IRREGULAR TRADE POLICY: REVERSALS, UNCERTAINTY AND
DIVERSION
by
YOUSSEF AIT BENASSER
A DISSERTATION
Presented to the Department of Economics
and the Division of Graduate Studies of the University of Oregon
in partial fulfillment of the requirements
for the degree of Doctor of Philosophy
June 2022
DISSERTATION APPROVAL PAGE
Student: Youssef Ait Benasser
Title: Irregular Trade Policy: Reversals, Uncertainty And Diversion
This dissertation has been accepted and approved in partial fulfillment of the require-
ments for the Doctor of Philosophy degree in the Department of Economics by:
Anca Cristea Chairperson
Jeremy Piger Core Member
Bruce Blonigen Core Member
Yeling Tan Institutional Representative
and
Krista Chronister Vice-Provost of Graduate Studies
Original approval signatures are on file with the University of Oregon Division of Grad-
uate Studies.
Degree awarded June 2022.
ii
Copyright © 2022 by YOUSSEF AIT BENASSER
All rights reserved.
iii
DISSERTATION ABSTRACT
Youssef Ait Benasser
Doctor of Philosophy
Department of Economics
June 2022
Title: Irregular Trade Policy: Reversals, Uncertainty And Diversion
Recent economic, political, and geostrategic developments are upending the global trade
framework, weakening global multilateralism and international coordination. The sta-
bility and predictability that characterized a slow-moving and consensus-based global
system of rules is ceding ground to increasingly more common deviations and irregular-
ities. This invites new questions about the properties of international trade policy. Using
a variety of econometric tools and data collection and construction tools, this disserta-
tion empirically investigates three such irregularities that are becoming more prevalent
or likely. I find that the economic costs and implications of such irregularities are rather
contained in scope and limited in size. This suggests that a transition out of the existing
international trade model and into a new framework might be conducted at reduced cost
to the continuity and stability of overall economic activity.
iv
CURRICULUM VITAE
NAME OF AUTHOR: Youssef Ait Benasser
GRADUATE AND UNDERGRADUATE SCHOOLS ATTENDED:
University of Oregon, Eugene, Oregon
Ecole Polytechnique, Palaiseau, France
Sciences Po, Paris, France
DEGREES AWARDED:
Doctor of Philosophy, Economics, 2022, University of Oregon
Master of Arts, Economics, 2019, University of Oregon
Master of Science, Economics and Public Policy, 2014, Sciences Po
Master of Science, Economics and Public Policy, 2014, Ecole Polytechnique
Bachelor, Political Science, 2012, Sciences Po
AREAS OF SPECIAL INTEREST:
International Economics
Trade Policy
Applied Econometrics
GRANTS, AWARDS, AND HONORS:
Graduate Teaching Award, University of Oregon, 2021
Kleinsorge Summer Research Fellowship, University of Oregon, 2019
v
ACKNOWLEDGMENTS
Someone once likened a doctoral program to being thrown against a wall, multiple times,
until you break. Every time you do not shatter, and as soon as you stand anew, you are
thrown against a different, sturdier wall. Over the past five years I started so many weeks
not knowing how I will climb over to the next. I ended many weeks in utter disbelief
of all that I had accomplished. And I saw many walls turn into short fences, and small
pebbles in the rear-view mirror of my academic journey. If I did not break, it was in
no small part thanks to my people. The people I brought with me, and the people I met
along the way.
First and foremost, my sisters, Basma and Yousra. The love that we share and the faith
that you have always had in me reignited my strength whenever it was dimming. I hope
I have made you proud. My also sisters, Imane, Hind, and Jemila, AKA Ze Hyper
Titounes. Thirteen years ago, we embarked on a joint path that would later fork into four,
sixteen, sixty-four paths... All inter-twinned, meandering, spiraling in intricate fractals
all around, surprising us, inspiring us, empowering us. How fortunate I have been to
walk my chosen path, different from yours, but always with you, and never alone. Two
hundred and fifty six, one thousand and twenty four; I can not wait to see these paths
multiply into ever larger powers of four, as we cultivate our own power of four. And
to another four, my forever besties: Youssef, Nicky, Mariam and Sarah, thank you for
infusing this journey with joy and lightness.
The Student Cooperative Association was the greatest gift of my graduate school expe-
rience. It gave me a landing point when I first arrived in Eugene, a roof over my head,
and it kept giving through the 3 years that I participated in it. The Janet Smith House’s
greatest gift was Stefanı́a. They shared my very first day in Eugene, and as I write these
words are flying back to share the last. They have seen me, listened to me, through the
entire process of becoming a doctor. They have been a friend, a confidant, a tree whose
shade I could collapse into when the blazing heat of it all was too much. As they go on
to pursue their own doctoral journey, I look forward to being those same things for them.
At the University of Oregon, I was extremely fortunate to cross paths with brilliant col-
leagues. Chandler, Jenni, and Cory, aka Evil Eyes Sympathizers, showed me boundless
kindness, and helped me light the way when it seemed darkest. They taught me that
professional relationships can be supportive, encouraging, validating, and empowering.
I am grateful for the togetherness that we shared over the past five years. Maria, and the
denizens of Lil House in The Prairie, Michelle, Nino, and Farah, were the most com-
passionate, most intelligent, and most hilarious neighbors I could have wished for. They
made PLC a pleasant place to be (a Herculean feat if you know the building). Anna, and
vi
Amani, who have been role models for me, and who continue to inspire me today. To all,
thank you. And let’s change this world!
Finally I must thank the members of my dissertation committee: Anca, Jeremy, Bruce,
Yeling. I promise to pay forward your generosity in whatever way I can to help the next
generation of scientists achieve their dreams as you have helped me achieve mine.
vii
To Mama & Baba who always inspire me. To Léane, Adam, and Naël, whom I hope to
one day inspire.
Contents
I Introduction 1
II Evaluating The Symmetry of Trade Policy: Evidence From Liberalization
Reversals 5
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.1 Agreements data . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.2 Trade data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.3 Stylized facts . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Methodology and Results . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3.1 Main specification . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3.2 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4 Robustness & Heterogeneity analysis . . . . . . . . . . . . . . . . . . 26
2.4.1 Strict Exogeneity . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4.2 Agreement Depth . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4.3 Geographic and political contexts . . . . . . . . . . . . . . . . 29
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
III Perception of Trade Policy Uncertainty and Stock Market Volatility : New
Measures 33
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2.1 Literature on Trade Policy Uncertainty . . . . . . . . . . . . . 35
3.2.2 Market Volatility: Stylized Facts . . . . . . . . . . . . . . . . . 38
3.3 Data and measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.3.1 Modelling financial volatility . . . . . . . . . . . . . . . . . . . 40
viii
3.3.2 A perception based measure of TPU: Twitter Component . . . . 42
3.3.3 A perception based measure of TPU: Institutional Component . 45
3.3.4 Discussion of proposed index . . . . . . . . . . . . . . . . . . 51
3.4 TPU and Stock Market Volatility . . . . . . . . . . . . . . . . . . . . . 55
3.4.1 Model and Estimation . . . . . . . . . . . . . . . . . . . . . . 56
3.4.2 Baseline Results . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.4.3 Out-of-Sample Analysis . . . . . . . . . . . . . . . . . . . . . 58
3.4.4 Portfolio Evaluation . . . . . . . . . . . . . . . . . . . . . . . 61
3.4.5 Heterogeneity Analysis . . . . . . . . . . . . . . . . . . . . . . 62
3.4.6 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.4.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
IV China’s Aid and Trade Diplomacy: A Zero-Sum Game? 71
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.3 Data & Stylized Facts . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.3.1 Trade and Investment agreements . . . . . . . . . . . . . . . . 76
4.3.2 Financial Aid Data . . . . . . . . . . . . . . . . . . . . . . . . 79
4.4 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.4.1 Official Financing (OF) Treatments . . . . . . . . . . . . . . . 82
4.4.2 Trade Agreements . . . . . . . . . . . . . . . . . . . . . . . . 83
4.4.3 Investment Agreements . . . . . . . . . . . . . . . . . . . . . . 84
4.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.5.1 Official Financing Results . . . . . . . . . . . . . . . . . . . . 85
4.5.2 Trade and BRI Agreements Results . . . . . . . . . . . . . . . 90
4.5.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
V Conclusion 98
Bibliography 100
ix
List of Figures
2.1 Country-pairs sharing a trade agreement, frequency 1950-2016 . . . . . 8
2.2 Distributions for GDP per capita, and trade to GDP ratios for treated
countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Exits by year . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4 PPML Estimation of Gravity Equation: Exit Effects . . . . . . . . . . . 25
2.5 PPML Estimation of Gravity Equation: FTA Exit Effects . . . . . . . . 29
2.6 PPML Estimation of Gravity Equation: PTA Exit Effects . . . . . . . . 30
2.7 PPML Estimation of Gravity Equation: Exit Effects by Geography . . . 31
3.1 Number of High Return Sessions Per Month . . . . . . . . . . . . . . . 39
3.2 Average Number of Days Between Two High Return Sessions . . . . . 40
3.3 Volatility Measures for the S&P500 Index . . . . . . . . . . . . . . . . 42
3.4 Word Cloud: Word Frequency in Selected TPU Tweets . . . . . . . . . 43
3.5 Twitter-Based Indicator of Trade Policy Uncertainty . . . . . . . . . . . 46
3.6 Institutional Signals Indicator of Trade Policy Uncertainty . . . . . . . 48
3.7 Proposed TPU Perception Index . . . . . . . . . . . . . . . . . . . . . 50
3.8 Proposed Index and Existing TPU measures . . . . . . . . . . . . . . . 52
3.9 Proposed Index and Existing Measures : Quantile Distributions . . . . . 53
3.10 Proposed Uncertainty Index and Trade Policy Volatility . . . . . . . . . 55
3.11 Responses to a Trade Policy Shock . . . . . . . . . . . . . . . . . . . . 58
3.12 Responses of Volatility Components to a Trade Policy Shock . . . . . . 59
3.13 Log Volatility: True value and Alternative Forecasts . . . . . . . . . . . 61
3.14 Certainty Equivalence Return of Alternative Portfolios . . . . . . . . . 63
3.15 Impulse Response Function of Volatility to TPU Shocks: By Sector . . 64
3.16 Impulse Response Function of Volatility to TPU Shocks: Individual Stocks 66
3.17 Impulse Response Function to a TPU Shock, using Baker et al’s Trade
Component of EPU . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.18 Impulse Response Function to an EPU Shock . . . . . . . . . . . . . . 68
x
4.1 Belt and Road Initiative Members, and Entry Years . . . . . . . . . . . 78
4.2 Composition of China’s Development Assistance to Africa (2017) . . . 81
4.3 Belt and Road Initiative Members, and Entry Years . . . . . . . . . . . 95
xi
List of Tables
2.1 Sample of treated units . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Summary trade data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Summary agreement data . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Trade liberalization reversals . . . . . . . . . . . . . . . . . . . . . . . 17
2.5 Symmetry cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.6 Pooled Estimation of Exit Symmetry . . . . . . . . . . . . . . . . . . . 24
2.7 Joint significance test of cumulative effects . . . . . . . . . . . . . . . 26
2.8 Panel Gravity Equation with Lead Treatment Variables . . . . . . . . . 28
3.1 Tweet Selection keywords . . . . . . . . . . . . . . . . . . . . . . . . 44
3.2 Forecast Error Variance Decomposition of S&P500 Monthly Realized
Volatility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3 List of High Exposure Stocks . . . . . . . . . . . . . . . . . . . . . . . 65
4.1 China’s Trade Agreement Diplomacy . . . . . . . . . . . . . . . . . . 77
4.2 China’s Development Assistance Flows To Africa (all categories) . . . . 80
4.3 Official Chinese Loans To African Countries (Excluding Trade Finance) 81
4.4 Trade Creation and Diversion Effects of Official Finance (OF) . . . . . 85
4.5 Trade Creation and Diversion Effects of Official Finance (OF) - Lagged
Treatment (Extensive) . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.6 Trade Creation and Diversion Effects of Official Finance (OF) - Lagged
Treatment (Intensive) . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.7 Effects of Official Finance (OF) By Category - Lagged Treatment (Ex-
tensive) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.8 Effects of Official Finance (OF) By Category - Lagged Treatment (Inten-
sive) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.9 Effects of Official Finance (OF) By Flow Type - Lagged Treatment (Ex-
tensive) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
xii
4.10 Effects of Official Finance (OF) By Flow Type - Lagged Treatment (In-
tensive) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.11 Trade Creation Effects of Preferential Tariff Treatment . . . . . . . . . 93
4.12 Trade Diversion Effects of Preferential Tariff Treatment . . . . . . . . . 93
xiii
Chapter I
Introduction
Every year, about 60% of the world’s output is traded internationally. In this context,
trade policy, understood as the rules and regulations that govern the transit of goods,
services, and financial flows across borders, is an important tool of government action
worldwide.
For close to a century, trade policy has been a matter of multilateral and multinational
coordination. The international rule-based trade system began as a transatlantic post-war
arrangement, wherein the United States traded the opening of its market for the disman-
tling of British colonial trade preferences (Zeiler (1998)). It has since morphed into a
complex set of rules and regulations that facilitate trade through a decidedly liberal lens.
The General Agreements on Tariffs and Trade (GATT), signed in 1947 between the US
and Western European nations, is now a 164-nation strong World Trade Organization
(WTO), complete with its own enforcement and dispute resolution mechanisms.
Today, as multilateralism weakens, trade policy–making is undergoing deep transforma-
tions. In recent years, North Atlantic countries have begun tearing apart the same system
they themselves once designed, and spent decades courting poorer economies into. By
the end of 2019, the United States had violated several WTO rules by reintroducing dis-
criminatory trade barriers, and brought the WTO dispute resolution system to a complete
halt by objecting to the renewal of its members. Earlier this year, the United Kingdom
all but withdrew from the European Union’s common market, the largest trading bloc in
the world. Accelerating this process of policy divergence is the Covid-19 pandemic. The
preference for national flexibility in trade policy–making has permeated the entirety of
1
the political spectrum. Whether rule-making bodies such as the WTO, let alone the rules
themselves, can withstand this pressure, is an open question. It is clear, however, that
trade policy–making is far less predictable than once thought, and that political affinity
for unilateral trade policy flexibility is gaining traction.
In the context of this increasingly dynamic, decentralized global system, trade policy
displays new properties. My dissertation focuses on three such properties of trade policy:
policy reversals, policy uncertainty, and the divergence of international policies.
Reversals: Evaluating The Symmetry of Trade Policy Shocks
As national economies begin to exercise increased flexibility in trade policy–making,
policy reversals are increasingly common. Periods of trade liberalization (lower trade
tariffs, fewer regulatory barriers) alternate with protectionist tides (increasing trade tar-
iffs, new barriers). In this context, I ask : how do gains from liberalization compare
to losses from protectionism? The question of ‘directional symmetry’ in policy reversals
has been studied in monetary policy as early as 1992 (Cover (1992)). This paper pioneers
the analysis of trade policy symmetry.
One of the challenges of empirical research is identifying the appropriate data necessary
to evaluate theoretical predictions. The value of this paper comprises in its successful
identification of observable cases of reversals, and its employment of both available and
constructed data variables to evaluate the question of symmetry in trade policy. The
paper’s findings suggest that the effects of protectionist versus liberalizing trade policy
reversals are not symmetrical, with liberalizing policies showing more lasting and larger
effects on trade volumes. This is a positive finding for policymakers worried about the
economic and social costs of increased trade policy flexibility.
Uncertainty: Measurement and Financial Impacts
Economic policy uncertainty has received significant attention in the economic litera-
ture. It is associated with greater stock market volatility and reduced investment and
employment in policy-sensitive sectors like defense, healthcare, finance and infrastruc-
ture construction (Baker et al. (2016a)). Attempts to isolate and evaluate the impact
of trade policy uncertainty (TPU) specifically, from other economic policy components,
have suffered from the absence of a methodology to measure perceptions of TPU (Feng
et al. (2017) Handley and Limão (2015a) Julio and Yook (2016)). I address this by de-
veloping a methodology for empirically gauging the public’s perception of trade policy
2
uncertainty, and employ the construct index to study the impact of TPU on stock market
performance.
The index extracts data signals from two sources. To approximate public attention, I
sample trade related tweets and observe the amplification they receive (retweets, likes,
quotes etc.). To anchor the measure in real policy developments, I incorporate data from
the US Federal Register on anticipated policy actions. The index appears very informa-
tive. An initial peak is observed following the 2016 elections, and a significant departure
from historical averages begins in late 2017, concurrent with the initial stages of the
US-China tariff war. The index begins to trend downwards after an initial agreement is
signed between the two trading rivals in late 2019. The proposed measure performs well
against two contemporaneously developed indices in Caldara et al. (2020a) and Baker et
al. (2019a) and is highly informative.
Using the index, I find that stock market volatility responds negatively to trade policy
uncertainty in the 2015–2020 study period. This is in direct negation of the positive
relationship between broader economic policy uncertainty and volatility documented in
the literature, suggesting that public response to trade policy uncertainty is distinct from
other policy-related risks. The chapter suggests that TPU deserves increased and singular
attention, and provides a critical tool to carry out further research in this area.
China’s Aid and Trade Diplomacy: A Zero-Sum Game?
Though trade tensions have been exacerbated by populist rhetoric, they do not resolve
with its retreat. Intellectual property, public subsidies to production, frictions from na-
tional security concerns (Berger et al. (2019)), regulatory divergence, and currency ma-
nipulation are all sources of trade policy tensions unlikely to relent. The international
rule-based trade system has successfully accommodated the competing Transatlantic in-
terests it was founded to serve for close to a century. It now appears unsuited for the
increasingly hegemonic Transpacific trade relationship.
One area of competition between large international trade participants concerns market
shares in poor and emerging markets. Of specific interest is China’s economic diplomacy
which mobilizes state resources to the benefit of business internationalization. Viewed
as a form of new-mercantilism wherein the government advances private sector interests,
it is widely decried in the West as a threat to existing trade relations (Mawdsley (2008)).
3
In my third and last chapter, I evaluate the extent to which China’s aid and trade diplo-
macy displaces existing trade links. To do this, I look at a combination of policy tools:
official financing (including official development aid, and other official flows), trade
agreements, and the Belt and Road Initiative, which one can think of as an investment
agreement. I focus on the effects of these policies in Africa, a region that has been par-
ticularly targeted by China’s activism.
In this chapter, I am guided by three questions: Do these policy treatments increase
countries’ exports to China? If such an increase is observed, is it happening at the cost of
pre-existing export flows? And how are different export industries affected differently?
This is a data intensive project. I mobilize different methods adapted to the variety of data
structures at hand: gravity methods, long-difference estimation equations, and synthetic
controls. My research will contribute to understanding whether trade diplomacy is setting
us on a path to polarization, and whether indeed these policy interventions are a zero-sum
game.
4
Chapter II
Evaluating The Symmetry of Trade
Policy: Evidence From Liberalization
Reversals
2.1 Introduction
Trade policy reversals, understood as back and forth changes in trade openness, are char-
acteristic of the competing tides of protectionism and liberalization. This tension was
recently put on full display in the context of the most intensive trade relation: China-US
trade. Bound by the rules of indiscriminate treatment under the WTO agreement since
2001, the two countries have restored significant trade barriers in the form of tariffs and
ad-hoc restrictions over the past few years. Some have been since removed, and others
are being evaluated for removal by the new administration. This back and forth invites a
policy conversation about the cost of these policy reversals and their symmetry.
Policy reversals are not specific to US-China trade or circumscribed to a context of
weakening global trade regulation. Within the multilateral trade system, countries ex-
ercise much of their trade policy flexibility through bilateral trade agreements. The 164
member governments of the World Trade Organization benefit from non-discriminatory
treatment from peers – shielding their trading firms from targeted tariff and regulatory
shocks. In return, they accept to treat all partners equally, losing the ability to adjust their
policy at the margins towards different trading partners. Trade policy flexibility is sig-
nificantly limited by this arrangement but not entirely lost. Indeed, the WTO agreement
allows countries to grant more favorable treatment to targeted partners via preferential
systems, free trade agreements, and customs unions, all indiscriminately referred to in
5
this paper as trade agreements. As a result, for all the growth in WTO membership, the
number of country pairs operating under trade agreements has grown much faster (see
figure 2.1). This stresses the degree to which trade agreements have become an essential
tool to extract and exercise trade policy flexibility within an otherwise restrictive environ-
ment within the global trade system. Countries exercise trade policy flexibility through
trade agreements by entering into them. They also exercise that flexibility by reviewing
and revising them, suspending or leaving them, and even at times, reentering them after
periods of retraction. Policy flexibility entails policy reversal, and from policy reversals
arise the question of policy symmetry.
In this paper, I ask: is trade policy symmetric? Since the inception of the General Agree-
ment on Tariffs and Trade (GATT) in 1947, the dominating trend in global trade has been
towards increased liberalization. While international trade liberalization is unlikely to be-
come undone, protectionist discourse and policies create more frequent exceptions to the
norm. Fluctuating political discourse promotes policy reversals. And new crises, such
as the Covid pandemic, bring traditional policy choices into focus, promising possible
departures from the trend. This has been evidenced by the tide of trade barrier reintro-
duction over the past years, from the US-China tariffs to the recent export restrictions
put in place by several countries to control price inflation in the context of global trade
tensions. Even as they reintroduce protectionist barriers, these policies are temporary by
design and invite future reversals. This paper aims to study one of the core properties of
the policy reversals - their symmetry.
Policy symmetry has been the topic of a large literature within monetary economics. This
extensive literature looking into the asymmetry of monetary policy shocks distinguishes
three dimensions: asymmetry related to the direction of the shock, the size of the shocks,
and asymmetry associated with the phase of the business cycle (Stockwell (2020)). This
paper focuses exclusively on directional asymmetry. It answers the question of symme-
try by leveraging liberalization agreement reversals as policy shocks and comparatively
studying the response of trade volumes to positive and negative policy shocks. In this
context, perfect trade policy symmetry would require that the response of trade flows to
a positive shock is equal in absolute value but has the opposite sign from the response to
a negative shock.
In my analysis, the identifying trade policy shocks are liberalization and protectionist
policy actions. I exploit the variation from trade agreements that have experienced sub-
sequent withdrawal. Thus, I am able to capture two policy shocks of equal size that affect
the same country pair, i.e., the entry into and exit from the agreement. Testing the sym-
metry hypothesis is achieved by jointly estimating and subsequently comparing country
pairs’ trade volume response to the liberalization and protectionist policies, respectively.
6
The treatment group is constructed from an exhaustive dataset collected for agreements
from 1950 to 2016 for all existing countries. Within this dataset, I identify and isolate
all instances of trade liberalization reversals and document several of their key charac-
teristics. The control group is a larger set of country pairs that do not experience any
policy reversals during the study period. Using a panel data generalization of the grav-
ity equation, I measure trade volume response to entry and exit for treated units against
the control group. The point estimates of the treatment parameters and the associated
confidence intervals allow me to compare the trade responses across policy directions.
The paper presents results from three main estimations: a pooled regression, an analysis
of deep trade policy shocks proxied by free trade agreements, and a focus on the sym-
metry properties of less comprehensive preferential trade agreements. In all three cases,
point estimates display signs of asymmetry: protectionist policies do not erase the gains
achieved from prior, equal-sized liberalization. However, at the 95% confidence level,
the symmetry of trade policy cannot be rejected.
The paper closest to this work is a recent study by USITC affiliates Daigle and LaRocca
(2019) which surveys country exits from agreement and describes some common pat-
terns and subsequently focuses on 6 case studies to illustrate geographic and situational
diversity of reversal decisions. This paper goes further by quantifying the impacts of
exits using econometric methods and firmly placing the analysis in the context of policy
symmetry. While the study of symmetry in the context of trade policy and trade agree-
ments is a new question, this exercise borrows from and builds on two trade literatures.
Identifying the impact of trade agreements, specifically, free trade agreements (FTAs),
has been a ubiquitous research agenda for many years. This has allowed for a constant
improvement of the econometric techniques required for accurate impact measurement.
Earlier research reached contradictory and “fragile” results (Ghosh and Yamarik (2004)).
Baier and Bergstrand (2007) seminal paper developed a framework that adequately han-
dles the issue of trade agreement endogeneity using panel methods. The specification is
anchored in theory and derived from the trade gravity equation. This workhorse model
is the starting point for my estimation equation. My approach will also reflect a recur-
rent concern in recent contributions to this literature, namely the heterogeneity in trade
response to policy shocks. Geographic, regulatory and economic characteristics have
been found to drive this heterogeneity (Eicher and Henn (2011), Kohl (2014), Kohl et al.
(2016)).
The study of symmetry conducted in this paper sheds light on the trade costs of policy
reversals and volatility. It also contributes to reemerging literature about protectionism
and liberalization reversals. Recent swings in trade policy have supplied multiple pro-
tectionist case studies that have been measured and analyzed. Auer et al. (2018) estimate
7
Figure 2.1: Country-pairs sharing a trade agreement, frequency 1950-2016
NSF-Kellogg’s Database on Economic Integration and WTO RTA database
that revoking NAFTA would have reduced US welfare by 0.2%. Fajgelbaum et al. (2019)
find that US retaliatory tariffs introduced in 2018 generated a loss to US consumers and
firms of 0.27% of GDP. Oberhofer and Pfaffermayr (2018) expect UK exports to the
common market to shrink by 45.7% following an exit from the common market. Ad-
verse shocks to non-tariff measures have also received increased attention (Miromanova
(2019), Haidar (2019)).
The rest of the paper is organized as follows. Section 2 presents the data, sample con-
struction, and the stylized facts surrounding policy reversals. Section 3 lays out the
empirical strategy and identification framework. The section then describes a set of the-
oretical underpinnings and candidate hypotheses and predictions to clarify what is meant
by symmetry and parse out the possible cases. I subsequently present baseline results.
Section 4 slices the sample along heterogeneity dimensions to reveal context-specific re-
sults. Section 5 concludes.
8
2.2 Data
2.2.1 Agreements data
Data on trade agreements is derived primarily from the NSF-Kellogg Institute Data Base
on Economic Integration Agreements (EIA). This dataset distinguishes six types of trade
agreements based on the depth of economic integration going from less to most inte-
grated. These are asymmetric preferential agreements, bi-directional preferential agree-
ments, free trade agreements, customs unions, common markets, and economic unions.
The latest release of the EIA covers all country pairs from 1953 to 2014.
To ensure that the study is as current as possible, I supplement the EIA data with two
additional years of observations using the World Trade Organization’s (WTO) database
on regional trade agreements. This extends the time-series dimension of the data until
2016 and allows me to expand the treatment group.
Symmetry is evaluated by comparing the response of trade volumes to a liberalization
agreement and to a withdrawal from the said agreement - which can be equivalently
understood as a protectionist policy shock to the extent that it involves the introduction
of new trade barriers. The comparison is performed within those country pairs that have
experienced a reversal. A trade policy reversal is defined as an initial liberalization that
is later followed by a withdrawal. The initial analysis will lump all withdrawals into the
same treatment - regardless of whether the withdrawal is towards the default state of no
agreement or a less integrated arrangement. The limited number of treated pairs dictates
this generalization. Identifying exit and agreement downgrading (respectively entry and
upgrading) as separate protectionist (respectively liberalization) policies results in a loss
of explanatory power. Therefore the initial analysis will adopt a more general definition
of policy shocks. Subsequent heterogeneity analysis will allow for differentiation, with
the caveat that statistical power will be reduced.
Table 2.1. lists all country pairs that I consider to be treated. There are 159 observed
reversals, and 33 of these have to be removed from the treatment sample due to unavail-
able data on the response variable, trade volumes. The 72 treated countries span a broad
geography, including African, European, American, and Asian countries. There is also a
diversity of income levels, trade openness, and trade volume levels. Figure 2.2 captures
the diversity of the treated sample by focusing on the varying levels of wealth and trade
openness. Panel A represents the distribution of GDP per capita for treated countries
and reveals that the treated sample spans the whole income distribution spectrum. It also
shows that poorer countries concentrate. Panel B plots the distribution of treated coun-
tries’ trade-to-GDP ratios, a measure of their trade openness. Again, the distribution is
9
wide, ranging from one of the world’s most closed countries, Eritrea, to one of the most
open to trade, Hong Kong.
Table 2.1: Sample of treated units
Pair Year Pre-Exit Post-Exit Agr. length Ex. length Reentry
BDI-LSO 1998 2 0 20 17 None
BDI-MOZ 1998 2 0 20 17 None
BDI-NAM 2005 2 0 13 10 None
BDI-TZA 2001 2 0 7 6 2008
BEN-MRT 2000 2 0 18 19 None
BFA-MRT 2000 2 0 18 19 None
BGR-BIH 2007 3 1 11 2 None
BGR-MDA 2007 3 0 7 3 2014
BIH-ROU 2007 3 1 11 3 None
BIH-SVN 2004 3 1 14 2 None
BOL-VEN 2012 4 3 6 17 None
BWA-SYC 2004 2 0 4 23 2008
CHL-POL 1999 1 0 5 10 2004
CHN-ITA 1990 1 0 28 9 None
CHN-MMR 2011 3 2 7 5 None
CHN-POL 2004 1 0 14 14 None
CIV-MRT 2000 2 0 18 19 None
COL-HTI 2004 2 0 14 2 None
COL-VEN 2012 4 3 6 17 None
COM-LSO 1998 2 0 20 17 None
COM-MOZ 1998 2 0 20 17 None
COM-NAM 2005 2 0 13 10 None
COM-TZA 2001 2 0 17 6 None
CPV-MRT 2000 2 0 18 19 None
ECU-VEN 2012 4 3 6 17 None
EGY-NAM 2005 2 0 13 6 None
ERI-TZA 2001 2 0 17 6 None
EST-CAN 2004 1 0 13 12 None
EST-UKR 2004 3 0 10 7 2014
ETH-LSO 1998 2 0 20 17 None
10
Table 2.1: Sample of treated units, continued
Pair Year Pre-Exit Post-Exit Agr. length Ex. length Reentry
ETH-MOZ 1998 2 0 20 17 None
ETH-NAM 2005 2 0 13 10 None
ETH-TZA 2001 2 0 17 6 None
GHA-MRT 2000 2 0 18 19 None
GIN-MRT 2000 2 0 18 19 None
GMB-MRT 2000 2 0 18 19 None
HKG-ITA 1998 1 0 20 17 None
HKG-POL 1990 1 0 28 9 None
HUN-BGR 1991 2 0 8 10 1999
HUN-CUB 1991 2 0 27 10 None
HUN-ITA 1990 1 0 2 1 1992
HUN-MNG 1991 2 0 27 10 None
HUN-POL 1991 2 0 1 10 1992
HUN-ROU 1991 2 0 6 10 1997
HUN-RUS 1991 2 0 13 10 2004
HUN-VNM 1991 2 0 27 10 None
KEN-LSO 1998 2 0 20 17 None
KEN-MOZ 1998 2 0 20 17 None
KEN-NAM 2005 2 0 13 10 None
KNA-POL 1999 1 0 5 10 2004
LKA-BGR 2007 1 0 11 8 None
LKA-HUN 2004 1 0 14 23 None
LKA-ITA 1990 1 0 28 9 None
LKA-POL 2004 1 0 14 23 None
MDA-ROU 2007 3 1 11 12 None
MDG-LSO 1998 2 0 3 17 2001
MDG-MOZ 1998 2 0 3 17 2001
MDG-SYC 2004 2 0 4 23 2008
MLI-MRT 2000 2 0 18 19 None
MOZ-ZAF 1997 2 0 4 3 2001
MRT-GNB 2000 2 0 18 19 None
MRT-LBR 2000 2 0 18 19 None
MRT-NER 2000 2 0 18 19 None
MRT-NGA 2000 2 0 18 19 None
MRT-SEN 2000 2 0 18 19 None
11
Table 2.1: Sample of treated units, continued
Pair Year Pre-Exit Post-Exit Agr. length Ex. length Reentry
MRT-SLE 2000 2 0 18 19 None
MRT-TGO 2000 2 0 18 19 None
MUS-LSO 1998 2 0 3 17 2001
MUS-MOZ 1998 2 0 3 17 2001
MUS-POL 1999 1 0 5 18 2004
MUS-SYC 2004 2 0 4 23 2008
MWI-SYC 2004 2 0 4 23 2008
NAM-DJI 2005 2 0 13 10 None
NAM-ERI 2005 2 0 13 10 None
NAM-RWA 2005 2 0 13 10 None
NAM-SDN 2005 2 0 13 10 None
NAM-SOM 2005 2 0 13 10 None
NAM-SYC 2004 2 0 4 9 2008
NAM-UGA 2005 2 0 13 10 None
PER-VEN 2012 4 3 6 17 None
PHL-BGR 2007 1 0 11 8 None
PHL-HUN 2004 1 0 14 23 None
PHL-ITA 1990 1 0 28 9 None
PHL-POL 2004 1 0 14 23 None
POL-BGR 1991 2 0 8 10 1999
POL-CUB 1991 2 0 27 10 None
POL-ITA 1990 1 0 2 1 1992
POL-MNG 1991 2 0 27 10 None
POL-ROU 1991 2 0 6 10 1997
POL-RUS 1991 2 0 13 10 2004
POL-VNM 1991 2 0 27 10 None
ROU-BGR 1991 2 0 8 10 1999
ROU-CAN 2007 1 0 11 26 None
ROU-CUB 1991 2 0 27 10 None
ROU-ITA 1990 1 0 3 9 1993
12
Table 2.1: Sample of treated units, continued
Pair Year Pre-Exit Post-Exit Agr. length Ex. length Reentry
ROU-MNG 1991 2 0 27 10 None
ROU-RUS 1991 2 0 27 10 None
ROU-VNM 1991 2 0 27 10 None
RWA-LSO 1998 2 0 20 17 None
RWA-MOZ 1998 2 0 20 17 None
RWA-TZA 2001 2 0 17 6 None
SDN-LSO 1998 2 0 20 3 None
SDN-MOZ 1998 2 0 20 17 None
SDN-TZA 2001 2 0 17 6 None
SVN-CAN 2004 1 0 13 12 None
SWZ-SYC 2004 2 0 4 23 2008
SYC-AGO 2004 2 0 4 23 2008
SYC-COD 2004 2 0 4 9 2008
SYC-POL 1999 1 0 5 18 2004
SYC-TZA 2004 2 0 4 9 2008
SYC-ZAF 2004 2 0 4 3 2008
SYC-ZMB 2004 2 0 4 9 2008
SYC-ZWE 2004 2 0 4 9 2008
SYR-POL 1999 1 0 5 18 2004
TUR-POL 1990 1 0 10 9 2000
TUR-SYR 2012 3 0 6 5 None
TZA-DJI 2001 2 0 17 6 None
TZA-LSO 1998 2 0 3 3 2001
TZA-MOZ 1998 2 0 3 3 2001
TZA-SOM 2001 2 0 17 6 None
UGA-LSO 1998 2 0 20 3 None
UGA-MOZ 1998 2 0 20 17 None
VEN-HTI 2004 2 0 14 2 None
VEN-POL 1990 1 0 14 9 2004
13
Table 2.1: Sample of treated units, continued
Pair Year Pre-Exit Post-Exit Agr. length Ex. length Reentry
ZMB-LSO 1998 2 0 3 3 2001
ZMB-MOZ 1998 2 0 3 17 2001
Regimes : 0. No agreement, 1. APA, 2. SPA, 3. FTA, 4. FTA + CU
Sources: NSF-Kellogg Institute & WTO, adjusted by author
There are different margins of treatment heterogeneity in the constructed data, and these
margins are coded in my dataset for later analysis. The first source of heterogeneity is the
variation in the length of the agreement’s enforcement period. Another source of hetero-
geneity is that a few country pairs experience more than one reversal. For example, over
the period of study, Romania and Bulgaria experienced an alternation of three shocks:
they started the period within an agreement, suspended it in 1991, and later entered into
a new agreement in 1999. A third and final source of treatment heterogeneity is the size
of the reversal. The trade barrier restoration implied by an exit from an FTA to an MFN
regime is larger than a downgrading from an FTA to a bilateral preferential agreement.
Coding these dimensions in the data allows me to account for them in subsequent het-
erogeneity analysis.
2.2.2 Trade data
I use import volumes as the response variable to agreement reversals. The empirical liter-
ature favors import over export flows as they are generally believed to be more precisely
measured by reporting countries. Import volumes are extracted from the UN’s Com-
mission on Trade’s database (COMTRADE). Analysis with exports yields very similar
results.
I choose to focus the analysis on the 31 years from 1986 to 2016. Starting the sample in
1986, instead of an earlier date, serves two purposes. First, it minimizes the multiplicity
of time breaks that would exist if the data started earlier. The empirical strategy de-
scribed in the next section assumes that unobserved bilateral characteristics are constant
over time, and this assumption becomes increasingly questionable as the time window
expands and the possibility for structural breaks increases. Second, as observed in the
data, withdrawals were very rare before the 1990s. For these two reasons, extending the
time horizon before 1986 is more likely to generate noise than to support identification
in the subsequent analysis.
14
Figure 2.2: Distributions for GDP per capita, and trade to GDP ratios for treated countries
CEPII Gravity Data, 2000
COMTRADE collects data in a way that has potential implications for empirical analysis.
On the one hand, it receives data from member countries, making it a reporting-based
database. On the other hand, it applies a filter to reported data and drops 0- and negligible
trade flows. This approach creates some data ambiguity: missing values can indicate
either truly missing reports or a zero-trade flow. Leaving this ambiguity unaddressed
can generate bias through measurement errors. To resolve this ambiguity, I proceed as
follows. A country that reports a total of 0 imports with all partner countries in year t is
considered to have failed to report, and all of its import flows for the year t are treated
as missing. Conversely, if a country reports a strictly positive total value of imports, its
unreported flows with other countries are treated as zeroes. This approach assumes that
a country that reports any flow in year t will report all flows in that year. This is a strong
but necessary assumption to extract additional information from a dataset that would be
lost with indiscriminate treatment. Table 2.2 shows the structure of the data and the result
of this filtering algorithm.
In addition to the 72 countries that experience policy reversals at some point in time, I
include 11 additional countries. These are chosen randomly among a list of countries
that have not experienced any policy reversals over the time of the study. Therefore, the
complete data consists of a panel of 83 countries, or 6806 pairs, over 31 years.
15
Table 2.2: Summary trade data
1990 1995 2000 2005 2010 2015
Reporting countries 40 58 76 74 73 60
Missing flows 4262 4184 2930 2566 2261 2878
Treated as na 2388 1581 526 540 514 1040
Treated as zero 1874 2603 2404 2026 1747 1838
Sources: NSF-Kellogg Institute & WTO, adjusted by author
2.2.3 Stylized facts
Next, I present some stylized facts observed in the constructed dataset. Within this sam-
ple, the disappointing track record of multilateral trade liberalization is reflected (and
perhaps reinforced) by the proliferation of bilateral and regional agreements, as captured
in table 2.3. From 1996 when the Uruguay round of tariff reductions (the latest negoti-
ation round to date) had been finalized, until 2016, the number of enforced agreements
amongst our sample of 83 countries went up by a staggering 70%. By 2016, about 18%
of the pairs in the sample operated under a negotiated preferential trade regime. Over the
same period, the number of FTAs increased 17-fold. This trend points to the importance
of bilateral arrangements in regulating global trade and shapes the geography of trade
flows, making the unraveling of these arrangements essential to track and evaluate.
Table 2.3: Summary agreement data
1986 1996 2006 2016
Number of countries 83 83 83 83
Number of possible pairs 6889 6889 6889 6889
Number of agreements 576 700 974 1194
Of which
PTAs 566 644 740 818
FTAs 2 16 170 297
CUs 7 40 65 80
Sources: NSF-Kellogg Institute & WTO, adjusted by author
In my sample, 72 countries were involved in an exit from an existing agreement between
1986 and 2016. Exits add up to 126 cases, summarized in table 2.4. 93% of exits are not
immediately replaced by an interim agreement and are complete reversals to the default
16
trade regime. The default trade regime is WTO’s Most Favored Nations (MFN) for most
of the country pairs. African countries are over-represented in the sample due to (a)
significant volatility in trade policy choices as countries have historically chosen to leave
regional trading blocs either to join an alternative region or to, later on, return to the
original one (b) their status as beneficiaries of asymmetric preferential trade agreements
that come with expiration criteria. The average length of an exit in our sample is 13
years, after which some exits are reversed, and renegotiated agreements are introduced.
We refer to these reintroductions as ”reentries.”
Table 2.4: Trade liberalization reversals
Number of exits 126
Of which
from FTA 12
to nothing 117
African Pairs 68
EU accession 17
Reversed 42
Average length of exit (years) 13
Maximum 28
Minimum 1
Sources: NSF-Kellogg Institute & WTO, adjusted by author
It is worth noting that the trade liberalization reversals span the whole time dimension of
the panel (figure 2.3). Peaks are registered around globally or regionally relevant dates.
The 1990-1991 peak reflects a string of preferential trade agreements amongst the Social-
ist Bloc countries that unraveled upon the collapse of the Soviet Union. The peak around
1998 demonstrates a series of realignments between different African trade blocs. The
third peak in 2004 is the consequence of the EU’s most significant enlargement to date
when 10 Eastern European countries joined the common market and consequently relin-
quished some of their existing agreement with third parties. Each of these three different
episodes is controlled separately in the rest of this study.
17
Figure 2.3: Exits by year
Source: NSF-Kellogg Institute & WTO, adjusted by author
2.3 Methodology and Results
2.3.1 Main specification
I follow the framework in Baier and Bergstrand (2007) for estimating the trade effect of
FTAs using the gravity equation. It starts with the gravity equation proposed by Anderson
(1979) and refined for panel estimation by Baier and Bergstrand (2007) :
A ω−θτ−θi i ij
Xij = ∑ − − Ej (II.1)θ
l Alωl τ
θ
lj
Exports from country i to country j are given as a function of several variables. Ej is total
expenditure by country j on all goods. Ai is a measure of the quality of goods in country
i, ωi is the wage in i; and τij is the iceberg trade cost between i and j. Origin charac-
teristics are weighted against characteristics from competing import sources, which are
captured by the aggregator in the denominator. θ is the constant trade elasticity.
Trade frictions τij,t are assumed to have an observable and an unobservable component
such that:
18
τij,t = Trade regimeij,t + ψij + νij,t (II.2)
Where νijt is a measure of time-varying costs and frictions not related directly to the
trading regime, and ψij summarizes pair-specific time-invariant characteristics. An iden-
tifying assumptions is that the unobserved time-varying frictions are uncorrelated with
the trade regime.
For the purpose of this study, the trade regime has two components: a base component
that captures whether the pair is operating under a trade agreement, and an additional
component that describes whether the ongoing trade regime is a result of a policy reversal
(i.e. agreement exit, or reentry into an agreement). To the extent that these components
do not have to contribute equal shares to trade frictions, we can rewrite equation (2.2) as
:
τij,t = α1Agrij,t + α2Exitij,t + α3Agrij,t × Reenteredij,t + νijt (II.3)
Where Agrij,t is an indicator variable that takes value 1 when the pair ij shares a trade
agreement in year t and 0 otherwise (preferential trade agreements, free trade agree-
ments, or more integrated trading regimes).
Similarly, Exitij,t assumes a value of 1 if the pair ij in year t is trading outside of an
agreement they previously shared. Note that because the sample includes ij pairs that
never share or exit an agreement, Exitij ̸= (1− agreementij).
Finally, Reentryij,t takes value if in year t the pair ij 1 is trading under an agreement
that was previously exited and then restored.
By combining (1) and (3), allowing for time variation, rewriting in exponential form
and replacing the endogeneous and non-observable variables with country-time, and pair
fixed effects, we obtain the following specification:
Xij,t = exp(ηi,t + ψj,t + γij + β1Agrij,t + β2Exitij,t
+ β3Agrij,t × Reenteredij,t) + εij,t (II.4)
Under this specification, the identifying assumption is that once we account for time-
varying import market characteristics (such as country size, market competition, and
total spending), time-varying exporter characteristics (product quality and price), and
19
time-invariant pair characteristics (distance, and other proximity measures), the only re-
maining source of variation among trade volume predictors is the bilateral cost of trade
τij,t. Additionally, my derivation assumes that the bilateral costs of trade are predicted
by the trade regime.
Finally, the specification is augmented with an additional term. The impact of exited
agreements prior to exit can be different from unexited agreements. To address this, I
incorporate a dummy that takes value 1 while an agreement that is subsequently disman-
tled is enforced. I denote this variable as Agr×Pre−Exit. Note that this specification
controls for individual and pairwise characteristics based on which selection into agree-
ments, exits and reentries happens.
Xij,t = exp(ηi,t + ψj,t + γij + β1Agrij,t + β2Agrij,t × Pre-Exitij,t
+ β3Exitij,t + β4Agrij,t × Reenteredij,t) + εij,t (II.5)
Given the specification above, the control group is implicitly defined as country pairs for
which all treatment dummies are 0. That is, country pairs that do not share an agreement,
that have not exited an agreement, and that have not reentered an agreement. It follows
that parameter β1 measures the effect of trade agreements on trade flows. β1 + β2 is the
trade effect of exited agreements prior-to the exit. β3 measures the effect of exits from an
agreement. A positive estimate for β3 would indicate that countries who exit agreements
continue to trade more intensively than the control group, even if they no longer share a
liberalizing agreement. A negative estimate for β3 would suggest that country pairs that
reverse a liberalization policy experience reductions in their trade flows and trade less
than countries that never liberalized at all. The parameter β4 measures the marginal effect
of reentry, such that β1+β4 is the total effect of reentered agreements. A positive estimate
of β4 would therefore indicate that reentry not only reestablishes original trade gains, but
contributes new ones. Conversely, a negative β4 would mean that re-liberalization is not
sufficient to reestablish the original gains, which would point to sustained and permanent
damages from the protectionist reversal.
To account for phase-in and phase-out periods of trade agreements, I also incorporate
lagged terms of each variable.
The final estimating equation is :
20
Xij,t = exp(ηi,t + ψj,t + γij + β1Agrij,t + β2Agrij,t × Pre-Exitij,t
+ β3Exitij,t + β4Agrij,t × Reenteredij,t + γ1Agrij,t−l
+ γ2Agrij,t−l × Pre-Exitij,t−l + γ3Exitij,t−l
+ γ4Agrij,t−l × Reenteredij,t) + εij,t (II.6)
2.3.2 Hypotheses
Table 2.5 summarizes the possible cases of symmetry using notation from equation (2.6),
in the limiting case where l = 0.
Table 2.5: Symmetry cases
Coefficients
Scenario β1 + β2 β3 β4
“Positive” Asymmetry + − > β2
Perfect Symmetry + 0 ≈ β2
“Negative” Asymmetry (liberalization shocks dominate) + + < β2
β1 + β2 is expected to assumed to take on positive values
Under this specification, if shocks to trade regimes are symmetric, we would expect the
gains materialized over the period of the agreement (β1+β2 > 0) to disappear following
an exit (β3 = 0), whereas, assuming identical scope across initial and renegotiated agree-
ments, reentry should restore trade to volumes close to those observed under the original
agreement (β1 + β4 = β1 + β2).
There are two possible cases of asymmetry. In the first one, liberalization policies dom-
inate protectionist ones. Under this scenario gains from liberalization are persistent,
meaning that exit does not completely reverse the gains, and that after dismantling the
agreements, pair’s trading volumes remain above baseline. Additionally, under this sce-
nario, reentry would not only restore the full gains but further build on them. I refer
to this case as “positive” asymmetry. A scenario where exits generate losses beyond a
return to baseline, and where reentry does not suffice to restore gains is also a case of
asymmetry, with protectionist policies dominating liberalization. I refer to this scenario
as “negative” asymmetry.
This terminology choice is not normative. Rather, it describes the net effect of the se-
quence of liberalization and protectionist policies. In the first case, the net trade effect
21
of the removal and subsequent restoration of barriers to trade is positive. In the sec-
ond case, the restoration of barriers to trade causes more damage than the erasure of the
liberalization gains, which results in a net negative impact on trade volumes.
What factors would drive each of these theoretical scenarios?
Half of the treated pairs are members of the WTO at the time of exit. This implies that
following exit, tariffs are restored to their most favored nation levels, thus eroding the
targeted comparative advantage that was extended to the partner country. Some non-
tariff barriers can also be restored, as the end of a trade regime can facilitate regulatory
divergence among the pair of countries. All else being equal, exiting an agreement should
return flows to their no-agreement baseline, leading to symmetry in the impact on the
trade volume of entry and exit from agreements.
This channel ignores the possible persistence of trade gains. The direct impact of trade
agreements might be tariff reductions, which are reversed upon exit. But much of the
progress achieved under an agreement might not be immediately reversed. Improved
knowledge of the partner’s market, new relations between suppliers and importers, and
value chain connections between the pair are likely to outlast the immediate return to pre-
agreement tariffs. If the contribution of these factors to trade creation is large enough,
we should expect to see some persistence that would support “positive” asymmetry, ie. a
case where an exit keeps trade flows above baseline.“Positive”asymmetry is a hypothesis
that matches our knowledge of gains from trade and liberalization. Indeed, the litera-
ture demonstrates that liberalization causes significant reallocations and changes to mar-
ket structures, which are likely to resist to a protectionist policy of similar magnitude.
Industry-wide and within firm productivity gains (Pavcnik (2002)) as well as changing
competitive pressures and exit dynamics (Melitz (2003)), durably alter the production
base. Sticky investment decisions generated by liberalization, such as FDI and vertical
integration, might not be entirely reversed by protectionism. Additionally, trade agree-
ments introduce significant regulatory overhauls (Rodrik (2018)). The undoing of the
regulatory convergence created by a protectionist policy might not result in straightfor-
ward divergence if the transition costs for firms have already been incurred and market
access benefits already achieved.
Conversely, a less intuitive hypothesis is that trade liberalization policies might be neg-
atively asymmetric. Targeted trade liberalization increases flows by giving the targeted
partner a competitive edge. From the gravity equation (2.1) we know that the import vol-
ume of country i from country j is increasing in quality and decreasing in price and trade
costs. The diversion potential of trade agreements is well documented in the literature
(Dai et al. (2014)). An upgrade of the trade regime will therefore increase trade at the
22
expense of partners who produce at better quality and equal cost. This distortion might
be entirely reversed when the price-competition advantage is removed. Could this effect
be as large as to depress trade volumes below baseline? Trade agreements also confer
other benefits on partners, including credibility, signaling and insurance (Fernández and
Portes (1998)). It is possible that the perceived uncertainty caused by the policy reversals
hinder credibility and reliability and lead market participants to divest from the affected
trade relationship.
A similar logic applies to reentries into exited agreements. In the symmetric case, the
coefficient on the reentry dummy should be equal in magnitude and significance to the
coefficient on the original agreements. However, if repeated changes in trade regimes
increase the perception of uncertainty by economic agents, the elasticity of their deci-
sions to liberalize might be lower, leading to “negative” asymmetry. If, on the other
hand, reentry builds on the remaining network and supply chain gains from the previous
agreement, the differential impact of reentry might exceed that of the original agreement,
translating into “positive” asymmetry.
2.3.3 Results
I estimate specification (5) using OLS and Poisson Pseudo-Maximum Likelihood (PPML).
OLS forces us to log linearize the estimating equation, with the obvious shortfall of in-
formation loss due to 0 flows being discarded. PPML, originally proposed by Silva and
Tenreyro (2006) and Santos Silva and Tenreyro (2011) and since adopted in gravity equa-
tion estimations, allows us to estimate the exponential version of the estimating equation.
PPML does not require that the dependent variable be Poisson-distributed, but only that
the model be correctly specified. It is well behaved in a wide range of situations and is
resilient to the presence of many zeros in the dataset. For these reasons, PPML is my
preferred estimation throughout this paper.
Tables 2.6-2.7 present the results of the estimation, and figure 2.4 visualizes the results
when the equation is estimated using PPML. The blue lines represent the 95% confidence
interval around the point estimate of the trade flow’s response to the initial liberalization
policy (β1+β2). The green light gives the same confidence interval for the response to the
exit (β3). And the turquoise line shows the cumulative response to reentry (β1+β4). The
model is multi-way clustered along importer, exporter, and year to account for standard
error correlation between groups. This method of clustering controls for correlation in
the error term within 6 clusters i, j, t, ij, it, jt. This is the most conservative approach to
clustering and will support me in making the most conservative inference (Larch et al.
2017). Results are given at three different lags : 0, 3 and 5.
23
Table 2.6: Pooled Estimation of Exit Symmetry
OLS PPML
(1) (2) (3) (4) (5) (6)
log(flow) log(flow) log(flow) flow flow flow
(Agreement) 0.209∗∗ 0.143∗ 0.184∗ 0.218∗∗∗ 0.154∗∗ 0.189∗∗t
(0.0716) (0.0697) (0.0714) (0.0646) (0.0580) (0.0627)
(Agreement× Pre− Exit)t 0.0815 0.0389 0.0695 -0.126 0.0329
(0.289) (0.265) (0.366) (0.361) (0.333)
Exit 0.114 0.116 0.0553 0.246 -0.426 -0.00342
(0.148) (0.307) (0.292) (0.156) (0.297) (0.178)
(Agreement×Reentred)t 0.402 0.0738 0.0293 0.205∗∗∗ -0.445 -0.0881
(0.336) (0.423) (0.438) (0.0564) (0.301) (0.173)
Agreementt−3 0.120 0.128∗∗∗
(0.0797) (0.0379)
(Agreement× Pre− Exit)t−3 -0.0543 -0.0134
(0.172) (0.332)
Exitt−3 0.0765 0.765∗
(0.203) (0.325)
(Agreement×Reentered)t−3 0.521 0.682∗
(0.269) (0.298)
Agreementt−5 0.0909 0.0996∗
(0.0811) (0.0503)
(Agreement× Pre− Exit)t−5 0.00914 -0.132
(0.128) (0.265)
Exitt−5 0.216 0.335
(0.190) (0.215)
(Agreement×Reentered) ∗∗ ∗t−5 0.692 0.333
(0.244) (0.158)
Constant 15.67∗∗∗ 15.65∗∗∗ 15.65∗∗∗
(0.0223) (0.0275) (0.0290)
Observations 80121 80121 80121 103600 103600 103600
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001. Reference group is country pairs not currently or previously bound by a trade agreement.
All regressions include exporter-time, importer-time, and pair fixed effects. Fixed-effects results not reported. Standard errors (in
parentheses) are multi-way clustered at importer, exporter, and year
24
Figure 2.4: PPML Estimation of Gravity Equation: Exit Effects
Estimates and confidence intervals of shock responses from equation (5)
25
Table 2.7: Joint significance test of cumulative effects
OLS PPML
Lag = 3 Lag = 5 Lag = 3 Lag = 5
Agreements Impact Pre-exit
Joint significance 0.143 0.092 0.681 0.616
Test P-value
Exit Impact
Joint significance 0.556 0.419 0.022 0.024
Test P-value
Reentry Impact
Joint significance 0.109 0.071 0.000 0.000
Test P-value
P-values for chi-square joint-significance tests
In figure 2.4, we see some weak signs of “positive” directional asymmetry. Initial
liberalization policies appear to have a positive but statistically insignificant impact on
trade flows. A pair that exits an agreement maintains a trade volume that is statistically
greater than the baseline. A subsequent liberalization policy has a larger positive impact
than the initial one, on average. This can be interpreted as country-pairs retaining some
of the gains from liberalization, beyond the enforcement period of the agreement.
The robust takeaway from figure 2.4 is, however, that trade policy symmetry is a hypothe-
sis that cannot be rejected at the 95% confidence level. Indeed, at that level, the repeated
policy shocks have statistically indistinguishable impacts on trade flows, regardless of
their direction.
2.4 Robustness & Heterogeneity analysis
The results presented in figure 2.4 can hide significant heterogeneity along many dimen-
sions. The exit variable does not discriminate between downgrades and complete exits,
it does not distinguish between different magnitudes of trade liberalization reversals: an
exit from an FTA signifies a larger reversal than the expiration of an asymmetric trade
agreement. The depth of the exited agreement therefore matters, as well as its length.
Longer agreements, upon dismantlement, could have longer lasting effects - and there-
26
fore, lead to less detrimental exits. These and other margins of heterogeneity are explored
in this section.
2.4.1 Strict Exogeneity
To confirm that there are no anticipatory effects or phase out effects that pollute the causal
estimates in previous regressions, I run one more regression using the fixed-effects spec-
ification. Baier and Bergstrand (2007) suggests that it is easy to test for the “strict exo-
geneity” of agreements. To do this, I add a lead variable of treatment (entry/exit/reentry)
into the equation. Strict exogeneity would require that the coefficients associated with
the lead regressors indexed ij, t + 1 are small and statistically insignificant. Results in
table 2.8 yield results that support the hypothesis of exogeneity of treatment to trade
flows. Conditional on treatment, lead variables in all three columns are statistically in-
distinguishable from 0. The anticipation effects of exited agreements are similar to those
of surviving agreements, both directionally and statistically. The signs suggest that, if
anything, firms temporarily delay imports -rather than increase them- ahead of a new
agreement, as found in Baier and Bergstrand (2007). They also temporarily - though not
significantly - increase imports ahead of an exit.
2.4.2 Agreement Depth
Do deep trade policies have different symmetry properties than the less comprehensive
ones?
To evaluate this proposition I estimate equation (2.6), for two groups separately : pairs
who experience reversals relating to comprehensive agreements such as FTAs or cus-
tom unions, and pairs who experience smaller reversals to preferential trade agreements
only. Preferential trade agreements are usually unilateral discriminatory trade prefer-
ences. Most PTAs are extended under North-South development programs, such as the
Generalized System of Preferences (GPS), the US’s African Growth and Opportunity
Act. PTAs can be reciprocal. In this case, they are distinguishable from FTAs in that
they do not aim to eliminate trade barriers - but merely lower them. Figures 2.5 and 2.6
visualise the results of these estimations.
Deep trade policy decisions show stronger signs of “positive” asymmetry. The liberal-
ization shocks have a large and significant impact on trade flows, and these gains persist,
and continue to grow beyond the negative exit effect. Reentries appear to be somewhat
symmetric to initial liberalization policy. PTAs present weak evidence of “positive” di-
rectional asymmetry. In both cases, however, and similarly to the baseline analysis, the
27
Table 2.8: Panel Gravity Equation with Lead Treatment Variables
(1) (2) (3)
log(flow) log(flow) log(flow)
(Agreement)t 0.249∗∗ 0.214∗ 0.211∗
(0.0849) (0.0839) (0.0845)
(Agreement)t+1 -0.0164 0.000680 -0.00420
(0.0783) (0.0833) (0.0800)
(AgreementxPre− Exit)t 0.0173 -0.117 -0.0270
(0.207) (0.244) (0.314)
(AgreementxPre− Exit)t+1 -0.239 -0.211 0.0318
(0.219) (0.230) (0.290)
(Exit)t -0.260 -0.158
(0.148) (0.397)
(Exit)t+1 0.0808 0.336
(0.140) (0.429)
(AgreementxReentered)t 0.167
(0.545)
(AgreementxReentered)t+1 0.294
(0.479)
Constant 15.67∗∗∗ 15.68∗∗∗ 15.66∗∗∗
(0.0207) (0.0224) (0.0273)
Observations 80121 80121 80121
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
28
Figure 2.5: PPML Estimation of Gravity Equation: FTA Exit Effects
Trade flow response to deep liberalization shocks, proxied by FTAs
hypothesis of trade policy symmetry cannot be rejected at the 95% confidence level.
2.4.3 Geographic and political contexts
To account for geographic disparities, I perform three exercises.
The Common European Market offers an appealing case study. Upon joining the com-
mon market, a new member must align its trade policy with the bloc. This creates a
series of agreement exits that fall within the purview of this study. There are 16 such
cases comprising the 2004 and 2007 Common Market expansion to Central and Eastern
European Countries (PECO). For instance, upon entering the European Union in 2004,
Estonia left a free-trade agreement with Ukraine that had been enforced since 1998.
I also slice the data sample to focus on African trade. African trade is documented to
be less responsive to liberalization policy shocks. The African continent is divided into
6 different trading blocs with different levels of trade liberalization and barrier removal.
Some of these areas, mainly the South African Development Community (SADC) and
the Economic Community of West African States (ECOWAS) have experienced a set of
member exits and suspensions, mostly, due to political reasons. The regional economic
29
Figure 2.6: PPML Estimation of Gravity Equation: PTA Exit Effects
Trade flow response to weak liberalization shocks, proxied by PTAs. See appendix
Table A.2
30
Figure 2.7: PPML Estimation of Gravity Equation: Exit Effects by Geography
Trade flow response to liberalization shocks in different contexts.
areas (REC) model of trade policies in Africa has failed to deliver growth in flows. In
2013, formal intra-Africa trade made up about 10% of total annual trade of each REC
(Geda and Seid (2015)). Numerous studies confirm the disappointing performance of
intra-regional trade within RECs (Geda and Kibret, 2008, Longo and Sekkat (2004)
Yeats, 1999). This is generally attributed to export supply constraints that hamper trade
development : weak infrastructure, productivity and trade facilitation. The applicability
of the gravity model to African trade is therefore itself a point of contention (Foroutan
and Pritchett (n.d.)).
Finally, I look at the trade policy reversals that surrounded the collapse of the Soviet
Union. Several socialist countries were bound by preferential trade agreements, many of
which were dismantled in a large wave of exits in or around 1991.
The results for all three contexts are presented in figure 2.7. The behavior of African
trade flows in response to liberalization shocks is unusual but not surprising in light of the
above mentioned literature. The statistical insignificance of policy liberalization shocks
between members of the Socialist Bloc is also commensurate with the particular structure
of their economies. Once again, there are signs of “positive” asymmetry among European
countries and their partners. Still, the hypothesis of perfect trade policy symmetry cannot
31
be ruled out.
2.5 Conclusion
This paper evaluates the cost of reversals in trade policy. It investigates whether the trade
response to liberalization and protectionist policies is symmetric. Evaluating the sym-
metry of trade policy is important in the context of economic retrenchment and resurgent
protectionism after what had seemed like an irreversible march towards increased liberal-
ization. Using an exhaustive sample of agreement exits between 1986-2016, I find weak
evidence of “positive” asymmetry in the impact of liberalization: leaving an agreement
with no immediate replacement - and reintroducing trade barriers - does not appear to
obliterate all liberalization gains, and trade between the treated pairs can remain signifi-
cantly above baseline. Furthermore, after a period of exit, a reentry can also be positively
asymmetrical and exceed the gains from the initial liberalization shock.
The weak evidence of “positive” asymmetry does not preclude the hypothesis of perfect
policy symmetry. At the 95% confidence level, the size of the response of trade flows
to liberalization policies is statistically indistinguishable from their response to protec-
tionist policies. This paper looked at different geographical and economic contexts and
analyzed the role of agreement depth in determining symmetry outcomes. In all consid-
ered cases, the hypothesis of perfect policy symmetry was not rejected.
The issue of policy symmetry deserves increased attention. While widely hailed amongst
economists, trade policy liberalization has become a polarizing theme in political dis-
course. In this context, policy reversals are likely to materialize frequently. I see three
possible extensions to this research agenda. This paper reveals significant heterogeneity
in treatment effects. Baier et al. (2018) developed an approach for estimating the trade
impact of FTAs heterogeneously and obtaining individual estimates by agreement, pair,
and direction. Applying this framework to evaluating exit impacts would elicit more spe-
cific - if less precise - estimates. A second empirical extension is to narrow the focus on
tariff policies. A large part of the liberalization reversals that we observed in recent years
centered around tariff increases. Estimating tariff elasticities of trade on the way up and
down can shed further light on tariff policy symmetry specifically. Finally, to enrich the
theoretical framework that lacks the question at hand, the findings elicited in this paper
can be integrated into a gravity model of a trade by adjusting the trade cost parameters
to reflect the persistence and dynamics suggested by this study and move away from the
symmetric elasticity generally imposed on trade costs.
32
Chapter III
Perception of Trade Policy Uncertainty
and Stock Market Volatility : New
Measures
3.1 Introduction
Does trade policy uncertainty increase volatility in financial markets? Affirmative an-
swers to this question pervaded stock market news coverage in 2018-2019 1. In the
context of rising trade and geostrategic tensions, the financial news sources widely ac-
credited trade policy for increasing market volatility, pointing out both policy actions and
the perceived uncertainty surrounding them as potential causality channels. In this paper,
I investigate this hypothesis by quantifying the contribution of trade policy uncertainty
(TPU) to excess financial volatility. I examine this relationship using a novel measure
of TPU that addresses concerns with prevalent news-based measures, and employ these
existing measures for robustness checks.
That trade policy could contribute to stock market volatility is a reasonable hypothesis.
Rational asset-pricing theory conceives of stock prices as the discounted value of the
stream of future dividends. Therefore, changes to the expectations of future returns or
the discount parameters will translate into larger variations that add to excess volatility.
1Example headlines include: ”Lagarde Expects More Market Volatility With Trade Trouble” (Bloomberg,
10/13/2018), ”Volatility Erupts Everywhere as Trade War Becomes a Currency War” (Bloomberg,
8/6/2019), ”Dow Plunges 760 points in worst day of 2019 as trade war intensifies” (CNBC, 8/4/2019),
”Markets close lower, hit by earnings and trade war uncertainty” (Los Angeles Times, 5/22/2019), ”How
the trade war became the stock market’s biggest driver” (MarketWatch, 9/9/2019), ”US-China trade ten-
sions lead to volatile markets” (Marketplace, 6/19/2018)
33
In this context, news, including trade policy news, can drive volatility. A news-based
equity volatility tracker proposed by Baker et al. (2019b) shows that the share of trade-
policy-related articles has increased from just over 2% in 1985 to 26% in 2015. Less
clear is the impact of the uncertainty that the public perceives when internalizing trade
policy trajectories. This paper bridges this gap in the literature by providing a timely
analysis of the impact of TPU on stock market volatility from 2015 to 2021.
Improving our understanding of TPU and its impacts is imperative. The predictability
of U.S. trade policy has weakened over the past few years. This reduced predictability
came from the Trump administration’s willingness to deviate from the World Trade Or-
ganization’s (WTO) multilateral rules to pursue its own policy goals on trade relations
with China and other partners. This widening trade policy space continues under the
Biden administration, which has endorsed an active industrial policy. In this context, the
weakening commitment to the multilateral trade system coupled with the WTO’s inabil-
ity to enforce its own rules, as its dispute resolution mechanisms grind to a halt, opens
the economy to new and repeated TPU shocks. Measuring the size of these shocks, and
their impact on sectors of the economy has therefore become a new focus of the trade
literature (Caldara et al. (2020b), Liu et al. (n.d.), Steinberg (2019), Feng et al. (2017),
Handley and Limao (2017), Handley and Limão (2015b)). This paper gives the most
extensive account to date on the impact of TPU on stock price volatility, distinguishing
short- and long-run dynamics as well as industry heterogeneity.
The market volatility / economic policy uncertainty nexus carries renewed importance in
this context. To the extent that policy uncertainty is a byproduct of a unilateral trade pol-
icy agenda and that uncertainty in policymaking is undesirable for investors, a trade-off
between the unilateral pursuit of trade policy goals and the stability of financial markets
might arise. The bull market has dominated U.S. equities since the end of the Great
Recession. It has come to be perceived as an indicator of the state of the economy, there-
fore, creating incentives for policymakers to protect the bull market against transitory
volatility and reversals. Attention to market volatility also has economic motivations,
given its role in asset-pricing and investment behaviors. Volatility shocks change the
compensations that shareholders require for bearing systematic risk (Guo (2002)), which
in turn impacts the cost of equity capital. Through its implications for investors and pub-
licly traded firms, volatility can spill over to the real economy. Campbell et al. (2001)
shows that stock market volatility leads volatility in other economic indicators and has
significant predictive power for real GDP growth. Further, and precisely because it leads
volatility in other economic series, volatility can be an early sign of financially induced
recessions (Chauvet et al. (2015)).
This paper employs time-series methods to measure the impact of TPU shocks on volatil-
34
ity and looks both at in-sample causality and out-sample forecasting gains. In addition to
overall market volatility, I look at the volatility of specific portfolios with varying degrees
of exposure to international trade. I also allow for heterogeneity in the outcome variable
by distinguishing transitory and persistent components of volatility.
Two contributions stand out. First, the methodological approach delivers a new measure
of TPU perceptions, which combines social network data and institutional signals. It
displays desired properties of consistency with known shocks and improves upon existing
news-based measures. It can be replicated and used for future research. Second, the
analysis delivers a new result that negates the causal effect of TPU on market volatility.
I find that TPU shocks do not create excess volatility, meaning that investors do not
respond systematically to trade policy uncertainty shocks. This result contrasts with the
finding in the literature that uses news-based measures to identify the volatility impact of
broad economic policy uncertainty (e.g. Brogaard and Detzel (2015), Phan et al. (2018),
Liu and Zhang (2015)). I argue that this discrepancy emphasizes the unique character
of trade as a component of policy uncertainty and reflects methodological differences in
proxying for uncertainty.
The rest of the paper is structured as follows—section 1 reviews both the literature on
trade policy uncertainty and stock market volatility. Section 2 describes data construction
methods and a new approach to measuring public perceptions of trade policy uncertainty.
Section 3 elicits causal relationships between market volatility and TPU using in and out-
of-sample analyses and discusses the results.
3.2 Background
3.2.1 Literature on Trade Policy Uncertainty
Economists distinguish notions of risk from volatility (Knight (1921)). An agent facing
risks can assign probabilities to different outcomes and optimize based on the resulting
expectations. Uncertainty results in a failure of the probabilistic approach: the probabil-
ities are unknown, or the outcome space itself is not known (Bloom (2014)).
In the context of policy, uncertainty is often used to refer to a mix of risk and uncer-
tainty proper. Economic policy uncertainty is an umbrella term covering uncertainty in
all policy fields: fiscal, monetary, regulatory, and otherwise. The contribution of pol-
icy to agents’ perception of economic uncertainty is a corollary of the increasing role
of governments in the economy. To the extent that they engage in decision-making that
relies on expectational optimization, economic agents bear the costs of economic policy
35
uncertainty. Increased uncertainty means that assigned probabilities are more likely to
be erroneous or incomplete, leading to optimization mistakes and economic and welfare
losses. Using firm-level data, Baker et al. (2016b) find that economic policy uncertainty
is associated with greater stock price volatility and reduced investment and employment
in policy-sensitive sectors like defense, healthcare, finance, and infrastructure construc-
tion.
Earlier research of trade policy uncertainty was theoretical. Handley and Limão (2015b)
theoretically model export market entry in the context of policy uncertainty. The policy
variable is the tariff level imposed by the foreign market, the policy space is comprised
of three different scenarios, and the switching probabilities follow an exogenous stochas-
tic process. Firms optimize entry decisions using beliefs about switching probabilities.
Unsurprisingly, the model predicts that entry is a decreasing function of switching prob-
abilities when tariffs are below their maximum level. The model is tested empirically
using Portuguese-Spanish trade in the wake of the countries’ accession to the European
Community. The evidence suggests that when uncertainty subsides following a trade
agreement, industries with higher potential profit loss in the worst-case scenario see the
most entries.
The early empirical studies of the impacts of TPU use policy events as a proxy for time-
variation in uncertainty. One such event is the US-China trade relation prior to the latter’s
WTO accession. Starting in 1980, the United States extended MFN treatment to China.
This decision was conditional on yearly Congressional renewal, which created a recur-
rent uncertainty shock every time the regime came up for a vote. Alessandria et al.
(2019) use these moments of increased uncertainty to examine trade response to uncer-
tainty shocks. The results suggest that trade increases in anticipation of uncertain future
increases in tariffs. These recurrent positive TPU shocks resolved in 2000 when China
received permanent MFN status as a WTO member. The resolution of the uncertainty
was associated with simultaneous entries and exits into export markets, favoring more
competitive firms (Feng et al. (2017)), and delivered higher growth to the industries that
had higher initial potential losses from uncertainty (Handley and Limao (2017)).
Brexit has also served as an experiment to study the impact of TPU. Steinberg (2019)
builds a three-country heterogeneous firm model using an input-output production struc-
ture, and where uncertainty about trade costs impacts the firm’s export entry decision.
Uncertainty is captured through a stochastic process for trade costs. The model is cali-
brated to match a pre-Brexit I-O Table. The predicted post-Brexit equilibrium indeed
delivers a significant welfare loss; however, uncertainty’s contribution to the loss is
marginal and accounts for less than a quarter of a percent of the overall welfare cost. In
sum, uncertainty in trade policymaking appears to impact the real economy negatively,
36
mainly through its impact on firms that participate in international trade.
Recently, the literature focused on developing empirical measures of TPU to study its
impact. This approach promises several advantages. The use of proxies in theoretical
and empirical studies focuses heavily on tariff-related uncertainty, consequently down-
playing other components of trade policy such as non-trade barriers. A direct empirical
measure can be more comprehensive in scope if it does not target a single aspect of TPU.
Further, directly measuring TPU is the first step towards studying its short and long-term
time-series dynamics and making more general statements about its impacts that are not
constricted to specific case studies. Finally, tractable time-variant indices can capture
the public perception of uncertainty and the ensuing implications of this perception for
agents’ behaviors and market outcomes, which is the intent of this paper.
Two attempts at building such empirical measures of TPU stand out. Following their
previously referenced 2016 seminal paper on economic policy uncertainty (EPU) mea-
surements, Baker, Bloom, and Davis have developed corollary break-downs of EPU com-
ponents as part of their Economic Policy Uncertainty project. Their approach to iden-
tifying the trade component of EPU relies on frequency counts of uncertainty-related
news articles. They select articles from a panel of 10 leading newspapers based on the
occurrence of uncertainty terms. Caldara et al. (2020b) introduces a similar news-based
index, with a variation in the selection of key terms, and also adds two new indices: a
firm-specific index using word counts from earnings call transcripts and a tariff-volatility
measure of TPU. The authors favor the news-based measure as an index of aggregate
trade policy uncertainty. Accordingly, variations of the news-based measures of TPU are
the most common empirical indices currently available to and employed by researchers.
Olasehinde-Williams (n.d.) uses the trade component of Baker et al. (2016b)’s EPU to
evaluate the ability of U.S. trade policy uncertainty to predict global output volatility.
Ongan and Gocer (2020) also uses the news-based index to analyze the role of uncer-
tainty in US-China trade balances.
Though they align well with general events of U.S. trade policy history, the news-based
measures currently used in the literature present some shortcomings. Perhaps the most
important is the conflation of realized volatility and future uncertainty. Realized volatility
relates to the size of realized policy shocks, such as a change in tariff. On the other hand,
future uncertainty is about the size of future shocks as expected by agents. The ability to
distinguish the two forms of shocks is important to the extent that recent literature reveals
that responses of macroeconomic variables - and possibly financial variables - to the two
classes of shocks can be very different. ? shows that while financial outcomes are sensi-
tive to volatility shocks, they do not respond to uncertainty. In their assessment of their
own measures, Caldara et al. (2020b) point out that their tariff volatility measure requires
37
changes in tariff rates to signal changes in tariff uncertainty – making it unresponsive to
negotiations and proposals. They also indicate that the news-based measure picks up
high tariff volatility episodes as uncertainty shocks. Additionally, a high coincidence be-
tween the index and the timeline of known policy actions can be a source of concern. A
news-based measure that mostly picks up policy breaks is more likely to measure policy
volatility, which, while a contributing factor to uncertainty, is only a functional input and
not a sufficient statistic. Finally, newspaper coverage might not exactly coincide with
market participants’ perceptions.
In this paper, I develop a novel trade policy uncertainty index that aims to precisely repli-
cate public perceptions of uncertainty and anticipation surrounding trade policy rather
than the actual policy breaks and ex-post responses. To do this, I employ two different
types of datasets: a social-media-based dataset consisting of tweets and twitter-based in-
teractions over policy uncertainty content and institutional data comprising the USTR’s
published notices and calls for comments. Section II discusses data strategy, including
the methodology for constructing the proposed TPU index.
3.2.2 Market Volatility: Stylized Facts
The subject of stock market volatility continues to receive ample attention in economics.
Schwert (1989)’s seminal paper revealed a ‘volatility puzzle.’ The author observes that
macroeconomic fundamentals and other economic variables cannot explain the time vari-
ation in stock volatility between 1857 and 1987. Attempts at resolving this puzzle have
sought to identify causalities by introducing new variables or applying new methods
(Christiansen et al. (2012), Choudhry et al. (2016), Asgharian et al. (2013), Chiu et al.
(2018)). Another equally prolific strand of the literature is methodological in nature and
aims to improve volatility forecasting by introducing new models (see Poon and Granger
(2003) for review). A precise forecasting model and a comprehensive list of determinants
continue to be elusive.
This broad attention to market volatility from economists has several justifications. When
market participants are risk-averse, volatility plays a role in asset-pricing and invest-
ment behaviors. Volatility shocks change the compensations that shareholders require
for bearing systematic risk (Guo (2002)), which in turn impacts the cost of equity capi-
tal. Through its implications for investors and publicly traded firms, volatility can spill
over to the real economy. Campbell et al. (2001) shows that stock market volatility
leads volatility in other economic indicators and has significant predictive power for real
GDP growth. Further, and precisely because it leads volatility in other economic series,
volatility can be an early sign of financially induced recessions (Chauvet et al. (2015)).
38
Figure 3.1: Number of High Return Sessions Per Month
High return session defined as trading days with a +/- 1% S&P500 close.
Figures 1-2 show recent trends in market volatility. Figure 3.1 displays the number of
high session days, defined as days with returns larger than ±1%, per month. 2020 stands
out as a high volatility year, with March consisting almost exclusively of high return ses-
sions. Before the pandemic, 2018 saw a significant increase in volatility after a two-year
retreat. Compared to 2016 and 2017, the number of high return sessions was significantly
up. Figure 3.2 also shows that 2018 was marked by rapid successions of high return days.
The financial press interpreted the rise in volatility as a consequence of the anti-globalist
pivot of U.S. trade policy and the implied uncertainty surrounding the U.S. and world
economic trends. However, this hypothesis has not been analyzed by the academic liter-
ature, a gap that this paper tries to fill. The following section describes the construction
of market volatility and trade policy uncertainty measures, later used in analyzing the
causal claim.
39
Figure 3.2: Average Number of Days Between Two High Return Sessions
High return session defined as trading days with a +/- 1% S&P500 close.
3.3 Data and measurements
3.3.1 Modelling financial volatility
I am interested in evaluating the impact of TPU on the persistent and transitory com-
ponents of stock market volatility. In constructing the outcome variables, I follow the
cyclical volatility model proposed by Harris et al. (2011) and outlined in Chiu et al.
(2018) paper on investor sentiment and financial market volatility. In this model, the
natural logarithm of the asset price at time s follows a continuous-time diffusion given
by :
dp(s) = σ2(s)dW (s) (III.1)
Where dW (s) is the increment of a Wiener process and σ2(s) is the instantaneous vari-
ance, which is strictly stationary and independent of dW (s).
Conditional on the sample path of σ2(s), the logarithmic return is normally distributed
with variance :
40
∫ t
σ2t = σ
2(s)ds (III.2)
t1
This framework assumes that the integrated standard deviation follows a two-factor dy-
namic structure of the form :
σt = qt + ct,
ct = αct−1 + ut
where qt is a long-run component, and ct is a transitory component of volatility. Sev-
eral papers in the finance literature corroborate this two factor approach to modelling
volatility (Lee and Engle (1999), Alizadeh et al. (2002), Brandt and Jones (2006)).
For empirical implementation, I follow Chiu et al. (2018). I proxy daily standard devia-
tion using daily returns defined as the logarithmic difference of close and open prices:
rt = pclose,t − popen,t (III.3)
I then apply the Hodrick and Prescott one-sided low-pass filter to the daily standard
deviation to estimate the daily transitory component of volatility. To avoid look-ahead
bias, the filter is applied to a rolling window of 134 days. The filter’s tuning parameter
is set to λ = 5, 760, 000 which is the recommended value for data of daily frequency
(Ravn and Uhlig (2002)). The resulting daily persistent component, and daily standard
deviation are aggregated to a monthly measure as :
(∑ )1/2
q Nt 2t = (∑i=1
qti)1/2
σ Nt 2t = i=1 rti
Finally, I calculate the transitory component of monthly volatility using the two factor
definition of volatility:
ct = σt − qt (III.4)
For the baseline estimation, I compute persistent and transitory volatility for the S&P500
index over the study period. The resulting volatility series for the S&P500 index are
41
Figure 3.3: Volatility Measures for the S&P500 Index
Left panel : Total monthly volatility (black) and persistent volatility (blue). Right panel
: persistent (blue) and transitory (green) components of monthly volatility.
presented in figure 3.3.
3.3.2 A perception based measure of TPU: Twitter Component
I propose a composite measure of TPU that tracks perceptions and policy developments
by combining two indicators. The first indicator measures the public’s attention to trade
policy and trade policy uncertainty, using Twitter data. A Twitter application program-
ming interface (API) allows me to achieve better precision and explore richer data than
news-based proxies. First, due to the limited character count, tweets are more concise
statements than articles. Keyword searches can therefore lead to fewer misclassification
of tweets as TPU-focused than of news articles. In other words, the joint appearance
of uncertainty-related and trade-related keywords in a non-TPU related news article is
more likely than in a non-TPU related tweet. Second, by using Twitter data, I can look
beyond press coverage and further into the responsiveness of the wider public to trade
policy news. Indeed, newspaper reporting on trade policy does inform public perception,
but the relationship is mediated by the size, attention, and interaction of the readership.
Newspaper-based indicators cannot capture these dimensions. Social media, due to their
42
Figure 3.4: Word Cloud: Word Frequency in Selected TPU Tweets
Based on a panel of 1543 trade policy uncertainty related tweets
network structure, allow us to measure reactions to TPU-content, specifically the size
of the audience and the share of recipients who deem the message worth spreading or
responding to.
How to convert tweets into data? In generating the Twitter-based TPU measure, I begin
by constructing a panel of 12 news sources2 and 15 economist commentators3 with a
significant presence on the social network platform. For the news sources, I pick the
most circulated national newspapers and referential sources in the finance and economics
spheres, such as Bloomberg and CNBC. I selected economists with large follower bases
who are more likely to represent and influence general perceptions. I also seek to build a
panel that is gender and background-balanced so as not to pick up the bias of a specific
subset of economists. I survey the timelines of each of the accounts in the panel for trade
2The panel’s 12 news sources are: Bloomberg, CNN, The Financial Times, The Washington Post, CNBC,
Newsweek, The Boston Globe, The Wall Street Journal, The New York Times, The Los Angeles Times,
USA Today, The Economist (US)
3The economist in the panel are: Dean Baker, Sandy Black, Leah Boustan, Beatrice Cherrier, Steeve
Hanke, Seema Jayachandran, Paul Krugman, Dina Pomeranz, Dani Rodrik, Nouriel Roubini, Claudia
Sahm, Betsey Stevenson, Joseph Stiglitz, Mark Thoma, Justin Wolfers
43
policy uncertainty content over the period extending from January 2015 to June 2021.
For tweet selection, I use a similar approach to Caldara et al. (2020b) and Baker et al.
(2014). To be selected into the panel, the tweet needs to contain one word each from a
set of trade policy terms and a set of uncertainty words. The keywords are listed in Table
3.1.
Table 3.1: Tweet Selection keywords
Trade Policy Uncertainty
Trade polic* Uncertain*
Trade agreement* Risk*
Trade deal* Potential
Tariff* Unclear
Import* Likely
Export*
Trade deficit*
Dumping
The asterisks indicate potential additional letters - mostly to account for plurals. Im-
port and Export must appear in conjunction with additional policy words (polic*, tax*,
quota*, fee*, limit*, restriction*, duty*, deal*, agreement*).
Using a Twitter API, I extract all trade-uncertainty-related tweets generated by the panel
members throughout January 2015 to June 2021. The proposed index is a perception-
based indicator that reflects the public’s attention to TPU and how much they interact
with trade uncertainty-related content. Let:
i = 1...N indexes the tweets,
j = 1, ..., K indexes users active in month t
t = 1...T , with T = 65 index the time.
I define the amplification weighted-frequ∑ency measure as :∑K N
i=1 InteractionsijtAWFt = (III.5)
followers
j=1 jt
Interactions are defined as the sum of retweets, likes, replies, and quotes. The histori-
cal follower data is not available via Twitter API and was recovered using the Internet
44
Archive’s Wayback Machine. The followers’ data is missing in 9 tweet-month observa-
tions, and the associated tweets are dropped. This generates a slight loss of information.
For this, and all other indices presented here, the series are standardized to unit-standard
deviation and then normalized to have a mean of 100.
Notice that this index is a weighted sum of published TPU tweets, where the associated
weights are the number of received interactions (normalized by follower base). Each
one of these interactions in itself amplifies the original message: retweets expand the
readership base, likes indicate the number of people who align themselves with the mes-
sage, and replies measure the number of readers who engage with the content favorably
or unfavorably. Together, they indicate the contemporary relevance of the tweet and de-
termine the size of the audience it will reach, thus propagating the uncertainty message.
For normalization purposes and to give a notion of scale, the total number of reactions is
divided by the total number of followers at the end of the given month.
One property of this indicator is that it will not count tweets that do not draw any re-
actions. This is a desirable property because such tweets are likely not indicative of
increased uncertainty perception by the public. Conversely, a tweet that elicits large
numbers of interactions receives a higher weight. In short, the indicator answers the fol-
lowing question: among the persons who were exposed to the TPU-related tweet, how
many deemed the topic relevant and/or worthy of amplification?
3.3.3 A perception based measure of TPU: Institutional Component
The second indicator uses institutional signals of trade policy uncertainty. Within the
architecture of the U.S. government, the United States Trade Representative (USTR) en-
joys a wide purview overrules and policies that regulate U.S. trade. The USTR is respon-
sible for negotiating, implementing, and reviewing the U.S. trade agreement, resolving
disputes, and shaping global trade policy in the different multilateral venues such as the
WTO. Under the Administrative Procedure Act (APA), the USTR’s decisions are subject
to public notice and call for comment procedures. Potential policy changes and actions
are signaled at a very early stage, as they become entertained by the administration. At
this stage, rules and policies are yet unformed and very much in progress: their scope,
implementation, and viability are uncertain. This process provides a rare opportunity
for capturing uncertainty about policymaking ex-ante instead of picking up the informed
response to fully formed policy changes ex-post.
I have reviewed and extracted data on all USTR public notices and calls for comments
from 2013 to 2021. I classified all releases by theme and by partner. In the federal
register, notices are associated with a release date and, if applicable, an expiration date.
45
Figure 3.5: Twitter-Based Indicator of Trade Policy Uncertainty
Designed using Twitter application programming interface.
46
For each notice, I research a resolution date, that is, the date at which the issue raised
by the notice is resolved. For instance, the resolution date of a notice announcing a
WTO dispute is the release date of the WTO panel report. In contrast, the resolution date
for a notice announcing an out-of-cycle country review under the Generalized System
of Preferences (GSP) is the publication of the review decision. Between its publication
and its resolution, each notice corresponds to a potential disruption in the way trade and
trade policy are conducted - generating additional uncertainty. A notice is considered
open at time t if and only if t is between the publication and resolution times. Under
this assumption, the count of open notices at any given time becomes a good indicator of
the trade policy uncertainty that market participants experience at any given time. This
indicator is simply defined as :
ORCt = the number of open calls and notices at time t (III.6)
where ORC stands for open review count. This indicator addresses a common issue with
existing indices and proxies of trade policy uncertainty. Previous studies of policy uncer-
tainty exploited policy changes, specifically tariff policy, as proxies for uncertainty, due
to the volatility they can generate in the time window following their enactment. Such
a proxy can be misleading. After a policy is enacted, agents only face the risks of im-
plementation. The state of the world, as relates to that policy component, is well known
to agents who can now update their optimization frameworks with knowledge of the sur-
rounding environment. This makes policy breaks an inadequate proxy to measure policy
uncertainty. My proposed indicator measures genuine uncertainty around trade policy: at
the time where a policy is added to the count in ORCt, there is uncertainty about whether
the policy will, in fact, change, and even more uncertainty about how it will change if
it does. As a result, the release of a new policy-related call for comments increases the
level of uncertainty that agents face by bringing into question their optimization environ-
ment without providing complete probabilities on the possibility, direction, and size of
the change.
Arguably, this measure can be further refined. While every additional call for comment
increases perceived uncertainty, not all calls for comments contribute in equal propor-
tions to agents’ perceptions depending on the scope of the announcement. For instance,
calls for comments could be weighted by the impacted trade volumes to account for these
disparities. I do without this refinement to reduce the number of inputs and facilitate the
replication of the index.
The two individual indicators capture different components of public perceptions of TPU.
They also each contain noisy signals that are not related to TPU. I employ a dynamic
47
Figure 3.6: Institutional Signals Indicator of Trade Policy Uncertainty
Designed using USTR notices and calls for comments.
48
factor model to extract TPU information from both series while reducing the amount
of noise they contain. One can decompose each of the time-series into two orthogonal
unobserved processes: the common component, driven by a shock that captures policy
uncertainty, and the idiosyncratic component, which is driven by shocks that are series-
specific or local. This approach reinforces the consistency of the final index in that it
elicits the underlying process that is common to both the institutional signals of TPU
and media coverage and discussions of the same while minimizing noisy signals. It also
improves upon existing methods in the literature that are limited to a single source and
type of uncertainty, and therefore more prone to imprecision.
Let ft be the single hidden factor (or common component). Each one of the indicators is
a function of the hidden factor such that :
Iit = λi,0ft + λi,1ft−1 + ...+ λi,pft−p + ei,t (III.7)
where Iit = [AFW ,ORC ]′t t and,
ft = ψ1ft−1 + ...+ ψrft−r + ηt (III.8)
where, p is the lag order of the measurement equation, ei,t is an idiosyncratic component
and r is the lag order of the common factor’s autoregressive process. The static version
of the model can be written as :
{
It = ΛFt + et (III.9)
Ft = Φ(L)Ft−1 +Gηt
The dynamic factor model is estimated via Principal Component Analysis (PCA). Being
a non-parametric technique, PCA does not require additional model specifications and
thus provides potential robustness against misspecification. This property allows me to
remain agnostic about underlying relationships in the model.
The final TPU perception index is presented in figure 3.6.
49
Figure 3.7: Proposed TPU Perception Index
The index is constructed as the common factor of the Twitter and Policy signal indica-
tors.
50
3.3.4 Discussion of proposed index
The Twitter amplification-weighted frequency measure aligns well with event-based pri-
ors on TPU. The series, presented in figure 3.3, is below average throughout 2015 and
for most of 2016. The first peak appears in the lead-up to the 2016 election. The index
remains consistently above the period’s average from March 2018, which coincides with
the early tariff decisions of the new U.S. administration. The February 2019 peak was
caused by highly-amplified tweets about the likely impact of tariffs on future inflation
and growth, a possible US-China trade deal, and renewed tensions in the US-EU trade
relations, including threats of tariffs on European car-makers. The August 2019 peak
picks up wide attention to Fed statements about TPU and trade policy in general, in-
cluding statements by Federal Reserve Chairman Jerome Powell and branch presidents
about monetary policy adjustments in the face of mounting trade tensions. The peak
also incorporates tweets about a possible change in the first tranche of tariffs imposed
on Chinese goods, the finalization of the second tranche, and threats of retaliation from
China. This month is also marked by a general interest in economist commentaries on
trade policy uncertainty as a risk factor for the economy. This is reflected in the unusu-
ally high interactions received by these tweets. A main advantage of this proposed index
is precisely its ability to go beyond the news cycle, and monitor the intensity and breadth
of TPU-related conversations amongst the public as they happen.
Similar to the first indicator, institutional signals of uncertainty increase after March 2018
and post-above-average measures for the remainder of the study period (figure 3.6). The
shape of the series also presents a clear break following the political transition of 2017.
Unlike its twitter-based counterpart, this indicator is more persistent, which appears con-
sistent with its institutional nature. It is not surprising that public attention should be
more volatile than the underlying policy signals of uncertainty.
Figure 3.8 plots the proposed index against two existing TPU indicators in the literature,
both proposed by Baker and co-authors. The dashed line represents their newspaper-
based TPU component of EPU index Baker et al. (2016b), whereas the dotted line is
the trade-component of their equity market volatility tracker Baker et al. (2019b). The
two indices are similar in construction, but the latter is normalized to match the period
average of the CBOE volatility index, the VIX.
My proposed index has a comparable shape to existing measures. However, it presents
desirable properties that separate it from them. First, it is noticeably smooth. Figure
3.9 depicts the quantile distribution of first-order differences. Large and sudden jumps
and troughs are fewer in the proposed index. This property results from the index’s
inclusion of fundamental policymaking developments by including the USTR data and
51
Figure 3.8: Proposed Index and Existing TPU measures
proposed TPU perception index (solid line) against Baker et al. 2016 TPU component of
EPU (dashed line) and Baker et al. 2019 TPU component of a news-based Equity Market
Volatility Tracker (dotted line).
52
Figure 3.9: Proposed Index and Existing Measures : Quantile Distributions
Quantile distribution of first difference series of compared indices : proposed TPU per-
ception index (solid line) against Baker et al. 2016 TPU component of EPU (dashed line)
and Baker et al. 2019 TPU component of a news-based Equity Market Volatility Tracker
(dotted line).
53
its comparatively small dependence on the fast-moving news cycle. Indeed, while we
expect to see some variability in the index, vast movements are not commensurate with
public perception of uncertainty, which should be highly serially correlated. We can
see an expression of the index smoothness towards the end of the study period. Policy
actions have significantly subsided in the second half of 2019 and going into 2020. The
landing of the perception index is phased and progressive in this period but much faster
in the alternative measures. News outlets might be quicker to turn the pages of the news
cycle than the informed public perceptions, and institutional process develop. The higher
persistence of the index is a desirable property for a measure of perceptions.
The USTR policy signals instill a forward-looking property into the index, which is a
desirable property. This is visible in the behavior of the indices in March 2018, a pe-
riod when the U.S. took several highly mediatized policy actions. The decision to slap
tariffs on aluminum imports was announced on March 1st. The unfair trade practices
review of China was released on March 22nd, laying the ground for the ensuing tariff
increases. Both of these developments were, however, already folded into the perception
index. The steel and aluminum tariffs enter the index as early as March 2016, when the
USTR issued a call for comments on the global steel and aluminum markets, teasing a
possible policy action. The out-of-cycle review of China’s trade practices under section
301 was announced in August 2017 and enter our index then. Neither the decision to
enforce tariffs nor the release of the report is clear positive shocks to uncertainty. In fact,
both of these developments resolve some uncertainty about the trajectory of U.S. trade
policy. They are primarily realized shocks to policy rather than a change in the size or
direction of expectations about the future. The forward-looking property of the index is
also demonstrated by the earlier and slower build-up towards the high uncertainty period.
In contrast, the alternative indices rapidly climb on the first reports of policy actions.
The indices carry different information. The comparatively high peaks recorded by the
alternative indices in December 2018 are much smaller in my proposed index. It is un-
clear what these peaks are associated with. In that month, the major U.S. trade policy
development was a tariff truce agreed to by the United States and China following a G-20
summit, arguably a negative shock to uncertainty. The large February 2019 peak in the
proposed index is conversely absent from the other series. Instead, the proposed index
picks up the U.S. last-minute decision to delay to an unspecified date a 15-percentage-
points tariff increase on $200 billion of imported Chinese goods. It is also heavily
boosted by the public attention to a January U.S. government threat against European
auto imports.
In an additional verification, I compare the proposed TPU index to an indicator mea-
suring the number of policy actions that the USTR has taken over the study period.
54
This indicator is a simple count of the number of trade policy actions, standardized and
normalized to a mean of 100. Figure 3.10 reveals that the proposed uncertainty index
contains different information than one would get from observing trade policy changes.
The correlation coefficient between the two series is low, at 0.164.
Figure 3.10: Proposed Uncertainty Index and Trade Policy Volatility
Note: Proposed TPU perception index (solid line) against a constructed measure of US
trade policy actions (dashed line).
In sum, the proposed index meets the design goals. It captures variations in TPU, as
shown by its similarity with alternative indices. It is forward-looking and does not track
moments of policy changes. It also does not align with policy actions. Finally, it suc-
cessfully integrates institutional and social media signals on trade policy uncertainty and
displays the stickiness that characterizes public perceptions.
3.4 TPU and Stock Market Volatility
Analyzing the impact of TPU on stock market volatility serves two purposes. First, it
attempts to assess the economic and financial cost of the recent high uncertainty episode.
The policy goals telescoped by the U.S. administration from 2017 to 2020 were narrowly
defined in terms of trade balance improvement with specific trading partners. This has
translated into deviations from past policy trends, revisions of existing agreements, and
suspension of WTO resolution mechanisms, all of which have significantly increased the
55
level of uncertainty as perceived by market participants, and as relayed by the media
coverage.
Nevertheless, due to the short track record of the policy shift, the literature has focused
less on counting the cost and more on simulating potential impacts4, understanding the
tariff and retaliation policy designs (Fetzer and Schwarz (n.d.)), or cataloging previous
episodes of trade conflicts (Mattoo and Staiger (2020)). Using the constructed TPU per-
ception index, we can begin to investigate one of the consequences of this policy change:
increased uncertainty. In this context, financial markets offer an excellent early case
study: they respond quickly to policy changes, they inform about investors’ attitudes
towards policymaking, and their dynamics can have relevance for the real economy.
Stock market volatility, according to Schwert (1989), reflects uncertainty about future
cash flows and discount rates, and thus informs on future economic activity. It also in-
creases the cost-of-capital, which can reduce future investment (Guo (2002)). Campbell
et al. (2001) shows that stock market volatility is a significant predictor of GDP growth.
Therefore, through its impact on volatility, we can make informed hypotheses on TPU’s
impact on aggregate economic performance. Furthermore, understanding the stock mar-
ket impact of uncertainty can support investors in designing adequate responses to adjust
their positions when they anticipate an uncertainty shock.
This section describes the approach for in-sample and out-of-sample analyses and dis-
cusses the results of this exercise.
3.4.1 Model and Estimation
The relationship between the TPU and market volatility series is explored using a struc-
tural vector autoregression (Sims (1980)). This is a common approach in the investiga-
tion of sentiment shocks and their market outcomes. I begin with the following VAR
equation:
∑p
Yt = A0 + AkYt−k + ut (III.10)
k=1
And,
Yt = [gt, πt, rt, tput, volt]
′ (III.11)
The inclusion of macroeconomic variables allows one to control for business-cycle-
4Bouët and Laborde (2018), Caceres et al. (2019) and Jeanne (2019) for theoretical modeling of trade wars
consequences
56
related shocks. The macroeconomic variables are from the St Louis Fed database, FRED.
gt is output growth, proxied by the log difference of the monthly industrial production
index, πt is monthly PCE inflation, and rt is the effective federal funds rate to capture
shocks to monetary policy. tput is the TPU perception index. The model is separately
estimated for different specifications of volt: total, persistent and transitory volatility.
The main estimation centers on the volatility response of the S&P 500 index, and I look
at heterogeneity responses of a selection of individual stocks in a following subsection.
The chosen lag for the VAR equation is k = 1, as suggested by the AIC criterion.
The structural shocks are derived using the Cholesky decomposition. The ordering of the
endogenous variables captures the restriction imposed on the system. The key restriction
is that TPU shocks do not propagate contemporaneously to the macroeconomic variables.
This restriction rests on the notion that the changes in investor and market participant be-
havior that lead to variation in macroeconomic variables lag changes in perception and
sentiment, as measured by the index (Baker and Wurgler (2006)).
3.4.2 Baseline Results
Figure 3.11 presents the resulting impulse response functions of macroeconomic funda-
mentals and S&P500 volatility following a positive shock to TPU as computed by the
Cholesky decomposition. A positive shock to TPU leads to a positive response of in-
dustrial production and inflation. The response appears short-lived. Monetary policy
responds with lower rates initially, but the response is not significant at the 95% confi-
dence level.
Figure 3.12 shows that the volatility components respond uniformly to TPU shocks: both
persistent and transitory volatilities decline slightly at the 95% confidence level. Tran-
sitory volatility is quicker to absorb the shock, whereas persistent volatility is slower to
adjust back from an initial dip. As a result, total index volatility responds with a quick
dip and a somewhat slow recovery. The persistent and transitory volatility responses
are larger at the peak than total volatility response. This reflects a negative correlation
between the persistent and transitory component. The Spearman correlation coefficient
between the two series over the study period is -0.18.
The negative and consistent response of stock market volatility suggests that TPU has
a chilling effect on market transactions overall, generating perhaps a wait-and-see atti-
tude that stabilizes stock prices. Bootstrapped confidence intervals show that we cannot,
however, rule out the null hypothesis that TPU shocks do not propagate to stock mar-
ket volatility. The volatility forecast-error variance decomposition in the context of the
57
specified SVAR, as reported in table 3.2, further reflects the weak contribution of TPU
shocks to volatility variations.
Figure 3.11: Responses to a Trade Policy Shock
Impulse response function of the SVAR system variables (industrial production growth,
inflation, interest rate, TPU and total monthly volatility of the S&P500 index) to a one
standard deviation shock to Trade policy uncertainty.
Table 3.2: Forecast Error Variance Decomposition of S&P500 Monthly Realized Volatil-
ity
Horizon Economic Growth Shock Inflation Shock Monetary Shock TPU Shock Other Shocks
1 0.0013 0.0656 0.2191 0.01024 0.70362
5 0.0020 0.0636 0.2165 0.01320 0.70455
10 0.0020 0.0636 0.2165 0.01323 0.70449
15 0.0020 0.0636 0.2165 0.01323 0.70447
19 0.0020 0.0636 0.2166 0.01323 0.70446
3.4.3 Out-of-Sample Analysis
Investors and market participants use historical trends and correlations to predict future
market dynamics better. The in-sample performance of TPU in explaining volatility
is weak. Can the inclusion of trade policy uncertainty in forecast models of volatility
improve their accuracy?
58
Figure 3.12: Responses of Volatility Components to a Trade Policy Shock
Impulse response function estimated using the structural VAR model separately using
transitory and persistent volatility components as fifth endogenous variables.
59
In keeping with the literature, I estimate an AR(6) benchmark forecasting model of
volatility. The benchmark forecasts are denoted Vt+m,B. I then use this benchmark to
evaluate the performance of a TPU-augmented forecast model given by:
∑6
V olt+m,A = αm + βp,mV olt+m−p + δmTPUt+m−1 + εt+m (III.12)
p=1
Forecasts are generated using a one-step-ahead recursive approach with an expanding
window, starting with an in-sample of 35 observations. At each step, from t = 35
onwards, the model’s parameters are estimated via OLS using historical data available
up to time t. The forecast is generated using the observed lag values of the predictive
variables. This process generates monthly volatility forecasts from July 2018 through
June 2021.
Following Wang et al. (2018), Rapach et al. (2009) and Campbell and Thompson (2007),
I evaluate model performance using out-of-sample R2 given by:
2 − MSPEB∆ROOS = 1 (III.13)MSPEA
where ∆R2OOS measures the percent reduction in mean squared predictive error (MSPE)
gained by transition from the benchmark to the augmented model. A positive value
therefore means that the augmented model improves upon the accuracy of the benchmark
model. Following Wang et al. (2018), I compute the ∆R2OOS for two versions of the
augmented model: with and without restrictions. The unrestricted model follows the
method laid out above. The restricted model is such that:
{ ∑6
Vˆ
α̂m +
ol = ∑p=1 β̂p,mV olt+m−p + δ̂mTPUt−1, if δm > 0t+m,A,R (III.14)
α̂ + 6m p=1 β̂p,mV olt+m−p, if δm < 0
The investor using the restricted model minimizes overfitting by excluding TPU from the
forecast when the associated coefficient is not consistent with their prior that TPU must
correlate positively with volatility.
Figure 3.13 reports the forecast performance results. The out-of-sample R-squared on
the unrestricted and restricted models are -49.19 and -16.20, respectively. The inclu-
sion of TPU into volatility forecast models does not improve upon the benchmark auto-
regressive model, further reinforcing the results from the in-sample analysis.
60
Figure 3.13: Log Volatility: True value and Alternative Forecasts
TPU-augmented forecast (pink) ∆R2OOS = −49.19 and restricted TPU-augmented fore-
cast (yellow) ∆R2OOS = −16.20 against the benchmark volatility forecast (green) and
the true volatility value (blue).
3.4.4 Portfolio Evaluation
Lastly, I identify the utility gain from the internalization of TPU dynamics in portfolio
allocation decisions. This exercise follows Wang et al. (2018) which builds on prior
literature (Neely et al. (2014), Rapach et al. (2009)).
Assume that a risk-averse investor has access to two assets: risk-free bonds and stock
equity. The utility function of the investor displays mean-variance preferences. It has
two components: the expected portfolio return, which enters positively, and a portfolio
volatility component that enters negatively with a weight that increases with the degree
of risk-aversion. The utility function can be written as:
1
U(rt) = Et(wtrt + rf,t)− γvart(wtrt + rf,t) (III.15)
2
61
Where rf,t is the risk-free rate on bonds, rt is the excess return on stocks, and wt is the
weight of stock equity in the portfolio. γ measures risk aversion. The optimal level for
the choice variable wt is a function of expected stock returns, volatility, and risk aversion.
( )
∗ 1 r̂t+1wt = (III.16)γ σ̂2t+1
Where r̂t+1 is a simple historical-average forecast of stock returns. The volatility forecast
is formed in three alternative ways using the benchmark forecast, the TPU-augmented
model, and the TPU-augmented restricted model σ̂2 = {Vˆol , Vˆol , VˆB A olA,R}. This
yields three possible portfolios indexed by p.
Once the weight is chosen, the portfolio return is given by:
Rt+1 = wt × rt+1 + rf,t+1 (III.17)
The performances of the three different portfolios can be compared by using the certainty
equivalence return (CER):
γ
CER = µ̂ + σ̂2p p p (III.18)2
Where µ̂p is the mean and σ̂ is the variance of the chosen portfolio over the entire period.
The CERs of the three alternative portfolios are shown in figure 3.14. Models that inter-
nalize TPU fail to improve the asset allocation and the associated returns.
3.4.5 Heterogeneity Analysis
The limited relevance of trade policy uncertainty shocks to market volatility broadly
measured through the S&P 500 might hide differences across sectors. To investigate
this possibility, I conduct a heterogeneity analysis using six sectoral exchange-traded
funds (ETFs) of the S&P-500. The chosen sectors have varying degrees of exposure to
trade policy and to international market conditions. On the one hand, the technology
(XLK) and industrials (XLE) sub-indices cover companies with high trade exposure due
to both a large share of revenues from non-US sales and global value chain linkages. On
the other hand, the utilities (XLU) and healthcare (XLV) ETFs represent more insulated
industries whose revenues are overwhelmingly domestically derived and are thus less ex-
posed to direct trade policy shocks. I also include two additional indices: energy (XLE)
62
Figure 3.14: Certainty Equivalence Return of Alternative Portfolios
Portfolios constructed using three different forecasting methods for the derivation of the
optimal equity weight: a benchmark forecast, a TPU augmented model and a restricted
TPU augmented model.
63
and consumer staples (XLP), intermediate sectors with majority domestic revenues and
moderate sensitivity to supply chain perturbations.
Figure 3.15 shows impulse-response functions for the total realized volatility to a shock
of TPU, across sectors. Consumer goods and manufacturing sectors appear to respond
with a slight uptick in volatility, but the increase is short-lived and quickly reversed.
The response of the different sectors displays similar magnitudes and trends and aligns
with the response of the aggregate S&P500 index. There is no evidence of sectoral
heterogeneity. This result emphasizes that rather than propagating the uncertainty shocks
in specific sectors, the stock market tends to respond with decreased volatility across the
board. It emphasizes that investors do not collectively and systematically change their
positions or substitute across sectors following a TPU shock.
Figure 3.15: Impulse Response Function of Volatility to TPU Shocks: By Sector
IRFs estimated using the specified SVAR in (10) using different measures of volatility
for specific sector-ETFs.
Another level of possible heterogeneity is across individual stocks. Of particular inter-
est are stocks of firms with significant trade exposure. These firms might indeed respond
more strongly and differently than general market trends. To test this hypothesis, I choose
7 US companies with some of the highest earning-exposures to China. Figure 15 shows
the total volatility response of these high China-exposure stocks to a TPU shock. Here
we observe both size and directional heterogeneity. Tesla’s strong volatility response
could be driven by the company’s reliance on China both for production capacity and
sales revenue. Still, revenue exposure does not appear to be a crucial discriminating fac-
64
tor in determining the size of the response. QCOM and MU, generate more than half
their earnings through sales to China but their stocks appear less volatile in response to
TPU shocks than IPGP and TSLA both of which are less reliant on exports to China.
Rather than eliciting a systemic volatility response of the whole market or along specific
segments, trade policy uncertainty appears to affect individual stocks idiosyncratically
without significantly destabilizing financial markets.
Table 3.3: List of High Exposure Stocks
Ticker Symbol China as Share of Revenues
QCOM 66%
MU 57%
TXN 44%
IPGP 43%
AMD 39%
VECO 35.8%
TSLA 21%
Financial data from Yahoo! Finance and CNBC
3.4.6 Robustness
To insure that the results of the analysis are not driven by properties of the constructed in-
dex, I run the same SVAR using Baker et al. (2016b) trade policy component of economic
policy uncertainty as a measure of TPU. The impulse response functions are presented in
figure 3.17. These results corroborate my own analysis. The impact of TPU on market
volatility continues to be insignificant initially, and then slightly negative.
3.4.7 Discussion
The above analysis demonstrates that trade policy uncertainty does not systematically
increase volatility on stock markets and that where a causal relation is significant, it is
usually negative. Trade policy uncertainty also has no predictive power for volatility, and
internalizing it in investment decisions does not improve portfolio performance. Trade
policy uncertainty does not contribute to market instability. Furthermore, the hetero-
geneity analysis emphasizes that the absence of a financial volatility response to TPU
cannot be explained away by the moderate degree of exposure to international trade of
U.S. markets.
65
Figure 3.16: Impulse Response Function of Volatility to TPU Shocks: Individual Stocks
IRFs estimated using the specified SVAR in (10) using different measures of volatility
for specific companies.
The findings of this paper stand in contrast with the existing literature on the volatility-
effect of economic policy uncertainty (EPU). Using Baker et al. (2016b) EPU index, Liu
and Zhang (2015) show that a one standard deviation increase in EPU leads to a 0.03%
increase in volatility and that EPU has a strong out-of-sample predictive power. Running
a structural VAR using my main specification in (10) but replacing TPU by Baker et al.
(2016b) EPU index yields similar results (figure 3.18). Asgharian et al. (2013) find that
a macroeconomic uncertainty index based on forecast dispersion significantly increases
long-run stock market volatility. This increased volatility causes a flight-to-quality be-
havior as evidenced by a reduced cross-correlation of stock and bond markets at times of
high macroeconomic uncertainty. Amengual and Xiu (2018) find that downward volatil-
ity jumps are associated with a resolution of monetary policy uncertainty, mostly through
statements from the FOMC and Fed chairman speeches.
What drives the difference in volatility response to EPU and TPU? Trade policy uncer-
tainty is a new form of uncertainty. For most of recent economic history, trade policy
has been stabilized by the commitment of the U.S. government to the WTO’s multilat-
eral framework that dictates the rules of global engagement. The transparency and anti-
discriminatory intent of the rules-based trading system and the alignment of consecutive
administrations with global trade liberalization enhanced the predictability and stabil-
66
Figure 3.17: Impulse Response Function to a TPU Shock, using Baker et al’s Trade
Component of EPU
IRFs estimated using the specified SVAR in (10), and using total S&P500 volatility as
stock volatility measure.
67
Figure 3.18: Impulse Response Function to an EPU Shock
IRFs estimated using the specified SVAR in (10), and using total S&P500 volatility as
stock volatility measure.
68
ity of the policy framework. The precipitated rise in trade policy uncertainty in recent
years is unusual. It combines institutional changes, global tensions, a volatile domestic
policy agenda, rendering it harder to navigate for the public. Faced with an unfamiliar
shock, investors might fail to develop a hedging strategy or reallocate assets - and choose
to hold positions in a wait-and-see attitude. In other words, investors choose to ignore
uncertainty when they do not know how to interpret it correctly.
This interpretation finds support in the theoretical literature on uncertainty and volatility.
According to Ederington and Lee (1993), unanticipated news are resolved by the market
through increased volatility. Indeed, ? show that economic variables respond to current
economic volatility, defined as the size of shocks that have just occurred. However, un-
certainty shocks, defined as changes to agents’ expectations of large future shocks, do
not affect economic variables. The authors show that investors paid premia that average
to zero to hedge shocks to uncertainty. Dew-Becker et al. (2021) supports this distinction
between realized shocks and perceived uncertainty. The former carries negative risk pre-
mia, whereas the latter does not. According to the authors, forward-looking uncertainty
shocks do not drive investor’s marginal utility - an argument that the asset allocation ex-
ercise in this paper supports. The previously cited studies that relate uncertainty shocks
to increased volatility rely on uncertainty proxies that might not narrowly capture pure
uncertainty. News indices are indeed much more likely to capture realized shocks to
policy rather than forward-looking uncertainty perceptions. It follows that attempts at
studying TPU using news-based indices are likely to lead to mistaken conclusions pre-
cisely because they are backward-looking and reflective of the size of realized shocks
rather than expectations of future shocks. Hedging behavior appears very limited when
trade policy uncertainty shocks are measured using forward-looking perception indices.
The stock market response is limited in scope and small in size across sectors.
Conclusion
The secular stability of trade policy brought about by the post-war multilateral trade sys-
tem has been disrupted by new geopolitical rivalries, populist discourses, and the supply-
chain challenges posed by a global pandemic. Protectionist policies are challenging the
sense of inevitability that cloaked multilateral liberalism. As a result, market participants
have to contend with a new and increasingly relevant source of uncertainty: trade policy
uncertainty (TPU).
I propose a new measure of trade policy uncertainty to support empirical research into
its drivers and economic consequences. The index aims at capturing market participants’
69
perceptions of TPU. It combines information from two distinct indicators measuring
public attitudes towards TPU on Twitter and institutional signals of trade policy changes.
The index reveals a large and sustained increase in TPU between March 2018 and early
2020.
Using the constructed index, I study the effects of TPU on macroeconomic variables
and stock market volatility. Whereas the literature shows a positive relationship between
economic policy uncertainty and returns volatility, I find that trade policy uncertainty, in
particular, does not increase overall market volatility. Heterogeneity analysis reveals that
increased uncertainty does not impact sectors differently. Investors do not appear to flee
to under-exposed sectors. The negative relation between TPU and excess returns stresses
the particularity of the uncertainty episode under study.
This paper advances the research on trade policy uncertainty in two significant regards.
On the one hand, it puts forth a new tractable, reproducible measure of TPU perception
that can support further empirical research of this increasingly relevant source of uncer-
tainty. On the other hand, it shows that the market’s response to TPU is markedly differ-
ent from its response to uncertainty stemming from other economic policy components.
The results of this paper invite questions about the significance of impacts of short-term
trade uncertainty episodes on the real economy. Its methodological contribution provides
a tool that can help address such questions.
70
Chapter IV
China’s Aid and Trade Diplomacy: A
Zero-Sum Game?
4.1 Introduction
In 125 BC, Zhang Qian returned to the Han Imperial capital of Chang’an from a 13-year
long trip, bearing news of lands “rich in unusual products whose people cultivated the
land and made their living in much the same way as the Chinese” (Sima (1993)). Zhang
Qian was China’s first government-mandated diplomat. Through successive missions
to western China and Central Asia, his travel diplomacy facilitated the establishment of
the Han empire’s silk road, for long the world’s largest trade network. Today, China’s
capitalism with Chinese characteristics continues to rely on diplomacy in the conduct of
international business. Active Chinese diplomacy continues to support business expan-
sion into new markets.
The ongoing attempt to recreate a 21st-century silk road, embodied in the Popular Re-
public of China (PRC)’s Belt and Road Initiative, is only the latest example of this prac-
tice. Since the 2000s, China has also increased its contributions to development aid and
official lending and multiplied its trade agreements with partners across the globe.
China’s African strategy is a good example. From 2000-to 2017, China initiated over
5800 development aid projects on the continent. It extended over USD 153.4 billion in
loans to African countries and canceled at least 1.9 billion in outstanding debt. Today
China is the largest bilateral lender in the continent. Since 2005, over 30 African coun-
tries have received zero-tariff treatment on exports. The Belt and Road Initiative, China’s
71
effort to place itself at the center of a modern trade network, now covers over 40 African
economies.
As official aid, loans, and agreements multiplied, so did the volumes of Sino-African
economic flows. Between 2003 and 2019, Chinese foreign direct investment in Nigeria,
Egypt, South Africa, and Algeria, the continent’s top four economies, grew by 67, 75,
136, and 310, respectively. Over the same period, Chinese imports from these markets
grew between 7 and 12 folds. Chinese exports to all African countries increased by a
staggering 1,016%. Imports increased by 964%. For comparison, U.S. exports to the
region grew by 146% over the same period, whereas imports increased by 94%.
The use of state resources to benefit business internationalization and the securing of
supply chains by China has been the focus of several studies. They emphasize the
role of government policies in shaping global supply chains in key sectors such as min-
ing (Humphreys (2013)), energy (Lind and Press (2018)), and agriculture (Belesky and
Lawrence (2019)). In all of these sectors, a solid state-business relationship allows
political-diplomatic action and firm operations to move in lockstep to achieve power and
profit’s combined political and economic goals. In the West, some have decried Chinese
mercantilist tendencies for their illiberal nature (Mawdsley (2008)). In Africa, others
point to the neocolonial nature of China’s interventions (Asongu et al. (2018)). These
critics perceive the PRC’s aid and trade diplomacy as a zero-sum game: for China’s
western rivals, it carries the threat of displacement and in the targeted economies of de-
pendence.
This paper evaluates this premise. From the perspective of China’s African partners, it
asks: does Chinese commercial diplomacy have export diversion effects? In other words,
does China’s commercial diplomacy rearrange the export network of targeted economies
to the exclusive benefit of China, or does it increase their export capabilities across the
board?
Whereas the literature on the growth and trade effects of China’s aid and trade diplomacy
has made significant progress in recent years, it has so far overlooked the question of di-
version. This paper aims to fill this gap. It identifies associations between China’s aid
and trade interventions and recipient countries’ exports to third-party countries. In the
process, it also contributes a more disaggregated analysis of the trade creation effects. I
look at four tools of Chinese commercial diplomacy: trade agreements, development aid,
72
government loans, and the Belt and Road Initiative, arguably the crown jewel of Chinese
commercial diplomacy. The choice of these tools is dictated by data availability, though
several other instruments could also be essential to study. For each of these policy in-
struments, I measure the trade diversion effects from the angle of the target country. The
analysis covers the period between 2000 and 2017.
The results suggest that China’s aid and trade interventions do not systematically dis-
place existing export flows and have net positive effects. China’s official development
assistance finance and preferential tariff treatments are associated with an increase in
African countries’ manufacturing exports. The Belt and Road Initiative, though a recent
phenomenon on the continent, does not present any signs of export diversion.
The rest of the paper is structured as follows: Section 2 discusses the literature on aid and
trade diplomacy and trade diversion in more length. Section 3 introduces the data, section
4 presents the methodology, and section 5 discusses the results. Section 6 concludes.
4.2 Literature Review
This paper evaluates the export creation and export diversion effects of China’s aid and
trade diplomacy in Africa. To do so, it considers two main types of treatments.
On the one hand, official finance (OF) comprises foreign aid and other forms of con-
cessional and non-concessional state financing. In a 2006 policy paper titled ”China’s
African Policy,” the country pledged more efforts to increase trade and investment and
provide economic assistance (Copper (2015)). The paper emphasized that ”in light of
its own financial capability and economic situation, China will do its best to provide and
gradually increase assistance to African nations with no political strings attached.” The
PRC’s OF takes different forms, spanning technical assistance and training programs,
concessional and non-concessional lending, debt relief, grants, and scholarships.
On the other hand, the second class of treatments relates to trade and investment agree-
ments. China is linked to over 30 African countries by preferential trade agreements
and has concluded over 40 memoranda of agreements under its flagship infrastructure
investment project: the Belt and Road Initiative.
The growth effect of China’s OF has been recently studied in Dreher et al. (2017). Using
the same treatment data as this paper but with a broader geographic focus, the authors
identify a positive effect of Chinese official financing on growth. They estimate that one
73
additional Chinese official development assistance (ODA) project produces 0.7 percent-
age points increase in economic growth two years after the pledge.
Two recent papers have studied the trade creation effects of China’s official finance in-
terventions in Africa. Liu and Tang (2018) investigate the impact of the US and China’s
foreign aid to countries on the continent on trade flows between donor and recipient
countries. They find that China’s development aid increases African exports to China,
whereas the same is not true of the US. Savin et al. (2020) corroborate this result. These
papers echo findings from previous research with broader geographic scope. For in-
stance, ? shows that bilateral aid is not only positively correlated with donor exports but
also positively associated with recipient exports to donors and that recipient exports of
strategic materials display a stronger association with bilateral aid. These findings chal-
lenge the prior theoretical work that emphasized the negative impact of aid on recipient
countries’ exports due to exchange rate appreciation.
Due to China’s timid participation in trade deal-making, few studies look at the country’s
trade agreements’ effects. Two studies have looked at the trade effects of China’s pref-
erential tariff policies in Africa. Sun and Omoruyi (2021) find that zero-tariff treatment
in favor of African partners significantly promoted export diversification from the man-
ufacturing industry. An older policy paper by Berhelemy (2011) analyzes the response
of African countries’ China imports in the early phase of the zero-tariff policy and finds
that they increase alongside imports from other partners.
To the exception of Berhelemy (2011) who tangentially brings up the question of import
diversion, the existing literature has overlooked the trade diversion effects of China’s OF
and trade agreements.
Yet, an old finding in the trade literature is that asymmetric trade policy can cause trade
diversion. First suggested by Viner (2014) in his research on customs unions, trade di-
version can often be a corollary to trade creation. The original theoretical framework that
motivates trade diversion reduces to a Cournot competition framework. Country A allo-
cates import decisions between countries B and C. If country A enters into a zero-tariff
trade agreement with country B, country B’s marginal cost of production for exports to
country A decreases. As a result, exports from B to A increase, all else equal. Addition-
ally, B receives a competitive advantage over C due to the asymmetric tariff treatment.
Thus, on the international trade network, part of the increased trade on the A −→ B edge
is diverted away from the C −→ A edge.
The empirical evidence on trade diversion varies on a case-by-case basis. In a highly
aggregated study, Dai et al. (2014) looks at the trade diversion effects of free trade agree-
ments using manufacturing export data from 1990 to 2002 for a total of 41 trading part-
74
ners. The results confirm that FTAs divert trade away from non-member countries. Fur-
thermore, Trade diversion is stronger for internal trade than external trade and imports
than exports. These aggregated results appear to hide discrepant dynamics at less aggre-
gated scales. In an econometric analysis of trade diversion under NAFTA, Fukao et al.
(2003) find evidence of US import diversion in specific industries. Out of 60 HS 2-digit
manufacturing lines, only 15, mainly textile and apparel products, experience significant
import diversion due to the entry into force of the regional liberalization agreement. Con-
trastingly, a study on the trade diversion effects of the ASEAN-China free-trade agree-
ment finds no evidence of trade diversion (Yang and Martinez-Zarzoso (2014)). Instead,
it identifies pure trade creation effects in exports and imports, both for countries within
and outside the trading bloc. This heterogeneity reveals the need for further empirical
localized studies of trade diversion dynamics, such as this one.
Trade diversion is not specific to trade agreements. Potentially any asymmetrical policy
intervention can also divert existing flows. Such is the case of anti-dumping (AD) and
countervailing (CV) measure actions. Prusa (2019) studies the trade impacts of 428
anti-dumping petitions filed between 1980 and 1988. The results show that in the year
following the filing of the petition, imports from non-named countries increase by 22%.
This growth in imports from non-named countries exceeds the reduction that affects the
named country during the investigation or after duties are levied. Bown and Crowley
(2007) document the causal nature of this dynamic, using the low-intensity US trade
war with Japan in the 1990s. The authors show that Japanese exports are ’deflected’
to third countries in response to remedial measures. The average anti-dumping duty on
Japanese exports leads to a 5–7% increase in Japanese exports of the same product to the
average third-country market. However, here too, sectoral heterogeneity is substantial.
While manufacturing flows appear to be reallocated in response to ”trade remedies,”
agricultural flows do not (Carter and Gunning-Trant (2010)), likely resulting from their
less liberalized nature. Besides AD and CV actions, voluntary export restrictions (VER)
have also been documented to reallocate trade flows (Hamilton (1985)).
Like AD and CV actions, or bilateral liberalization policies, China’s OF actions and
trade agreements, and investment deals under the BRI, are an asymmetric trade pol-
icy instrument targeting specific countries, with possible diversion effects. In fact, this
distortionary effect is often assumed to be true. In an analysis of China’s energy mercan-
tilism, Lind and Press (2018) argues that market considerations heavily inform China’s
use of political and diplomatic clout. Political resources are deployed to gain control
or influence over key suppliers, diversify products, suppliers, and transport routes, cre-
ate inventories, and provide security to protect vulnerable assets. This mercantilist logic
extends to other non-energy markets where the Chinese government seeks a stronger
75
presence. The use of these instruments and the ensuing reinforcement of China’s eco-
nomic presence in target countries has been portrayed in Western media as exclusive and
antagonistic to the interests of such third-party partners like the European Union and the
United States (Mawdsley (2008)). The depiction of economic diplomacy interventions
as a zero-sum game implicitly presumes a distortionary diverting effect. The goal of this
paper is to evaluate these effects econometrically.
Whereas it has not provided a quantitative evaluation of the role of economic diplomacy
in restructuring trade networks, the previously described literature offers helpful pointers
to guide our work. A first takeaway is that it is essential to adopt a disaggregated ap-
proach to allow for sectoral heterogeneity. Furthermore, different policy instruments
must be studied separately, even if seemingly similar (Moons and Bergeijk (2017)).
Lastly, the literature provides some candidate econometric models for this study. The
gravity model of trade, as adapted by Baier and Bergstrand (2007), can be readily used
to estimate both creation and diversion effects as in Dai et al. (2014) and Yang and
Martinez-Zarzoso (2014). However, data structure and variation will impede the imple-
mentation of a standard gravity model à la Baier and Bergstrand (2007). In such cases,
I will use alternative specifications such as diff-in-diff (Nitsch (2007)) or cross-sectional
approaches (Rose (2007)). The estimation equations chosen for this paper are discussed
further in the methodology section.
4.3 Data & Stylized Facts
I analyze the trade creation and diversion effects of two broad classes of treatments:
official financing, trade agreements, and an investment agreement: the BRI. This sec-
tion describes the data on these different tools and provides some stylized facts on their
structure.
4.3.1 Trade and Investment agreements
Trade agreements are the quintessential trade diplomacy instrument. Governments en-
tering into these arrangements seek to foster trade between their countries, possibly to
the detriment of third-party countries. Data on trade agreements is derived primarily
from the NSF-Kellogg Institute Database on Economic Integration Agreements (EIA).
The latest release of the EIA covers all country pairs from 1953 to 2014. To ensure
that the study is as current as possible, I supplement the EIA data with three additional
years of observations using the World Trade Organization’s (WTO) database on regional
trade agreements. Thus, the panel’s time dimension runs through 2017. Due to inaccura-
76
cies in the EIA’s China trade regimes 1, I make corrections to the dataset using Sun and
Omoruyi (2021)’s China’s customs data. Table 4.1 presents the profiles and counts of
trade agreements binding China with third-party countries.
China’s free-trade strategy has until recently been cautious. The country is part of a
limited number of free-trade agreements, the largest of which are ASEAN-China and the
Asia-Pacific Trade Agreement. Indeed, China’s most significant FTAs center around its
immediate geographic environment. China signed its first African trade agreement with
Mauritius in 2019, and it entered into force in January 2021 and is thus outside of the
time scope of this study. The China-Mauritius trade agreement remains the last one of
its kind in Africa to date.
Table 4.1: China’s Trade Agreement Diplomacy
All Countries African Countries
Year No Agreement Asymmetric Preferential FTA No Agreement Asymmetric Preferential FTA
2000 192 0 0 0 51 0 0
2001 187 0 5 0 51 0 0
2002 187 0 5 0 51 0 0
2003 187 0 5 0 51 0 0
2004 186 0 5 1 51 0 0
2005 162 24 5 1 27 24 0
2006 153 24 4 11 27 24 0
2007 149 26 4 13 25 26 0
2008 146 28 4 14 23 28 0
2009 145 28 4 15 23 28 0
2010 141 29 5 17 22 29 0
2011 135 35 6 16 21 30 0
2012 136 34 6 16 22 29 0
2013 137 33 6 16 23 28 0
2014 137 33 6 16 23 28 0
2015 137 33 5 17 23 28 0
2016 137 33 5 17 23 28 0
2017 135 35 5 17 21 30 0
China has more heavily relied on asymmetric preferences to facilitate trade. Under the
Forum on China-Africa Cooperation (FOCAC), China extended zero-tariff treatment to
thirty least-developed countries (LDCs) in Africa in 2005. The roll-out has been progres-
sive. As of July 1st, 2010, 30 countries 2 in Africa benefited from zero tariffs on 95% of
1The EIA database considers a trade cooperation agreement between Egypt and China as a preferential
system - even though it lacks tariff measures. Additionally, it only incorporates zero-tariff treatments
under the Forum on China-Africa Cooperation (FOCAC) from 2011 - once registered with the WTO.
2Ethiopia, Angola, Benin, Burundi, Equatorial Guinea, Togo, Eritrea, Cape Verde, Democratic Republic
of the Congo (DRC), Djibouti, Guinea, Guinea-Bissau, Comoros, Lesotho, Liberia, Rwanda, Madagas-
car, Mali, Mauritania, Mozambique, Niger, Sierra Leone, Senegal, Sudan, Somalia, Tanzania, Uganda,
Zambia, Chad, and the Central African Republic. Note that the analysis excludes Sudan and South Sudan
due to the 2012 partition.
77
Figure 4.1: Belt and Road Initiative Members, and Entry Years
products under China’s Duty-free treatment for LDCs.
In addition to trade agreements, I also look at investment agreements signed under the
Belt and Road Initiative (BRI), a cornerstone instrument of China’s trade diplomacy.
The BRI envisions the emergence of a China-centric trade network in the process. ”The
Silk Road Economic Belt focuses on bringing together China, Central Asia, Russia, and
Europe (the Baltic); linking China with the Persian Gulf and the Mediterranean Sea
through Central Asia and West Asia; and connecting China with Southeast Asia, South
Asia, and the Indian Ocean. The 21st-Century Maritime Silk Road is designed to go
from China’s coast to Europe through the South China Sea and the Indian Ocean in one
route, and from China’s coast through the South China Sea to the South Pacific in the
other”.
By 2019, 38 African countries had signed MoUs under the Belt and Road Initiative,
including the continent’s largest economies of Nigeria, Egypt, South Africa, and African
LDCs (see figure 1). As a result of the peripheral location of most African countries
relative to the Belt and Road, the level of scheduled and disbursed investments varies
significantly from one country to the other. In the absence of cross-continental projects
like Eurasia’s China-Mongolia-Russia or Bangladesh-China-India-Myanmar Economic
Corridor, the roll-out of the BRI in Africa is at the bilateral level.
78
Total BRI engagements in Sub-Saharan Africa towered at over 25 billion USD in 2019,
second only to investment commitments in East Asia (Green Finance and Development
Center). These engagements are unevenly distributed across countries, with some seeing
significant investments (Nigeria, Egypt, Ethiopia), while others have yet to identify a
joint project. Data on investment engagements and project construction under the BRI is
not centrally collected, and ad-hoc sources can be discrepant, with some reports failing to
disentangle investments within the BRI framework from other general flows. Therefore,
I use MoU signing as the treatment proxy. Therefore, the results cannot be interpreted
on the intensive margin but only on the extensive margin. Additionally, over half of the
African members joined in 2018, which means that the data will not reveal long-term
dynamics and will only provide treatment response to a one-year horizon.
4.3.2 Financial Aid Data
Despite contributing billions of dollars in aid annually and being the world’s top bilateral
creditor (Horn et al. (2021)), China does not participate in existing global reporting sys-
tems, such as the OECD’s Creditor Reporting System (CRS) and the International Aid
Transparency Initiative (IATI). As a result, there is a significant gap in the availability of
internationally comparable statistics on Chinese official financing.
This paper uses William & Mary’s AidData research lab Global Chinese Official Fi-
nance Dataset, Version 2.0 (Custer and Zhang (2021)). The dataset records the known
universe of projects supported by official financial and in-kind commitments (or pledges)
from China from 2000 to 2017. It does so by synthesizing and standardizing vast amounts
of unstructured, open-source, project-level information published by governments, in-
tergovernmental organizations, companies, non-governmental organizations, journalists,
and research institutions. The database includes official development assistance (ODA)
and other official flows (OOF)s. The distinction follows the OECD rules. Under these
rules, ODAs must be concessional (i.e., grants and soft loans) and administered to pro-
mote the economic development and welfare of developing countries as the main objec-
tive. OOFs are official sector transactions that do not meet the concessionality criterion.
The dataset reveals that China has spent $843 billion on financial aid between 2000 and
2017. This sum is roughly equivalent to the amount spent by the US, the world’s largest
donor of foreign aid. This volume breaks into 13,427 Chinese development ’projects’
officially pledged, committed, in implementation, or completed between 2000 and 2021.
The dataset identifies different official finance flow types: grants, free-standing technical
assistance, scholarships and training, loans, debt relief, and export credit.
For analysis, the data is restricted geographically to African recipients and economically
79
to projects that have at least been officially committed to. Once this filtering is applied,
the data consists of 5,152 projects. Table 4.2 provides summary statistics on allocations,
and figure 2 shows the disaggregated composition. It is worth noting that virtually all
African countries are receiving some form of development aid from China by the end of
the period. Correspondingly, the number of new development projects increased five-fold
from 79 in 2000 to 502 in 2017.
Table 4.2: China’s Development Assistance Flows To Africa (all categories)
Development Assistance
Commitment Year Number of Recipients Number of Projects
2000 36 79
2001 37 105
2002 42 126
2003 41 140
2004 41 145
2005 45 199
2006 45 270
2007 47 333
2008 47 276
2009 49 327
2010 47 289
2011 48 385
2012 47 349
2013 47 343
2014 49 350
2015 49 415
2016 50 481
2017 51 502
Table 4.3 provides a yearly breakdown of new loan and debt relief flows. In terms of vol-
umes, China issued countries in Africa over 153 billion USD in new official loans. Over
the same period, 49 African countries received at least one loan, with the yearly number
of recipients oscillating between 12 and 27. Over the same period, China restructured
at least 2 billion USD in existing debt to benefit a total of 40 countries. These figures
highlight the scope of China’s aid and trade activism in Africa, and explain the media
attention and suspicion that it fuels in the West (Mawdsley (2008)).
80
Figure 4.2: Composition of China’s Development Assistance to Africa (2017)
Table 4.3: Official Chinese Loans To African Countries (Excluding Trade Finance)
New Loans Debt Cancellation
Year Num. Recipients Total Volume (Million USD) Num. Beneficiaries Total Volume (Million USD)
2000 12 99.49 1 Not Available
2001 13 152.64 24 798.11
2002 13 231.12 6 124.42
2003 10 922.99 4 2.44
2004 12 642.32 3 Not Available
2005 16 803.91 3 51.0
2006 22 2,329.25 5 60.77
2007 23 1,760.84 28 619.02
2008 22 5,282.04 1 Not Available
2009 23 4,162.77 2 2.54
2010 25 5,809.31 5 10.34
2011 26 5,608.60 7 98.98
2012 27 10,600.94 3 44.65
2013 27 11,015.16 0 0
2014 21 9835.87 0 0
2015 24 5,250.71 0 0
2016 23 2,5347.16 7 57.19
2017 23 1,3199.05 6 83.07
81
4.4 Methodology
4.4.1 Official Financing (OF) Treatments
To evaluate the trade creation and diversion effects of official financing, I estimate a
gravity equation augmented with an official finance treatment variable. The use of gravity
equation to estimate the trade impact of financial flows has so far been the standard
approach in the literature (Pettersson and Johansson (2013), Savin et al. (2020), Liu and
Tang (2018)). The log linear version of the model has the following form:
log(Xijt) =α0 + α1log(gdpit) + α2log(gdpjt) + α3log(popit) + α4log(popjt)
+ α6RTAijt + β1treatmentit + β2(treatmentit × NotChinaj) + ηt (IV.1)
+ ηij + εijt
Where Xij,t is the value of exports from country i to country j in year t. Gross domestic
products (gdp) and populations (pop) are explicitly included as predictors, whereas time-
invariant bilateral gravity variables are absorbed in γij , which also captures multilateral
resistance. Time fixed effects absorb variations over the time dimension. The treatment
variable will take value 1 if country i has received official financing from China in year t
and 0 otherwise. Because I seek to check for differential response to treatment of exports
to China and exports to third party countries, I interact the treatment variable with a
dummy variable NotChina that takes value 1 when the importer is not China, and 0
otherwise.
In this specification, β1 measures the increase in export flows to China for countries that
receive financial aid. β1+β2 measures the response of export flows of recipient countries
to third-party trading partners. Thus, values of the parameters such that β2 + β1 < 0
would point at an export diversion effect of China’s OF.
Equation (4.1) can suffer from simultaneity bias if a feedback exists between export
flows and OF allocations, and from reverse causality. To address this concern, I specify
the following preferred model:
log(Xijt) =α0 + α1log(gdpit) + α2log(gdpjt) + α3log(popit) + α4log(popjt)
+ α6RTAijt + β3treatmenti,t (IV.2)1−3 + γ3(treatmenti,t1−3 × NotChinaj)
+ ηt + ηij + εijt
82
I estimate this specification using two different definitions of treatment. Following Savin
et al. (2020), I define treatment as a binary variable that takes value 1 if exporter i has
been a recipient of China’s official financing (OF) between years -1 and -3, and 0 other-
wise. I will refer to this as the extensive margin of treatment. In a second estimation, I
define treatment as the number of OF flows received; i.e. the intensive margin of treat-
ment. In this latter case, the specification is augmented with additional dummy variables
on recipient status. This allows me to evaluate the impact of treatment intensity condi-
tional on treatment.
Equations (4.1) and (4.2) are estimated for aggregate exports, as well as for three broad
sectors: manufacturing, raw materials, and agriculture. The differential response to eco-
nomic diplomacy treatments across industries is established in the literature (Moons and
Bergeijk (2017)). Additionally, in subsequent estimations, the treatments are disaggre-
gated into categories and types, following the literature on aid and growth which finds
significant differences in growth effects across flow types (Pettersson and Johansson
(2013)).
4.4.2 Trade Agreements
The export creation and diversion effects of China’s preferential treatments for African
countries are estimated separately, via two equations:
log(X Cij,t) = α0 + β1 × ATAij,t + β2 × ATA + ηij,t + ηit + ηjt + εij,t (IV.3)
log(XRoWi,t ) = α + γ ATA
C
0 1 i,t + δ1popi,t + δ2gdpi,t + δ3wtoi,t + ηi + ηt + εi,t (IV.4)
Equation (4.3) follows the literature on the effects of trade agreements initially proposed
in (Baier and Bergstrand (2007)) seminal paper. Export flows are a function of a series
of interacted fixed effects and of the trade regime captured in the dummy ATA which
stands for asymmetric trade agreement and takes value 1 if the flow from i → j benefits
from preferential treatment, and 0 otherwise. For estimation, the data is restricted to
country pairs that have no trade agreements, or that share an asymmetric preferential
agreement. Dropping deeper agreements (PTAs, FTAs, etc.) allows me to compare the
performance of China’s asymmetric treatments against a control groups of pairs that
do not have any agreement. In this context, γ1 measures the average performance of
83
asymmetric preferential treatments extended by importer countries to African exporters.
γ2 captures the additional effect of preferential treatments by China. A positive γ2 would
mean that China’s preferential treatment is better at supporting African exports to China
than are other systems of preferences.
Equation (4.5) estimates the diversion effects of China’s zero-tariff policies. The out-
come variable is total export flows to the rest of the world (RoW). Because China’s pref-
erential tariff program is tied to LDC status, which is defined based on income per capita,
GDPi,t and POPi,t are added into the equation to control for selection into treatment.
Furthermore, the regression controls for the World Trade Organization membership sta-
tus. Individual and time fixed effects control for other sources of variation.
My approach to the estimation of the trade diversion effect deviates from the existing lit-
erature. In a study of the trade-diversion effects of free trade agreements Dai et al. (2014)
use a gravity specification similar to equation (4.4), augmented with two dummies that
take value 1 when the exporter or importer have trade agreements with other countries,
respectively. Their specification captures the export and import diversion effects of the
agreements - which they find to be significant. Adapting this approach to the isolated
case of Chinese agreements is not possible due to limited variations that do not allow for
the inclusion of a full set of fixed effects.
4.4.3 Investment Agreements
To be able to exploit variations in trade dynamics, analysis of the impact of BRI member-
ship is confined to countries that have joined in 2015: South Africa, Cameroon. While
Somalia also joined in 2015, its trade flows are generally unstable and hard to predict
given the country’s perpetual instability, and very small economic size, which makes it
a bad candidate for a comparative estimation approach. Most other members joined the
initiative in 2018 - which is outside the time span of this study.
Given the short treatment period and low number of treated units, which limit the appli-
cability of a panel data, I rely on a synthetic controls approach. In this method, the export
flows of each treated country post-treatment are compared to those of a synthetic counter-
factual that is specific to each treated unit. The synthetic counterfactual is a weighted sum
of comparable, non-treated, African economies. The weights are constructed to match
the pre-treatment trends as closely as possible. This addresses concerns of selection into
treatment, and relaxes the parallel trend assumptions (Abadie et al. (2010)).
To build the synthetic controls, I use the Generalized Synthetic Control Method devel-
oped by Xu (2017). This method allows me to mach units on pre-treatment observables
while also including unobserved time-varying heterogeneities using interactive fixed ef-
84
fects. The matching is performed over the outcome only (export flows), to avoid bias.
Using a relatively long pre-treatment period (2000-2014) additionally allows me to get a
better pre-treatment fit.
4.5 Results
4.5.1 Official Financing Results
Table 4.4: Trade Creation and Diversion Effects of Official Finance (OF)
Dependent variable:
All exports Manufacturing Agriculture Raw materials
Control: Importer’s GDP 0.589∗∗∗ 0.813∗∗∗ 0.705∗∗∗ 0.533∗∗∗
(0.0345) (0.0401) (0.0455) (0.0505)
Control: Importer’s Population 0.940∗∗∗ 0.796∗∗∗ 0.391∗∗ 0.599∗∗∗
(0.0921) (0.114) (0.122) (0.142)
Control: Exporter’s GDP 0.391∗∗∗ 0.212∗∗∗ 0.319∗∗∗ 0.485∗∗∗
(0.0324) (0.0379) (0.0436) (0.0470)
Control: Exporter’s Population -0.955∗∗∗ -1.069∗∗∗ -1.708∗∗∗ -1.552∗∗∗
(0.134) (0.167) (0.209) (0.219)
Control: RTA (yes/no) 0.328∗∗∗ 0.158∗∗ 0.437∗∗∗ 0.337∗∗∗
(0.0448) (0.0512) (0.0542) (0.0641)
Treatment: OF received (yes/no) 0.954∗∗∗ 0.976∗∗∗ 0.981∗∗ 1.291∗∗∗
(0.271) (0.293) (0.334) (0.312)
Treatment: OF received × Not China -0.997∗∗∗ -0.991∗∗∗ -0.948∗∗ -1.178∗∗∗
(0.272) (0.295) (0.336) (0.314)
Constant -4.424∗∗ -4.795∗∗ 6.317∗∗ 3.188
(1.440) (1.775) (2.170) (2.366)
Observations 88329 71037 49506 52805
Fixed-Effects T, T, T, T,
EX×IM EX×IM EX×IM EX×IM
Joint significant P-value 0.13 0.64 0.35 0.0051
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
Table 4.4 presents results from the estimation of equation (4.1). The evidence suggests
that receiving official financing from China is significantly associated with higher exports
to China, and this effect is particularly large for raw materials. However, this increased
trade does not appear to happen at the expense of existing flows with third-party partners,
as reflected by the diversion coefficient’s lower magnitude and the combined effects’
statistical insignificance.
85
Equation (4.1) imposes contemporaneity on the treatment and exports’ response and is
likely prone to simultaneity and reverse causality biases. Using lagged treatment, equa-
tion (4.2) addresses this concern. The results from the estimation are reported in table
4.5. Official Finance does not appear to associate significantly with exports to China or
third-party countries in this setup.
Table 4.5: Trade Creation and Diversion Effects of Official Finance (OF) - Lagged Treat-
ment (Extensive)
Dependent variable:
All Exports Manufacturing Agriculture Raw Materials
Control: Importer’s GDP 0.653∗∗∗ 0.869∗∗∗ 0.807∗∗∗ 0.541∗∗∗
(0.0421) (0.0496) (0.0557) (0.0619)
Control: Importer’s Population 0.441∗∗∗ 0.449∗∗ 0.174 0.571∗∗
(0.121) (0.154) (0.163) (0.193)
Control: Exporter’s GDP 0.316∗∗∗ 0.0838 0.299∗∗∗ 0.419∗∗∗
(0.0394) (0.0476) (0.0529) (0.0585)
Control: Exporter’s Population -0.443∗∗ -0.338 -0.966∗∗∗ -1.462∗∗∗
(0.156) (0.202) (0.260) (0.264)
Control: RTA 0.260∗∗∗ 0.0381 0.353∗∗∗ 0.0896
(0.0576) (0.0664) (0.0677) (0.0820)
Treatment: OF Received (dummy) 0.688 1.120 1.727∗ -0.275
(0.560) (0.633) (0.799) (0.786)
Treatment: OF Received (dummy) x Not China -0.871 -1.467∗ -1.825∗ 0.123
(0.564) (0.638) (0.803) (0.793)
Constant -4.165∗ -6.767∗∗ 0.148 4.127
(1.767) (2.232) (2.732) (2.989)
Observations 71702 56616 38684 41282
Fixed-Effects T, T, T, T,
EX×IM EX×IM EX×IM EX×IM
Joint significant P-value 0.0062 0.0001 0.255 0.17
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
Using specification (2) but adjusting treatment variables to the number of OF flows re-
ceived, I can look at the intensive margin of treatment effects. Results in Table 4.6
suggest that the intensive treatment margin is more relevant to export response than the
extensive margin. Countries targeted with a higher number of assistance projects can
experience more significant changes in their export flow structures. Thus, an additional
project is associated with a 2% increase in exports to China that concerns all sectors.
More importantly, there is no evidence of trade diversion on the intensive margin. Indeed,
overall exports and manufacturing and raw material exports experience a significant but
marginal increase of less than a percentage point. The results in table 4.6 are robust to
conditioning whether a country receives financing from China.
86
Table 4.6: Trade Creation and Diversion Effects of Official Finance (OF) - Lagged Treat-
ment (Intensive)
Dependent variable:
All Exports Manufacturing Agriculture Raw Materials
Control: Importer GDP 0.634∗∗∗ 0.766∗∗∗ 0.728∗∗∗ 0.481∗∗∗
(0.0439) (0.0489) (0.0557) (0.0616)
Control: Importer GDP 0.412∗∗ 0.623∗∗∗ 0.209 0.477∗∗
(0.130) (0.148) (0.158) (0.185)
Control: Exporter GDP 0.268∗∗∗ 0.0244 0.280∗∗∗ 0.441∗∗∗
(0.0406) (0.0474) (0.0526) (0.0583)
Control: Importer GDP -0.597∗∗∗ -0.682∗∗∗ -1.152∗∗∗ -1.755∗∗∗
(0.162) (0.201) (0.261) (0.264)
RTA (yes/no) 0.200∗∗∗ 0.0776 0.389∗∗∗ 0.143
(0.0585) (0.0649) (0.0679) (0.0806)
Treatment: Cumulative OF 0.0229∗∗ 0.0512∗∗∗ 0.0381∗∗∗ 0.0415∗∗∗
(0.00733) (0.00866) (0.0113) (0.00941)
Treatment: Cumulative OF x Not China -0.0200∗∗ -0.0439∗∗∗ -0.0366∗∗ -0.0360∗∗∗
(0.00735) (0.00866) (0.0113) (0.00942)
Constant -1.363 -2.918 3.041 8.085∗∗
(1.881) (2.213) (2.746) (2.963)
Observations 66452 60008 41402 44293
Fixed-Effects T, T, T, T,
EX×IM EX×IM EX×IM EX×IM
Joint significant P-value 0.0069 0.0000 0.3031 0.0006
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
87
Disaggregating OF into flow categories, model 2 yields the results presented in tables 7
and 8. The data consists of 3 broad categories: Official Development Aid (ODA) flows,
which have a significant concessional component, and Other Official Financing (OOF)
that do not. In addition, the dataset contains some financing flows that are ambiguous or
not yet clearly determined. Being a recipient of ODA flows appear to have little bear-
ing on export volumes. However, conditional on being a recipient, the intensity of ODA
treatments, measured by the number of projects received over three years, does asso-
ciate with higher exports to China. One additional ODA project increases total exports
to China by an average of 3% and reduces exports to other countries by an average of
0.7% (significant at 1% confidence level). This effect is driven mainly by the manufac-
turing sector. Contrastingly, receiving other official finance flows is associated with a net
increase in exports, but the impact does not fluctuate by treatment intensity.
Table 4.7: Effects of Official Finance (OF) By Category - Lagged Treatment (Extensive)
Dependent variable:
All Exports Manufacturing Agriculture Raw Materials
Control: Importer GDP 0.635∗∗∗ 0.882∗∗∗ 0.768∗∗∗ 0.556∗∗∗
(0.0437) (0.0526) (0.0578) (0.0656)
Control: Importer Population 0.401∗∗ 0.504∗∗ 0.225 0.560∗∗
(0.129) (0.172) (0.174) (0.215)
Control: Exporter GDP 0.284∗∗∗ 0.0500 0.240∗∗∗ 0.435∗∗∗
(0.0400) (0.0489) (0.0531) (0.0597)
[1em] Control: Exporter Population -0.482∗∗ -0.292 -0.896∗∗∗ -1.667∗∗∗
(0.160) (0.212) (0.271) (0.277)
RTA (dummy) 0.202∗∗∗ -0.0180 0.294∗∗∗ 0.0477
(0.0586) (0.0698) (0.0686) (0.0857)
Treatment: Other Official Finance (dummy) 0.328∗ 0.439∗ 0.311 0.519∗∗
(0.162) (0.177) (0.212) (0.190)
Treatment: OOF x Not China -0.276 -0.366∗ -0.215 -0.409∗
(0.163) (0.178) (0.213) (0.192)
Treatment: Official Development Aid (dummy) 0.392 0.576 0.951 -0.581
(0.508) (0.562) (0.639) (0.672)
Treatment: ODA (dummy) x Not China -0.546 -0.842 -0.996 0.509
(0.511) (0.566) (0.642) (0.676)
Treatment: Ambiguous (dummy) 0.279 0.556∗∗ 0.461∗ 0.398
(0.184) (0.198) (0.218) (0.209)
Treatment: Ambiguous (dummy) x Not China -0.140 -0.473∗ -0.372 -0.316
(0.185) (0.199) (0.219) (0.210)
Constant -2.476 -7.406∗∗ 0.856 5.775
(1.849) (2.400) (2.876) (3.210)
Observations 66753 51250 34902 37254
Fixed-Effects T, T, T, T,
EX×IM EX×IM EX×IM EX×IM
Joint significant P-value : OOF 0.0081 0.0014 0.0001 0.0002
Joint significant P-value : ODA 0.0060 0.0001 0.4821 0.3769
Joint significant P-value : AMB 0.0000 0.0005 0.0005 0.0049
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
88
Table 4.8: Effects of Official Finance (OF) By Category - Lagged Treatment (Intensive)
Dependent variable:
All Exports Manufacturing Agriculture Raw Materials
Control: Importer’s GDP 0.622∗∗∗ 0.855∗∗∗ 0.755∗∗∗ 0.539∗∗∗
(0.0440) (0.0531) (0.0583) (0.0662)
Control: Importer’s Population 0.409∗∗ 0.550∗∗ 0.241 0.618∗∗
(0.130) (0.173) (0.175) (0.216)
Control: Exporter’s GDP 0.247∗∗∗ 0.00775 0.296∗∗∗ 0.387∗∗∗
(0.0407) (0.0504) (0.0546) (0.0613)
Control: Exporter’s Population -0.810∗∗∗ -1.056∗∗∗ -0.835∗∗ -1.923∗∗∗
(0.170) (0.228) (0.288) (0.297)
Control: RTA (dummy) 0.200∗∗∗ -0.0315 0.277∗∗∗ 0.0257
(0.0586) (0.0700) (0.0685) (0.0857)
Conditioning var: OOF (dummy) 0.185 0.259 0.185 0.396
(0.175) (0.191) (0.238) (0.207)
Conditioning var: OOF (dummy) x Not China -0.119 -0.165 -0.0618 -0.262
(0.176) (0.192) (0.239) (0.209)
Treatment: Cumulative OOF 0.00472 0.0355 0.0513 0.0101
(0.0153) (0.0189) (0.0350) (0.0200)
Treatment: Cumulative OOF x Not China 0.00719 -0.0275 -0.0666 0.00528
(0.0154) (0.0190) (0.0351) (0.0203)
Conditioning var: ODA (dummy) 0.0704 0.314 0.961 -0.767
(0.528) (0.584) (0.663) (0.687)
Conditioning var: ODA (dummy) x Not China -0.135 -0.541 -0.970 0.803
(0.531) (0.588) (0.666) (0.692)
Treatment: Cumulative ODA 0.0276∗ 0.0298∗ -0.000428 0.0197
(0.0122) (0.0139) (0.0175) (0.0150)
Treatment: Cumulative ODA x Not China -0.0344∗∗ -0.0295∗ -0.00351 -0.0322∗
(0.0123) (0.0140) (0.0176) (0.0151)
Conditioning var: Cumulative AMB 0.235 0.311 0.430 0.311
(0.206) (0.222) (0.363) (0.233)
Conditioning var: Cumulative AMB x Not China -0.166 -0.332 -0.306 -0.324
(0.207) (0.224) (0.364) (0.235)
Treatment: Cumulative AMB -0.0258 0.0732 -0.0442 0.00842
(0.0477) (0.0503) (0.185) (0.0536)
Treatment: Cumulative AMB x Not China 0.0659 -0.00936 0.0324 0.0485
(0.0479) (0.0507) (0.185) (0.0542)
Constant 1.247 0.347 -0.509 8.760∗∗
(1.938) (2.537) (3.006) (3.357)
Observations 66452 50965 34664 36911
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
89
Disaggregating by type, loans stand out as the most effective tool in impacting export
volumes to China but do not divert existing flows. Borrowing from China is associated
with a 77% increase in exports to China and a 3% increase in exports to third-party
partners. On the intensive margin, one additional loan increases manufacturing exports
to China by 13%. Other types of financing, including export credit, debt relief, and
technical assistance, do not significantly correlate with variations in exports to China or
third-party partners.
4.5.2 Trade and BRI Agreements Results
The esults in table 4.11 show that the export-creation effects of China’s preferential treat-
ment program for African LDCs are, on average, larger than other iterations of the Gen-
eralized System of Preferences, specifically in the manufacturing and raw materials sec-
tors. Receiving zero-tariff treatment increases manufacturing exports to China 4-fold,
whereas receiving preferential treatment from non-China importers has no significant
effect on recipients’ manufacturing exports. This might be a reflection of the different
designs of these preferential frameworks. Indeed, the US’s AGOA applies tariff reduc-
tion selectively on products, and specifically excludes import-sensitive products in the
manufacturing and agriculture sector (Frazer and Biesebroeck (2010)). The discrepancy
can also be due to differential utilization rate of agreements, given that preferences can
be significantly underutilized by exporting countries. For instance, Brenton (2006) the
EU’s GSP system is particularly underutilized.
In conjunction with table 4.12, the results above show that China’s African trade agree-
ments are net trade creators. Indeed, there is no significant evidence of export diversion,
in aggregate and across industries. Controlling for a country’s gross domestic prod-
uct, population, and WTO membership status, benefiting from preferential treatment by
China does not significantly lower exports to the rest of the world. This is true for aggre-
gate exports (column 1) and of industry-level flows (columns 2-4).
The net export creation effects of China’s zero-tariff policies echo findings from prior
literature. Sun and Omoruyi (2021) demonstrate that these policies significantly support
export diversification, especially in the manufacturing sector. An earlier working paper
by Engel (2014) also finds that Chinese LDC preferences favorably impact exports at
both the intensive and extensive margins.
Lastly, I turn to the trade creation and diversion effects of BRI agreements. The esti-
mation method uses synthetic controls, and obtains treatment effects by comparing the
performance of a treated country’s export flows to those of a synthetic counterfactual
constructed from observed data. The results are presented in figure 4. Three countries
90
Table 4.9: Effects of Official Finance (OF) By Flow Type - Lagged Treatment (Extensive)
Dependent variable:
All Exports Manufacturing Agriculture Raw Materials
Control: Importer’s GDP 0.634∗∗∗ 0.888∗∗∗ 0.762∗∗∗ 0.565∗∗∗
(0.0438) (0.0528) (0.0579) (0.0657)
Control: Importer’s Population 0.396∗∗ 0.474∗∗ 0.256 0.550∗
(0.130) (0.173) (0.175) (0.215)
Control: Exporter’s GDP 0.253∗∗∗ 0.0330 0.216∗∗∗ 0.381∗∗∗
(0.0412) (0.0503) (0.0540) (0.0615)
Control: Importer’s Population -0.601∗∗∗ -0.432∗ -1.231∗∗∗ -1.797∗∗∗
(0.162) (0.214) (0.276) (0.280)
Control: RTA (dummy) 0.202∗∗∗ -0.0205 0.278∗∗∗ 0.0389
(0.0586) (0.0700) (0.0685) (0.0855)
Treatment: Export Credit (dummy) 0.0676 0.180 0.497∗ 0.225
(0.187) (0.203) (0.227) (0.213)
Treatment: Ex. Credit (dummy) x Not China 0.0181 -0.0836 -0.425 -0.144
(0.187) (0.204) (0.228) (0.215)
Treatment: Debt Relief (dummy) -0.170 -0.0579 -0.0371 -0.328
(0.163) (0.177) (0.212) (0.189)
Treatment: Debt Relief (dummy) x Not China 0.212 0.0181 -0.0192 0.335
(0.164) (0.178) (0.213) (0.190)
Treatment: Technical Assistance (dummy) 0.176 -0.0259 0.154 -0.0843
(0.260) (0.272) (0.293) (0.283)
Treatment: Tech. Ass. (dummy) x Not China -0.119 0.000887 -0.0976 0.175
(0.261) (0.273) (0.294) (0.285)
Treatment: Loan (dummy) 0.579∗∗∗ 0.824∗∗∗ 0.577∗∗ 0.449∗
(0.168) (0.184) (0.218) (0.196)
Treatment: Loan (dummy) x Not China -0.552∗∗ -0.802∗∗∗ -0.480∗ -0.402∗
(0.169) (0.185) (0.219) (0.197)
Treatment: Other OF (dummy) 0.220 0.484 0.441 -0.0715
(0.360) (0.390) (0.517) (0.436)
Treatment: Other OF (dummy) x Not China -0.270 -0.551 -0.451 0.0185
(0.362) (0.393) (0.519) (0.439)
Constant -0.942 -5.770∗ 4.214 7.784∗
(1.894) (2.463) (2.964) (3.279)
Observations 66522 51039 34720 36986
Fixed-Effects T, T, T, T,
EX×IM EX×IM EX×IM EX×IM
Joint significant P-value : Ex. Credit 0.0001 0.0001 0.0100 0.0095
Joint significant P-value : Debt Relief 0.383 0.0891 0.0290 0.8138
Joint significant P-value : Tech. Ass. 0.0377 0.42 0.0716 0.0157
Joint significant P-value : Loan 0.1632 0.3221 0.0001 0.0998
Joint significant P-value : Other 0.2351 0.1627 0.8441 0.3623
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
91
Table 4.10: Effects of Official Finance (OF) By Flow Type - Lagged Treatment (Inten-
sive)
Dependent variable:
All Exports Manufacturing Agriculture Raw Materials
Conditioning and control variables omitted from table
Cumulative Export Credit (count) 0.0124 0.0247 0.0408 0.0140
(0.0205) (0.0207) (0.0437) (0.0221)
Cumulative Export Credit (count) x Not China -0.00383 -0.0257 -0.0569 -0.00861
(0.0207) (0.0209) (0.0439) (0.0224)
Cumulative Debt Relief (count) 0.0235 -0.404 -0.123 0.111
(0.247) (0.253) (0.258) (0.279)
Cumulative Debt Relief (count) x Not China -0.0701 0.440 0.0807 -0.172
(0.248) (0.255) (0.260) (0.281)
Cumulative Technical Assistance (count) 0.0928 0.105 -0.105 0.0411
(0.0558) (0.0614) (0.0804) (0.0667)
Cumulative Tech. Ass. (count) x Not China -0.109 -0.116 0.108 -0.0693
(0.0561) (0.0619) (0.0809) (0.0673)
Cumulative Lending (count) 0.00676 0.118∗∗∗ 0.0606 0.0269
(0.0191) (0.0316) (0.0487) (0.0330)
Cumulative Lending (count) x Not China 0.00662 -0.0891∗∗ -0.0681 -0.0154
(0.0192) (0.0318) (0.0489) (0.0333)
Cumulative Other (count) 0.0104 0.00491 -0.0111 0.0247
(0.0151) (0.0175) (0.0208) (0.0186)
Cumulative Other (count) x Not China -0.0150 -0.00269 0.00463 -0.0301
(0.0152) (0.0176) (0.0209) (0.0188)
Constant 0.171 -1.787 3.140 6.978∗
(1.949) (2.535) (3.040) (3.361)
Observations 66452 50965 34664 36911
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
92
Table 4.11: Trade Creation Effects of Preferential Tariff Treatment
Dependent variable:
All Exports Manufacturing Agriculture Raw Materials
Preferential Treatment (Dummy) 0.0714 -0.0912 0.249 -0.0210
(0.392) (0.437) (0.259) (0.367)
Preferential Treatment x China 1.051 1.622∗ 0.313 1.206∗
(0.521) (0.605) (0.524) (0.554)
Constant 11.61∗∗∗ 10.29∗∗∗ 11.48∗∗∗ 11.93∗∗∗
(0.0615) (0.0762) (0.0550) (0.0762)
Observations 31743 24333 14963 16183
Fixed-Effects EX×T, IM×T, EX×T, IM×T, EX×T, IM×T, EX×T, IM×T,
EX×IM EX×IM EX×IM EX×IM
Joint significant P-value 0.203 0.0113 0.2421 0.0188
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
Table 4.12: Trade Diversion Effects of Preferential Tariff Treatment
Dependent variable:
All Exports Manufacturing Agriculture Raw Materials
Exporter’s Population 0.169 0.178 -0.506 0.717
(0.415) (0.669) (0.797) (0.717)
Exporter’s GDP 0.539∗∗∗ 0.157 -0.354∗ 0.498∗∗
(0.102) (0.151) (0.180) (0.161)
Exporter’s WTO 0.156 -0.637 1.237∗∗ 1.975∗∗∗
(0.214) (0.340) (0.404) (0.365)
PTA with China (dummy) -0.148 -0.0451 0.203 0.0153
(0.0815) (0.120) (0.143) (0.128)
Constant 8.735∗ 12.58∗ 25.09∗∗∗ 1.663
(3.535) (5.648) (6.725) (6.049)
Observations 896 847 841 847
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
93
are retained for analysis: South Africa (ZAF), Cameroon (CMR) and Somalia (SOM).
These are the earliest African members in the BRI, with entry years spanning 2014-2015.
For each of the three countries three outcomes are analyzed: flows to China, flows to the
United States, and flows to the European Union. This yields 9 estimated treatment ef-
fects. The point estimates of the treatment effect are represented by the white circles.
The segments represent the 95% confidence interval around the point estimate.
As shown on the figure, the treatment effects on trade with China, the US, and Europe,
are all indistinguishable from zero. It is clear that we cannot reject the hypothesis that
BRI does not divert existing trade flows with China’s Western rivals. These results might
however be sensitive to the recent nature of the initiative in Africa and the peripheral
status of member countries in the region in the overall vision and financial commitments
of the BRI. Indeed Sebastian Ibold’s BRI data collection project Ibold (2022) only iden-
tifies one project that tangentially targets Cameroon consisting in the Chad-Cameroon
railway, and no committed projects for South Africa or Somalia.
4.5.3 Discussion
The trade creation results are broadly in line with prior findings in the literature: receiv-
ing more Official Financing from China is associated with increases in recipients’ exports
to China (table 4.6). This relationship is primarily due to the role played by official de-
velopment assistance (ODA). The fact that ODA flows are more beneficial to recipients’
exports is supported by results from Savin et al. (2020) who show that African exports
to China associate positively with ODA flows, with manufacturing industries leading the
response. The finding that manufactured African exports are most sensitive to ODA is
also in line with Savin et al. (2020) results. Pettersson and Johansson (2013) on the other
hand, points to a particularly strong link between ODA and recipient’s exports of strate-
gic material. This does not seem to materialize in the context of China’s aid to Africa, as
the manufacturing sector repeatedly appears as the most responsive across specifications.
This paper sheds light on which type of official finance (OF) flows drives the associa-
tion between official finance and trade. It shows that lending is the primary mechanism
through which official finance impacts recipients’ exports. This result is particularly
relevant in a context where China has become the first bilateral lender to Africa.
Furthermore, the results above demonstrate that how we define treatment matters to our
study of the response of African exports to China’s official finance. The impact of ODA
depends on its intensity, and being a recipient of official flows does not elicit significant
responses from export flows. However, receiving more official finance, particularly more
ODA, leads to a positive response of the recipient’s exports. Conversely, increasing OOF
94
Figure 4.3: Belt and Road Initiative Members, and Entry Years
95
flows - generally more commercially oriented - does not significantly increase trade.
On trade diversion, the hypothesis that China’s OF in Africa displaces existing trade
flows is not supported in the aggregate analysis (Table 6). In fact, in most cases, OF
treatments coincide with a net increase in exports: to China and third-party countries.
However, a closer look at disaggregated treatment reveals that ODA financing diverts
raw materials exports to third-party countries. Indeed, one additional ODA flow de-
creases raw material exports to non-China partners by a statistically significant 2%. Due
to the predominance of commodity exports in Africa, this translates into a total export di-
version of 0.6%. While not large, this figure brings into focus the aid/commodity nexus
in China’s Africa strategy. Previous empirical work on trade creation effects saw no
”there there,” but trade diversion effects indicate that there might yet be a ”there there.”
Like much of the empirical research on China’s aid and trade diplomacy, results of
this paper dispel export-diversion and concentration criticism about China’s approach
to Africa. China’s official finance, trade agreements, and investment memoranda with
African countries do not appear to increase her market share in Africa at the expense
of other partners. There is also no evidence of these intervention reinforcing a ”dutch
disease”. Exports of raw materials experience increases smaller than the manufactur-
ing sector, and where there is diversion it remains very low. These results are similar
to findings in the recent literature on China’s aid and trade diplomacy. Dreher et al.
(2017) test the claim that significant financial support from China impairs the effective-
ness of grants and loans from Western donors and lenders, and find no support for this
claim. Bon and Cheng (2021) show that China’s increasing involvement in debt restruc-
turing, rather than being obscure, is similar in its approach to the Paris Club. Bräutigam
and Gallagher (2014) analyze China’s resource-secured or commodity-backed loans, and
find that contrary to many of the claims in the popular press, Chinese finance is not out
of line with interest rates found in global capital markets, and does not bring windfall
commodity profits to China. Dreher and Fuchs (2015) argue, after an analysis of China’s
aid allocation, that its patterns are comparable to Western donors’ and that it appears to
be independent of recipients’ natural resource endowment - and conclude that denoting
Chinese aid as ”rogue aid” seems unjustified. Many elements that underlie the suspi-
cion towards the ”Beijing Model” do not appear to find support in empirical economic
analysis.
Instead, the evidence in this paper lends support to the argument that China’s African
trade involvement is mutually beneficial. The net export creation effect in manufactur-
ing shows that China’s involvement can support export diversification, not only through
preferential treatment, as showed in Sun and Omoruyi (2021), but alsoalso contribute
to addressing supply-side issues through ODA and ODA-adjacent financing, leading to
96
an improvement in production capacities and competitiveness of recipient economies -
which rather than a zero-sum game could be a win-win strategy.
4.6 Conclusion
Today, China is Africa’s largest trading partner, investor, and creditor, making it an es-
sential player in its economy. Her engagement is promoted through active aid and trade
diplomacy. The absence of rigorous empirical evidence on the economic impacts of these
interventions has allowed the policy debate around it to become polarized and suspicious.
This paper aims to fill this gap and inform this debate by evaluating the export creation
and diversion effects of China’s aid and trade diplomacy in Africa.
The results show that China’s economic diplomacy activism in Africa does not signif-
icantly divert exports away from existing links. In addition, it does not appear that it
pushes African economies towards more specialization in natural resource exports. In-
stead, manufacturing industries appear to benefit from ODA flows and trade agreements,
both of which generate significant net trade creation effects. Early evidence on the Belt
and Road Initiative shows no sign of trade diversion. These results do not preclude the
existence of potential negative economic externalities of Chinese engagement and do not
presume that their welfare or distributional impacts are positive on balance.
This paper also offers methodological insights into future research on trade diplomacy
effects. It suggests that the margin of treatment needs to be carefully defined, and it also
shows that heterogeneity analysis must be applied systematically to explore different
sub-types of treatment and to distinguish the response of different sector-level outcomes.
The paper’s findings invite further empirical research into the mechanisms that account
for the patterns that have been identified. A more granular approach at the geography
and industry level could generate new data that isolates the causality channels. Is this
a supply-side effect that improves the production capabilities of target countries? Addi-
tionally, as the implementation of the BRI advances, richer data will become available
to study its trade and economic effects in Africa. Finally, as our macroeconomic un-
derstanding of the impacts of China’s engagement in Africa improves, another question
becomes more accessible and more urgent to address. Research into China’s Africa strat-
egy will have to turn to its welfare and distributional consequences, the actual litmus test
of policy interventions.
97
Chapter V
Conclusion
The World Trade Organization enumerates five foundational principles of the multilat-
eral trade system: non-discrimination, predictability and transparency, competitiveness,
benefits for less developed countries, and environmental protections. Events in the past
few years have brought the interplay of trade policy-making frameworks and an evolving
political and geopolitical context into sharp focus. Economically populist discourse, with
protectionist undertones, has pervaded trade policy-making, shifting priorities, reviving
old paradigms such as industrial policy, and directly threatening the stability of the trade
policy order epitomized by the World Trade Organization.
Without arguing the importance of these principles, or the WTO’s ability to deliver on
the associated promises, we can economically and quantitatively anticipate the potential
losses from a weakening of the rule-based system by studying instances where these
five principles are relaxed. Such is the goal of this volume. The findings minimize the
economic and financial cost of recent developments in trade-policy making. The trade
policy irregularities that came under study in this work do not generate the prohibitive
losses needed to advocate for the preservation of the trade system as it exists, at all
costs. Reversals in trade policy do not appear to have any spillover or externality effects
beyond their first-order impact. Fluctuations in trade policy uncertainty do not generate
the large movements in financial flows that other forms of uncertainty can cause. And
maligned ”mercantilist” policies appear to be trade increasing and supportive of export
diversification.
This analysis is by no means definitive, and much remains to be evaluated when it comes
to the potential costs of the emergence of a decentralized atomic trade system. I have
here focused on macroeconomic implications. Further research needs to be conducted
98
into the individual and social welfare. But even as we, as economists, contemplate the
gains from what exists, and the costs of transitions, we must too imagine alternative
designs able to generate benefits that can exceed the identified costs.
99
Bibliography
Abadie, Alberto, Alexis Diamond, and Jens Hainmueller, “Synthetic Control Meth-
ods for Comparative Case Studies: Estimating the Effect of California’s Tobacco Con-
trol Program,” Journal of the American Statistical Association, 2010, 105 (490), 493–
505.
Alessandria, George A., Shafaat Y. Khan, and Armen Khederlarian, “Taking Stock
of Trade Policy Uncertainty: Evidence From China’s Pre-WTO Accession,” 2019.
Alizadeh, Sassan, Michael W Brandt, and Francis X Diebold, “Range-Based Estima-
tion of Stochastic Volatility Models,” The Journal of finance (New York), 2002, 57 (3),
1047–1091.
Amengual, Dante and Dacheng Xiu, “Resolution of policy uncertainty and sudden
declines in volatility,” Journal of Econometrics, 2018, 203 (2), 297–315.
Anderson, James E., “A Theoretical Foundation for the Gravity Equation,” The Ameri-
can Economic Review, 1979, 69 (1), 106–116.
Asgharian, Hossein, Ai Jun Hou, and Farrukh Javed, “The Importance of the
Macroeconomic Variables in Forecasting Stock Return Variance: A GARCH-MIDAS
Approach,” Journal of Forecasting, 2013, 32 (7), 600–612.
Asongu, Simplice A, Jacinta C Nwachukwu, and Gilbert A A Aminkeng, “Lessons
from a Survey of China’s Economic Diplomacy,” Journal of world trade, 2018, 52 (5),
789–814.
Auer, Raphael, Barthélémy Bonadio, and Andrei A Levchenko, “The Economics and
Politics of Revoking NAFTA,” Working Paper 25379, National Bureau of Economic
Research December 2018.
100
Baier, Scott L. and Jeffrey H. Bergstrand, “Do free trade agreements actually increase
members’ international trade?,” Journal of international Economics, 2007, 71 (1), 72–
95.
Baier, Scott L, Jeffrey H Bergstrand, and Matthew W Clance, “Heterogeneous ef-
fects of economic integration agreements,” JOURNAL OF DEVELOPMENT ECO-
NOMICS, nov 2018, 135, 587–608.
Baker, Malcolm and Jeffrey Wurgler, “Investor Sentiment and the Cross-Section of
Stock Returns,” The Journal of Finance, 2006, 61 (4), 1645–1680.
Baker, Scott R, Nicholas Bloom, and Steven J Davis, “Measuring economic policy
uncertainty.(Report),” Quarterly Journal of Economics, 2016, 131 (4), 1593.
, , and , “Measuring economic policy uncertainty.(Report),” Quarterly Journal of
Economics, 2016, 131 (4), 1593.
, , Brandice Canes-Wrone, Steven J Davis, and Jonathan Rodden, “Why has US
policy uncertainty risen since 1960?(Author abstract),” American Economic Review,
2014, 104 (5), 56–60.
, , Steven J Davis, and Kyle Kost, “Policy News and Stock Market Volatility,”
SSRN Electronic Journal, 2019.
, , , and , “Policy News and Stock Market Volatility,” SSRN Electronic Journal,
2019.
Belesky, Paul and Geoffrey Lawrence, “Chinese state capitalism and neomercantilism
in the contemporary food regime: contradictions, continuity and change,” The Journal
of Peasant Studies, 2019, 46 (6), 1119–1141.
Berger, David, Ian Dew-Becker, and Stefano Giglio, “Uncertainty Shocks as Second-
Moment News Shocks,” The Review of Economic Studies, 04 2019, 87 (1), 40–76.
Berhelemy, Jean-Claude, “China’s Engagement and Aid Effectiveness in Africa,”
African Development Bank, 2011, (129).
Bloom, Nicholas, “Fluctuations in Uncertainty,” The Journal of Economic Perspectives,
2014, 28 (2), 153–175.
Bon, Gatien and Gong Cheng, “Understanding China’s role in recent debt relief oper-
ations: A case study analysis,” International economics (Paris), 2021, 166, 23–41.
101
Bouët, Antoine and David Laborde, “US trade wars in the twenty-first century with
emerging countries: Make America and its partners lose again,” World economy, 2018,
41 (9), 2276–2319.
Bown, Chad P and Meredith A Crowley, “Trade deflection and trade depression,”
Journal of international economics, 2007, 72 (1), 176–201.
Brandt, Michael W and Christopher S Jones, “Volatility Forecasting With Range-
Based EGARCH Models,” Journal of business & economic statistics, 2006, 24 (4),
470–486.
Brian, Samantha DeCarlo Daigle and Gregory LaRocca, “When Trade Agreements
End: A Survey of Cases 1949–2019.,” International Commerce and Economics, 2019.
Brogaard, Jonathan and Andrew Detzel, “The asset-pricing implications of govern-
ment economic policy uncertainty,” Management Science, 2015, 61 (1), 3.
Bräutigam, Deborah and Kevin P Gallagher, “Bartering Globalization: China’s
Commodity-backed Finance in Africa and Latin America,” Global policy, 2014, 5
(3), 346–352.
Caceres, Carlos, Diego A Cerdeiro, and Rui C Mano, “Trade Wars and Trade Deals:
Estimated Effects using a Multi-Sector Model,” IMF working paper, 2019.
Caldara, Dario, Matteo Iacoviello, Patrick Molligo, Andrea Prestipino, and Andrea
Raffo, “The economic effects of trade policy uncertainty,” Journal of Monetary Eco-
nomics, 2020, 109, 38–59.
, , , , and , “The economic effects of trade policy uncertainty,” Journal of
Monetary Economics, 2020, 109, 38–59.
Campbell, John Y. and Samuel B. Thompson, “Predicting Excess Stock Returns Out
of Sample: Can Anything Beat the Historical Average?,” The Review of Financial
Studies, 11 2007, 21 (4), 1509–1531.
Campbell, John Y, Martin Lettau, Burton G Malkiel, and Yexiao Xu, “Have Individ-
ual Stocks Become More Volatile? An Empirical Exploration of Idiosyncratic Risk,”
The Journal of finance (New York), 2001, 56 (1), 1–43.
Carter, Colin A and Caroline Gunning-Trant, “U.S. trade remedy law and agriculture:
trade diversion and investigation effects: U.S. trade remedy law and agriculture,” The
Canadian journal of economics, 2010, 43 (1), 97–126.
102
Chauvet, Marcelle, Zeynep Senyuz, and Emre Yoldas, “What does financial volatil-
ity tell us about macroeconomic fluctuations?,” Journal of Economic Dynamics and
Control, 2015, 52, 340–360.
Chiu, Ching-Wai, Richard D F Harris, Evarist Stoja, and Michael Chin, “Financial
market Volatility, macroeconomic fundamentals and investor Sentiment,” Journal of
Banking & Finance, 2018, 92, 130–145.
Choudhry, Taufiq, Fotios I. Papadimitriou, and Sarosh Shabi, “Stock market volatil-
ity and business cycle: Evidence from linear and nonlinear causality tests,” Journal of
Banking & Finance, May 2016, 66, 89–101.
Christiansen, Charlotte, Maik Schmeling, and Andreas Schrimpf, “A comprehen-
sive look at financial volatility prediction by economic variables,” Journal of Applied
Econometrics, 2012, 27 (6).
Copper, John F., China’s Foreign Aid and Investment Diplomacy, Volume III: Strategy
Beyond Asia and Challenges to the United States and the International Order, New
York: Palgrave Macmillan, 2015.
Cover, James Peery, “Asymmetric Effects of Positive and Negative Money-Supply
Shocks,” The Quarterly journal of economics, 1992, 107 (4), 1261–1282.
Dai, Mian, Yoto V Yotov, and Thomas Zylkin, “On the trade-diversion effects of free
trade agreements,” Economics letters, 2014, 122 (2), 321–325.
Dew-Becker, Ian, Stefano Giglio, and Bryan Kelly, “Hedging macroeconomic and
financial uncertainty and volatility,” Journal of Financial Economics, 2021.
Dreher, Axel and Andreas Fuchs, “Rogue aid? An empirical analysis of China’s aid
allocation,” The Canadian journal of economics, 2015, 48 (3), 988–1023.
, , Bradley Parks, Austin M. Strange, and Michael J. Tierney, “Aid, China, and
Growth: Evidence from a New Global Development Finance Dataset,” SSRN Elec-
tronic Journal, 2017.
Ederington, Louis H. and Jae HA Lee, “How Markets Process Information: News
Releases and Volatility,” The Journal of Finance, 1993, 48 (4), 1161–1191.
Eicher, Theo S and Christian Henn, “In search of WTO trade effects: Preferential trade
agreements promote trade strongly, but unevenly,” JOURNAL OF INTERNATIONAL
ECONOMICS, mar 2011, 83 (2), 137–153.
103
Fajgelbaum, Pablo D, Pinelopi K Goldberg, Patrick J Kennedy, and Amit K Khan-
delwal, “The Return to Protectionism*,” The Quarterly Journal of Economics, 11
2019, 135 (1), 1–55.
Feng, Ling, Zhiyuan Li, and Deborah L Swenson, “Trade policy uncertainty and ex-
ports: Evidence from China’s WTO accession,” Journal of International Economics,
2017, 106, 20.
Fernández, Raquel and Jonathan Portes, “Returns to Regionalism: An Analysis of
Nontraditional Gains from Regional Trade Agreements,” The World Bank economic
review, 1998, 12 (2), 197–220.
Fetzer, Thiemo and Carlo Schwarz, “Tariffs and Politics: Evidence from Trump’s
Trade Wars,” SSRN Electronic Journal.
Foroutan, Faezeh and Lant Pritchett, “Intra-Sub-Saharan African Trade: is it too Lit-
tle?,” Journal of African Economies, 2 (1), 74–105.
Frazer, Garth and Johannes Van Biesebroeck, “Trade Growth Under The African
Growth and Opportunity Act,” The Review of Economics and Statistics, 2010, 92 (1),
128–144.
Fukao, Kyoji, Toshihiro Okubo, and Robert M Stern, “An econometric analysis of
trade diversion under NAFTA,” The North American Journal of Economics and Fi-
nance, 2003, 14 (1), 3–24.
Geda, Alemayehu and Edris Hussein Seid, “The potential for internal trade and re-
gional integration in Africa,” Journal of African Trade, 2015, 2 (1-2), 19–50.
Ghosh, Sucharita and Steven Yamarik, “Does trade creation measure up? A reexam-
ination of the effects of regional trading arrangements,” Economics letters, 2004, 82
(2), 213–219.
Guo, Hui, “Stock Market Returns, Volatility, and Future Output,” Review, 2002, 84 (5).
Haidar, J. I., “Sanctions and export deflection: Evidence from Iran,” Economic Policy,
2019, 90 (32), 319–355.
Hamilton, Carl, “Voluntary Export Restraints and Trade Diversion,” Journal of common
market studies, 1985, 23 (4), 345–355.
Handley, Kyle and Nu Limao, “Policy Uncertainty, Trade, and Welfare: Theory and
Evidence for China and the United States.(Author abstract),” American Economic Re-
view, 2017, 107 (9), 2731–2783.
104
and Nuno Limão, “Trade and Investment under Policy Uncertainty: Theory and Firm
Evidence,” American Economic Journal: Economic Policy, 2015, 7 (4), 189–222.
and , “Trade and Investment under Policy Uncertainty: Theory and Firm Evidence,”
American Economic Journal: Economic Policy, 2015, 7 (4), 189–222.
Harris, Richard D F, Evarist Stoja, and Fatih Yilmaz, “A cyclical model of exchange
rate volatility,” Journal of banking & finance, 2011, 35 (11), 3055–3064.
Hoppe, Mombert Brenton Paul;, “The African Growth and Opportunity Act, Exports,
and Development in Sub-Saharan Africa,” 2006, (No. 3996).
Horn, Sebastian, Carmen M. Reinhart, and Christoph Trebesch, “China’s overseas
lending,” Journal of International Economics, 2021, 133, 103539.
Humphreys, David, “New mercantilism: A perspective on how politics is shaping world
metal supply,” Resources Policy, 2013, 38 (3), 341–349.
Ibold, Christian, “BRI Projects,” 2022.
Jeanne, O., “Currency Wars, Trade Wars and Global Demand,” in “in” 2019.
Jouanjean, Marie-Agnes Engel Jakob;, “More than a gesture? Examining the export
impact of Chinese and Indian duty-free quotafree trade preferences for Least Devel-
oped Countries,” 2014.
Julio, Brandon and Youngsuk Yook, “Policy uncertainty, irreversibility, and cross-
border flows of capital,” Journal of International Economics, 2016, 103 (C), 13.
Knight, Frank, Risk, Uncertainty and Profit, Houghton Mifflin, 1921.
Kohl, Tristan, “Do we really know that trade agreements increase trade?,” Review of
World Economics / Weltwirtschaftliches Archiv, 2014, 150 (3), 443–469.
, Steven Brakman, and Harry Garretsen, “Do Trade Agreements Stimulate Interna-
tional Trade Differently? Evidence from 296 Trade Agreements,” The World Economy,
2016, 39 (1), 97–131.
Lee, Gary GJ and Robert Engle, “A permanent and transitory component model of
stock return volatility,” in R. Engle and H. White, eds., Cointegration, Causality, and
Forecasting: A Festschrift in Honor of Clive W.J. Granger, New York: Oxford Uni-
versity Press, 1999, pp. 475–497.
105
Lind, Jennifer and Daryl G Press, “Markets or Mercantilism? How China Secures Its
Energy Supplies,” International Security, 2018, 42 (4), 170–204.
Liu, Ailan and Bo Tang, “US and China aid to Africa: Impact on the donor-recipient
trade relations,” China Economic Review, 2018, 48, 46–65.
Liu, Li and Tao Zhang, “Economic policy uncertainty and stock market volatility,”
Finance Research Letters, 2015, 15, 99–105.
Liu, Qing, Fei Pei, Huaqing Wu, and Xianfeng Zhang, “Trade Policy Uncertainty,
Firm Heterogeneity and Export Mode,” Emerging Markets Finance and Trade.
Longo, Robert and Khalid Sekkat, “Economic Obstacles to Expanding Intra-African
Trade,” World Development, 2004, 32 (8), 1309–1321.
Mattoo, Aaditya and Robert W Staiger, “Trade Wars: What do they mean? Why are
they happening now? What are the costs?,” Economic policy, 2020.
Mawdsley, Emma, “Fu Manchu versus Dr Livingstone in the Dark Continent? Repre-
senting China, Africa and the West in British broadsheet newspapers,” Political Geog-
raphy, 2008, 27 (5), 509–529.
Melitz, Marc J, “The Impact of Trade on Intra-Industry Reallocations and Aggregate
Industry Productivity,” Econometrica, 2003, 71 (6), 1695–1725.
Miromanova, Anna, “Quantifying the Trade Reducing Effect of Embargoes: Firm Level
Evidence from Russia,” Working Paper, 2019.
Moons, Selwyn J V and Peter A G Bergeijk, “Does Economic Diplomacy Work? A
Meta-analysis of Its Impact on Trade and Investment,” World economy, 2017, 40 (2),
336–368.
Neely, Christopher J., David E. Rapach, Jun Tu, and Guofu Zhou, “Forecasting the
Equity Risk Premium: The Role of Technical Indicators,” Management Science, 2014,
60 (7), 1772–1791.
Nitsch, Volker, “State Visits and International Trade,” World economy, 2007, 30 (12),
1797–1816.
Oberhofer, Harald and Michael Pfaffermayr, “Estimating the Trade and Welfare Ef-
fects of Brexit: A Panel Data Structural Gravity Model,” January 2018, (6828).
106
Olasehinde-Williams, Godwin, “Is US trade policy uncertainty powerful enough to pre-
dict global output volatility?,” Journal of International Trade & Economic Develop-
ment.
Ongan, Serdar and Ismet Gocer, “Does trade policy related uncertainty affect interna-
tional trade? Evidence from the US-China commodity trade,” China Economic Jour-
nal, 2020.
Pavcnik, Nina, “Trade Liberalization, Exit, and Productivity Improvements: Evidence
from Chilean Plants,” The Review of Economic Studies, 2002, 69 (1), 245–276.
Pettersson, Jan and Lars Johansson, “Aid, Aid for Trade, and bilateral trade: An em-
pirical study,” Journal of International Trade and Economic Development, 2013, 22
(6), 866–894.
Phan, Dinh Hoang Bach, Susan Sunila Sharma, and Vuong Thao Tran, “Can eco-
nomic policy uncertainty predict stock returns? Global evidence.(Report),” Journal of
International Financial Markets, Institutions & Money, 2018, 55, 134.
Poon, Ser-Huang and Clive W.J. Granger, “Forecasting Volatility in Financial Mar-
kets: A Review,” Journal of Economic Literature, June 2003, 41 (2), 478–539.
Prusa, Thomas J, “The Trade Effects of U.S. Antidumping Actions,” in “The Effects of
U.S. Trade Protection and Promotion Policies,” Chicago: University of Chicago Press,
2019, pp. 191–214.
Rapach, David E., Jack K. Strauss, and Guofu Zhou, “Out-of-Sample Equity Pre-
mium Prediction: Combination Forecasts and Links to the Real Economy,” The Review
of Financial Studies, 08 2009, 23 (2), 821–862.
Ravn, Morten O and Harald Uhlig, “On Adjusting the Hodrick-Prescott Filter for the
Frequency of Observations,” The review of economics and statistics, 2002, 84 (2),
371–376.
Rodrik, Dani, “What Do Trade Agreements Really Do?,” Journal of Economic Perspec-
tives, May 2018, 32 (2), 73–90.
Rose, Andrew K, “The Foreign Service and Foreign Trade: Embassies as Export Pro-
motion,” World economy, 2007, 30 (1), 22–38.
S., Dreher A.-Elston T.B. Fuchs A. Ghose S. Lin J. Malik A. Parks B.C. Russell B.
Solomon K. Strange A. Tierney M.J. Walsh K. Zaleski L. Custer and S. Zhang,
107
“Tracking Chinese Development Finance: An Application of AidData’s TUFF 2.0
Methodology,” 2021.
Santos Silva, J M C and Silvana Tenreyro, “Further simulation evidence on the per-
formance of the Poisson pseudo-maximum likelihood estimator,” Economics Letters,
2011, 112 (2), 220–222.
Savin, Ivan, Marta Marson, and Marina Sutormina, “How different aid flows affect
different trade flows: Evidence from Africa and its largest donors,” Structural Change
and Economic Dynamics, 2020, 55, 119–136.
Schwert, G William, “Why Does Stock Market Volatility Change Over Time?,” The
Journal of finance (New York), 1989, 44 (5), 1115–1153.
Silva, J M C Santos and Silvana Tenreyro, “The Log of Gravity,” The Review of Eco-
nomics and Statistics, 2006, 88 (4), 641–658.
Sima, Qian, “Records of the grand historian. Qin dynasty,” 1993.
Sims, Christopher A., “Macroeconomics and Reality,” Econometrica, 1980, 48 (1), 1–
48.
Steinberg, Jb, “Brexit and the macroeconomic impact of trade policy uncertainty,” Jour-
nal Of International Economics, 2019, 117, 175–195.
Stockwell, Thomas, “Accommodative Monetary Policy Shocks Do Not Stimulate Out-
put,” Technical Report September 2020.
Sun, Zhina and Ehizuelen Michael Mitchell Omoruyi, “Effect of China’s Zero-Tariff
Treatment under Forum on China-Africa Cooperation (FOCAC) on Export Diversifi-
cation in Beneficiary Countries in Africa,” Journal of African Trade, 2021, 8, 23–32.
Viner, Jacob, The customs union issue, New York: Oxford University Press, 2014.
Wang, Yudong, Yu Wei, Chongfeng Wu, and Libo Yin, “Oil and the short-term pre-
dictability of stock return volatility,” Journal of Empirical Finance, 2018, 47, 90–104.
Xu, Yiqing, “Generalized Synthetic Control Method: Causal Inference with Interactive
Fixed Effects Models,” Political Analysis, 2017, 25 (1), 57–76.
Yang, Shanping and Inmaculada Martinez-Zarzoso, “A panel data analysis of trade
creation and trade diversion effects: The case of ASEAN–China Free Trade Area,”
China Economic Review, 2014, 29, 138–151.
108
Zeiler, THOMAS W, “Managing Protectionism: American Trade Policy in the Early
Cold War,” Diplomatic history, 1998, 22 (3), 337–360.
109