dummies. For the Panel-Anderson-Wincoop
estimates, the CountryFixedEffects are the importer and exporter country dummies 3i
and 3j. In the estimates with country-pair fixed-effects, the CountryFixedEffects are 3ij
and all time-invariant variables like Distij, Borderij and Languageij are absorbed by these
APPENDIX 2: DATA
Our estimations are based on trade flows between 38 countries – all of the EU
countries, three EFTA countries (Iceland, Norway and Switzerland), the CEFTA
countries, the Baltic States (Estonia, Lithuania and Latvia), six countries taking part
in the Barcelona process (Turkey, Jordan, Israel, Egypt, Tunisia, Morocco), as well as
the United States, Canada, China, Japan and Australia, and were carried out on the
basis of total trade, manufacturing trade, and intermediate goods trade for the years
1995–99. The underlying sources of data were as follows. Trade data were derived
from the UN COMTRADE databank. The values for exports, measured in US
dollars, are deflated using the US GDP deflator for the panel estimates. GDPs in
current and constant US$ are from the WDI 2003 database. Data on population are
obtained from the IMF International Financial Statistics CD-ROM. ‘Great Circle
Distances’ are calculated from data on latitudes and longitudes of capital cities
available at www.wcrl.ars.uda.gov/cec/java/capitals.htm. Intermediate goods trade
was derived by aggregating the trade flows at the 2-digit HS level on the basis of the
BEC classification of industries. The MFN and preferential tariff rates (simple average)
are derived from TRAINS – UNCTAD.
When considering diagonal cumulation, one is considering the relationship between
three countries or country groupings: the exporting country, the importing country,
and those countries which are part of the system of diagonal cumulation (in this case
the PECS). Given this three-part relationship the ROO dummy takes a value of 1,
if the importing country has a preferential trading agreement with the EU without
diagonal cumulated rules of origin with the exporting country, and a value of 0
otherwise. In the regressions themselves we then distinguish between two different ROO
dummies, and this is done slightly differently for the panel difference-in-difference
approach, and the cross-section and panel Andersen and Wincoop approaches.
With regard to all of these, however, we start with an aggregate ROO which is then
appropriately decomposed. For the panel difference-in-difference estimations also the
control groups are defined with respect to the aggregate ROO. Table A1 therefore
gives the precise details for this aggregate ROO as well as indicating which flows were
RULES OF ORIGIN 615
Table A1. Determination of ‘aggregate ROO’ dummy by classification of
bilateral trade flows
Aggregate ROO Dummy
123 95 96 9798 99
Hub-Hub trade flows
Hub-Spoke trade flows (both directions)
EU CEFTA + Balts, EFTA**
Med (w/o Turkey) Turkey
Spoke-Spoke trade flows benefiting from PECS (treatment group, Turkey from ’99)
CEFTA + Balts CEFTA + Balts
CEFTA + BaltsEFTA
CEFTA + BaltsTurkey
EFTACEFTA + Balts
Turkey CEFTA + Balts
Spoke-Spoke trade flows NOT benefiting from PECS
Turkey Med (w/o Turkey)
CEFTA + Balts Med (w/o Turkey)
Switzerland Norway + Iceland
Norway + IcelandSwitzerland
EFTAMed (w/o Turkey)
SloveniaMed (w/o Turkey)
Med (w/o Turkey)CEFTA + Balts
Med (w/o Turkey)EFTA
Med (w/o Turkey) Med (w/o Turkey)
Med (w/o Turkey) Turkey
Med (w/o Turkey)Slovenia
Trade flows involving exports to Rest of World (RoW)
RW CEFTA, Balts, EFTA, Med
Trade flows involving exports from Rest of World (RoW)
CEFTA + Balts RW
** Full cumulation is currently operated by the European Economic Area (EEA comprises the Community,
Iceland, Liechtenstein and Norway). These countries apply full cumulation between themselves and diagonal
cumulation with other Pan-European countries. In our empirical test, we do not distinguish between full and
616 PATRICIA AUGIER, MICHAEL GASIOREK AND CHARLES LAI-TONG
included in the three control groupings used. Table A2 then details the decomposition
into ROOImpact and ROOImpact+TD which is used in the difference-in-difference regressions,
and Table A3 the decomposition into ROO and ROO+TD used in the remaining
The determination of ROOImpact and ROOImpact+TD
With the difference-in-difference approach, we are measuring the impact of the intro-
duction of the PECS system. In the first instance therefore we take 1–ROOij, where
ROOij is as defined in Table A1, and we then distinguish between ROOImpact and
ROOImpact+TD. The details are given in Table A2.
The decomposition of the aggregate ROO dummy into ROO and ROO+TD
In the cross-section and panel Andersen and Wincoop regressions, we show the
extent to which trade flows are lower between non-cumulating countries. However,
as well as picking up on the supply switching effect of the ROOs it is also possible
that, for certain trade flows, the aggregate ROO dummy is picking up on trade
diversion. We therefore decompose the aggregate ROO into two components:
ROO+TD and ROO. The former captures those trade flows where the conflation of
supply switching and trade diversion may be possible – i.e. where the importing
countries have a symmetric agreement with the EU but where there is no agreement
between i and j. The latter then captures those trade flows which should be picking
up on the pure cumulation effect.
APPENDIX 3: CUMULATION
As mentioned in the text, the PECS involves so-called diagonal cumulation, which is
one of the three main forms of cumulation. This appendix explains the differences in
Three types of cumulation are:
Bilateral. Between two trading partners. Materials originating in one country are
considered as originating in the other partner country (and vice versa). All PTAs
allow for bilateral cumulation.
Diagonal. Between three or more trading partners linked by FTAs with identical
rules of origin. Materials ‘originating’ in one country are also considered as
originating in all the other countries.
Full or total. Between three or more countries, but with more flexibility than diagonal
cumulation. Allows intermediate processing to be split in any way between the
parties to the PTA provided that when added together all the materials/processing
used throughout the area are sufficient to meet the origin rule.
RULES OF ORIGIN617 Download full-text
Table A2. ROOImpact and ROOImpact+TD dummies used in the panel regressions
(difference-in-difference) by classification of bilateral trade flows
Exporting countries (i)ROO Impact ROO Impact+TD
9596 97989995 96 9798 99
Hub-Hub trade flows
Hub-Spoke trade flows (both directions)
EUCEFTA + Balts, EFTA**
Med (w/o Turkey) Turkey
Spoke-Spoke trade flows benefiting from PECS (treatment group, Turkey from ’99)
CEFTA 98* CEFTA 98
Turkey CEFTA 98
Spoke-Spoke trade flows NOT benefiting from PECS
Turkey Med (w/o Turkey)
CEFTA + BaltsMed (w/o Turkey)
Switzerland Norway + Iceland
Norway + IcelandSwitzerland
EFTAMed (w/o Turkey)
SloveniaMed (w/o Turkey)
Med (w/o Turkey)CEFTA + Balts
Med (w/o Turkey)EFTA
Med (w/o Turkey)Med (w/o Turkey)
Med (w/o Turkey) Turkey
Med (w/o Turkey)Slovenia
Trade flows involving exports to Rest of World (RoW)
RW CEFTA, Balts, EFTA,