Production Networks and Rules of Origin:

Moving From NAFTA to USMCA

Alejandra López Espino
March 17, 2024

read the latest version here

Abstract
Free Trade Agreements (FTAs) give firms within the member countries duty-free access
to each other’s markets. But with a catch. Non-bloc workers and capitalists are the primary
beneficiaries if these firms rely heavily on suppliers outside the bloc for their upstream inputs.
So to limit this diversion of factor demand, FTAs include elaborate rules—known as Rules of
origin (RoOs)—that dictate how much of a product’s value must be created within the bloc for
it to be “in compliance,” i.e., enjoy duty-free access to member country markets.
Because value chains are often lengthy, it is difficult to discern which firms RoOs favor, which
firms they penalize, and by how much. I use comprehensive value-added tax records from Mexico
to shed new light on these issues. In particular, I study the effects of the North American Free
Trade Agreement (NAFTA) and its replacement, the United States-Mexico-Canada Agreement
(USMCA), on the Mexican automotive sector.
Four main findings emerge. First, Mexican auto sector value chains exhibit strong intercon-
nectedness, with 30 percent of firms serving ten or more assemblers and contributing to a third of
the transaction volume. Second, when the USMCA replaced NAFTA, car parts producers were
the most affected group within the value chain. These firms experienced a threefold decrease
in compliance rates compared to car assemblers. Third, however, the steep increase in regional
content requirements (RCRs) was alleviated by a new ”super-core roll-up” provision, which
allowed some producers to round up their suppliers’ domestic content to 100 percent. Finally,
if the super-core roll-up provision had not been implemented, the fraction of Mexican parts
producers qualifying for duty-free treatment within the bloc would have halved. Instead, the
qualifying fraction dropped by only 14 percentage points.

1 Introduction

In recent decades, Free Trade Agreements (FTAs) have become increasingly popular.1 These

agreements facilitate trade between the member countries by granting preferential tariffs to goods

produced within them. But as production chains have become increasingly globalized, implementing

these FTAs has become increasingly complicated. Large chunks of the FTA documents are now
1 See Maggi (2014)[22], for a comprehensive review of the economics of Free Trade Agreement (FTA)s, their motives,

design, and evaluation.

1
devoted to codifying the conditions under which a good is considered originary, i.e., produced inside

the FTA territory. These conditions, known as rules of origin (RoOs), play a crucial role in preventing

trade circumvention and attracting economic activity to the FTA region.

The complexity of these rules, which vary significantly across trade deals and industries, has

presented a challenge for governments, who often design them without sufficient guidance on their

overall economic impact. As highlighted by Kniahi and de Melo (2022)[18], this complexity has made

it difficult for researchers to study the effects of Rules of Origin (RoOs) and hindered governments’

ability to regulate trade effectively. Accordingly, in this paper, I develop a methodology that allows

governments to assess the incidence of their RoOs and explore the impact of alternative policy

regimes. I implement this methodology using value-added tax records from Mexico, which provide

transactions-level information on each firm’s entire value chain. To my knowledge, I am the first to

use this type of data to analyze RoOs.

A case study of Mexico

I focus my analysis on the Mexican auto sector for three reasons. First, automotive production

in North America provides a compelling case study of global value chains (GVCs) and effectively

demonstrates the impact of RoOs on value chains (VCs). Throughout NAFTA’s 26 years, the

automotive GVC became intricately integrated. Subsequently, with the replacement of NAFTA

by USMCA in January 2020, there was a significant rise in the regional value content (RVC)

requirements. These modifications have facilitated an insightful natural experiment, allowing us to

evaluate compliance under both regulatory frameworks while isolating these effects from adjustments

in firms’ sourcing strategies. As a result, this approach allows for a deeper understanding of how

RoOs impact origin compliance and RVC while also enabling the isolation of these effects from

adjustments in firms’ sourcing strategies.

Second, as evidenced by its prominence in both NAFTA and the recently negotiated USMCA,

the automotive sector holds significant importance in Mexico’s trade relationships. It accounts for

7 out of 10 articles within the RoOs chapter of the USMCA and 90 out of 270 pages in the RoOs

chapter outline specific rules for the automotive industry, providing further proof of its significance.

2
 Finally, as the seventh-largest automobile producer in the world and the largest in Latin America2 ,

Mexico’s auto sector contributes significantly to the country’s economy. It represents 18% of

manufacturing GDP and 3.6% of national GDP. In addition, Mexico is a crucial player in the

international auto market, being a prominent exporter of automotive products: approximately 32%

of the country’s manufacturing exports are of automobile products. Moreover, Mexico is the fourth

largest auto-parts exporter globally and the top US supplier.

Contributions

My analysis yields three main contributions. The first is a comprehensive dataset on the Mexican

production network, specifically focusing on the auto sector in 2017. A unique aspect of this dataset

is the inclusion of Mexican firm-to-firm connections inferred from value-added tax (VAT) records.

By combining them with customs data and other complementary sources, I recover the domestic

connections of Mexican firms and their connections abroad. While previous studies have utilized VAT

data to establish firm-to-firm connections in countries like Belgium, Japan, Chile, and Costa Rica,3

this is the first paper to employ a similar approach for Mexico. Using these data, I characterize the

value chain of 13 auto assemblers in Mexico, revealing the significant interconnectivity across their

value chains for the first time. Remarkably, most firms serve at least one assembler. Additionally,

such well-established suppliers also possess substantial shares of the trade volume. For instance,

firms that serve at least five assemblers exchange more than half of the transaction volume.

My second contribution is the creation of an open-source origin calculator that evaluates a firm’s

RoO compliance based on the product it exports, its inputs’ origin, and its production costs. This

calculator constitutes the most detailed representation of RoOs for North America in the literature, a

valuable tool for any researcher that studies RoOs in NAFTA or USMCA. I outline the methodology

for developing this calculator in Section 4.

My third contribution regards the insights I elicit by combining the created dataset and calculator.

I assess the potential differences in compliance if the USMCA’s RoOs were applied during the sample

year instead of the RoOs from NAFTA, which were in force at the time. I estimate a 9.65% decrease
2 These rankings are in terms of units.
3 See Dhyne et al. 2021 [9]; Kikkawa et al. 2020[17], Bernard et al. 2016[4]; Furusawa et al. 2017[13], Huneeus 2018
[15], Alfaro-Ureña et al. 2018 [1], to mention a few.

3
in assemblers’ compliance rate under USMCA but, most notably, the estimated reduction for auto

part producers is 18.33%.

My findings regarding a controversial interpretation of the USMCA’s RoOs are of particular

interest. In 2022, the United States unilaterally decided to implement a different method for

calculating the RVC of super-core parts in the production of passenger vehicles or light trucks.

However, the Dispute Settlement Panel rejected their method, which did not allow core parts

producers to round up (”roll up”) the regional factor content of their inputs. I show that if the

dispute settlement panel had ruled in favor of the US’ interpretation, the compliance rate would have

decreased by 50%, in contrast to the estimated 18% decrease when the super-core roll-up is allowed.4

In sum, this paper offers valuable insights into the intricate features of RoOs, uncovering the

complexities and interplays of compliance rates and RCRs along the value chain, focusing on the

context of the transition from NAFTA to USMCA regulations.

Relation to the literature

My work contributes to and builds upon the large empirical literature on FTAs. The most closely

related papers study the same region and agreements. In this group, Conconi (2018) [6] and Head

et al. (2022) [14] stand out, especially the latter, which also analyzes the effects of the USMCA

on Mexican auto firms. Its methodology and objectives are somewhat different, however. In their

project, inputs are viewed as a continuum, analyzing the impact of the new agreement on sourcing

strategies. The approach implicitly assumes that inputs obtained within the region are exclusively

made using regional resources.5 In contrast, I treat firms as a discrete finite set and the value

chain as a fixed graph, focusing on patterns of compliance along the entire upstream value chain.

Given these differences, our results are not directly comparable. Nevertheless, in Section 6.2, I

show how compliance is severely underestimated when non-RVC components of RoOs are omitted,

when the upstream compliance is not accounted in RVC calculations, and when we assume a flat

RCR across all product codes. Conconi et al. (2018)[6], on the other hand, explore the impact
4 Notably, the implications of cumulation in RoOs have also been investigated by Bombarda and Gamberoni

(2013)[5].
5While I relax this implicit assumption, it entails the introduction of other simplifying assumptions. I aim to

transparently outline these differences and similarities between our approaches in the subsequent sections, recognizing
the potential for learning from the complementarity of our perspectives.

4
of RoOs within FTAs on trade patterns of intermediate goods. Their study aims to uncover the

implications of these rules on sourcing decisions and trade flows. Specifically, the authors concentrate

on NAFTA’s RoOs defined by changes in tariff classification, also known as tariff shift, rather than

RCR requirements. By leveraging variations across countries and products, they demonstrate that

NAFTA’s RoOs incentivized a redirection of intermediate imports from third-party countries towards

NAFTA partners.

Several other empirical studies merit mention. In particular, Kniahin & de Melo(2022)[18],

provides a detailed characterization of RoOs, including provisions on regional value, changes in

tariff classifications, cumulation, as well as product-specific allowances and exceptions. However,

specific provisions related to specific processes are not studied due to a lack of detail in the available

data. Freund (2019)[12] and Febelmayr et al. (2019)[11] highlight the need for improvements in

the design of RoOs. Additionally, Yang (2022)[30] has shed light on the ambiguous effects of RoOs,

showing that more stringent requirements often lead to lower regional content. If regional content

requirements are too high, RoOs can have consequences opposite to those intended, effectively acting

as barriers to trade. However, these papers do not consider another deterrent to compliance: the

complexity of RoOs, which makes it difficult for businesses to comply and results in higher costs and

time delays. This can particularly affect small and medium-sized firms, as they may not have the

resources to navigate the complex rules and may choose to avoid utilizing FTAs altogether. Evidence

from Colombia, as presented by Krishna et al. (2022)[20], supports this notion. They demonstrate

that meeting RoO requirements increases production costs and imposes significant fixed costs in

the form of documentation expenses. Additionally, their findings indicate that RoOs can negatively

impact competition, favoring larger and more experienced firms more likely to utilize preferential

benefits.

One feature of modern supply chains is that they exhibit high levels of specialization. Therefore,

suppliers often customize their products to suit specific downstream uses. Thus, increasing trade

barriers between two countries with integrated GVC imposes significant and heterogeneous costs on

all firms in the FTA territory and in other countries. (See De Gortari 2019 [8], Antràs and de Gortari

2019[3]). I aim to further our understanding of the effects of RoO along the value chain, specifically

focusing on the Mexican automobile industry and the transition from NAFTA to USMCA.

5
 While my work is empirical, it is related to the theoretical literature analyzing the economic

effects of RoOs. Notable papers include Rosellón (2000)[25], Falvey & Reed (2002)[10], Mukunoki

(2017)[23], Krishna (2005)[19], Ju & Krishna (2005)[16], and Tsirekidze (2021)[28], who provide

valuable insights into various aspects of RoOs.

The papers by Ju and Krishna[16] and Krishna[19] examine the impact of RoOs in FTAs. Ju and

Krishna identify contrasting effects of RoO, with different impacts on imports of intermediate and

final goods, affecting the welfare of producers inside and outside the FTA. Both studies contribute to

understanding how RoO affect trade behavior and market access, shedding light on the complexities

exporters face in navigating trade agreements.

Rosellón (2000)[25] demonstrates that a stricter rule of origin can boost the demand for domestic

factors if the substitution effect outweighs the effects of reduced operational scale and output

reallocation. It advocates for policy decisions on RoOs that aim to maximize welfare and foster

domestic technological advancement by aligning the interests of stakeholders in the production chain.

Falvey & Reed (2002)[10] delve into the complexities of applying RoOs, emphasizing the challenges

and variations in determining product origin. They highlight the potential discrepancies in origin

determinations within the same country, influenced by the underlying objectives of the implemented

laws. The paper underscores the supportive role of RoOs alongside other policy instruments,

elucidating their impact on trade dynamics and the potential consequences for importing nations.

Mukunoki (2017)[23] explores how the formation of a FTA alters the patterns of Foreign Direct

Investment (FDI) by non-FTA country firms, particularly in connection to the restrictiveness of

RoOs. The study outlines potential effects such as deterring outside firms’ FDIs, the substitution

of efficient firms with less proficient ones, or the elimination of pre-existing FDIs before the FTA

conclusion, complicating the welfare assessment of FTAs and potentially reducing consumer surplus.

Tsirekidze (2021)[28] provides a game theory perspective showing the essential role of RoOs in

achieving global free trade. In the symmetric countries case, global free trade stability diminishes as

countries engage more in GVCs, primarily due to free riding issues. In asymmetric countries settings,

RoOs can stabilize global free trade by curbing the benefits countries derive from others’ free trade

agreements, underscoring the crucial role of RoOs in realizing global free trade.

One last strand of literature related to this project pertains to the study of boolean networks—a

6
specific type of network modeling that proves highly useful when dealing with many components

and complex interactions. Boolean networks are widely employed in Physics, Biology, and Computer

Science (Saadatpour & Albert (2013) [26] and Schwab et al. (2020)[27] are examples of this

literature) to model intricate systems, where the state of each node (in this case, firms) is determined

by other variables in the network, such as the origin of inputs (i.e., upstream RoO compliance).

Network modeling, with components represented as nodes (firms) and interactions denoted as edges

(input purchases), represents a powerful method for structurally analyzing and modeling complex

systems. Deciphering the structure and interactions within these networks lays the groundwork for

understanding the system’s overall behavior—in this instance, the Mexican auto VC.

In summary, treating the transition from NAFTA to USMCA as a natural experiment, this paper

offers valuable insights into the intricacies of RoOs by uncovering the complexities and interplays of

compliance rates and RCRs along the value chain.

Structure of the paper

The remaining sections of this paper are structured as follows: Section 2 provides an overview of the

key concepts utilized in this paper to establish a solid foundation for the analysis. Section 3 outlines

the methodology employed to identify and examine the Mexican auto value chain (VC), offering novel

insights into its structure and dynamics. Section 4 presents the methodology adopted to develop

the origin calculator, which is a crucial tool in evaluating the effects of RoOs (RoOs). Section 5

investigates the changes in RoOs from the North American Free Trade Agreement (NAFTA) to the

United States-Mexico-Canada Agreement (USMCA), shedding light on the modifications and their

potential implications. Section 6 carries out an empirical exercise, unveiling the main findings and

outcomes derived from the analysis. Finally, in Section 7, I conclude the paper and present our

closing remarks, summarizing the key takeaways and implications of this research.

2 Rules of Origin (RoOs)

In this section, we establish the key definitions needed to understand the methodology employed.

We also introduce the terminology used to classify RoOs based on their structure. These definitions

7
have been sourced from two primary references: the NAFTA and USMCA documents, and the Rules

of Origin Facilitator (ROF) project. The latter is a collaborative initiative between the International

Trade Centre (ITC), the World Customs Organization (WCO), and the World Trade Organization

(WTO).6

Free Trade Agreements aim to promote international trade by eliminating barriers and extending

preferential access to markets among member countries. These agreements typically cover trade in

goods, services, and investment provisions. In addition, they may also include customs cooperation,

trade facilitation, harmonization of standards, and regulatory cooperation measures. To ensure

fair trade among member countries, preferential rules of origin are applied, preventing non-member

countries from taking advantage of preferential tariffs without offering reciprocal benefits.

Harmonized System (HS) Classification, also known as the HS Nomenclature, is an interna-

tional customs classification system created by the WCO. It assigns unique 6-digit codes to different

groups of products. These codes help customs authorities identify products and determine the

appropriate import duty, taxes, and trade measures. Understanding the HS classification is necessary

to determine the applicable RoOs under any FTA, as different commodity codes carry different rules.

The 6-digit codes of the classification scheme simultaneously represent chapters (the first 2 digits),

headings (the first 4 digits), and subheadings (the full 6 digits). In the case of USMCA partners,

these codes are further disaggregated into 8-digit items, also known as commodity codes or national

tariff lines (NTLs).

Rules of origin (RoOs) To determine eligibility for preferential tariffs offered under a trade

agreement, goods must comply with a set of criteria known as rules of origin (RoOs). These criteria

are used to establish whether a good is considered to have originated in the region of the trade

agreement. RoOs are specific to each product and are typically based on the HS classification scheme.

This means that every HS code eligible for preferential tariff under a trade agreement has its own

unique rule of origin. These rules can be set at different levels, ranging from an entire chapter to a
6 To ensure conciseness and relevance, some details have been excluded if they are not directly relevant to NAFTA

or USMCA. For a complete understanding of the concepts discussed in the following sections, please consult the official
website of the ROF at findrulesoforigin.org.

8
very specific product within an HS or NTL. RoOs are negotiated and included as part of the main

agreement between trading partners, usually in the form of a protocol or annex. Because RoOs are

specific to each FTA, they can vary significantly across agreements.

In an FTA region, the origin(s) of a given good can be determined in two ways. The first way

by which a good would be considered “originary” to (i.e., originating from) a region is if the good

were obtained from entirely within that one single region. Examples in this category are livestock,

agricultural products, or extracted minerals. Note that goods in this group are “originary” only if

produced without the addition of any non-originary materials. The second way for determining that

a good is “originary” to a given FTA region is if the good undergoes a substantial transformation

within that region. In the auto sector, all RoOs fall into the latter category. This latter type of RoO

is explained in the following subsection.

Substantial Transformation refers to a type of RoO that requires a good to go through a specific

process in order to be considered originary to a particular country. There are three main avenues

by which substantial transformation is considered to occur. The first is known as change in tariff

classification (CTC), which requires that the final good be classified under a tariff classification group

that differs from the classification group(s) of all non-originating materials used in the production

process. The classification group can be a chapter, a heading, a subheading, or an item, which

corresponds to the HS codes at two, four, six, and eight digits, respectively. Consider a situation

where a manufacturer produces tractor spark plugs with the NTL code 85111002. The rule concerning

these spark plugs requires a change in the classification subheading from any other subheading. Put

plainly, for these tractor spark plugs to be recognized as originating products, their non-originating

inputs should not be categorized under subheading 851110. This means the producer may source

originary spark plugs from subheading 851110, but non-originary components in the manufacturing

process cannot fall under this category for the tractor spark plugs to meet the requirements.

The second avenue is the RVC calculation, which requires that a percentage of the good’s total

value must have been added within the FTA region. When substantial transformation conditions

are defined in this way, they are called regional content requirements (RCRs). Lastly, the specific

processing (SP) rule stipulates that a particular process must be carried out at a certain stage of

9
production in order for the good to be considered originary to where this processing took place.

For a given product, a RoO can be defined using a single type of substantial transformation or a

combination of them (e.g., CTC and RVC) in determining its origin. Furthermore, there may be

exceptions and allowances within each type, specifying certain conditions for particular products

and permitting relaxations under special circumstances. Section 4 explains in detail how these three

types are combined in NAFTA and the USMCA.

De minimis. This provision allows for the use of a small amount of non-originating materials in

the production of a good without impacting its origin status. Essentially, this acts as a relaxation for

strict rules of origin. The threshold for both NAFTA and the USMCA is set at 10%.

Roll-up. Under this provision, a part or an intermediate material that gains originating status

under a FTA is considered 100% originating contingent upon further processing. This applies even if

non-originating inputs were used to produce the part or the intermediate material. Essentially, this

provision means that the value of non-originating materials used in the production of a good get

disregarded and is another way of making RoOs more lenient. If roll-up is not allowed, firms are

required to monitor the non-original content fraction for each input.

Net cost refers to the total cost of production minus any cost lost to sales promotion, marketing,

after-sales service, royalties, shipping, packing, and non-allowable interest that may be included in

the total cost.

Transaction value means the customs value, as determined in accordance with the Customs

Valuation Agreement; that is, the price actually paid or payable for a good or material.

Regional value content. In the case of NAFTA and the USMCA’s HS chapter designation for

automobile products, the regional value content can be calculated following two methods:

• Net cost
NC − V NM
RV C = × 100 (1)
NC

where N C is the net cost of the good, and V N M is the value of non-originating materials.

10
 • Transaction Value
TV − V NM
RV C = × 100 (2)
TV

where T V is the transaction value of the good.

In the automotive industry, the majority of product-specific RoOs only permit the use of the net

cost method. Therefore, in our origin calculator, we will exclude the transaction value method.

3 Identifying the Value Chain (VC)

The identification of the Mexican VC marks a significant and original contribution of this paper. To

this end, we conceptualize the world economy as a production network, where a directed weighted

graph is used to represent the relationships among firms. In this network, the nodes represent

individual firms, the edges symbolize input purchases, and the weight of each edge indicates the

trade volume between the buyer and the seller. Each firm specializes in producing a distinct product,

allowing us to visualize the VCs as paths or sets of paths within this production network.

Although one may feel tempted to think of VCs linearly, VCs look more like a tree where the

root node represents the final good’s producer, and all edges point towards the root instead of away
7
from it. A VC can be classified as global when it extends across multiple countries, highlighting

the interconnectedness of economies on a global scale. Lastly, we can view the production network as

the union of all the VCs within the economy. These VCs overlap and intertwine, creating intricate

patterns, such as cliques and cycles, and giving rise to complex interactions within the network.

3.1 Identifying auto assemblers

Our initial dataset captures a comprehensive network that illustrates interactions among all Mexican

manufacturing firms, including their domestic exchanges and interactions with foreign firms. The

foreign interactions are consolidated at the country-product level, ensuring that a thorough under-

standing of domestic manufacturing activity is maintained. It is important to note that this dataset

represents a cross-section of the network and specifically covers the year 2017, a period during which
7 Pol Antràs famously referred to these structures as snakes and spiders, respectively, in his 2018 Ohlin Lecture. [2]

11
NAFTA was in effect. Four different sources of information were combined to construct this dataset.

The first source is VAT receipts that contain connections between Mexican firms. The second source

is the firms’ annual tax declarations, which provide details about their sales revenue, production costs,

administrative costs, profits, and main activity. The third source is customs records that provide

information about imports and exports at the product level. I have linked these three datasets

using anonymized firm identifiers provided by the Secretarı́a de Hacienda y Crédito Público (SHCP,

Secretariat of Finance and Public Credit). Lastly, the Mexican Economic Census of 2019, conducted

by the Instituto Nacional de Estadı́stica y Geografı́a (INEGI, National Statistics and Geography

Institute), provides information about firms’ production labor. For a more in-depth description of

the data, please refer to appendix A in the current paper.

I begin by identifying the nodes of auto assemblers within the dataset. Initially, we focus on firms

with North American Industry Classification System (NAICS) codes beginning with 3361, the number

sequence that corresponds to “Motor Vehicle Manufacturing.” From this category, we further analyze

two subcategories: 336110, corresponding to “Automobiles and Pickup Trucks Manufacturing,” and

336120, which corresponds to “Truck and Truck Tractors Manufacturing.” A total of 61, 028 firms

produce goods that fall into these activity codes. From within this set of auto assemblers, in order

to conduct a comprehensive analysis of compliance under different sets of RoOs, we specifically

concentrate on those assemblers that export at least a portion of their production. We exclude firms

classified as retailers as opposed to producers. When exporting, companies are required to report

the NTL code and provide a description of the goods being shipped, which assists us in validating

these firms as assemblers. To avoid the inclusion of any firms misclassified as assemblers, we only

consider those that exported more than 500 units in 2017. This criterion appears reasonable, as the

company just above this threshold exports 732 vehicles, while the one just below exports only 93

vehicles. Applying this filtering process, we are left with a total of 13 auto assemblers, 10 of which

are engaged in light-duty vehicle production, together accounting for 58.77% of the light vehicle

exports reported in the official statistics for 2017. The remaining 3 assemblers produce heavy-duty

vehicles and contribute to 45.91% of 2018’s exports in the heavy truck segment.8 It is important
8 Source: INEGI’s Registro Administrativo de la Industria Automotriz de Vehı́culos Ligeros (RAIAVL, Administrative

Registry of the Automotive Industry for Light Vehicles) and Registro Administrativo de la Industria Automotriz de
Vehı́culos Pesados (RAIAVP, Administrative Registry of the Automotive Industry for Heavy Vehicles).

12
to note that heavy vehicle statistics are only available for 2018 onward. However, when comparing

the light vehicle production figures for 2018 with those of 2017, they are nearly identical, with the

former representing 99.63% of the latter. table 10 shows the assemblers’ exports by NTL code.

Table 1: Identified Assemblers

Assembler Vehicle Type Exports Avg. Price Units

Alpha light 5,738.13 14,516.51 483,154

Beta light 5,569.27 22,364.24 308,810

Gamma light 4,533.09 24,659.12 232,011

Delta light 3,371.91 15,931.40 194,483

Epsilon light 2,199.79 14,635.18 174,174

Zeta light 3,178.02 24,520.22 169,828

Theta light 907.26 12,325.55 87,206

Iota light 2,515.52 59,277.46 73,792

Kappa light 1,274.90 30,175.20 65,328

Lambda heavy 5,202.56 70,124.87 55,302

Mu heavy 772.60 131,664.97 9,579

Nu heavy 370.11 100,136.57 3,629

Xi light 44.79 69,273.66 732

Note: Exports are reported in million dollars.

After pinning down the assemblers, the next step is to trace back their VC. This involves recording

the assemblers’ providers, their providers’ providers, and so on. Throughout the remainder of this

section, we will refer to an illustrative example of a production network depicted in Figure 1. Figure 2

provides a visual representation of the steps for identifying each VC, via the methods used in this

analysis.

13
 Figure 1: An example of a production network

While tracing an assembler’s VC might seem straightforward in principle, two factors contribute

to the task’s complexity. Firstly, the sheer size of the data complicates the endeavor; even after

filtering the dataset to include only manufacturing firms and their connections, there are still 174,261

nodes and 3,043,287 edges. Manually tracing each VC is not feasible due to scalability issues. To

address this challenge, this study relies on the concept of an “in-component”. An in-component in

a directed graph refers to the set of vertices that can be reached by following directed edges in an

inward direction from a given “root” vertex — in this case, an assembler. Thus, the in-component of

an assembler represents the subset of firms that are part of the assembler’s VC. In Panel 2 of Figure 2,

the highlighted orange nodes indicate the firms within the in-component of vertex a. Note that firm

k is excluded from this set, as it is part of firm j’s downstream but is not part of the VC of assembler

a. Component-finding algorithms are widely used in computer science and readily available in various

programming languages. They are known for their efficiency, as they determine the components of

a finite graph in linear time in terms of its number of nodes and edges, using either breadth-first

search (BFS) or depth-first search (DFS) approaches. These algorithms start at a specified node

14
 9
(root) and examine its edges, initiating a new search whenever an unlabeled node is encountered.

Since these search algorithms were designed to recognize tree structures in graphs, and since trees

are inherently directed and acyclic, labeling the in-component of a VC alone cannot fully trace the

VC’s entirety.10 The search for edges will overlook cycles and thus cannot be considered exhaustive.

I use Python’s graph-tool library[24] for labeling each assembler’s in-component as well as for the

remaining steps in this section.

This study’s approach is first to reduce the network by retaining only the nodes that belong to

the in-component of at least one assembler.

9 In BFS, the search covers all nodes within a given level of depth before progressing to the next level. In DFS, the

search extends as far as possible along each branch before initiating a new search.
10 See Figure 26 for examples of search algorithms and how they perform in our production network example.

15
 (1) Identified assembler (2) In-component

(3) Sprouted cycles (4) Topological sort

Figure 2: Value Chain (VC) trace-back procedure

The presence of cycles in production networks poses our second challenge, for several reasons.

Firstly, cycles can complicate our analysis and interpretation of the VC, in that they create inter-

dependencies among firms that can be difficult to disentangle and understand. This can lead to

feedback loops and self-reinforcing effects, which hinder our ability to identify the unique impacts of

different RoO on compliance. Hence, addressing the presence of cycles is crucial for our analysis.

Previous work in the computer science literature has addressed this issue through various strategies

and techniques. One commonly-used approach involves aggregating a network into a directed acyclic

graph (DAG), transforming cyclic dependencies into sequential ones. This aggregation process

16
includes identifying and eliminating cycles by collapsing interconnected nodes, thereby simplifying

the network structure. For instance, Coscia (2018)[7] introduced a method that analyzes strongly

connected components (SCCs) within a network, revealing the network’s underlying hierarchical

structure. While these strategies are useful in many applications, the interpretation of a collapsed

cycle in the context of RoO remains unclear. For this study, I propose an alternative approach better

suited to our analysis: instead of removing or collapsing cycles, we unfold them by identifying the

edge within the cycle that is furthest from the root, duplicating the source node, and transferring the

edge origin to the duplicated node. These are the pink edges in Figure 2’s panels (2) and (3), and

the duplicated nodes are labeled a∗ , b∗ , and c∗ in panel (3). This method yields a tree representation

of the value chain, where the node copies function as leaf nodes. Given this characteristic, I have

termed this approach sprouting. appendix C describes the algorithm used to back-trace and sprout

the VC simultaneously.

Since we cannot observe the sourcing decisions of foreign firms, back-tracing along a path stops

whenever a foreign node is reached. Consequently, any nodes located upstream from this point are

excluded from our analysis. The algorithm stops whenever all edges collected as part of the VC

either are foreign or have no supplying nodes.

After successfully identifying the nodes and edges within a given VC, our final step involves

arranging the nodes in sequential order, from the most upstream to the most downstream. This

ordering is crucial, as it directly impacts suppliers’ compliance, determining their origin and subse-

quently affecting their clients’ RVC. Consequently, it has a direct bearing on the RoO compliance

of the clients themselves. To accomplish this, we employ graph−tool’s topological sort function on

a modified version of the VC, where the edges have been reversed. The topological sort algorithm

establishes a linear ordering of the vertices in a DAG, ensuring that if an edge (u, v) is present in

the graph, u precedes v in the ordering. As we aim to evaluate origin in an upstream–downstream

direction, we need to reverse the direction of edges before sorting them, in order to ensure that

assemblers’ origin is assessed last. This step is depicted in Figure 2’s panel (4).

17
3.2 Auto VC Features

The high degree of interconnectedness in the VC is immediately apparent. Here, a firm is considered

to “serve” an assembler if that firm is part of the assembler’s in-component; meaning, it is positioned

upstream of the assembler. As shown in table 2, 93.38% of the VC firms serve more than one

assembler, with 80% of the firms serving 5 or more assemblers. Remarkably, 10% of the firms serve

nearly all assemblers. It is noteworthy that popular providers also hold significant trade volume

shares. For example, firms that serve at least 11 assemblers collectively account for 21% of the trade

volume in the VC.

Table 2: Firms present in multiple VCs

Assemblers Mean Sales Firms in Value Chain Trade Volume

Served Revenue Count % Cum. % % Cum. %

13 1 0.03 0.03 17.02 0.01 0.01

12 315 10.42 10.45 32.41 5.87 5.88

11 998 33.01 43.46 37.34 21.42 27.30

10 244 8.07 51.53 31.90 4.47 31.77

9 77 2.55 54.08 43.76 1.94 33.71

8 86 2.84 56.92 44.08 2.18 35.89

7 89 2.94 59.86 40.49 2.07 37.96

6 148 4.90 64.76 76.73 6.53 44.49

5 448 14.82 79.58 60.70 15.63 60.12

4 100 3.31 82.89 65.15 3.74 63.86

3 291 9.63 92.52 91.39 15.29 79.15

2 26 0.86 93.38 114.47 1.71 80.86

1 200 6.62 100.00 166.56 19.15 100.00

18
 Figure 3: VC Intersections

998
1000

Intersection
size 500
434
315
283
195 219
57 16
4 1 23 2 1 5 3 42 31 15 9 2 1 5 5 2 2 47 46 27 10 10 6 2 43 18 13 9 4 2 33 24 15 9 4 1 2 2 25 1
0
Γ
Μ
Λ
Κ
∆
Ε
Α
Ι
Ξ
19

Β
Ζ
Θ
Ν
10K 5K 0
VC size
(firm count)
 Figure 3 offers supporting evidence of interconnectedness in the VC. An intersection in this figure

is defined by a subset of assemblers, and it is displayed if at least one firm serves all assemblers in the

subset. Firms in an intersection are denoted by a black dot. If a firm is not part of an intersection,

we see a gray shadow instead. The bars on the y-axis represent the number of firms upstream of each

assembler, while the bars on the x-axis show the number of firms that are part of each intersection.

We can see that, although the intersection including all assemblers contains a single firm, the second

and third broadest intersections (in terms of the number of assemblers) are also among the most

populated (in terms of number of upstream firms). Additionally, many intersections have a size larger

than 10 firms.

Figure 4 illustrates the distribution of firms at each level of assemblers upstream. The x axis

displays the assemblers listed in table 1. The numbers in each cell indicate the firm count at a specific

VC level of a given assembler. The VC level indicates number of links in the shortest path connecting

an assembler to a given upstream firm. One notable observation is the emergence of a funnel pattern

within the VC; broadly speaking, the number of firms directly serving assemblers is smaller, followed

by a larger count at Levels 2 to 4, and then a decrease in the subsequent upstream layers.

Figure 4: Upstream Distribution by Source

17 13
Firm
16 672 Count
15 833 52 708 24 20000
14 792 32 786 169

13 975 52 35 1099 148

12 1037 52 32 168 1001 437 476

15000
11 1398 32 177 596 903 189 1056
VC Level

10 1518 242 369 994 1384 1476 1095

9 1624 990 798 1500 1925 1729 1384

8 2187 392 1129 1473 112 193 1743 2777 2160 1600 113 10000
7 2879 1520 2303 2446 595 1511 2127 3302 2630 2591 478 838

6 3743 3101 2972 2829 2004 2537 2761 3975 3193 2874 1708 2050

5 4988 4811 4041 4059 3453 4661 4656 5882 4010 2515 3128 3511
5000
4 6526 7358 5896 6281 5411 7251 6853 6873 5663 1841 5237 5191

3 8410 12901 8354 8896 9800 11993 8964 8172 7599 1689 9459 8685

2 7725 11667 201 11017 9917 18924 14784 10286 4277 8509 1020 20553 15307

1 1834 2244 947 2692 3300 4484 2667 2172 1153 1537 1084 4569 7056

Α Β Γ ∆ Ε Ζ Θ Ι Κ Λ Μ Ν Ξ
Assembler

20
 We segregate the distribution based on domestic and foreign provenance to delve deeper into

this characteristic. What becomes apparent is the the fact that this pattern persists. The funnel

shape likely signifies that these firms are vertically integrated, and the densely populated zones in

the VC possibly represent the juncture at which this integration begins. It is plausible that certain

tax IDs are assigned to parts manufacturing, while others are designated for assembly within the

same administration. Nonetheless, it is important to recognize that conclusive verification would

require additional data analysis or corroborative evidence beyond the existing dataset.

Figure 5: Upstream Distribution by Source

Trade 13
Trade
16 2
Volume 16 670 Volume
15 6 7 Share Share
15 827 52 708 17
14 14 4 11 12
0.05 14 778 28 775 157
13 13 5 51 33 0.5
13 962 52 30 1048 115
12 68 4 9 23 4 52
12 969 52 28 159 978 433 424
11 118 4 5 26 60 5 212
0.04 11 1280 28 172 570 843 184 844
0.4
10 226 6 10 40 133 29 147
10
VC Level

VC Level

1292 236 359 954 1251 1447 948

9 277 25 40 81 187 108 247
9 1347 965 758 1419 1738 1621 1137
8 275 3 83 68 3 129 511 292 333 0.03
0.3
8 1912 389 1046 1405 112 190 1614 2266 1868 1267 113
7 281 26 177 166 11 9 204 804 488 1031 1 11
7 2598 1494 2126 2280 584 1502 1923 2498 2142 1560 477 827
6 673 116 366 281 25 69 223 798 659 1083 8 33
0.02 6 3070 2985 2606 2548 1979 2468 2538 3177 2534 1791 1700 2017
0.2
5 1549 582 694 716 117 266 781 2347 963 1463 69 162
5 3439 4229 3347 3343 3336 4395 3875 3535 3047 1052 3059 3349
4 2861 2628 2004 2213 669 2155 2829 3630 2333 916 454 871
4 3665 4730 3892 4068 4742 5096 4024 3243 3330 925 4783 4320
3 5482 9005 4531 5371 4703 7669 5503 5906 4385 760 4141 4036
0.01 0.1
3 2928 3896 3823 3525 5097 4324 3461 2266 3214 929 5318 4649
2 6063 9574 8839 7642 15534 12143 8297 2930 6569 691 16725 12037
2 1662 2093 201 2178 2275 3390 2641 1989 1347 1940 329 3828 3270
1 1222 1645 65 1853 2679 3582 2442 1523 540 1282 111 3648 5797
1 612 599 882 839 621 902 225 649 613 255 973 921 1259

Α Β Γ ∆ Ε Ζ Θ Ι Κ Λ Μ Ν Ξ
Α Β Γ ∆ Ε Ζ Θ Ι Κ Λ Μ Ν Ξ
Assembler
Assembler

(a) Domestic (b) Foreign

Now that we have reviewed some key features of the automotive VC in Mexico, our next step

involves characterizing the products manufactured at different stages of the VC. This characterization

will aid in narrowing down the set of RoOs analyzed and coded into the origin calculator.

To this purpose, I consider an HS if it there is a firm in the VC having such HS code or if the HS

code is subject to an auto sector-specific RoO. table 3 provides an overview of the distribution of

tariff item codes (products) considered in this project. Notably, Columns 3 and 7 showcase the share

of trade volume in the VC, while Columns 4 and 8 highlight the share of Mexican exports to the US

and Canada by product category.11

11 It should be noted that the former pertains only to firms within the studied VC, whereas the latter encompasses

all Mexican exports, including those from firms not contained within the studied subset.

21
 Several insightful observations arise from this data. For instance, Column 8 indicates that 19%

of Mexican exports to its North American partners consist of products in the auto sector, with the

majority of these exports being parts. If we were to analyze the impact of RoO on Mexican firms

and we only considered their effect on auto assemblers, we would be significantly underestimating

their effects.

Furthermore, half of the auto parts exports are classified as super-core parts, which are crucial

components within the VC. Notably, all vehicle and super-core parts are subject to RoOs under either

NAFTA or the USMCA. This finding likely underscores the significance of the recent controversy

surrounding within-firm super-core roll-up.

An additional insight from Columns 3 and 7 is that the majority of products in the VC are

subject to RoOs, emphasizing the pervasive influence of RoOs across various stages of production.

Moreover, despite the predominance of non-auto firms within the automotive VC, they contribute a

modest 23% of the VC’s trade volume, as indicated in Columns 6–7.

Table 3: Product Distribution by Type

Not subject to RoOs Subject to RoOs

Product Product
Product type Product % VC % EX % Product % VC % EX %
Count Count

(1) (2) (3) (4) (5) (6) (7) (8)

Vehicles: - - - - 48 0.38 52.89 14.82

passenger vehicle - - - - 7 0.06 40.41 7.64

light truck - - - - 5 0.04 1.06 4.64

heavy truck - - - - 6 0.05 10.54 2.27

other vehicle - - - - 30 0.24 0.89 0.27

Parts: 1 0.01 0.00 0.01 436 3.48 23.58 18.89

super-core - - - - 159 1.27 15.20 9.48

other parts 1 0.01 0.00 0.01 280 2.24 8.37 9.41

Non-auto: 406 3.24 0.08 2.83 11632 92.89 23.44 25.84

22
4 Origin Calculator

This section describes the creation of an origin calculator that streamlines the process of assessing

firms’ RoO compliance. In what follows, we will use the terminology in table 4 to describe the

structure of the auto sector RoOs found in NAFTA and the USMCA.

Table 4: The building blocks present in NAFTA and the USMCA

Rule Definition
CC The originating status is conferred to a good classified in a different HS
chapter than the non-originating inputs.
CTH The originating status is conferred to a good classified in a different HS
heading than the non-originating inputs.
CTSH The originating status is conferred to a good classified in a different HS
subheading than the non-originating inputs.
CTI The originating status is conferred to a good classified in a different HS
tariff item than the non-originating inputs.
ALW The originating status is allowed to be conferred from non-originating inputs
of specific HS codes.
ECT The originating status cannot be conferred to a good if the non-originating
inputs are from HS codes listed under an exception.
RVC A good obtains originating status if a defined RVC percentage has been
reached.
SP A good originates in the country where a defined technical requirement, i.e.,
a specified working or processing, has taken place.

The terms in table 4 are explained in detail in Kniahin & de Melo (2022)[18], along with other

“building blocks” not present in the sector and agreements studied here.

The calculator has been created in two main steps and can assess compliance with NAFTA’s and

the USMCA’s RoO. These steps transform the PDF documents defining the RoO into an R function

that computes compliance.

First, I scrape each document and extract all product-specific RoO into a text file containing one

RoO per line. Leveraging the uniform format in RoO statements, I transform them into a structured

table. Specifically, I use phrases as markers and organize the statements into a spreadsheet. To

better illustrate, consider USMCA’s RoO for HS code 8701.10, which pertains to single-axle tractors:

“A change to a good of subheading 8701.10 from any other heading, provided there is a

23
 regional value content of not less than 60 percent under the net cost method.”

The statement consists of two parts: the first states that non-originary inputs must be from a

different HS heading; the second states that a regional value content of at least 60 percent is required.

Now consider USMCA’s RoO for subheading 8701.20:

“ A change to a good of subheading 8701.20 from any other heading, provided there is

a regional value content of not less than 70 percent under the net cost method. ”

Where the highlighted parts in the statement are identical to those for the subheading 8701.10,

there are 6 other RoO with this same pattern. Below, we can see the tokenized statements, where

phrases used as markers are highlighted, and the delimiter “|” separates each token:

“ A change to | a good of subheading 8701.10 | from any other heading, | provided there

is a regional value content | of not less than 60 percent | under the net cost method .”

“ A change to | a good of subheading 8701.20 | from any other heading, | provided there

is a regional value content | of not less than 70 percent | under the net cost method .”

I use the markers as patterns that let me introduce the delimiters in the intended position

using regular expressions. Regular expressions, or regex/regexp, are match patterns interpreted by
12
string-searching algorithms, often used in “find and replace” operations.

By tokenizing each statement and using regular expressions, we can summarize them as follows:

“CTC | 8701.10 | heading | VC | 60 | net cost”

“CTC | 8701.20 | heading | VC | 70 | net cost”

Then, using “|” as delimiters, I collect all RoO statements into a CSV file. Using R, we further

process and rearrange the information on each RoO statement.

Through this process, I identified 13 different RoO types for NAFTA and twenty for the USMCA.
Tables 5 and 6 summarize these findings.
12 See https://regexr.com for learning, building, and testing regular expressions.

24
 Table 5: NAFTA RoOs

rule RVC RCR comb CTC level ECT ALW count EX % VC %

1 CC & RVC60 yes 60.00 AND yes chapter no no 1 0 % 0.08 %
2 CC + ALW50 yes 50.00 · yes chapter no yes 5 0.14 % 0.79 %
3 CTH no · · yes heading no no 11 5.56e-05 % 2.8 %
4 CTH & RVC60 yes 60.00 AND yes heading no no 24 7.2 % 11 %
5 CTH & RVC62.5 yes 62.00 AND yes heading no no 27 43 % 43 %
6 CTH + ALW50 yes 50.00 · yes heading no yes 130 16 % 11 %
7 CTH + ALW60 yes 60.00 · yes heading no yes 9 2.1 % 0.65 %
8 CTH + ECT no · · yes heading yes no 55 2.6 % 1.2 %
9 CTH + ECT50 yes 50.00 · yes heading yes no 60 8.4 % 2.1 %
10 CTH + ECT50 yes 50.00 · yes heading no no 11 2 % 2.1 %
11 CTH | RVC50 yes 50.00 OR yes heading no no 26 3.5 % 2.7 %
12 CTH | RVC60 yes 60.00 OR yes heading no no 38 8 % 7.6 %
13 CTSH + ALW50 yes 50.00 · yes heading no yes 2 6.6 % 1.4 %
Note: Column EX % shows the share of Mexican auto exports to the US and Canada for which a given
RoO applies. Similarly, Column VC % shows the automotive VC trade volume share.

The names of these rule types follow a pattern: following the terminology in table 4, if an RoO

requires a change in classification, the name start with CC, CTH, CTSH, or CTI, indicating whether

the change is required at the chapter, heading, subheading or tariff item-level, respectively. Similarly,

if it requires a value-added calculation, the name will start with “RVC” followed by 2 digits indicating

the RCR. If the RoO contains both an RVC and a CTC component, these components are separated

by “&” or “|” to indicate how these rules are combined.

25
 Table 6: USMCA RoOs

rule RVC RCR comb CTC level ECT ALW count EX % VC %

1 CC | RVC70 yes 70.00 OR yes chapter no no 1 0.085 % ·
2 CTH & RVC50 yes 50.00 AND yes heading no no 2 1.57e-04 % 0%
3 CTH & RVC60 yes 60.00 AND yes heading no no 17 0.12 % 11 %
4 CTH & RVC62.5 yes 62.00 AND yes heading no no 6 0.069 % 43 %
5 CTH & RVC70 yes 70.00 AND yes heading no no 4 7.3 % ·
6 CTH & RVC75 yes 75.00 AND yes heading no no 19 40 % ·
7 CTH + ALW70 yes 70.00 - yes heading no yes 32 4.3 % ·
8 CTH + ECT no - - yes heading yes no 44 1 % 1.2 %
9 CTH + ECT65 yes 65.00 - yes heading yes no 66 3.9 % ·
10 CTH + ECT70 yes 70.00 - yes heading yes no 8 3.8 % ·
11 CTH | RVC65 yes 65.00 OR yes heading no no 45 6.4 % ·
12 CTH | RVC70 yes 70.00 OR yes heading no no 45 4.7 % ·
13 CTH | RVC75 yes 75.00 OR yes heading no no 38 8.2 % ·
14 CTI no - - yes heading no no 14 0.27 % ·
15 CTSH no - - yes heading no no 27 1.9 % 0%
16 CTSH + ECT no - - yes heading yes no 5 1.5 % ·
17 CTSH + ECT65 yes 65.00 - yes heading yes no 2 6.8 % ·
18 CTSH + ECT70 yes 70.00 - yes heading yes no 11 0.25 % ·
19 CTSH | RVC75 yes 75.00 OR yes heading yes no 1 0.24 % ·
20 RVC75 yes 75.00 - yes - no no 64 9.2 % ·
Note: Column EX % shows the share of Mexican auto exports to the US and Canada for which a given
RoO applies. Similarly, Column VC % shows the automotive VC trade volume share.

In addition, certain rule types include the “ECT” term to indicate exceptions to the CTC rule.

However, these types also include a caveat; if a listed input falls under the exception, a good may

still be considered originary if its RVC is above the RCR indicated by the 2 digits next to it. If the

RoO just has “ECT” with no RCR, it means a good will not be considered originary if it falls under

the exceptions, regardless of its RVC.

There are some instances wherein a rule includes details that cannot be observed in the data. For

example, it may include different RCRs for two goods within the same NTL code.13 See appendix E

for a list of simplifying assumptions in programming the calculator.

The RoO types in Tables 5 and 6 can be further aggregated into groups to simplify the calculator

algorithm. Such groups are shown in Tables 6a and 6b.

13 In 2020, the Mexican government published the Ley de los Impuestos Generales de Importación y de Exportación

(LIGIE, Import and Export General Tax Law), which adds two new digits to the NTL code called Número de
Identificación Comercial (NICO, Commercial Identification Number). Such an addition will enable the removal of
most simplifications in future calculator implementations.

26
 Figure 6: RoO Groups

(a) NAFTA (b) USMCA

Group Count EX % VC % Group Count EX % VC %

1 CTC 11 6e-05 % 2.8 % 1 CTC 41 2.2 % 2.8 %
2 CTC & RVCXX 52 51 % 53 % 2 CTC & RVCXX 48 48 % 53 %
3 CTC + ALWXX 146 25 % 14 % 3 CTC + ALWXX 32 4.3 % 14 %
4 CTC + ECT 55 2.6 % 1.2 % 4 CTC + ECT 49 2.5 % 1.2 %
5 CTC + ECTXX 71 10 % 2.1 % 5 CTC + ECTXX 87 15 % 2.1 %
6 CTC | RVCXX 64 11 % 10 % 6 CTC | RVCXX 130 20 % 10 %

7 RVCXX 64 9.2 % ·%

After scraping the agreements’ PDFs and organizing the statements into a table, I move on to

the second step, which involves creating a mapping between product codes and their corresponding

RoO group for each agreement. The resulting pseudo-code in appendix F illustrates how each RoO

group is mapped into the origin calculator algorithm to assess firms’ compliance.

Coding RoO statements at the group level offers a more efficient and accurate approach than

converting each RoO into an if/then statement. While an alternative compliance calculator is publicly

available on the Mexican Economy Secretariat’s website,14 it is not suitable for research purposes as

it requires manual input and cannot be used on a large scale. However, this tool has helped test the

accuracy of my algorithm.

5 Rules of Origin (RoOs): From NAFTA to the USMCA

The implementation of the USMCA has brought about significant changes in RoO for the automotive

industry. Under NAFTA, automobiles and their parts were required to have at least 50% of their

content made by member countries in order to qualify for zero tariffs. However, the new RoO under

the USMCA have raised this requirement substantially, with an average of 62% now mandated. This

is likely to have a strong impact on Mexican auto exports, with 95% of these now required to have a

regional content of 70% or higher.

14 For more information, visit: https://www.snice.gob.mx/cs/avi/snice/hce.calc.origen2020.html

27
 Table 9: Changes to RCR

USMCA
50 60 62.5 65 70 75 No RoO
NAFTA
50 · · · 2.89 3.26 2.85 2.89
60 0.00 0.04 0.02 · 10.54 8.33 ·
62.5 · · 0.00 · · 41.99 0.53
No RoO · · 0.30 0.28 0.72 · 1.83
Note: values in each cell show the VC trade volume share for each RCR pair.

The structural changes in RoO between NAFTA and the USMCA are outlined in table 9. A

closer look reveals that the majority of modifications involved an increase in RCR, from 50, 60, or

62.5 percent to either 60, 62.5, 65, 70, or 75 percent. Additionally, the USMCA introduced new RoO

for HSclassifications that previously had none and also replaced Chapter-specific CTC rules with

new RVC, set at a high RCR of 75 percent.

Another significant modification in RoO regards the roll-up of super-core parts and components.

This provision gives producers the option to assess the RVC of super-core systems separately and

roll-up their content. That is, if the system satisfies the RCR, then 100% of its value can be counted

as originary for the value calculation of the vehicle. Although the provision allowing producers to roll

up super-core parts was present under NAFTA, it underwent expansion under USMCA: whilst the

RCR for super-core parts and their components rose from 62.5% to 75%, the scope of the systems to

which these rules apply expanded. Previously, these rules only applied to engines and transmission

systems; however, under the USMCA, they were extended to include axle, suspension, steering, body,

and battery systems. Section 6, discusses this provision in detail.

As RoO become increasingly stringent, the issue of enforcement gains significant prominence.

Specifically, the enforcement of RCRs poses a substantial challenge for governments, due to their

reliance on self-reported documents. In order to certify their origin, firms present a certificate of origin,

which is not bound by a prescribed format as long as it contains a minimum amount of information to

identify the certifier’s identity (which can be the producer, the importer, or the exporter), the product

tariff item, the criteria it satisfies for origin, and supporting documentation.15 By default, there is no

additional check, but authorities in the importing country may request supplementary documentation

and are required to “maintain criminal, civil, or administrative penalties for violations of its laws
15 refer to appendix G for a detailed description of these requirements

28
 Figure 7: Value Chain Trade Volume: Distribution by RoOs

NAFTA
No
CC CTH CTI CTSH
RoO
| & + | + |
RVC ALW ECT RVC ECT RVC
75 70 50 60 62.5 70 75 70 65 70 65 70 75 65 70 75

& RVC 60 0
CC
+ ALW 50 0 0

60 0 0 0 0.11
& RVC
62.5 0 0.42 0.01
USMCA

50 0.03 0.01 0 0.02 0 0 0 0

ALW
CTH 60 0.01
+
0.01 0 0 0
ECT
0 0.01 0 0 0
50
0.02
| RVC
60 0.08

CTSH + ALW 50 0

No
0 0 0 0 0.01 0 0.01 0 0
RoO

29
and regulations related to [the Origin procedures].” In essence, firms self-report their origin and may

face audits by the importing country, yet there is no standardized approach to enforcement. This

introduces an inherent risk of inaccuracies and misrepresentations, further complicating the regulatory

oversight process. The significance of relying on self-reported origin certificates is especially apparent

in the case of the US, since it does not collect VAT, unlike Canada and Mexico. VAT systems play a

critical role in tax collection because they integrate into the production and distribution process,

thereby bolstering transparency and compliance. If all member countries were to adopt a VAT system,

the origin could be authenticated through VAT records. Consequently, alternative mechanisms for

ensuring accurate reporting and compliance are imperative in this context.

6 Compliance Exercise

Equipped with the VC data and the origin calculator, I conducted the first exercise. The objective is

to evaluate the origin of all products within the value chain under both sets of RoOs: those defined in

NAFTA and those specified in the USMCA. The topological sort method, described in Section 3, is

helpful in this stage. Employing this approach, we ensure that, firms in the upstream have previously

been assessed for their origin at each production stage.

The baseline exercise involves evaluating compliance under both sets of rules for the value chain

formed by our group of assemblers. It is important to note that we utilize data from 2017, a period

during which NAFTA was in force. Therefore, the nature of this exercise can, at best, emulate

the scenario that would have unfolded if accountants of each firm were required to assess the same

production structure under two sets of rules. Despite our current omission of changes in the network,

this exercise yields insightful outcomes. It contributes to our understanding of how RoO across

products and along the VC interact within the production network, taking into account its topological

characteristics. Figures 8 and 9 show how the compliance and RVC differ with NAFTA and USMCA

RoOs. This implementation of the calculator closely aligns with both agreements, to the extent that

the available data permit. Assumptions and simplifications are delineated in detail in appendix E.

We then contrast these outcomes across three distinct sets: the complete VC, solely firms within the

VC that are engaged in the production of either a car parts or a vehicle (i.e., the assemblers), and a

30
subset comprising the 13 auto assemblers identified in Section 3.

Figure 8: Compliance Rate

Entire Value Chain Auto Firms Assemblers Only

100
−4.6 −9.4

−14.1
Compliance Ratem (CR)

80

60

98.3% 93.6% 97.9% 88.4%

40

76.9% 62.8%
20

NAFTA USMCA NAFTA USMCA NAFTA USMCA

FTA: NAFTA USMCA CR change relative to NAFTA in p.p.

As anticipated, compliance levels are lower under the USMCA RoOs, with a nearly ten percentage

points decrease observed for assemblers. However, the impact on parts producers is notably more

pronounced, as their compliance decreases almost twofold. It is worth noting that, as discussed in

Section 3.2, nearly 20% of Mexican exports to its partners comprise vehicle parts. Notice also that,

as a result of the super-core provision expansion, the RVC under the USMCA is higher than that

under NAFTA, although not sufficient to drive compliance rates upward.

31
 Figure 9: Regional Content Requirement

Entire Value Chain Auto Firms Assemblers Only

100

+2.8
Reginal Value Content (RVC)

80

−0.9 −1.4
60

40

66.5% 65.6% 72.7% 75.5% 66.1% 64.7%

20

NAFTA USMCA NAFTA USMCA NAFTA USMCA

FTA: NAFTA USMCA RVC change relative to NAFTA in p.p.

6.1 The Roll-up Dispute

In May 2021, the Mexican government requested formal discussions on the interpretation of the

accumulation provision for super-core systems in the new rules of origin. This was due to the

US unilaterally changing the way RVC is calculated, deducting any foreign content from the RVC

calculation. After a controversy-resolution panel conducted its review in December 2022, it was

concluded that automakers may continue to “roll up” the value of super-core parts and their car

parts, allowing the whole of a system to count as originating in the region if the good contains the

required percentage of regional content. While this increase in regional content requirements has

made RoO stricter, the roll-up provision has also granted more flexibility. These conflicting impacts

make the true overall effect of the new RoOs ambiguous ex-ante. In this section, we delve into the

implications of the controversy.

According to USMCA regulations, these parts must originate within a given USMCA country for

the vehicle or truck to be considered as having originated there. Moreover, Canada and Mexico have

32
contended that the agreement permitted the inclusion of rolled-up content for originating super-core

parts in the final vehicle RVC calculation. Conversely, the US has argued that the overall vehicle

RVC calculation and the core parts origination requirement were distinct and separate calculations.

Mexico and Canada have argued that the USMCA negotiations indicated that, once an essential

auto part is considered originary (by meeting a minimum of 75% of RVC), its regional value

incorporated into the total RVC of the vehicle must be 100%, having already fulfilled the requirements.

Contrastingly, the US interpretation suggested that even when an essential auto part qualifies as

originary, its regional value when incorporated into the total RVC of the vehicle should not necessarily

be 100%, but rather the percentage that has allowed it to meet the originating part requirements

(which can be between 75% and 100%) — thus, making compliance with RoO more challenging for

producers on the Mexican side of the border.[29] As Head et al. (2022)[14] show, stricter RoO affect

Mexico and Canada more than the US, due to the fact that most of Mexico’s and Canada’s output is

exported to the US, while the US auto output is mainly sold domestically.

Figure 10: Super-core Roll-up Effect on Compliance Rate

Entire Value Chain Auto Firms Assemblers Only

with with with

without without without
super−core roll−up super−core roll−up super−core roll−up

−4.7 −4.8 −9.5 −9.8

100
Compliance Rate (CR)

80
−14.1 −38.4

60

98.3% 93.6% 93.5% 97.9% 88.4% 88.1%

40
76.9% 62.8%
20 38.5%

NAFTA USMCA USMCA NAFTA USMCA USMCA NAFTA USMCA USMCA

FTA: NAFTA USMCA CR change relative to NAFTA in p.p.

Tables 10 and 11 show the result from my estimated compliance rates, assuming opposite

33
resolution panel rulings. Upon comparison of the compliance rates for both USMCA scenarios, it

becomes evident that the super-core provision crucially dampened the effect of higher RCRs; while

the compliance rate for auto assemblers would have experienced a twofold decrease, comparing

compliance rates for auto firms under both scenarios leads to the conclusion that the parts producers

are the party who benefited the most from the panel’s resolution. Finally, we also note that when we

consider all firms in the VC, the effect on auto-specific firms does not appear as pronounced.

Figure 11: Super-core Roll-up Effect on Regional Value Content

Entire Value Chain Auto Firms Assemblers Only

with with with

without without without
super−core roll−up super−core roll−up super−core roll−up
100
Reginal Value Content (RVC)

80 +2.8
−2.6
−0.9 −0.9 −1.4 −1.5
60

40
72.7% 75.5% 68.7%
66.5% 65.6% 65.6% 66.1% 64.7% 64.6%
20

NAFTA USMCA USMCA NAFTA USMCA USMCA NAFTA USMCA USMCA

FTA: NAFTA USMCA RVC change relative to NAFTA in p.p.

The findings in Figure 11 serve to reinforce the notion that the super-core roll-up provision

significantly influences compliance rates through firms’ RVC. This observation leads us to consider

the possibility that policymakers dedicate a considerable amount of resources to formulating intricate

regulations, which, in practice, appear to have far less impact on practical outcomes compared to a

single paragraph discreetly situated in an appendix within each agreement.

34
6.2 Alternative RoO specifications

We examine four hypothetical scenarios to understand further how each component of RoO interacts

and translates into firms’ outcomes. In the baseline scenario, we hold the RoO unchanged, but

we remove the roll-up option. The second scenario involves assessing the compliance of all firms

in the VC, under the assumption that a producer located in the region automatically satisfies the

origin criteria; this approach is referred to as the “no upstream compliance” assumption while

focusing exclusively on the RVC portion of the RoO and assuming a uniform RCR, equal to the

mean requirement specified in the agreements. Finally, the fourth scenario entails evaluating the

compliance of upstream firms while retaining RVC-only RoO with a flat RCR.

Figure 12: Compliance Rate Under Alternative Specifications — Entire Value Chain

Entire Value Chain

NAFTA USMCA

−33 −28.8
100 −0.1 +0.4 −49.4 −39.3

80
Compliance Rate (CR)

60

98.3% 98.3% 98.3% 93.6% 93.5% 94%

40
65.3% 69.5%
54.3%
20 44.2%

Baseline No Roll−Up No Upstream Flat RVC and Flat RVC Baseline No Roll−Up No Upstream Flat RVC and Flat RVC
Compliance No Upstream Compliance No Upstream

FTA: NAFTA USMCA CR change relative to NAFTA in p.p.

35
 Figure 13: Compliance Rate Under Alternative Specifications — Auto Firms

Auto Firms
NAFTA USMCA

100

−7.7
80
+6.4
Compliance Rate (CR)

−24.3 −9
60

40
76.9% 76.9% 76.9% 76.9%
69.2% 69.2%
62.8% 62.8%
53.8%
20 38.5%

Baseline No Roll−Up No Upstream Flat RVC and Flat RVC Baseline No Roll−Up No Upstream Flat RVC and Flat RVC
Compliance No Upstream Compliance No Upstream

FTA: NAFTA USMCA CR change relative to NAFTA in p.p.

Figure 14: Compliance Rate Under Alternative Specifications — Only Assemblers

Assemblers Only
NAFTA USMCA

−34.3 −28.9
100
−0.3 +0.8 −46.3 −35.4

80
Compliance Rate (CR)

60

97.9% 97.9% 97.9%

88.4% 88.1% 89.2%
40

63.6% 69%
53%
20 42.1%

Baseline No Roll−Up No Upstream Flat RVC and Flat RVC Baseline No Roll−Up No Upstream Flat RVC and Flat RVC
Compliance No Upstream Compliance No Upstream

FTA: NAFTA USMCA CR change relative to NAFTA in p.p.

Figures 12–14 display the compliance rate results for the entire VC, for auto firms, and for

assemblers, respectively. Similarly, Figures 15–17 illustrate the variations in RVC for the three sets

of firms. Notably, and surprisingly, the absence of roll-up has no significant impact on the outcome

for any group, except for a modest 1.4 percentage points (6.8 p.p) decrease in the regional content of

auto firms under NAFTA (USMCA) rules (See Figure 16). This unanticipated observation may be

inferred from the similarity of outcomes under the no-upstream compliance scenario. This suggests

that upstream compliance had limited influence on downstream firms under NAFTA, possibly due to

36
lower RCRs and fewer systems being eligible for a super-core roll-up.

Figure 15: RVC Under Alternative Specifications — Entire Value Chain

Entire Value Chain

NAFTA USMCA

100

80
Regional Value Content (RVC)

+0.3 −3.1 +0.3 +1.2 −5.6 +1.2

60

40
66.5% 66.5% 66.8% 63.4% 66.8% 65.6% 65.6% 66.8% 66.8%
60%
20

Baseline No Roll−Up No Upstream Flat RVC and Flat RVC Baseline No Roll−Up No Upstream Flat RVC and Flat RVC
Compliance No Upstream Compliance No Upstream

FTA: NAFTA USMCA RVC change relative to NAFTA in p.p.

Figure 16: RVC Under Alternative Specifications — Auto Firms

Auto Firms
NAFTA USMCA

100

−6.8 −2.7 −8.2 −2.7

80 −1.4 +0.1 −5.8 +0.1
Regional Value Content (RVC)

60

40
72.7% 71.3% 72.8% 72.8% 75.5% 72.8% 72.8%
66.9% 68.7% 67.3%

20

Baseline No Roll−Up No Upstream Flat RVC and Flat RVC Baseline No Roll−Up No Upstream Flat RVC and Flat RVC
Compliance No Upstream Compliance No Upstream

FTA: NAFTA USMCA RVC change relative to NAFTA in p.p.

37
 Figure 17: RVC Under Alternative Specifications — Only Assemblers

Assemblers Only
NAFTA USMCA

100

80
Regional Value Content (RVC)

+0.7 −3.4 +0.7 −0.1 +2.1 −5.5 +2.1

60

40

66.1% 66.1% 66.8% 62.7% 66.8% 64.7% 64.6% 66.8% 66.8%

59.2%
20

Baseline No Roll−Up No Upstream Flat RVC and Flat RVC Baseline No Roll−Up No Upstream Flat RVC and Flat RVC
Compliance No Upstream Compliance No Upstream

FTA: NAFTA USMCA RVC change relative to NAFTA in p.p.

Interestingly, the elimination of the roll-up provision affects the auto-firms almost exclusively.

When comparing this group with the assemblers’ group, we find that the impact is mainly influenced

by parts producers. It is worth noting that the reduction in the average RVC from NAFTA to

USMCA for auto firms is just under 10%, yet this adjustment holds significant implications for the

origin of the firms, as it positions the average auto firm content below the 75% baseline, which

coincides with the super-core RCR under USMCA.

7 Concluding Remarks

This paper introduces innovative tools for analyzing RoOs, made feasible by the recent availability of

new firm-to-firm trade datasets. Leveraging this data, I investigate the patterns within the automotive

industry value chain, which is a central sector for the Mexican economy. The study reveals three

primary findings: Firstly, the auto value chain demonstrates extensive interconnectedness, with

the majority of firms serving multiple assemblers, and approximately half of these firms servicing

ten or more assemblers. This inter-connectivity is not only reflected in the firm count but also

in trade volume, with a third of the VC’s trade volume concentrated among firms serving ten or

more assemblers. Secondly, the most significant changes to RoOs, in terms of their trade shares,

were increases of RCR while preserving the underlying RoO type. Parts producers, rather than

car assemblers, are the firms most affected by these changes. Furthermore, as RCRs become more

38
stringent, the significance of upstream compliance escalates. Thirdly, the recent dispute between

USMCA partners centered on one of the most pivotal provisions in the agreement: the super-core

systems roll-up. The compliance rate of auto firms without this provision is nearly three times lower

than the rate estimated when the super-core roll-up is permitted. In conclusion, this work aims at

advancing our knowledge of global value chains and their relationship to trade policy. In combination

with the origin calculator, it offers potential avenues to improve policy design and evaluation efforts.

It also prompts further exploration of numerous research questions using the methodologies presented

here.

References

[1] Alonso Alfaro-Ureña et al. Costa Rican Production Network: Stylized Facts. Official Document.

Banco Central de Costa Rica, Dec. 2018, p. 29.

[2] Pol Antras. “Global Value Chains: The Economics of Spiders and Snakes” (Seoul National

University Asia Center Samik Hall (Room 220), Room 240). 2018. url: https://www.eaerweb.

org / selectConferenceArticleInfo . do ? article _ a _ no = JE0002 _ 2019 _ 20194 _ 30 & ano =

JE0002_2019_20194_30.

[3] Pol Antràs and Alonso de Gortari. On the Geography of Global Value Chains. Working Paper

w23456. Cambridge, MA: National Bureau of Economic Research, 2019. doi: 10.3386/w23456.

[4] Andrew B. Bernard, Andreas Moxnes, and Yukiko U. Saito. “Production Networks, Geography,

and Firm Performance”. In: Journal of Political Economy 127.2 (Apr. 1, 2016), pp. 639–688.

issn: 0022-3808. doi: 10.1086/700764.

[5] Pamela Bombarda and Elisa Gamberoni. “Firm Heterogeneity, Rules of Origin, and Rules of

Cumulation*”. In: International Economic Review 54.1 (2013), pp. 307–328. issn: 1468-2354.

doi: 10.1111/j.1468-2354.2012.00734.x.

[6] Paola Conconi et al. “From Final Goods to Inputs: The Protectionist Effect of Rules of Origin”.

In: The American Economic Review 108.8 (Aug. 2018), pp. 2335–2365. issn: 00028282. doi:

10.1257/aer.20161151.

39
 [7] Michele Coscia. “Using Arborescences to Estimate Hierarchicalness in Directed Complex

Networks”. In: PLOS ONE 13.1 (Jan. 30, 2018). Ed. by Constantine Dovrolis, e0190825. issn:

1932-6203. doi: 10.1371/journal.pone.0190825.

[8] Alonso de Gortari. Disentangling Global Value Chains. Working Paper w25868. Cambridge,

MA: National Bureau of Economic Research, May 2019. doi: 10.3386/w25868.

[9] Emmanuel Dhyne et al. “Trade and Domestic Production Networks”. In: The Review of

Economic Studies 88.2 (Mar. 22, 2021), pp. 643–668. issn: 0034-6527, 1467-937X. doi: 10.

1093/restud/rdaa062.

[10] Rod Falvey and Geoff Reed. “Rules of Origin as Commercial Policy Instruments”. In: Interna-

tional Economic Review (May 2002).

[11] Gabriel Felbermayr, Feodora Teti, and Erdal Yalcin. “Rules of Origin and the Profitability of

Trade Deflection”. In: Journal of International Economics 121 (Nov. 1, 2019), p. 103248. issn:

0022-1996. doi: 10.1016/j.jinteco.2019.07.003.

[12] Caroline Freund. “The Need for Better Disciplines on Rules of Origins in the WTO: Evidence

from NAFTA”. In: The Shifting Landscape of Global Trade Governance. Ed. by Manfred

Elsig, Michael Hahn, and Gabriele Spilker. 1st ed. Cambridge University Press, Aug. 8, 2019,

pp. 107–120. isbn: 978-1-108-75768-3 978-1-108-48567-8 978-1-108-70744-2. doi: 10.1017/

9781108757683.005.

[13] Taiji Furusawa et al. “Global Sourcing and Domestic Production Networks”. In: EAER

International Conference. Seoul National University Asia Center Room 210 and 230, 2017,

pp. 54–67. doi: 10.11644/KIEP.EAER.Conf.2018.29.

[14] Keith Head, Thierry Mayer, and Marc Melitz. The Laffer Curve for Rules of Origin. Working

Paper. CEPR, Oct. 9, 2022, p. 49.

[15] Federico Huneeus. Production Network Dynamics and the Propagation of Shocks. Working

Paper. Princeton University, Dec. 2018.

[16] Jiandong Ju and Kala Krishna. “Firm Behaviour and Market Access in a Free Trade Area with

Rules of Origin”. In: Canadian Journal of Economics/Revue canadienne d’économique 38.1

(2005), pp. 290–308. issn: 1540-5982. doi: 10.1111/j.0008-4085.2005.00281.x.

40
[17] Ayumu Ken Kikkawa, Glenn Magerman, and Emmanuel Dhyne. Imperfect Competition in

Firm-to-Firm Trade. Working Paper. NBER, Nov. 2020, p. 71. url: https://drive.google.

com/file/d/0B61Yc8nfDbDWNlhZR0ZfbHQwckE/view?usp=sharing.

[18] Dzmitry Kniahin and Jaime de Melo. “A Primer on Rules of Origin as Non-Tariff Barriers”. In:

Journal of Risk and Financial Management 15.7 (June 28, 2022), p. 286. issn: 1911-8074. doi:

10.3390/jrfm15070286.

[19] Kala Krishna. Understanding Rules of Origin. Feb. 1, 2005. url: https://papers.ssrn.com/

abstract=669449. preprint.

[20] Kala Krishna et al. Learning to Use Trade Agreements. Working Paper. NBER, Nov. 2022.

[21] Alejandra López Espino. Characterizing the Mexican Production Network. Working Paper. The

Pennsylvania State University, Nov. 2023.

[22] Giovanni Maggi. “International Trade Agreements”. In: Handbook of International Economics.

Vol. 4. Elsevier, 2014, pp. 317–390. isbn: 978-0-444-54314-1. doi: 10.1016/B978-0-444-54314-

1.00006-9.

[23] Hiroshi Mukunoki. “The Welfare Effect of a Free Trade Agreement in the Presence of Foreign

Direct Investment and Rules of Origin”. In: Review of International Economics 25.4 (2017),

pp. 733–759. issn: 1467-9396. doi: 10.1111/roie.12282.

[24] Tiago P. Peixoto. “The Graph-Tool Python Library”. In: figshare (2014). doi: 10.6084/m9.

figshare.1164194.

[25] Juan Rosellón. “The Economics of Rules of Origin”. In: Journal of International Trade

& Economic Development 9.4 (Dec. 2000), pp. 397–425. issn: 09638199. doi: 10 . 1080 /

096381900750056849.

[26] Assieh Saadatpour and Réka Albert. “Boolean Modeling of Biological Regulatory Networks: A

Methodology Tutorial”. In: Methods 62.1 (July 2013), pp. 3–12. issn: 10462023. doi: 10.1016/

j.ymeth.2012.10.012.

41
[27] Julian D. Schwab et al. “Concepts in Boolean Network Modeling: What Do They All Mean?”

In: Computational and Structural Biotechnology Journal 18 (2020), pp. 571–582. issn: 20010370.

doi: 10.1016/j.csbj.2020.03.001.

[28] David Tsirekidze. “Global Supply Chains, Trade Agreements and Rules of Origin”. In: The

World Economy 44.11 (Nov. 2021), pp. 3111–3140. issn: 0378-5920, 1467-9701. doi: 10.1111/

twec.13137.

[29] USMCA Panel Rules against US Position in Automotive Origin Dispute. Tax Insights from

Customs and International Trade. PwC, Jan. 2023. url: https://www.pwc.com/us/en/

services/tax/library/usmca- panel- rules- against- us- position- in- auto- origin-

dispute.html.

[30] Chenying Yang. Rules of Origin and Auto-Parts Trade. Working Paper. Singapore Management

University, July 2022.

Appendices

A Data Sources

In this section, I describe the construction of the dataset used in our analysis. I combine four

sources to reconstruct the Mexican auto VC for 2017. I defer a discussion of the imputation of firms’

production labor to appendix B.

(i) Domestic Firm-to-firm Linkages.– We retrieve connections among Mexican firms from Online

Tax Receipts provided by the SHCP, called commercial CFDI for their acronym in Spanish, CFDIs.

We use the commercial prepend to distinguish these receipts from payroll CFDIs. The unit of

observation in commercial CFDIs is a month and a pair of firms, the buyer and the seller. For each

entry in the dataset, we observe the monthly value of transactions, the number of receipts issued

in that month, and the taxes retained, one of them being the VAT. Firms source different inputs

at different periodicities based on the nature of the inputs, supplier availability, market conditions,

business needs, and external factors. This and the fact that firms declare taxes annually impose a

42
natural yearly restriction on what constitutes one realization of the production network. Therefore we

aggregate all monthly data at the yearly level. Thus, an observation in this dataset is characterized

by a buyer, a seller, and a (yearly) transaction value.

(ii) State and Main Activity.– This dataset specifies the state where the firms’ tax ID is registered

and the product 6-digit NAICS code for the firm’s main activity, obtained from the firms’ annual tax

declarations. The main activity of a firm is defined by the product accounting for the largest share

of the firm’s sales revenue.

(iii) Imports and Exports.– We retrieve firms’ foreign linkages from custom records, which include

all imports and exports at the product level (NTL code). The unit of observation is given by each

transaction at the product level and the corresponding importer-seller or exporter-buyer firm pairs;

however, to ensure consistency and mitigate the impact of noisy recording of foreign firm identities, I

consolidate all foreign transactions into annual country-product pairs.

The datasets in (i)-(iii) can be linked using anonymized firm identifiers. I use (i)-(ii) to measure

firms’ sales revenue and materials purchases.

Production Labor.– The last data source is the Mexican Economic Census of 2019, which I use to

infer firms’ production labor. I compute the mean ratio of input purchases to production labor for

firms within each 6-digit NAICS code. Then, I utilize these ratios to estimate the production labor

for each firm by multiplying it with their materials purchases.16

Constructing an appropriate dataset to characterize the Mexican production network poses one

of the key challenges. This is primarily due to the vast size of the datasets involved, such as the

commercial CFDI data and the customs records. In 2017 alone, the commercial CFDI data consisted

of a staggering 1, 719, 086, 142 observations, with each observation containing 21 fields of information.

Similarly, the customs records comprised over 110, 235, 291 transactions at the product level, spread

across 14 different tables, and encompassing more than 150 fields of data.

Furthermore, before being able to generate descriptive statistics of the network, a significant

amount of cleaning and validation of these datasets was necessary. The intricate details of this

process are outlined comprehensively in LopezEspino (2022)[21]. Due to the sheer volume of data,

handling, and processing, it necessitated the use of an high-performance computing (HPC) cluster,
16 An objective for future projects is to merge the Economic Census with the other firm-level datasets, which will

allow the investigation of a broader set of economic questions.

43
along with an Apache Spark engine. R served as the main programming interface for this stage of

the analysis.

To minimize the impact of purchases that are not directly utilized in production, I adopt a

reasonable assumption and focus solely on manufacturing firms and their transactions. By doing so,

I create a refined network dataset that comprises 174261 nodes and 3043287 edges, representing a

subset of firms and their inter-dependencies within the production network.

Table 10: Assemblers’ exports by product code

NTLC Vehicle Type Duty Exports

87012001 heavy truck heavy 5,276,988

87012002 heavy truck heavy 3,000

87032101 passenger vehicle light 300,163

87032102 passenger vehicle light 58

87032199 passenger vehicle light 352,685

87032201 passenger vehicle light 5,412,845

87032202 passenger vehicle light 125

87032301 passenger vehicle light 23,378,731

87032302 passenger vehicle light 1,597

87032401 passenger vehicle light 1,097,565

87032402 passenger vehicle light 1,064

87042104 light truck light 65

87042199 light truck light 4,944

87042299 heavy truck heavy 169,450

87042399 heavy truck heavy 778,797

87043199 light truck light 767,499

87043299 heavy truck heavy 117,031

Note: exports are reported in thousand dollars.

44
 Table 11: Manufacturing sectors in the Auto Value Chain

NAICS Value Chain Trade Volume Firms in Value Chain

Description
Subsector Cum. Cum.
USD Share Count Share
Share Share

336 Transportation equipment manufacturing 1.72e+10 30 % 30% 515 9.94 % 9.94%

468 Retail trade of motor vehicles, parts, fuels and lubricants 3.2e+09 5.6 % 36% 259 5% 14.94%

435 Wholesale trade of agricultural, industrial, commercial and 2.92e+09 5.1 % 41% 670 12.93 % 27.87%

services machinery, equipment and furniture, and other gen-

eral purpose machinery and equipment

326 Plastic and rubber industry 2.59e+09 4.5 % 45% 180 3.47 % 31.34%

335 Electric appliances, accessories and electric power generation 2.11e+09 3.7 % 49% 134 2.59 % 33.93%

equipment manufacturing

332 Metal products manufacturing 2.08e+09 3.6 % 53% 363 7% 40.93%

434 Wholesale trade of agricultural, forestry and industrial raw 1.36e+09 2.4 % 55% 444 8.57 % 49.50%

materials, and waste materials

327 Nonmetallic mineral products manufacturing 1.32e+09 2.3 % 57% 41 0.79 % 50.29%

333 Machinery and equipment manufacturing 1.03e+09 1.8 % 59% 289 5.58 % 55.87%

331 Basic metal industry 9.55e+08 1.7 % 61% 100 1.93 % 57.80%

339 Other manufacturing industries 9.25e+08 1.6 % 62% 157 3.03 % 60.83%

334 Manufacturing of computer, communications, and measuring 1.9e+08 0.33 % 63% 54 1.04 % 61.87%

equipment, and other electronic equipment, components and

appliances manufacturing

325 Chemical industry 1.67e+08 0.29 % 63% 30 0.58 % 62.45%

436 Wholesale trade of trucks and new parts for automobiles, 1.13e+08 0.2 % 63% 13 0.25 % 62.70%

pickup trucks and trucks

467 Retail trade of hardware and glass 4.45e+07 0.077 % 63% 40 0.77 % 63.47%

314 Textile products manufacturing, except apparel 2.85e+07 0.049 % 63% 3 0.06 % 63.53%

211 Oil and gas extraction 2.55e+07 0.044 % 63% 5 0.1 % 63.63%

337 Furniture, mattresses and blinds manufacturing 2.09e+07 0.036 % 63% 32 0.62 % 64.25%

324 Petroleum and coal products manufacturing 2.04e+07 0.035 % 63% 1 0.02 % 64.27%

313 Textile inputs manufacturing, and textiles finishing 7,774,067 0.014 % 63% 2 0.04 % 64.31%

321 Wood industry 6,782,282 0.012 % 63% 7 0.14 % 64.45%

B Imputation of Production Labor and Robustness Checks

The potential source of measurement error in this analysis is limited to the net costs. As the labor

costs associated with production are not directly observable in the dataset, I employ summary

45
statistics from the 2019 Mexican Economic Census to estimate the net cost.

From the economic census, I derive the input purchase-to-net cost ratio for all firms at the activity

code level (six-digit NAICS). Subsequently, I utilize these ratios to infer the approximate net costs

for individual firms.

To validate the accuracy of this approximation method, I cross-verify it using an alternative data

source. Some firms in the value chain are required to present annual tax declarations, allowing me to

estimate net costs by leveraging the accounting structure of these declarations.

Figure 18: Accounting Items in Annual Tax Declarations

I calculate the net cost as the difference between sales and the sum of indirect production costs. In

Figure 25, I present the distribution of net costs derived from both sources, as well as a third approach.

In this third method, I obtain the input purchase-to-net cost ratio from firms’ declarations and then

utilize it to estimate net costs based on their manufacturing input purchases. Upon examining the

subset of firms for which data is available from both sources, the distributions closely resemble each

46
 Figure 19: Sales Revenue by Data Source

0.4

0.3
Density

0.2

0.1

0.0
102 104 106 108 1010
Sales Revenue

Source: Tax Declarations VAT Records

other. Nevertheless, it is conceivable that the distribution of net costs from annual tax declarations

is marginally skewed to the right, given that only larger firms are mandated by law to submit an

annual tax declaration.

By comparing figures 24 and 25, we see that the source of discrepancy for net costs stems from

input purchases. This is due to the fact that I only consider manufacturing purchases, to reduce the

likelihood of accounting for non-production related purchases inside the net costs. See table 11 for a

list of sectors considered.

47
 0.4

0.3
Density

0.2

0.1

0.0
102 104 106 108 1010
Sales Revenue

Source: Tax Declarations VAT Records

Figure 20: Sales Revenue by Data Source, intersecting firms

Figure 21: Production Labor by Data Source

0.4
Density

0.2

0.0
102 104 106 108
Production Labor

Source: Payroll Receipts + Ec. Census Tax Declarations

48
 Figure 22: Production Labor by Data Source, intersecting firms

0.4
Density

0.2

0.0
102 104 106 108
Production Labor

Source: Payroll Receipts + Ec. Census Tax Declarations

Figure 23: Input Purchases by Data Source

0.4

0.3
Density

0.2

0.1

0.0
100 102 104 106 108 1010
Input Purchases

Source: Tax Declarations VAT Records

49
 Figure 24: Input Purchases by Data Source, intersecting firms

0.4

0.3
Density

0.2

0.1

0.0
100 102 104 106 108 1010
Input Purchases

Source: Tax Declarations VAT Records

Figure 25: Net Cost Comparison

0.4

0.3
Density

0.2

0.1

0.0
100 102 104 106 108 1010
Net Cost

Tax VAT Records VAT Records

Source: Declarations + Ec. Census + Tax Declarations

50
C Trace-back and Sprout Algorithm

Algorithm 1 Sprouting function

1: function sprout(G, branch) ▷ G:graph, branch:edge

2: s ← source of edge branch

3: t ← target of edge branch

4: t∗ ← copy node t of graph G

5: target of branch ← t∗

6: G ← remove node t form graph G

7: return G

8: end function

Algorithm 2 Traceback function

1: function Traceback(G, root) ▷ G:graph, root:node (assembler)

2: ic ← root’s in-component in graph G

3: g ← filter G to keep only nodes in ic

4: while g is not DAG do

5: c ← get a cycle in graph g

6: branch ← find the edge furthest away from root

7: g ← sprout(g, branch)

8: end while

9: return g

10: end function

Figure 26 shows how this is the case, with BFS, as well as other popular search and tree finding

algorithms.

51
 (a) BFS (b) A∗ algorithm

(c) Predecessor tree (d) Dominator tree

Figure 26: Back tracing with alternative algorithms

52
 Figure 27: Domestic Firm Level Distribution: Auto Parts

Core Part Principal Part Complementary Part

Trade
16 1
Volume
15 1 1 1
Share
14 1 1 2 1 2
13 1 2 3 1
12 3 1 1 3 1 6 2 1
11 3 2 1 2 1 1 1 2 1
VC Level

10 2 2 3 2 1 3 3 4 3 1 1 1 2 0.06
9 2 1 4 2 4 2 3 3 3 4 5 1 4 1
8 2 5 3 7 3 5 6 10 7 6 3 3 1 1 4 2
7 6 3 5 1 10 3 4 10 1 1 4 15 11 8 2 7 1 2 2 4 7 2
0.04
6 9 8 7 1 3 11 1 1 20 8 19 12 3 6 7 24 23 4 2 3 10 6 14 4 1 5 2 5 10 6 1 4
5 32 23 18 6 14 18 1 2 4 37 31 27 21 7 28 26 48 28 8 5 15 26 23 20 2 5 13 2 5 18 6 3 6
4 39 58 40 24 77 25 23 34 64 98 65 60 48 107 97 72 39 9 34 56 41 64 28 6 29 68 12 16 40 3 27 32
3 42 101 31 118159 25 125 89 69 173 63 108201213159 79 50 5 181116 45 108 57 18 115134 26 15 29 3 106 93 0.02
2 17 24 31 96 38 25 123 52 34 52 59 90 132 79 62 29 45 207114 29 51 49 23 122 74 16 5 24 1 149 85
1 5 21 43 3 16 60 32 8 5 16 29 48 9 23 12 13 1 83 59 7 2 12 7 24 4 8 25 28

Α Β Γ ∆ Ε Ζ Θ Ι Κ ΛΜΝ Ξ Α Β Γ ∆ Ε Ζ Θ Ι Κ ΛΜΝ Ξ Α Β Γ ∆ Ε Ζ Θ Ι Κ ΛΜΝ Ξ

Assembler

Figure 28: Foreign Firm Level Distribution: Auto Parts

Core Part Principal Part Complementary Part

1 5
Trade
16 78 92 89 Volume
15 47 84 1 57 2 119 1 39
Share
14 1 68 5 155 13 111 1 31 21

13 19 168 1 5 128 5 288 1 1 63 20

12 52 1 3 2 275 1 5 3 49 46 61 299 11 1 5 58 31 60

11
VC Level

293 4 1 713 6 16 154 100 13 143 439 3 12 64 98 52 231

0.09
10 362 1 369 18 950 49 23 394 409 746 300 608 27 31 95 202 554 357

9 46 43 463 76 419 286 215 992 844 1035 251 607 407 57 343 347 919 332

8 409 1 106 1 1 391 42 2 1083 28 407 754 5 15 350 1389 1182 607 8 1016 34 319 216 13 38 199 620 950 563 13

7 544 59 420 8 90 734 69 3 16 1571 643 1578 2327 78 518 375 1379 1822 496 58 189 1320 653 1485 715 64 756 300 664 1788 513 51 220 0.06
6 1948 699 1309 151 407 1068 557 110 146 4331 2250 3479 1388 692 1525 969 3846 2587 1154 595 819 3702 2518 2746 687 748 1972 667 1490 2409 867 728 875

5 1927 2981 2623 1592 2970 1523 57 1041 1387 5441 8603 5492 3564 3888 8891 5910 5170 3877 509 3078 3787 5093 8305 4763 1635 3974 8659 2409 1944 3810 323 2988 3441

4 2249 6874 2084 6347 6780 1136 4 7205 5788 6943 18566 6629 8114 17726 20026 9075 7870 4893 445 17509 13404 6334 16071 6643 3004 14860 19150 3565 2961 4680 413 14996 11717

0.03
3 1318 2932 1913 5777 4887 1783 8 7013 4038 4735 9413 7552 7209 18324 14442 9375 2507 5746 386 23166 13835 4117 7728 6834 3059 17479 11338 3171 1096 4401 401 21804 12659

2 167 33 3 467 1321 655 289 1 1465 969 884 794 12 1661 3040 5405 2309 1557 545 1190 81 7295 5338 1069 952 24 1535 1108 5528 2128 550 185 1011 72 7565 5408

1 98 76 476 214 360 28 45 475 466 554 128 134 645 306 509 478 37 140 93 54 585 599 784 95 105 326 203 63 208 25 19 19 41 319 311 380

Α Β Γ ∆ Ε Ζ Θ Ι Κ ΛΜΝ Ξ Α Β Γ ∆ Ε Ζ Θ Ι Κ ΛΜΝ Ξ Α Β Γ ∆ Ε Ζ Θ Ι Κ ΛΜΝ Ξ

Assembler

53
 Figure 29: RoO Transition Matrix

NAFTA
No
CC CTH CTI CTSH
RoO
| & + | + |
RVC ALW ECT RVC ECT RVC
75 70 50 60 62.5 70 75 70 65 70 65 70 75 65 70 75

& RVC 60 1
CC
+ ALW 50 0.8 0.2

60 0.08 0.71 0.04 0.17

& RVC
62.5 0.11 0.7 0.19
USMCA

50 0.37 0.16 0.03 0.15 0.16 0.09 0.03 0.01

ALW
CTH 60 1
+
0.76 0.18 0.04 0.02
ECT
0.03 0.15 0.68 0.11 0.03
50
1
| RVC
60 1

CTSH + ALW 50 1

No
0.01 0.02 0.05 0.3 0.15 0.14 0.17 0.13 0.01
RoO

54
D Additional Statistics

Figure 30: Mexican Exports to FTA partners: Distribution by RoOs

NAFTA
No
CC CTH CTI CTSH
RoO
| & + | + |
RVC ALW ECT RVC ECT RVC
75 70 50 60 62.5 70 75 70 65 70 65 70 75 65 70 75

& RVC 60 0
CC
+ ALW 50 0 0

60 0 0 0 0.02
& RVC
62.5 0 0.12 0.01
USMCA

50 0.02 0.01 0 0.01 0.01 0 0 0

ALW
CTH 60 0.01
+
0 0 0 0
ECT
0 0.01 0.01 0.01 0
50
0.01
| RVC
60 0.03

CTSH + ALW 50 0.02

No
0 0 0 0 0.01 0 0 0 0
RoO

55
E RoO simplifying assumptions

Table 12: Origin calculator’s simplifying assumptions

TC Assumption

8702.10, 8702.90 A change to a motor vehicle for the transport of 16 or more persons

8703.21-8703.90 A change to a passenger vehicle

8704.21, 8704.31 A change to a light truck

8704.22, 8704.23 A change to a heavy truck

8704.32, 8704.90 A change to a vehicle that is solely or principally for off-road

87.06, 87.07 A good for use in a passenger vehicle or light truck

8708.10, 8708.21, A good for use in a passenger vehicle, light truck, or heavy truck

8708.70, 8708.93,

8708.95

8708.29 A good, different from body stamping, for use in a passenger vehicle, light truck,

or heavy truck

8708.30 A good for use in a passenger vehicle, light truck or heavy truck

8708.40, 8708.50, A good for use in a passenger vehicle or light truck

8708.80, 8708.94

8708.91 A good for use in a passenger vehicle, light truck or heavy truck

8708.92 A change to silencers (mufflers) or exhaust pipes for use in a passenger vehicle,

light truck, or heavy truck

8708.99 Chassis frames are for use in a passenger vehicle or light truck

F Origin Calculator Algorithm

Algorithm 3 Regional Value Content Function

1: function content( vnm , nc)
2: rvc ← (nc − sum( vnm $value))/nc
3: return rvc
4: end function

56
Algorithm 4 de Minimis Function
Require: ect, alw ∈ {true, false}
1: function minimis(ntlc, vmn , tv, roo , ect, alw)
2: ctclist ← substring( HS $code, 1, roo $level) == substring(ntcl, 1, roo $level)
3: if ect == alw then
4: exlist ← ctlist
5: else if alw then
6: f rom ← roo $f rom ▷ from is a list of allowed NTL codes
7: alwlist ← filter HS $code where
HS $code == substring(ntcl, 1, level)
8: exlist ← ntlc ∈ ctclist \ alwlist
9: else
10: except ← roo$except ▷ list of NTL code exceptions
11: exlist ← ntlc ∈ except ∪ ctclist
12: end if
13: vmn exc ← filter vmn where
substring( vmn $pcode, 1, roo $level) ∈ exlist
14: m ← 1 − (tv − sum( vmn exc $value))/tv
15: return m
16: end function

57
Algorithm 5 origin calculator
1: function Calculator(ntlc, f ta, inputs , nc, tv, f lat)
Require: f lat ∈ {true, false}
2: import( roos naf ta , roos usmca )
3: region ←“CAN”,“MEX”,“USA”
4: if f ta ==“NAFTA” then
5: rooi ← roos naf ta where ntlc == roos naf ta$product
6: else
7: rooi ← roos usmca where ntlc == roos usmca$product
8: end if
9: vmn ← inputs where inputs $country ∈ / region
10: rvc ←rvc(vmn, nc)
11: if f lat or rooi$type ==“RVCXX” then
12: if rvc > rooi$rcr f lat then
13: roo ← 1
14: else
15: roo ← 0
16: end if
17: else if rooi$type ∈ {“CTC”,“CTC + ECT”} then
18: min ← minimis(ntlc, vmn , vc, roo , ect, alw)
19: if min < 0.1 then
20: roo ← 1
21: else
22: roo ← 0
23: end if
24: else if rooi$type ∈ {“CTC + ECTXX”} then
25: min1 ←minimis(ntlc, vmn , vc, rooi , rooi $ect,false)
State min2 ←minimis(ntlc, vmn , vc, rooi ,false, rooi $alw)
26: if min1 < 0.1 then
27: roo ← 1
28: else if min2 < 0.1 and rvc > rooi $rcr then
29: roo ← 1
30: else
31: roo ← 0
32: end if
33: else ▷ rooi$type ∈ {“CTC & RVCXX”,“CTC | RVCXX”}
34: min ←minimis(ntlc, vmn , vc, roo , ect, alw)
35: if rooi $comb ==“OR” then
36: if min < 0.1 or rvc > rooi $rcr then
37: roo ← 1
38: else
39: roo ← 0
40: end if
41: else
42: if min < 0.1 and rvc > rooi $rcr then
43: roo ← 1
44: else
45: roo ← 0
46: end if
47: end if 58
48: end if
49: return roo
50: end function
G Certificate of Origin

Under NAFTA, there were official documents that producers had to fill out to certify their product’s

origin.17 In an attempt to reduce red tape and modernize compliance, member countries have included

three main modifications to USMCA. First, in addition to producers, exporters, and importers can

now present the CO. Second, these formats are no longer required as long as a minimum amount of

information is included. These are:

1. whether the certifier is the exporter, producer, or importer.

2. certifier’s name, title, address (including country), telephone number, and email address

3. exporter’s name, address (including country), e-mail address, and telephone number if different

from the certifier

4. producer’s name, address (including country), e-mail address, and telephone number, if different

from the certifier or exporter or, if there are multiple producers

5. importer’s name, address, e-mail address, and telephone number. The address of the importer

shall be in a Party’s territory

6. description and HS Tariff Classification

7. origin criteria under which the good qualifies.

8. blanket period: the period of the certification covers multiple shipments of identical goods for

a specified period of up to 12 months

9. the certification must be signed and dated by the certifier and accompanied by the following

statement:

I certify that the goods described in this document qualify as originating and that the

information contained in this document is true and accurate. I assume responsibility for

proving such representations and agree to maintain and present, upon request or to make

available during a verification visit, documentation necessary to support this certification.

17 Such formats are available for all partners in the US Customs and Border Protection (CBP) website: https:
//www.cbp.gov/trade/nafta/nafta-certificate-origin

59
H List of Abbreviations

BFS breadth-first search

CBP Customs and Border Protection

CFDI Comprobante Fiscal Digital por Internet (Online Digital Tax Receipt)

CTC change in tariff classification

DAG directed acyclic graph

DFS depth-first search

FDI Foreign Direct Investment

FTA Free Trade Agreement

FTA Free Trade Agreement

GDP gross domestic product

GVC global value chain

HPC high-performance computing

HS Harmonized System

INEGI Instituto Nacional de Estadı́stica y Geografı́a (National Statistics and Geography Institute)

ITC International Trade Centre

LIGIE Ley de los Impuestos Generales de Importación y de Exportación (Import and Export

General Tax Law)

NAFTA North American Free Trade Agreement

NAICS North American Industry Classification System

NICO Número de Identificación Comercial (Commercial Identification Number)

60
NTL national tariff line

RAIAVL Registro Administrativo de la Industria Automotriz de Vehı́culos Ligeros (Administrative

Registry of the Automotive Industry for Light Vehicles)

RAIAVP Registro Administrativo de la Industria Automotriz de Vehı́culos Pesados (Administrative

Registry of the Automotive Industry for Heavy Vehicles)

RCR regional content requirement

ROF Rules of Origin Facilitator

RoO rule of origin

RoO rule of origin

RVC regional value content

SCC strongly connected component

SHCP Secretarı́a de Hacienda y Crédito Público (Secretariat of Finance and Public Credit)

SP specific processing

USMCA United States-Mexico-Canada Agreement

VAT value-added tax

VC value chain

WCO World Customs Organization

WTO World Trade Organization

61