Democracy and Trade Policy at the Product Levellowast

Soubhik Bararidagger In Song KimDagger

May 14 2020

Abstract

Despite widespread recognition that trade policies vary significantly across products andtrading partners granular trade policy data is often inaccessible to scholars due to the dif-ficulties in collecting linking and structuring data from multiple sources In this paper weovercome these challenges and construct a dataset of 57 billion observations of applied tar-iff rates between any pair of 136 countries over 20 years Using this data we offer the firstproduct-level analysis of the impact of political regimes on trade liberalization To illustratethe severity of aggregation bias we estimate a series of Bayesian multilevel models at differentlevels of disaggregation at the Harmonized System (HS) 2- 4- and 6-digit levels We findthat democracies tend to enact lower trade barriers than non-democracies However our anal-ysis shows significant variations across products even within the same industry revealing forexample that democracies are often more likely to protect consumer goods Our findings sug-gest that trade politics depends heavily on heterogeneous preferences of individual producersacross specific products and the types of political institutions within which these preferencesare realized

Keywords democracy trade liberalization tariff-line data big data heterogeneous effectsagricultural protection

lowastWe thank the IDB-team at the World Trade Organization for their approval of the re-dissemination of tariff-linedata Financial support from the Amazon Web Services (AWS) Cloud Credits for Research Program is acknowledgedWe thank Weihuang Wong for his valuable insights and contributions to improve the paper Thanks also to TimBuumlthe Devin Caughey Volha Charnysh Raymond Hicks David Lake Lisa Camner McKay Asya MagazinnikEdward Mansfield Helen Milner Rich Nielsen Pablo Pinto Peter Rosendorff Sujeong Shim Alastair Smith RandyStone Jan Stuckatz Rory Truex Ariel White and Bernardo Zacka as well as the seminar participants at PrincetonUniversity Rutgers University Technische Universitaumlt Muumlnchen Washington University in St Louis University ofWisconsin-Madison UCLA (Computational IRWorkshop) University of California Riverside the 2017 InternationalPolitical Economy Society (IPES) and the 2018 European Political Science Association annual meetings An earlierversion of this article was circulated under the title ldquoTrade Liberalization and Regime Type Evidence from a NewTariff-line Datasetrdquo We are grateful for Samir Dutta and Hyungsuk Yoon for their excellent research assistance

daggerPhD Student Department of Government Harvard University Cambridge MA 02138 Emailsbararigharvardedu URL httpwwwsoubhikbarariorg

DaggerAssociate Professor Department of Political Science Massachusetts Institute of Technology Cambridge MA02139 Email insongmitedu URL httpwebmiteduinsongwww

1 Introduction

A rich body of theoretical research on trade liberalization has identified numerous factors that affect

countriesrsquo trade policies Traditionally this literature has focused on the distributional consequences

of trade across industries in which underlying demands for liberalization are driven by factor en-

dowments or sector-specific skills (eg Scheve and Slaughter 2001 Mayda and Rodrik 2005 Luuml

Scheve and Slaughter 2012) More recent firm-level studies of trade politics predict that political

cleavages will arise across firms even within the same industry (eg Kim 2017 Kim and Osgood

2019) In this paper we conduct an empirical evaluation of the related question of whether and how

this demand-side interacts with the characteristics of domestic political institutions which has long

been a source of controversy among social scientists (Rodrik 1995 Morrow Siverson and Tabares

1998 Mansfield Milner and Rosendorff 2000 2002 Milner and Kubota 2005 Kono 2006)

Although canonical political economy models of trade often predict that trade policies will be

highly various and determined endogenously by complex interactions across various political actors

(eg Mayer 1984 Grossman and Helpman 1994 Maggi and Rodriguez-Clare 2007) empirical anal-

ysis of trade policies across different regime types has been constrained by poor data as researchers

have been limited to using high-level aggregate measures of trade policies when evaluating how

underlying trade preferences are translated into policy outcomes As we demonstrate below this

is due in large part to the enormous difficulties in collecting linking and structuring product-level

trade policy data from multiple sources Thus despite the significant variations that indeed exist in

trade policy across products and trading partners many studies often employ Most Favored Nation

(MFN) applied tariff rates or non-tariff barrier ldquocoverage ratiordquo that are averaged across products

(Mansfield and Busch 1995 Gawande and Hansen 1999 Kono 2006) In fact the resulting single

number for a given importer-year observation has often been used to examine how trade policies

differ across regime types (eg Milner and Kubota 2005)

The first contribution of this paper is to address this discrepancy by constructing a dataset of

over 57 billion observations of product-level applied tariff rates that countries apply to their trading

partners We do this by incorporating the universe of preferential rates and the Generalized System

of Preferences (GSP) at the tariff-line level ndash the level at which tariff policy is actually set We

develop a replicable automated procedure to (1) retrieve tariff data from multiple web data sources

1

(2) identify the partner-specific tariff rates for each product and (3) resolve any discrepancies that

arise We then combine our product-level trade policy data for each directed dyad with numerous

country- dyad- and directed dyad-level datasets available in the literature such as measures of

political institutions GATTWTO membership and product-level bilateral trade volume at the

Harmonized System (HS) 6-digit level To the best of our knowledge this is the first database that

combines bilateral trade policies and trade volume at the most granular level comparable across

136 countries over 30 years

Using this data we show that researchers may commit ecological fallacy in studying trade

politics if empirical studies are not done at the appropriate unit of analysis For example within

the United States agricultural sector the 2016 US applied MFN tariff on Onions which is identified

by a HS 6-digit subheading code 071220 is 256 However at the HS 4-digit heading level onions

are classified as Dried vegetables (0712) and given an average tariff rate of 651 More severely

if the analyst chooses the HS 2-digit chapter code classification as the unit of aggregation onions

are simply classified as Vegetables (07) and given the average tariff rate of 435 To date most

researchers only consider the average country-level applied rate in which dried onions and all other

vegetable products would be equally assigned a 334 tariff rate Based on this ldquolowrdquo average

applied tariff rate scholars often believe that the US has mostly eliminated trade barriers and

some may go even so far as to falsely conclude that tariffs are no longer relevant to study trade

politics

Our second contribution is to empirically examine how tariff policy varies by products for coun-

tries with different regime types To exploit the rich structure of this massive dataset we develop a

Bayesian multilevel estimator that explicitly models the correlations across industries while over-

coming computational challenges in estimation and statistical inference with variational approxima-

tions (Jordan Ghahramani Jaakkola et al 1999) To illustrate the severity of potential aggregation

bias when researchers employ aggregated measures of trade policies we estimate the proposed sta-

tistical model at different degrees of disaggregation at the HS2 HS4 and HS6 levels In doing so

we offer the first analysis that yields granular estimates across numerous products and industries

enabling researchers to re-evaluate some of key assumptions in the literature about the relationships

between underlying trade preferences and trade policy-making

We begin our analysis by comparing the MFN trade policies between democracies and non-

2

democratic nations Consistent with Milner and Kubota (2005) we find that democracies are

associated with lower tariffs than non-democracies on average However we find a high level of

heterogeneity across products even within narrowly defined industries Specifically we find that

democracies are more likely to protect consumer goods such as food textile and manufactured

products Meanwhile we find less variability across products within some of industries such as

cotton Furthermore industries such as wood and metal industries as well as intermediate products

tend to get significantly lower tariff rates in democracies compared to non-democracies Our find-

ing provides evidence for highly heterogenous effects of democratic political institutions on trade

liberalization across products consistent with the firm-level theoretical framework In particular

it suggests that trade politics depends heavily on the differences across individual products and

producers within industries We also find that democratic institutions with larger selectorate than

autocracies do not necessarily empower consumers relative to producers thereby calling into ques-

tion whether the underlying assumption of many studies in the Open Economy Politics (OEP)

literature that individual trade preferences (eg given by their factoral or sectoral interests) are

translated into actual trade policies is indeed valid (Lake 2009)

Finally we undertake a dyadic analysis to examine whether the interaction of regime types

between trading partners affects the depth of trade liberalization As Mansfield Milner and

Rosendorff (2000) note measures of bilateral trade barriers across all combinations of country-

pairs are notoriously difficult to collect at the product level thereby constraining researchers to

use bilateral trade volume as a proxy measure for partner-specific trade policy Using our novel

dataset we are able to conduct the most rigorous dyadic analysis of tariff levels to date We con-

sider a total of 90 bilateral Free Trade Agreements (FTAs) that were signed between 1991 and

2012 We compute differences in average applied tariff rates before and after each agreement at

HS2 and HS4 levels We then compare the difference-in-differences between the two institutional

combinations In contrast to existing studies we find little evidence that pairs of democratic nations

tend to undergo deeper trade liberalization than mixed pairs (Mansfield Milner and Rosendorff

2000) However the direction of trade liberalization matters We show that a non-democratic im-

porter engages in shallower trade liberalization when negotiating with a democratic exporter than a

democratic importer does when negotiating with another democracy Democratic importers mean-

while give even deeper reductions in tariffs to non-democratic negotiating partners than they do

3

to other democracies Overall our findings shed new lights on the claim that democratic political

institutions facilitate unilateral and bilateral trade liberalization

The rest of the paper is organized as follows In the next section we provide a detailed descrip-

tion of computational challenges around the usage of tariff-line data and our automated dataset

compilation pipeline to address them Specifically we show that numerous discrepancies exist be-

tween two primary databases that have been widely used in the literature and we explain how we

construct a new dataset that resolves these discrepancies Section 3 presents the empirical findings

from the monadic and dyadic analyses The final section concludes The bilateral trade policy

and volume data at the HS6 level will be fully made available via the Journalrsquos Dataverse reposi-

tory To facilitate future research the source code for constructing the bilateral product-level tariffs

database as well as the estimated product-varying effects of political institutions and their posterior

distributions will be made publicly available at httppoltradegithubio

2 New Database Bilateral Product-level Applied Tariffs

In this section we introduce the new bilateral product-level tariffs database We illustrate the

importance of accounting for the heterogeneity in trade policies across products and trading part-

ners This database will lay an important empirical foundation for our empirical evaluation of the

interaction between democratic political institutions and trade policy-making at the product level

Variation in Tariffs Across Products and Trading Partners While overall tariff rates have

decreased substantially through a slew of bilateral and multilateral trade agreements over the past

decades countries continuously spend enormous resources on negotiating tariff rates In fact tariff

still serves as an important revenue source for many countries even among developed economies

(Hansen 1990 Bastiaens and Rudra 2016) For example US tariff revenue was approximately

$40 billion in 2015 (before the significant increase of applied tariffs under the Trump administra-

tion) which was similar to the revenue from the federal capital gains tax on corporations (Betz

and Pond 2020)1 To be sure there exist various non-tariff barriers (NTB) to trade However

with its simplicity in application and international standardization tariffs serve as highly valuable1In the US about 60 of products are still subject to non-zero tariffs and the mean MFN tariff rate for

these ldquodutiable goodsrdquo is approximately 73 With the tariffs under the Trump administration the applied rates

sometimes go up to 25 or even higher across many products

4

(2) identify the partner-specific tariff rates for each product and (3) resolve any discrepancies that

arise We then combine our product-level trade policy data for each directed dyad with numerous

country- dyad- and directed dyad-level datasets available in the literature such as measures of

political institutions GATTWTO membership and product-level bilateral trade volume at the

Harmonized System (HS) 6-digit level To the best of our knowledge this is the first database that

combines bilateral trade policies and trade volume at the most granular level comparable across

136 countries over 30 years

Using this data we show that researchers may commit ecological fallacy in studying trade

politics if empirical studies are not done at the appropriate unit of analysis For example within

the United States agricultural sector the 2016 US applied MFN tariff on Onions which is identified

by a HS 6-digit subheading code 071220 is 256 However at the HS 4-digit heading level onions

are classified as Dried vegetables (0712) and given an average tariff rate of 651 More severely

if the analyst chooses the HS 2-digit chapter code classification as the unit of aggregation onions

are simply classified as Vegetables (07) and given the average tariff rate of 435 To date most

researchers only consider the average country-level applied rate in which dried onions and all other

vegetable products would be equally assigned a 334 tariff rate Based on this ldquolowrdquo average

applied tariff rate scholars often believe that the US has mostly eliminated trade barriers and

some may go even so far as to falsely conclude that tariffs are no longer relevant to study trade

politics

Our second contribution is to empirically examine how tariff policy varies by products for coun-

tries with different regime types To exploit the rich structure of this massive dataset we develop a

Bayesian multilevel estimator that explicitly models the correlations across industries while over-

coming computational challenges in estimation and statistical inference with variational approxima-

tions (Jordan Ghahramani Jaakkola et al 1999) To illustrate the severity of potential aggregation

bias when researchers employ aggregated measures of trade policies we estimate the proposed sta-

tistical model at different degrees of disaggregation at the HS2 HS4 and HS6 levels In doing so

we offer the first analysis that yields granular estimates across numerous products and industries

enabling researchers to re-evaluate some of key assumptions in the literature about the relationships

between underlying trade preferences and trade policy-making

We begin our analysis by comparing the MFN trade policies between democracies and non-

2

democratic nations Consistent with Milner and Kubota (2005) we find that democracies are

associated with lower tariffs than non-democracies on average However we find a high level of

heterogeneity across products even within narrowly defined industries Specifically we find that

democracies are more likely to protect consumer goods such as food textile and manufactured

products Meanwhile we find less variability across products within some of industries such as

cotton Furthermore industries such as wood and metal industries as well as intermediate products

tend to get significantly lower tariff rates in democracies compared to non-democracies Our find-

ing provides evidence for highly heterogenous effects of democratic political institutions on trade

liberalization across products consistent with the firm-level theoretical framework In particular

it suggests that trade politics depends heavily on the differences across individual products and

producers within industries We also find that democratic institutions with larger selectorate than

autocracies do not necessarily empower consumers relative to producers thereby calling into ques-

tion whether the underlying assumption of many studies in the Open Economy Politics (OEP)

literature that individual trade preferences (eg given by their factoral or sectoral interests) are

translated into actual trade policies is indeed valid (Lake 2009)

Finally we undertake a dyadic analysis to examine whether the interaction of regime types

between trading partners affects the depth of trade liberalization As Mansfield Milner and

Rosendorff (2000) note measures of bilateral trade barriers across all combinations of country-

pairs are notoriously difficult to collect at the product level thereby constraining researchers to

use bilateral trade volume as a proxy measure for partner-specific trade policy Using our novel

dataset we are able to conduct the most rigorous dyadic analysis of tariff levels to date We con-

sider a total of 90 bilateral Free Trade Agreements (FTAs) that were signed between 1991 and

2012 We compute differences in average applied tariff rates before and after each agreement at

HS2 and HS4 levels We then compare the difference-in-differences between the two institutional

combinations In contrast to existing studies we find little evidence that pairs of democratic nations

tend to undergo deeper trade liberalization than mixed pairs (Mansfield Milner and Rosendorff

2000) However the direction of trade liberalization matters We show that a non-democratic im-

porter engages in shallower trade liberalization when negotiating with a democratic exporter than a

democratic importer does when negotiating with another democracy Democratic importers mean-

while give even deeper reductions in tariffs to non-democratic negotiating partners than they do

3

to other democracies Overall our findings shed new lights on the claim that democratic political

institutions facilitate unilateral and bilateral trade liberalization

The rest of the paper is organized as follows In the next section we provide a detailed descrip-

tion of computational challenges around the usage of tariff-line data and our automated dataset

compilation pipeline to address them Specifically we show that numerous discrepancies exist be-

tween two primary databases that have been widely used in the literature and we explain how we

construct a new dataset that resolves these discrepancies Section 3 presents the empirical findings

from the monadic and dyadic analyses The final section concludes The bilateral trade policy

and volume data at the HS6 level will be fully made available via the Journalrsquos Dataverse reposi-

tory To facilitate future research the source code for constructing the bilateral product-level tariffs

database as well as the estimated product-varying effects of political institutions and their posterior

distributions will be made publicly available at httppoltradegithubio

2 New Database Bilateral Product-level Applied Tariffs

In this section we introduce the new bilateral product-level tariffs database We illustrate the

importance of accounting for the heterogeneity in trade policies across products and trading part-

ners This database will lay an important empirical foundation for our empirical evaluation of the

interaction between democratic political institutions and trade policy-making at the product level

Variation in Tariffs Across Products and Trading Partners While overall tariff rates have

decreased substantially through a slew of bilateral and multilateral trade agreements over the past

decades countries continuously spend enormous resources on negotiating tariff rates In fact tariff

still serves as an important revenue source for many countries even among developed economies

(Hansen 1990 Bastiaens and Rudra 2016) For example US tariff revenue was approximately

$40 billion in 2015 (before the significant increase of applied tariffs under the Trump administra-

tion) which was similar to the revenue from the federal capital gains tax on corporations (Betz

and Pond 2020)1 To be sure there exist various non-tariff barriers (NTB) to trade However

with its simplicity in application and international standardization tariffs serve as highly valuable1In the US about 60 of products are still subject to non-zero tariffs and the mean MFN tariff rate for

these ldquodutiable goodsrdquo is approximately 73 With the tariffs under the Trump administration the applied rates

sometimes go up to 25 or even higher across many products

4

democratic nations Consistent with Milner and Kubota (2005) we find that democracies are

associated with lower tariffs than non-democracies on average However we find a high level of

heterogeneity across products even within narrowly defined industries Specifically we find that

democracies are more likely to protect consumer goods such as food textile and manufactured

products Meanwhile we find less variability across products within some of industries such as

cotton Furthermore industries such as wood and metal industries as well as intermediate products

tend to get significantly lower tariff rates in democracies compared to non-democracies Our find-

ing provides evidence for highly heterogenous effects of democratic political institutions on trade

liberalization across products consistent with the firm-level theoretical framework In particular

it suggests that trade politics depends heavily on the differences across individual products and

producers within industries We also find that democratic institutions with larger selectorate than

autocracies do not necessarily empower consumers relative to producers thereby calling into ques-

tion whether the underlying assumption of many studies in the Open Economy Politics (OEP)

literature that individual trade preferences (eg given by their factoral or sectoral interests) are

translated into actual trade policies is indeed valid (Lake 2009)

Finally we undertake a dyadic analysis to examine whether the interaction of regime types

between trading partners affects the depth of trade liberalization As Mansfield Milner and

Rosendorff (2000) note measures of bilateral trade barriers across all combinations of country-

pairs are notoriously difficult to collect at the product level thereby constraining researchers to

use bilateral trade volume as a proxy measure for partner-specific trade policy Using our novel

dataset we are able to conduct the most rigorous dyadic analysis of tariff levels to date We con-

sider a total of 90 bilateral Free Trade Agreements (FTAs) that were signed between 1991 and

2012 We compute differences in average applied tariff rates before and after each agreement at

HS2 and HS4 levels We then compare the difference-in-differences between the two institutional

combinations In contrast to existing studies we find little evidence that pairs of democratic nations

tend to undergo deeper trade liberalization than mixed pairs (Mansfield Milner and Rosendorff

2000) However the direction of trade liberalization matters We show that a non-democratic im-

porter engages in shallower trade liberalization when negotiating with a democratic exporter than a

democratic importer does when negotiating with another democracy Democratic importers mean-

while give even deeper reductions in tariffs to non-democratic negotiating partners than they do

3

to other democracies Overall our findings shed new lights on the claim that democratic political

institutions facilitate unilateral and bilateral trade liberalization

The rest of the paper is organized as follows In the next section we provide a detailed descrip-

tion of computational challenges around the usage of tariff-line data and our automated dataset

compilation pipeline to address them Specifically we show that numerous discrepancies exist be-

tween two primary databases that have been widely used in the literature and we explain how we

construct a new dataset that resolves these discrepancies Section 3 presents the empirical findings

from the monadic and dyadic analyses The final section concludes The bilateral trade policy

and volume data at the HS6 level will be fully made available via the Journalrsquos Dataverse reposi-

tory To facilitate future research the source code for constructing the bilateral product-level tariffs

database as well as the estimated product-varying effects of political institutions and their posterior

distributions will be made publicly available at httppoltradegithubio

2 New Database Bilateral Product-level Applied Tariffs

In this section we introduce the new bilateral product-level tariffs database We illustrate the

importance of accounting for the heterogeneity in trade policies across products and trading part-

ners This database will lay an important empirical foundation for our empirical evaluation of the

interaction between democratic political institutions and trade policy-making at the product level

Variation in Tariffs Across Products and Trading Partners While overall tariff rates have

decreased substantially through a slew of bilateral and multilateral trade agreements over the past

decades countries continuously spend enormous resources on negotiating tariff rates In fact tariff

still serves as an important revenue source for many countries even among developed economies

(Hansen 1990 Bastiaens and Rudra 2016) For example US tariff revenue was approximately

$40 billion in 2015 (before the significant increase of applied tariffs under the Trump administra-

tion) which was similar to the revenue from the federal capital gains tax on corporations (Betz

and Pond 2020)1 To be sure there exist various non-tariff barriers (NTB) to trade However

with its simplicity in application and international standardization tariffs serve as highly valuable1In the US about 60 of products are still subject to non-zero tariffs and the mean MFN tariff rate for

these ldquodutiable goodsrdquo is approximately 73 With the tariffs under the Trump administration the applied rates

sometimes go up to 25 or even higher across many products

4

to other democracies Overall our findings shed new lights on the claim that democratic political

institutions facilitate unilateral and bilateral trade liberalization

The rest of the paper is organized as follows In the next section we provide a detailed descrip-

tion of computational challenges around the usage of tariff-line data and our automated dataset

compilation pipeline to address them Specifically we show that numerous discrepancies exist be-

tween two primary databases that have been widely used in the literature and we explain how we

construct a new dataset that resolves these discrepancies Section 3 presents the empirical findings

from the monadic and dyadic analyses The final section concludes The bilateral trade policy

and volume data at the HS6 level will be fully made available via the Journalrsquos Dataverse reposi-

tory To facilitate future research the source code for constructing the bilateral product-level tariffs

database as well as the estimated product-varying effects of political institutions and their posterior

distributions will be made publicly available at httppoltradegithubio

2 New Database Bilateral Product-level Applied Tariffs

In this section we introduce the new bilateral product-level tariffs database We illustrate the

importance of accounting for the heterogeneity in trade policies across products and trading part-

ners This database will lay an important empirical foundation for our empirical evaluation of the

interaction between democratic political institutions and trade policy-making at the product level

Variation in Tariffs Across Products and Trading Partners While overall tariff rates have

decreased substantially through a slew of bilateral and multilateral trade agreements over the past

decades countries continuously spend enormous resources on negotiating tariff rates In fact tariff

still serves as an important revenue source for many countries even among developed economies

(Hansen 1990 Bastiaens and Rudra 2016) For example US tariff revenue was approximately

$40 billion in 2015 (before the significant increase of applied tariffs under the Trump administra-

tion) which was similar to the revenue from the federal capital gains tax on corporations (Betz

and Pond 2020)1 To be sure there exist various non-tariff barriers (NTB) to trade However

with its simplicity in application and international standardization tariffs serve as highly valuable1In the US about 60 of products are still subject to non-zero tariffs and the mean MFN tariff rate for

these ldquodutiable goodsrdquo is approximately 73 With the tariffs under the Trump administration the applied rates

sometimes go up to 25 or even higher across many products

4

Figure 1 Variations in Ad Valorem Applied Tariff Rates across Trading Partners andIndustries This figure demonstrates how our tariff-line data captures both partner-specific and industry-varyingtrade policies Importers are plotted down each column and exporters are plotted across each row For a given countryand partner our data distinguishes precise tariff rates on more than 100 specific products in various industries (coloredwithin plot) from 1989 to 2015 Increases in applied tariff rates may be attributed to the conversion of specific tariffrates into ad valorem equivalents or actual temporary increases due to ldquobinding overhangrdquo (Pelc 2013 92)

measures for researchers to make systematic comparisons of trade policies across various countries

and products2 In this section we describe the challenges in compiling a large-scale dataset of

bilateral tariff rates We discuss the variation in applied tariff rates our data compilation process

the discrepancies in available data sources and the ways we organize the data for our empirical

analyses

The need for a dataset that captures bilateral tariffs at the product level stems from the substan-

tial heterogeneity in trade polices across industries and partners as Figure 1 shows For example2Getting disaggregated NTB measures is notoriously difficult Also it can be debated whether regulatory policies

that are designed to protect public health and safety should be deemed as barriers of trade further complicating the

use of NTB measures for evaluating trade policies across countries with different norms and cultural backgrounds

5

the first row shows that across industries and over time the MFN tariff rates applied by the US

on imports from China (both members of the WTO) are very different from the preferential rates

applied on imports from Mexico (both members of NAFTA) The columns show that exporters (in

this example China and Mexico) face markedly different tariffs on their products with different

trading partners This heterogeneity exists despite broad membership in the WTO because WTO

members are permitted to enter regional trade agreements under Article XXIV of GATT Enabling

Clause and to lower tariffs for the least developed countries with the GSP That is the rule of ldquonon-

discriminationrdquo does not hold in practice For example in 2013 the US tariffs on cars (Harmonized

Tariff Schedule subheading 87039000) exported by FTA partner South Korea was 15 whereas

it was 25 (the MFN rate) for cars originating from other WTO members Moreover even the

GSP rate for specific products can vary across GSP beneficiaries for strategic reasons As Carnegie

(2015 60) finds Pakistan was partially suspended from the US GSP program in 1996 due to its

violations of workersrsquo rights Indeed we find that the applied rates on gloves (HTS subheading

39262030) given to Pakistan was 3 (the MFN rate) in 1997 instead of the GSP rate of 0 even

though Pakistan remained a GSP beneficiary and still received benefits for many other products

To fully examine the political sources of such heterogeneity researchers must use partner-specific

tariff-line data rather than aggregate tariff measures

Automated Data Collection We develop an automated procedure to create a dataset of bi-

lateral trade policy for each tariff-line product and partner To create our dataset we begin with

two data sources (1) the WTO Integrated Database (IDB) and (2) UNCTAD Trade Analysis

Information System (TRAINS) Both contain applied tariff rates on a variety of products for all

WTO countries from 1996 to 2016 However there are three challenges that limit the use of these

databases by researchers in practice

First to download all product-level tariffs each database requires users to submit numerous

queries to the system for each importer-year pair which in our case amounts to more than 2188

queries (Step 1 in Figure 2) To overcome this difficulty we develop software that automates the

data retrieval process gathering more than 100 gigabytes (GB) of product-level tariff data

The secondmdashand more crucialmdashchallenge is to identify the correct partner-specific rates Specif-

ically both databases specify only the ldquotyperdquo or category of tariff rate that a given importer applies

6

year

imp

code

type

rate

2007

US

A01

0190

20

ldquoMos

t Fav

oure

d N

atio

nrdquo6

8

2007

US

A0

year

imp

code

type

rate

2007

US

A01

0190

20

ldquoMos

t Fav

oure

d N

atio

nrdquo6

8

2007

US

A0

year

imp

exp

code

desc

riptio

nra

teso

urce

2007

US

AJA

M09

1040

40ldquoC

BER

A

pref

eren

tialrdquo

0TR

AIN

S

year

imp

code

type

rate

2007

US

A01

0190

20ldquoM

ost F

avou

red

Nat

ionrdquo

68

2007

US

A0

year

imp

code

type

rate

2007

US

A01

0190

20ldquoM

ost F

avou

red

Nat

ionrdquo

68

2007

US

A0

year

imp

exp

code

desc

riptio

nra

teso

urce

2007

US

AJA

M09

1040

40ldquoC

BER

A

pref

eren

tialrdquo

0TR

AIN

S

year

imp

exp

code

type

rate

2007

US

AS

GP

0101

9020

ldquoMFN

ap

plie

d ra

terdquo

68

2007

US

AS

GP

0101

9020

ldquoGen

eral

du

tyrdquo

15

2007

US

AS

GP

0101

9020

ldquoFre

e-tr

ade

for

Sin

gapo

rerdquo

0

year

imp

code

type

rate

2007

US

A01

0190

20ldquoM

FN

appl

ied

rate

rdquo6

8

2007

US

A01

0190

20ldquoG

ener

al

duty

rdquo 15

2007

US

A01

0190

20ldquoF

ree-

trad

e fo

r0

year

imp

exp

code

type

rate

2007

US

AS

GP

0101

9020

ldquoMFN

ap

plie

d ra

terdquo

68

2007

US

AS

GP

0101

9020

ldquoGen

eral

du

tyrdquo

15

2007

US

AS

GP

0101

9020

ldquoFre

e-tr

ade

for

Sin

gapo

rerdquo

0

year

imp

code

type

rate

2007

US

A01

0190

20ldquoM

FN

appl

ied

rate

rdquo6

8

2007

US

A01

0190

20ldquoG

ener

al

duty

rdquo 15

2007

US

A01

0190

20ldquoF

ree-

trad

e fo

r0

Ste

p 1

Web

scr

ape

prod

uct-

leve

l tar

iffs

Ste

p 2

D

isag

greg

ate

topa

rtne

r-sp

ecifi

c du

ties

Ste

p 3

M

erge

two

sour

ces

IDB query form TRAINS query form

year

imp

code

desc

riptio

nra

te

1998

US

A09

1040

40ldquoM

FNrdquo

38

1998

US

A09

1040

40ldquoG

ener

alrdquo

25

1998

US

A09

1040

40ldquoC

BER

A

pref

eren

tialrdquo

0

year

imp

exp

code

desc

riptio

nra

te

2007

US

AIT

AJA

MJO

R

0910

4040

ldquoMFN

rdquo3

8

2007

US

AIT

AJA

MJO

R

0910

4040

ldquoGen

eral

rdquo25

2007

US

AC

UB

JAM

DM

A

0910

4040

ldquoCB

ERA

pr

efer

entia

lrdquo0

year

imp

code

desc

riptio

nra

te

1998

US

A09

1040

40ldquoM

FNrdquo

38

1998

US

A09

1040

40ldquoG

SP

rdquo3

8

38 m

illio

n pr

oduc

t-le

vel d

utie

s

106

mill

ion

prod

uct-

leve

l dut

ies

47

bill

ion

part

ner-

spec

ific

dutie

s

41

bill

ion

part

ner-

spec

ific

dutie

s

year

imp

exp

code

desc

riptio

nra

teso

urce

2007

US

AJA

M09

1040

40ldquoC

BER

A

pref

eren

tialrdquo

0ID

B

57

billi

on

mer

ged

dutie

s

gt2 th

ousa

nd

impo

rter

-yea

r que

ries

helliphellip helliphellip helliphellip

year

imp

exp

code

desc

riptio

nra

te

2007

US

AIT

AJA

MJO

R

0910

4040

ldquoMFN

rdquo3

8

2007

US

AH

TIJA

MZW

E

0910

4040

ldquoGS

Prdquo

38

helliphellip helliphellip

Figure2

Tariff-lineDataset

CreationThisfig

ureillustrates

theprocessof

creating

ourindu

stry-le

velp

artner-spe

cific

tariffda

taset

Asan

exam

ple

weshow

how

weprod

ucethe1998

USd

utyon

Jamaicanim

portsof

Gingersaffron

turmeric(curcuma)thyme

bayleavesc

urry

andotherspices

(HTSsubh

eading

0910

4040

)Firstwescrape

tariffs

across

allavailableim

portersan

dyearsusingthepu

blic

web

form

sforID

Ban

dTRAIN

SThen

weusethetariffbe

neficiary

description(sho

wnas

desc

ript

ion)

tofin

dallt

ariffswho

sebe

neficiary

grou

pinclud

esJa

maicaAsshow

neach

databa

seismissing

adu

tythat

theothercontainsID

Bcontains

anMFN

dutya

generald

uty

andapreferential

dutyb

utno

ttheGSP

dutyT

RAIN

Scontains

aMFN

duty

andaGSP

dutybu

tno

tthepreferential

dutyFinallyto

select

thedu

tyam

ongstthesecand

idates

mostlik

elyap

pliedin

practicewe

useacustom

merging

algorithm

describe

din

App

endixA2In

this

caseJa

maica

enjoys

azero

tariffdu

eto

apreferential

Caribbe

anBasin

Econo

mic

RecoveryAct

(CBERA)du

tyw

hich

supe

rsedes

both

theMFN

andGSP

rates

7

Issue Year-Importer-Exporter-HS(Product description)

WTO IDB report UNCTAD TRAINS report(asymp AVE)

Solution N obs()

2013-China-India-09041200(Crushed or ground Piper pepper)

10 3 noneMissingReport 1991-Japan-Korea-140490499

(Cod fish)none 10 3

Usenon-missing

235 billion(418)

1997-Australia-Singapore-22082010(Grape wine)

3 3 + $3112L(asymp 127) 3

2005-Canada-Australia-22084010(Rum)

2456centlitre of alcohol 2456centlitre of alcohol(asymp 143) 3

Use ad valoremequivalent (AVE)computed byUNCTAD

2004-Argentina-Paraguay-87083110(Motor vehicle brakes)

0 3 14

ConflictingReports

1996-USA-Mexico-87033100(Cars of le 1500 cc cylinder capacity)

25 0 3

Use lower(preferential)rate

024 billion(425)

Table 1 Solutions to Tariff Data Issues This table illustrates examples of specific issues that arise whenattempting to find the correct applied rate for a tariff-line using the IDB and TRAINS databases In each exampleour algorithm selects the source believed to be the more precise applied rate For instance for Australiarsquos 1997 tariffon Grape wine from Singapore IDB reports only a 3 ad valorem rate while TRAINS accounts for an additional$3112 per litre of wine in its ad valorem equivalent (AVE) rate We provide full details of the merging algorithm inAppendix A2

to its partners For example IDB reports that in 1998 the United States applied a 38 tariff rate

on Ginger saffron turmeric (curcuma) thyme bay leaves curry and other spices (HTS subheading

69120090) for all partners belonging to the United States Generalized System of Preferences but

does not specify the specific countries included (eg Albania Angola and so on) We use a mix

of hand-coding from WTO and World Bank reference materials and string matching algorithms

applied to country names and regional trade agreement titles in order to map each unique ldquotyperdquo

appearing in the original data to its corresponding set of disaggregated country ISO codes Even

when the tariff ldquotyperdquo clearly applies to one country an additional step is needed to link the tex-

tual description to the relevant country code Step 2 in Figure 2 illustrates this process using an

example tariff-line Appendix A1 describes our data collection and processing in full detail

Finally there exist a number of inconsistencies between the two data sources Table 1 illustrates

two issues that we identify First we find significant differences in data coverage This is problematic

given that researchers tend to rely primarily on either one of the widely used data sources for

empirical research but usually not both Data for 127 importer-years appear only in IDB (but not

TRAINS) while data for about 842 importer-years appear only in TRAINS (but not IDB) As a

result we find that at least 235 billion observations are missing from one of the databases we

make sure to utilize the available data whenever possible Second IDB returns duties as they are

originally reported (eg 2456centlitre of alcohol) while TRAINS uses a method to estimate

8

an ad valorem equivalent (AVE) for any reported non-ad valorem rate (eg 2456centlitre of

alcohol asymp 143) TRAINS also uses this method to convert mixed or compound duties (eg 3

+ $3112L asymp 127)3 In both cases our algorithm chooses TRAINS since it is the more precise

and informative source to use Third preferential rates may be available from only one source

As shown in the last row of Table 1 TRAINS shows the correct 1996 NAFTA duty-free rate for

United States-Mexico trade in Cars of le 1500 cc cylinder capacity while IDB does not Likewise

IDB shows Argentinarsquos duty-free rate for Motor vehicle brakes imports from MERCOSUR trade

bloc partner Paraguay while TRAINS does not Our algorithm picks the correct partner-specific

preferential rate for both tariff-lines After resolving these issues of missing data and discrepancy

between the two sources we create a dataset of over 57 billion observations of bilateral trade policy

at the product level Appendix A2 details each step in our resolution algorithm In total our

dataset covers 2476 WTO importer-year tariff profiles (3080 importer-year profiles overall) from

1989 to 2015 including comprehensive policies for the top 50 trading countries beginning in 1995

(see Figure A1)

3 The Effects of Regime Type on Trade Policy

In this section we demonstrate the significance of employing our granular data Specifically we

examine differences in import tariff policy between democracies and non-democracies We begin

by analyzing unilateral trade policies (monadic analysis) across countries and products using MFN

applied tariff rates We then utilize our bilateral tariff data to investigate whether pairs of democ-

racies engage in deeper trade liberalization than other pairs do (dyadic analysis) We emphasize

that the goal of this section is to empirically demonstrate high variability in the effects of regime

type on trade policy rather than to propose a new theoretical framework3For a given non-ad valorem tariff tariff UNCTAD calculates an ad valorem equivalent (AVE) by estimating the

unit value of a product using volume statistics The type of statisticsmdasheither tariff-line level statistics from TRAINS

HS6 statistics from UN Comtrade or HS6 statistics aggregated across OECD countriesmdashdepends on data availability

for each product The unit value is then used to approximate a () tariff rate In cases where only an IDB report

is available for a compound rate we impute an AVE using only the ad valorem component of the duty rate

9

31 Monadic Analysis

Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-

Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of

factors with which their country is abundantly endowed and therefore one should expect that trade

liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged

across products they find that democratization in labor-abundant developing countries is associated

with lower trade barriers Other scholars have argued the reverse however suggesting that the need

to win elections makes democratic politicians sensitive to the demands of interest groups who offer

support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile

need appeal to only a very small segment of society to secure their power and therefore might be

less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield

2006)

We begin by examining whether trade policy varies between democracies and non-democracies

across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in

which competing economic interests across sectors determine industry-level trade policy (eg Mayer

1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of

trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting

in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show

political activities of industries such as lobbying interact with economic heterogeneity in import-

penetration and demand elasticity Consequently the canonical model of trade policy predicts

differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)

We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels

to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade

policies Our product-level analysis is also motivated by the possibility that underlying preferences

of various economic actors may interact differently with political institutions at the product-level

a prediction that is consistent with the firm-level theories in international political economy (Kim

2017) The hierarchical structure of the Harmonized System of product classification facilitates

aggregation and disaggregation at different levels of detail For example Brisket cuts a product

in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code

10

02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine

animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine

animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while

our full dataset captures trade policy for individual products at the tariff-line level we disaggregate

up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level

explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products

compared to HS2 or HS4 For example in the United States variance between HS2 groupings of

products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4

groupings of products explained 52 and between-HS6 variance explained 68 Second further

disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained

categorization that is internationally comparable as each country may set their own tariff lines using

8-digit or sometimes 10-digit codes independently4

311 Methodology

To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit

model of the observed MFN tariff rate τipt for importer i and product p in year t

τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt

τipt =

τ lowastipt if τ lowastipt ge 0

0 otherwise

(1)

where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero

otherwise We perform separate analyses aggregating products at either the HS2 industry level or

at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate

τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and

use logged values to address the high skewness of tariffs To compare our empirical findings against

existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV

score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner

and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries

although they tend to describe highly similar goods under the common HS6 category which is already highly specific

11

(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients

Zit represents a vector of country-level time-varying confounders of regime type and trade policy

log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and

an intercept5 To address missingness in covariate data we create multiple imputed datasets and

conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt

represents log import volume (at the same level of product aggregation as the tariff rate) All

covariates are lagged by 1 year

We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed

comparative advantages which captures technological differences across countries industries h (that

product p belongs to) and time t7 For example this allows us to account for the possibility

that developing and developed countries may use different production technologies even when they

produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and

εipt is idiosyncratic error assumed to be drawn from a Normal distribution

ηiiidsim N (0Ση) θt

iidsim N (0Σθ) εiptiidsim N (0 σ2

ε ) (2)

To be sure countries may have different domestic institutions that aggregate trade preferences

across various sectors meaning that trade policies of certain sectors tend to be highly correlated

For example the US Congress established the Agricultural Policy Advisory Committee (APAC)

and other advisory committees within the Department of Agriculture to provide advice on the

administration and implementation of US trade policy To account for heterogeneous political

processes across products we model the product-varying effects hierarchically Specifically we

allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg

Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use

ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log

trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade

volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally

intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total

exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2

level

12

the first row shows that across industries and over time the MFN tariff rates applied by the US

on imports from China (both members of the WTO) are very different from the preferential rates

applied on imports from Mexico (both members of NAFTA) The columns show that exporters (in

this example China and Mexico) face markedly different tariffs on their products with different

trading partners This heterogeneity exists despite broad membership in the WTO because WTO

members are permitted to enter regional trade agreements under Article XXIV of GATT Enabling

Clause and to lower tariffs for the least developed countries with the GSP That is the rule of ldquonon-

discriminationrdquo does not hold in practice For example in 2013 the US tariffs on cars (Harmonized

Tariff Schedule subheading 87039000) exported by FTA partner South Korea was 15 whereas

it was 25 (the MFN rate) for cars originating from other WTO members Moreover even the

GSP rate for specific products can vary across GSP beneficiaries for strategic reasons As Carnegie

(2015 60) finds Pakistan was partially suspended from the US GSP program in 1996 due to its

violations of workersrsquo rights Indeed we find that the applied rates on gloves (HTS subheading

39262030) given to Pakistan was 3 (the MFN rate) in 1997 instead of the GSP rate of 0 even

though Pakistan remained a GSP beneficiary and still received benefits for many other products

To fully examine the political sources of such heterogeneity researchers must use partner-specific

tariff-line data rather than aggregate tariff measures

Automated Data Collection We develop an automated procedure to create a dataset of bi-

lateral trade policy for each tariff-line product and partner To create our dataset we begin with

two data sources (1) the WTO Integrated Database (IDB) and (2) UNCTAD Trade Analysis

Information System (TRAINS) Both contain applied tariff rates on a variety of products for all

WTO countries from 1996 to 2016 However there are three challenges that limit the use of these

databases by researchers in practice

First to download all product-level tariffs each database requires users to submit numerous

queries to the system for each importer-year pair which in our case amounts to more than 2188

queries (Step 1 in Figure 2) To overcome this difficulty we develop software that automates the

data retrieval process gathering more than 100 gigabytes (GB) of product-level tariff data

The secondmdashand more crucialmdashchallenge is to identify the correct partner-specific rates Specif-

ically both databases specify only the ldquotyperdquo or category of tariff rate that a given importer applies

6

year

imp

code

type

rate

2007

US

A01

0190

20

ldquoMos

t Fav

oure

d N

atio

nrdquo6

8

2007

US

A0

year

imp

code

type

rate

2007

US

A01

0190

20

ldquoMos

t Fav

oure

d N

atio

nrdquo6

8

2007

US

A0

year

imp

exp

code

desc

riptio

nra

teso

urce

2007

US

AJA

M09

1040

40ldquoC

BER

A

pref

eren

tialrdquo

0TR

AIN

S

year

imp

code

type

rate

2007

US

A01

0190

20ldquoM

ost F

avou

red

Nat

ionrdquo

68

2007

US

A0

year

imp

code

type

rate

2007

US

A01

0190

20ldquoM

ost F

avou

red

Nat

ionrdquo

68

2007

US

A0

year

imp

exp

code

desc

riptio

nra

teso

urce

2007

US

AJA

M09

1040

40ldquoC

BER

A

pref

eren

tialrdquo

0TR

AIN

S

year

imp

exp

code

type

rate

2007

US

AS

GP

0101

9020

ldquoMFN

ap

plie

d ra

terdquo

68

2007

US

AS

GP

0101

9020

ldquoGen

eral

du

tyrdquo

15

2007

US

AS

GP

0101

9020

ldquoFre

e-tr

ade

for

Sin

gapo

rerdquo

0

year

imp

code

type

rate

2007

US

A01

0190

20ldquoM

FN

appl

ied

rate

rdquo6

8

2007

US

A01

0190

20ldquoG

ener

al

duty

rdquo 15

2007

US

A01

0190

20ldquoF

ree-

trad

e fo

r0

year

imp

exp

code

type

rate

2007

US

AS

GP

0101

9020

ldquoMFN

ap

plie

d ra

terdquo

68

2007

US

AS

GP

0101

9020

ldquoGen

eral

du

tyrdquo

15

2007

US

AS

GP

0101

9020

ldquoFre

e-tr

ade

for

Sin

gapo

rerdquo

0

year

imp

code

type

rate

2007

US

A01

0190

20ldquoM

FN

appl

ied

rate

rdquo6

8

2007

US

A01

0190

20ldquoG

ener

al

duty

rdquo 15

2007

US

A01

0190

20ldquoF

ree-

trad

e fo

r0

Ste

p 1

Web

scr

ape

prod

uct-

leve

l tar

iffs

Ste

p 2

D

isag

greg

ate

topa

rtne

r-sp

ecifi

c du

ties

Ste

p 3

M

erge

two

sour

ces

IDB query form TRAINS query form

year

imp

code

desc

riptio

nra

te

1998

US

A09

1040

40ldquoM

FNrdquo

38

1998

US

A09

1040

40ldquoG

ener

alrdquo

25

1998

US

A09

1040

40ldquoC

BER

A

pref

eren

tialrdquo

0

year

imp

exp

code

desc

riptio

nra

te

2007

US

AIT

AJA

MJO

R

0910

4040

ldquoMFN

rdquo3

8

2007

US

AIT

AJA

MJO

R

0910

4040

ldquoGen

eral

rdquo25

2007

US

AC

UB

JAM

DM

A

0910

4040

ldquoCB

ERA

pr

efer

entia

lrdquo0

year

imp

code

desc

riptio

nra

te

1998

US

A09

1040

40ldquoM

FNrdquo

38

1998

US

A09

1040

40ldquoG

SP

rdquo3

8

38 m

illio

n pr

oduc

t-le

vel d

utie

s

106

mill

ion

prod

uct-

leve

l dut

ies

47

bill

ion

part

ner-

spec

ific

dutie

s

41

bill

ion

part

ner-

spec

ific

dutie

s

year

imp

exp

code

desc

riptio

nra

teso

urce

2007

US

AJA

M09

1040

40ldquoC

BER

A

pref

eren

tialrdquo

0ID

B

57

billi

on

mer

ged

dutie

s

gt2 th

ousa

nd

impo

rter

-yea

r que

ries

helliphellip helliphellip helliphellip

year

imp

exp

code

desc

riptio

nra

te

2007

US

AIT

AJA

MJO

R

0910

4040

ldquoMFN

rdquo3

8

2007

US

AH

TIJA

MZW

E

0910

4040

ldquoGS

Prdquo

38

helliphellip helliphellip

Figure2

Tariff-lineDataset

CreationThisfig

ureillustrates

theprocessof

creating

ourindu

stry-le

velp

artner-spe

cific

tariffda

taset

Asan

exam

ple

weshow

how

weprod

ucethe1998

USd

utyon

Jamaicanim

portsof

Gingersaffron

turmeric(curcuma)thyme

bayleavesc

urry

andotherspices

(HTSsubh

eading

0910

4040

)Firstwescrape

tariffs

across

allavailableim

portersan

dyearsusingthepu

blic

web

form

sforID

Ban

dTRAIN

SThen

weusethetariffbe

neficiary

description(sho

wnas

desc

ript

ion)

tofin

dallt

ariffswho

sebe

neficiary

grou

pinclud

esJa

maicaAsshow

neach

databa

seismissing

adu

tythat

theothercontainsID

Bcontains

anMFN

dutya

generald

uty

andapreferential

dutyb

utno

ttheGSP

dutyT

RAIN

Scontains

aMFN

duty

andaGSP

dutybu

tno

tthepreferential

dutyFinallyto

select

thedu

tyam

ongstthesecand

idates

mostlik

elyap

pliedin

practicewe

useacustom

merging

algorithm

describe

din

App

endixA2In

this

caseJa

maica

enjoys

azero

tariffdu

eto

apreferential

Caribbe

anBasin

Econo

mic

RecoveryAct

(CBERA)du

tyw

hich

supe

rsedes

both

theMFN

andGSP

rates

7

Issue Year-Importer-Exporter-HS(Product description)

WTO IDB report UNCTAD TRAINS report(asymp AVE)

Solution N obs()

2013-China-India-09041200(Crushed or ground Piper pepper)

10 3 noneMissingReport 1991-Japan-Korea-140490499

(Cod fish)none 10 3

Usenon-missing

235 billion(418)

1997-Australia-Singapore-22082010(Grape wine)

3 3 + $3112L(asymp 127) 3

2005-Canada-Australia-22084010(Rum)

2456centlitre of alcohol 2456centlitre of alcohol(asymp 143) 3

Use ad valoremequivalent (AVE)computed byUNCTAD

2004-Argentina-Paraguay-87083110(Motor vehicle brakes)

0 3 14

ConflictingReports

1996-USA-Mexico-87033100(Cars of le 1500 cc cylinder capacity)

25 0 3

Use lower(preferential)rate

024 billion(425)

Table 1 Solutions to Tariff Data Issues This table illustrates examples of specific issues that arise whenattempting to find the correct applied rate for a tariff-line using the IDB and TRAINS databases In each exampleour algorithm selects the source believed to be the more precise applied rate For instance for Australiarsquos 1997 tariffon Grape wine from Singapore IDB reports only a 3 ad valorem rate while TRAINS accounts for an additional$3112 per litre of wine in its ad valorem equivalent (AVE) rate We provide full details of the merging algorithm inAppendix A2

to its partners For example IDB reports that in 1998 the United States applied a 38 tariff rate

on Ginger saffron turmeric (curcuma) thyme bay leaves curry and other spices (HTS subheading

69120090) for all partners belonging to the United States Generalized System of Preferences but

does not specify the specific countries included (eg Albania Angola and so on) We use a mix

of hand-coding from WTO and World Bank reference materials and string matching algorithms

applied to country names and regional trade agreement titles in order to map each unique ldquotyperdquo

appearing in the original data to its corresponding set of disaggregated country ISO codes Even

when the tariff ldquotyperdquo clearly applies to one country an additional step is needed to link the tex-

tual description to the relevant country code Step 2 in Figure 2 illustrates this process using an

example tariff-line Appendix A1 describes our data collection and processing in full detail

Finally there exist a number of inconsistencies between the two data sources Table 1 illustrates

two issues that we identify First we find significant differences in data coverage This is problematic

given that researchers tend to rely primarily on either one of the widely used data sources for

empirical research but usually not both Data for 127 importer-years appear only in IDB (but not

TRAINS) while data for about 842 importer-years appear only in TRAINS (but not IDB) As a

result we find that at least 235 billion observations are missing from one of the databases we

make sure to utilize the available data whenever possible Second IDB returns duties as they are

originally reported (eg 2456centlitre of alcohol) while TRAINS uses a method to estimate

8

an ad valorem equivalent (AVE) for any reported non-ad valorem rate (eg 2456centlitre of

alcohol asymp 143) TRAINS also uses this method to convert mixed or compound duties (eg 3

+ $3112L asymp 127)3 In both cases our algorithm chooses TRAINS since it is the more precise

and informative source to use Third preferential rates may be available from only one source

As shown in the last row of Table 1 TRAINS shows the correct 1996 NAFTA duty-free rate for

United States-Mexico trade in Cars of le 1500 cc cylinder capacity while IDB does not Likewise

IDB shows Argentinarsquos duty-free rate for Motor vehicle brakes imports from MERCOSUR trade

bloc partner Paraguay while TRAINS does not Our algorithm picks the correct partner-specific

preferential rate for both tariff-lines After resolving these issues of missing data and discrepancy

between the two sources we create a dataset of over 57 billion observations of bilateral trade policy

at the product level Appendix A2 details each step in our resolution algorithm In total our

dataset covers 2476 WTO importer-year tariff profiles (3080 importer-year profiles overall) from

1989 to 2015 including comprehensive policies for the top 50 trading countries beginning in 1995

(see Figure A1)

3 The Effects of Regime Type on Trade Policy

In this section we demonstrate the significance of employing our granular data Specifically we

examine differences in import tariff policy between democracies and non-democracies We begin

by analyzing unilateral trade policies (monadic analysis) across countries and products using MFN

applied tariff rates We then utilize our bilateral tariff data to investigate whether pairs of democ-

racies engage in deeper trade liberalization than other pairs do (dyadic analysis) We emphasize

that the goal of this section is to empirically demonstrate high variability in the effects of regime

type on trade policy rather than to propose a new theoretical framework3For a given non-ad valorem tariff tariff UNCTAD calculates an ad valorem equivalent (AVE) by estimating the

unit value of a product using volume statistics The type of statisticsmdasheither tariff-line level statistics from TRAINS

HS6 statistics from UN Comtrade or HS6 statistics aggregated across OECD countriesmdashdepends on data availability

for each product The unit value is then used to approximate a () tariff rate In cases where only an IDB report

is available for a compound rate we impute an AVE using only the ad valorem component of the duty rate

9

31 Monadic Analysis

Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-

Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of

factors with which their country is abundantly endowed and therefore one should expect that trade

liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged

across products they find that democratization in labor-abundant developing countries is associated

with lower trade barriers Other scholars have argued the reverse however suggesting that the need

to win elections makes democratic politicians sensitive to the demands of interest groups who offer

support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile

need appeal to only a very small segment of society to secure their power and therefore might be

less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield

2006)

We begin by examining whether trade policy varies between democracies and non-democracies

across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in

which competing economic interests across sectors determine industry-level trade policy (eg Mayer

1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of

trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting

in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show

political activities of industries such as lobbying interact with economic heterogeneity in import-

penetration and demand elasticity Consequently the canonical model of trade policy predicts

differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)

We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels

to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade

policies Our product-level analysis is also motivated by the possibility that underlying preferences

of various economic actors may interact differently with political institutions at the product-level

a prediction that is consistent with the firm-level theories in international political economy (Kim

2017) The hierarchical structure of the Harmonized System of product classification facilitates

aggregation and disaggregation at different levels of detail For example Brisket cuts a product

in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code

10

02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine

animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine

animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while

our full dataset captures trade policy for individual products at the tariff-line level we disaggregate

up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level

explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products

compared to HS2 or HS4 For example in the United States variance between HS2 groupings of

products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4

groupings of products explained 52 and between-HS6 variance explained 68 Second further

disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained

categorization that is internationally comparable as each country may set their own tariff lines using

8-digit or sometimes 10-digit codes independently4

311 Methodology

To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit

model of the observed MFN tariff rate τipt for importer i and product p in year t

τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt

τipt =

τ lowastipt if τ lowastipt ge 0

0 otherwise

(1)

where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero

otherwise We perform separate analyses aggregating products at either the HS2 industry level or

at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate

τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and

use logged values to address the high skewness of tariffs To compare our empirical findings against

existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV

score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner

and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries

although they tend to describe highly similar goods under the common HS6 category which is already highly specific

11

(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients

Zit represents a vector of country-level time-varying confounders of regime type and trade policy

log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and

an intercept5 To address missingness in covariate data we create multiple imputed datasets and

conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt

represents log import volume (at the same level of product aggregation as the tariff rate) All

covariates are lagged by 1 year

We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed

comparative advantages which captures technological differences across countries industries h (that

product p belongs to) and time t7 For example this allows us to account for the possibility

that developing and developed countries may use different production technologies even when they

produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and

εipt is idiosyncratic error assumed to be drawn from a Normal distribution

ηiiidsim N (0Ση) θt

iidsim N (0Σθ) εiptiidsim N (0 σ2

ε ) (2)

To be sure countries may have different domestic institutions that aggregate trade preferences

across various sectors meaning that trade policies of certain sectors tend to be highly correlated

For example the US Congress established the Agricultural Policy Advisory Committee (APAC)

and other advisory committees within the Department of Agriculture to provide advice on the

administration and implementation of US trade policy To account for heterogeneous political

processes across products we model the product-varying effects hierarchically Specifically we

allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg

Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use

ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log

trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade

volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally

intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total

exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2

level

12

year

imp

code

type

rate

2007

US

A01

0190

20

ldquoMos

t Fav

oure

d N

atio

nrdquo6

8

2007

US

A0

year

imp

code

type

rate

2007

US

A01

0190

20

ldquoMos

t Fav

oure

d N

atio

nrdquo6

8

2007

US

A0

year

imp

exp

code

desc

riptio

nra

teso

urce

2007

US

AJA

M09

1040

40ldquoC

BER

A

pref

eren

tialrdquo

0TR

AIN

S

year

imp

code

type

rate

2007

US

A01

0190

20ldquoM

ost F

avou

red

Nat

ionrdquo

68

2007

US

A0

year

imp

code

type

rate

2007

US

A01

0190

20ldquoM

ost F

avou

red

Nat

ionrdquo

68

2007

US

A0

year

imp

exp

code

desc

riptio

nra

teso

urce

2007

US

AJA

M09

1040

40ldquoC

BER

A

pref

eren

tialrdquo

0TR

AIN

S

year

imp

exp

code

type

rate

2007

US

AS

GP

0101

9020

ldquoMFN

ap

plie

d ra

terdquo

68

2007

US

AS

GP

0101

9020

ldquoGen

eral

du

tyrdquo

15

2007

US

AS

GP

0101

9020

ldquoFre

e-tr

ade

for

Sin

gapo

rerdquo

0

year

imp

code

type

rate

2007

US

A01

0190

20ldquoM

FN

appl

ied

rate

rdquo6

8

2007

US

A01

0190

20ldquoG

ener

al

duty

rdquo 15

2007

US

A01

0190

20ldquoF

ree-

trad

e fo

r0

year

imp

exp

code

type

rate

2007

US

AS

GP

0101

9020

ldquoMFN

ap

plie

d ra

terdquo

68

2007

US

AS

GP

0101

9020

ldquoGen

eral

du

tyrdquo

15

2007

US

AS

GP

0101

9020

ldquoFre

e-tr

ade

for

Sin

gapo

rerdquo

0

year

imp

code

type

rate

2007

US

A01

0190

20ldquoM

FN

appl

ied

rate

rdquo6

8

2007

US

A01

0190

20ldquoG

ener

al

duty

rdquo 15

2007

US

A01

0190

20ldquoF

ree-

trad

e fo

r0

Ste

p 1

Web

scr

ape

prod

uct-

leve

l tar

iffs

Ste

p 2

D

isag

greg

ate

topa

rtne

r-sp

ecifi

c du

ties

Ste

p 3

M

erge

two

sour

ces

IDB query form TRAINS query form

year

imp

code

desc

riptio

nra

te

1998

US

A09

1040

40ldquoM

FNrdquo

38

1998

US

A09

1040

40ldquoG

ener

alrdquo

25

1998

US

A09

1040

40ldquoC

BER

A

pref

eren

tialrdquo

0

year

imp

exp

code

desc

riptio

nra

te

2007

US

AIT

AJA

MJO

R

0910

4040

ldquoMFN

rdquo3

8

2007

US

AIT

AJA

MJO

R

0910

4040

ldquoGen

eral

rdquo25

2007

US

AC

UB

JAM

DM

A

0910

4040

ldquoCB

ERA

pr

efer

entia

lrdquo0

year

imp

code

desc

riptio

nra

te

1998

US

A09

1040

40ldquoM

FNrdquo

38

1998

US

A09

1040

40ldquoG

SP

rdquo3

8

38 m

illio

n pr

oduc

t-le

vel d

utie

s

106

mill

ion

prod

uct-

leve

l dut

ies

47

bill

ion

part

ner-

spec

ific

dutie

s

41

bill

ion

part

ner-

spec

ific

dutie

s

year

imp

exp

code

desc

riptio

nra

teso

urce

2007

US

AJA

M09

1040

40ldquoC

BER

A

pref

eren

tialrdquo

0ID

B

57

billi

on

mer

ged

dutie

s

gt2 th

ousa

nd

impo

rter

-yea

r que

ries

helliphellip helliphellip helliphellip

year

imp

exp

code

desc

riptio

nra

te

2007

US

AIT

AJA

MJO

R

0910

4040

ldquoMFN

rdquo3

8

2007

US

AH

TIJA

MZW

E

0910

4040

ldquoGS

Prdquo

38

helliphellip helliphellip

Figure2

Tariff-lineDataset

CreationThisfig

ureillustrates

theprocessof

creating

ourindu

stry-le

velp

artner-spe

cific

tariffda

taset

Asan

exam

ple

weshow

how

weprod

ucethe1998

USd

utyon

Jamaicanim

portsof

Gingersaffron

turmeric(curcuma)thyme

bayleavesc

urry

andotherspices

(HTSsubh

eading

0910

4040

)Firstwescrape

tariffs

across

allavailableim

portersan

dyearsusingthepu

blic

web

form

sforID

Ban

dTRAIN

SThen

weusethetariffbe

neficiary

description(sho

wnas

desc

ript

ion)

tofin

dallt

ariffswho

sebe

neficiary

grou

pinclud

esJa

maicaAsshow

neach

databa

seismissing

adu

tythat

theothercontainsID

Bcontains

anMFN

dutya

generald

uty

andapreferential

dutyb

utno

ttheGSP

dutyT

RAIN

Scontains

aMFN

duty

andaGSP

dutybu

tno

tthepreferential

dutyFinallyto

select

thedu

tyam

ongstthesecand

idates

mostlik

elyap

pliedin

practicewe

useacustom

merging

algorithm

describe

din

App

endixA2In

this

caseJa

maica

enjoys

azero

tariffdu

eto

apreferential

Caribbe

anBasin

Econo

mic

RecoveryAct

(CBERA)du

tyw

hich

supe

rsedes

both

theMFN

andGSP

rates

7

Issue Year-Importer-Exporter-HS(Product description)

WTO IDB report UNCTAD TRAINS report(asymp AVE)

Solution N obs()

2013-China-India-09041200(Crushed or ground Piper pepper)

10 3 noneMissingReport 1991-Japan-Korea-140490499

(Cod fish)none 10 3

Usenon-missing

235 billion(418)

1997-Australia-Singapore-22082010(Grape wine)

3 3 + $3112L(asymp 127) 3

2005-Canada-Australia-22084010(Rum)

2456centlitre of alcohol 2456centlitre of alcohol(asymp 143) 3

Use ad valoremequivalent (AVE)computed byUNCTAD

2004-Argentina-Paraguay-87083110(Motor vehicle brakes)

0 3 14

ConflictingReports

1996-USA-Mexico-87033100(Cars of le 1500 cc cylinder capacity)

25 0 3

Use lower(preferential)rate

024 billion(425)

Table 1 Solutions to Tariff Data Issues This table illustrates examples of specific issues that arise whenattempting to find the correct applied rate for a tariff-line using the IDB and TRAINS databases In each exampleour algorithm selects the source believed to be the more precise applied rate For instance for Australiarsquos 1997 tariffon Grape wine from Singapore IDB reports only a 3 ad valorem rate while TRAINS accounts for an additional$3112 per litre of wine in its ad valorem equivalent (AVE) rate We provide full details of the merging algorithm inAppendix A2

to its partners For example IDB reports that in 1998 the United States applied a 38 tariff rate

on Ginger saffron turmeric (curcuma) thyme bay leaves curry and other spices (HTS subheading

69120090) for all partners belonging to the United States Generalized System of Preferences but

does not specify the specific countries included (eg Albania Angola and so on) We use a mix

of hand-coding from WTO and World Bank reference materials and string matching algorithms

applied to country names and regional trade agreement titles in order to map each unique ldquotyperdquo

appearing in the original data to its corresponding set of disaggregated country ISO codes Even

when the tariff ldquotyperdquo clearly applies to one country an additional step is needed to link the tex-

tual description to the relevant country code Step 2 in Figure 2 illustrates this process using an

example tariff-line Appendix A1 describes our data collection and processing in full detail

Finally there exist a number of inconsistencies between the two data sources Table 1 illustrates

two issues that we identify First we find significant differences in data coverage This is problematic

given that researchers tend to rely primarily on either one of the widely used data sources for

empirical research but usually not both Data for 127 importer-years appear only in IDB (but not

TRAINS) while data for about 842 importer-years appear only in TRAINS (but not IDB) As a

result we find that at least 235 billion observations are missing from one of the databases we

make sure to utilize the available data whenever possible Second IDB returns duties as they are

originally reported (eg 2456centlitre of alcohol) while TRAINS uses a method to estimate

8

an ad valorem equivalent (AVE) for any reported non-ad valorem rate (eg 2456centlitre of

alcohol asymp 143) TRAINS also uses this method to convert mixed or compound duties (eg 3

+ $3112L asymp 127)3 In both cases our algorithm chooses TRAINS since it is the more precise

and informative source to use Third preferential rates may be available from only one source

As shown in the last row of Table 1 TRAINS shows the correct 1996 NAFTA duty-free rate for

United States-Mexico trade in Cars of le 1500 cc cylinder capacity while IDB does not Likewise

IDB shows Argentinarsquos duty-free rate for Motor vehicle brakes imports from MERCOSUR trade

bloc partner Paraguay while TRAINS does not Our algorithm picks the correct partner-specific

preferential rate for both tariff-lines After resolving these issues of missing data and discrepancy

between the two sources we create a dataset of over 57 billion observations of bilateral trade policy

at the product level Appendix A2 details each step in our resolution algorithm In total our

dataset covers 2476 WTO importer-year tariff profiles (3080 importer-year profiles overall) from

1989 to 2015 including comprehensive policies for the top 50 trading countries beginning in 1995

(see Figure A1)

3 The Effects of Regime Type on Trade Policy

In this section we demonstrate the significance of employing our granular data Specifically we

examine differences in import tariff policy between democracies and non-democracies We begin

by analyzing unilateral trade policies (monadic analysis) across countries and products using MFN

applied tariff rates We then utilize our bilateral tariff data to investigate whether pairs of democ-

racies engage in deeper trade liberalization than other pairs do (dyadic analysis) We emphasize

that the goal of this section is to empirically demonstrate high variability in the effects of regime

type on trade policy rather than to propose a new theoretical framework3For a given non-ad valorem tariff tariff UNCTAD calculates an ad valorem equivalent (AVE) by estimating the

unit value of a product using volume statistics The type of statisticsmdasheither tariff-line level statistics from TRAINS

HS6 statistics from UN Comtrade or HS6 statistics aggregated across OECD countriesmdashdepends on data availability

for each product The unit value is then used to approximate a () tariff rate In cases where only an IDB report

is available for a compound rate we impute an AVE using only the ad valorem component of the duty rate

9

31 Monadic Analysis

Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-

Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of

factors with which their country is abundantly endowed and therefore one should expect that trade

liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged

across products they find that democratization in labor-abundant developing countries is associated

with lower trade barriers Other scholars have argued the reverse however suggesting that the need

to win elections makes democratic politicians sensitive to the demands of interest groups who offer

support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile

need appeal to only a very small segment of society to secure their power and therefore might be

less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield

2006)

We begin by examining whether trade policy varies between democracies and non-democracies

across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in

which competing economic interests across sectors determine industry-level trade policy (eg Mayer

1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of

trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting

in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show

political activities of industries such as lobbying interact with economic heterogeneity in import-

penetration and demand elasticity Consequently the canonical model of trade policy predicts

differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)

We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels

to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade

policies Our product-level analysis is also motivated by the possibility that underlying preferences

of various economic actors may interact differently with political institutions at the product-level

a prediction that is consistent with the firm-level theories in international political economy (Kim

2017) The hierarchical structure of the Harmonized System of product classification facilitates

aggregation and disaggregation at different levels of detail For example Brisket cuts a product

in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code

10

02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine

animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine

animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while

our full dataset captures trade policy for individual products at the tariff-line level we disaggregate

up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level

explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products

compared to HS2 or HS4 For example in the United States variance between HS2 groupings of

products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4

groupings of products explained 52 and between-HS6 variance explained 68 Second further

disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained

categorization that is internationally comparable as each country may set their own tariff lines using

8-digit or sometimes 10-digit codes independently4

311 Methodology

To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit

model of the observed MFN tariff rate τipt for importer i and product p in year t

τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt

τipt =

τ lowastipt if τ lowastipt ge 0

0 otherwise

(1)

where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero

otherwise We perform separate analyses aggregating products at either the HS2 industry level or

at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate

τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and

use logged values to address the high skewness of tariffs To compare our empirical findings against

existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV

score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner

and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries

although they tend to describe highly similar goods under the common HS6 category which is already highly specific

11

(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients

Zit represents a vector of country-level time-varying confounders of regime type and trade policy

log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and

an intercept5 To address missingness in covariate data we create multiple imputed datasets and

conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt

represents log import volume (at the same level of product aggregation as the tariff rate) All

covariates are lagged by 1 year

We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed

comparative advantages which captures technological differences across countries industries h (that

product p belongs to) and time t7 For example this allows us to account for the possibility

that developing and developed countries may use different production technologies even when they

produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and

εipt is idiosyncratic error assumed to be drawn from a Normal distribution

ηiiidsim N (0Ση) θt

iidsim N (0Σθ) εiptiidsim N (0 σ2

ε ) (2)

To be sure countries may have different domestic institutions that aggregate trade preferences

across various sectors meaning that trade policies of certain sectors tend to be highly correlated

For example the US Congress established the Agricultural Policy Advisory Committee (APAC)

and other advisory committees within the Department of Agriculture to provide advice on the

administration and implementation of US trade policy To account for heterogeneous political

processes across products we model the product-varying effects hierarchically Specifically we

allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg

Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use

ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log

trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade

volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally

intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total

exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2

level

12

Issue Year-Importer-Exporter-HS(Product description)

WTO IDB report UNCTAD TRAINS report(asymp AVE)

Solution N obs()

2013-China-India-09041200(Crushed or ground Piper pepper)

10 3 noneMissingReport 1991-Japan-Korea-140490499

(Cod fish)none 10 3

Usenon-missing

235 billion(418)

1997-Australia-Singapore-22082010(Grape wine)

3 3 + $3112L(asymp 127) 3

2005-Canada-Australia-22084010(Rum)

2456centlitre of alcohol 2456centlitre of alcohol(asymp 143) 3

Use ad valoremequivalent (AVE)computed byUNCTAD

2004-Argentina-Paraguay-87083110(Motor vehicle brakes)

0 3 14

ConflictingReports

1996-USA-Mexico-87033100(Cars of le 1500 cc cylinder capacity)

25 0 3

Use lower(preferential)rate

024 billion(425)

Table 1 Solutions to Tariff Data Issues This table illustrates examples of specific issues that arise whenattempting to find the correct applied rate for a tariff-line using the IDB and TRAINS databases In each exampleour algorithm selects the source believed to be the more precise applied rate For instance for Australiarsquos 1997 tariffon Grape wine from Singapore IDB reports only a 3 ad valorem rate while TRAINS accounts for an additional$3112 per litre of wine in its ad valorem equivalent (AVE) rate We provide full details of the merging algorithm inAppendix A2

to its partners For example IDB reports that in 1998 the United States applied a 38 tariff rate

on Ginger saffron turmeric (curcuma) thyme bay leaves curry and other spices (HTS subheading

69120090) for all partners belonging to the United States Generalized System of Preferences but

does not specify the specific countries included (eg Albania Angola and so on) We use a mix

of hand-coding from WTO and World Bank reference materials and string matching algorithms

applied to country names and regional trade agreement titles in order to map each unique ldquotyperdquo

appearing in the original data to its corresponding set of disaggregated country ISO codes Even

when the tariff ldquotyperdquo clearly applies to one country an additional step is needed to link the tex-

tual description to the relevant country code Step 2 in Figure 2 illustrates this process using an

example tariff-line Appendix A1 describes our data collection and processing in full detail

Finally there exist a number of inconsistencies between the two data sources Table 1 illustrates

two issues that we identify First we find significant differences in data coverage This is problematic

given that researchers tend to rely primarily on either one of the widely used data sources for

empirical research but usually not both Data for 127 importer-years appear only in IDB (but not

TRAINS) while data for about 842 importer-years appear only in TRAINS (but not IDB) As a

result we find that at least 235 billion observations are missing from one of the databases we

make sure to utilize the available data whenever possible Second IDB returns duties as they are

originally reported (eg 2456centlitre of alcohol) while TRAINS uses a method to estimate

8

an ad valorem equivalent (AVE) for any reported non-ad valorem rate (eg 2456centlitre of

alcohol asymp 143) TRAINS also uses this method to convert mixed or compound duties (eg 3

+ $3112L asymp 127)3 In both cases our algorithm chooses TRAINS since it is the more precise

and informative source to use Third preferential rates may be available from only one source

As shown in the last row of Table 1 TRAINS shows the correct 1996 NAFTA duty-free rate for

United States-Mexico trade in Cars of le 1500 cc cylinder capacity while IDB does not Likewise

IDB shows Argentinarsquos duty-free rate for Motor vehicle brakes imports from MERCOSUR trade

bloc partner Paraguay while TRAINS does not Our algorithm picks the correct partner-specific

preferential rate for both tariff-lines After resolving these issues of missing data and discrepancy

between the two sources we create a dataset of over 57 billion observations of bilateral trade policy

at the product level Appendix A2 details each step in our resolution algorithm In total our

dataset covers 2476 WTO importer-year tariff profiles (3080 importer-year profiles overall) from

1989 to 2015 including comprehensive policies for the top 50 trading countries beginning in 1995

(see Figure A1)

3 The Effects of Regime Type on Trade Policy

In this section we demonstrate the significance of employing our granular data Specifically we

examine differences in import tariff policy between democracies and non-democracies We begin

by analyzing unilateral trade policies (monadic analysis) across countries and products using MFN

applied tariff rates We then utilize our bilateral tariff data to investigate whether pairs of democ-

racies engage in deeper trade liberalization than other pairs do (dyadic analysis) We emphasize

that the goal of this section is to empirically demonstrate high variability in the effects of regime

type on trade policy rather than to propose a new theoretical framework3For a given non-ad valorem tariff tariff UNCTAD calculates an ad valorem equivalent (AVE) by estimating the

unit value of a product using volume statistics The type of statisticsmdasheither tariff-line level statistics from TRAINS

HS6 statistics from UN Comtrade or HS6 statistics aggregated across OECD countriesmdashdepends on data availability

for each product The unit value is then used to approximate a () tariff rate In cases where only an IDB report

is available for a compound rate we impute an AVE using only the ad valorem component of the duty rate

9

31 Monadic Analysis

Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-

Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of

factors with which their country is abundantly endowed and therefore one should expect that trade

liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged

across products they find that democratization in labor-abundant developing countries is associated

with lower trade barriers Other scholars have argued the reverse however suggesting that the need

to win elections makes democratic politicians sensitive to the demands of interest groups who offer

support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile

need appeal to only a very small segment of society to secure their power and therefore might be

less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield

2006)

We begin by examining whether trade policy varies between democracies and non-democracies

across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in

which competing economic interests across sectors determine industry-level trade policy (eg Mayer

1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of

trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting

in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show

political activities of industries such as lobbying interact with economic heterogeneity in import-

penetration and demand elasticity Consequently the canonical model of trade policy predicts

differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)

We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels

to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade

policies Our product-level analysis is also motivated by the possibility that underlying preferences

of various economic actors may interact differently with political institutions at the product-level

a prediction that is consistent with the firm-level theories in international political economy (Kim

2017) The hierarchical structure of the Harmonized System of product classification facilitates

aggregation and disaggregation at different levels of detail For example Brisket cuts a product

in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code

10

02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine

animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine

animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while

our full dataset captures trade policy for individual products at the tariff-line level we disaggregate

up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level

explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products

compared to HS2 or HS4 For example in the United States variance between HS2 groupings of

products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4

groupings of products explained 52 and between-HS6 variance explained 68 Second further

disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained

categorization that is internationally comparable as each country may set their own tariff lines using

8-digit or sometimes 10-digit codes independently4

311 Methodology

To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit

model of the observed MFN tariff rate τipt for importer i and product p in year t

τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt

τipt =

τ lowastipt if τ lowastipt ge 0

0 otherwise

(1)

where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero

otherwise We perform separate analyses aggregating products at either the HS2 industry level or

at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate

τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and

use logged values to address the high skewness of tariffs To compare our empirical findings against

existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV

score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner

and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries

although they tend to describe highly similar goods under the common HS6 category which is already highly specific

11

(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients

Zit represents a vector of country-level time-varying confounders of regime type and trade policy

log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and

an intercept5 To address missingness in covariate data we create multiple imputed datasets and

conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt

represents log import volume (at the same level of product aggregation as the tariff rate) All

covariates are lagged by 1 year

We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed

comparative advantages which captures technological differences across countries industries h (that

product p belongs to) and time t7 For example this allows us to account for the possibility

that developing and developed countries may use different production technologies even when they

produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and

εipt is idiosyncratic error assumed to be drawn from a Normal distribution

ηiiidsim N (0Ση) θt

iidsim N (0Σθ) εiptiidsim N (0 σ2

ε ) (2)

To be sure countries may have different domestic institutions that aggregate trade preferences

across various sectors meaning that trade policies of certain sectors tend to be highly correlated

For example the US Congress established the Agricultural Policy Advisory Committee (APAC)

and other advisory committees within the Department of Agriculture to provide advice on the

administration and implementation of US trade policy To account for heterogeneous political

processes across products we model the product-varying effects hierarchically Specifically we

allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg

Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use

ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log

trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade

volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally

intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total

exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2

level

12

an ad valorem equivalent (AVE) for any reported non-ad valorem rate (eg 2456centlitre of

alcohol asymp 143) TRAINS also uses this method to convert mixed or compound duties (eg 3

+ $3112L asymp 127)3 In both cases our algorithm chooses TRAINS since it is the more precise

and informative source to use Third preferential rates may be available from only one source

As shown in the last row of Table 1 TRAINS shows the correct 1996 NAFTA duty-free rate for

United States-Mexico trade in Cars of le 1500 cc cylinder capacity while IDB does not Likewise

IDB shows Argentinarsquos duty-free rate for Motor vehicle brakes imports from MERCOSUR trade

bloc partner Paraguay while TRAINS does not Our algorithm picks the correct partner-specific

preferential rate for both tariff-lines After resolving these issues of missing data and discrepancy

between the two sources we create a dataset of over 57 billion observations of bilateral trade policy

at the product level Appendix A2 details each step in our resolution algorithm In total our

dataset covers 2476 WTO importer-year tariff profiles (3080 importer-year profiles overall) from

1989 to 2015 including comprehensive policies for the top 50 trading countries beginning in 1995

(see Figure A1)

3 The Effects of Regime Type on Trade Policy

In this section we demonstrate the significance of employing our granular data Specifically we

examine differences in import tariff policy between democracies and non-democracies We begin

by analyzing unilateral trade policies (monadic analysis) across countries and products using MFN

applied tariff rates We then utilize our bilateral tariff data to investigate whether pairs of democ-

racies engage in deeper trade liberalization than other pairs do (dyadic analysis) We emphasize

that the goal of this section is to empirically demonstrate high variability in the effects of regime

type on trade policy rather than to propose a new theoretical framework3For a given non-ad valorem tariff tariff UNCTAD calculates an ad valorem equivalent (AVE) by estimating the

unit value of a product using volume statistics The type of statisticsmdasheither tariff-line level statistics from TRAINS

HS6 statistics from UN Comtrade or HS6 statistics aggregated across OECD countriesmdashdepends on data availability

for each product The unit value is then used to approximate a () tariff rate In cases where only an IDB report

is available for a compound rate we impute an AVE using only the ad valorem component of the duty rate

9

31 Monadic Analysis

Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-

Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of

factors with which their country is abundantly endowed and therefore one should expect that trade

liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged

across products they find that democratization in labor-abundant developing countries is associated

with lower trade barriers Other scholars have argued the reverse however suggesting that the need

to win elections makes democratic politicians sensitive to the demands of interest groups who offer

support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile

need appeal to only a very small segment of society to secure their power and therefore might be

less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield

2006)

We begin by examining whether trade policy varies between democracies and non-democracies

across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in

which competing economic interests across sectors determine industry-level trade policy (eg Mayer

1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of

trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting

in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show

political activities of industries such as lobbying interact with economic heterogeneity in import-

penetration and demand elasticity Consequently the canonical model of trade policy predicts

differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)

We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels

to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade

policies Our product-level analysis is also motivated by the possibility that underlying preferences

of various economic actors may interact differently with political institutions at the product-level

a prediction that is consistent with the firm-level theories in international political economy (Kim

2017) The hierarchical structure of the Harmonized System of product classification facilitates

aggregation and disaggregation at different levels of detail For example Brisket cuts a product

in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code

10

02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine

animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine

animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while

our full dataset captures trade policy for individual products at the tariff-line level we disaggregate

up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level

explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products

compared to HS2 or HS4 For example in the United States variance between HS2 groupings of

products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4

groupings of products explained 52 and between-HS6 variance explained 68 Second further

disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained

categorization that is internationally comparable as each country may set their own tariff lines using

8-digit or sometimes 10-digit codes independently4

311 Methodology

To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit

model of the observed MFN tariff rate τipt for importer i and product p in year t

τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt

τipt =

τ lowastipt if τ lowastipt ge 0

0 otherwise

(1)

where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero

otherwise We perform separate analyses aggregating products at either the HS2 industry level or

at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate

τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and

use logged values to address the high skewness of tariffs To compare our empirical findings against

existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV

score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner

and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries

although they tend to describe highly similar goods under the common HS6 category which is already highly specific

11

(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients

Zit represents a vector of country-level time-varying confounders of regime type and trade policy

log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and

an intercept5 To address missingness in covariate data we create multiple imputed datasets and

conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt

represents log import volume (at the same level of product aggregation as the tariff rate) All

covariates are lagged by 1 year

We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed

comparative advantages which captures technological differences across countries industries h (that

product p belongs to) and time t7 For example this allows us to account for the possibility

that developing and developed countries may use different production technologies even when they

produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and

εipt is idiosyncratic error assumed to be drawn from a Normal distribution

ηiiidsim N (0Ση) θt

iidsim N (0Σθ) εiptiidsim N (0 σ2

ε ) (2)

To be sure countries may have different domestic institutions that aggregate trade preferences

across various sectors meaning that trade policies of certain sectors tend to be highly correlated

For example the US Congress established the Agricultural Policy Advisory Committee (APAC)

and other advisory committees within the Department of Agriculture to provide advice on the

administration and implementation of US trade policy To account for heterogeneous political

processes across products we model the product-varying effects hierarchically Specifically we

allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg

Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use

ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log

trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade

volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally

intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total

exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2

level

12

31 Monadic Analysis

Do democratic political institutions facilitate unilateral trade liberalization Applying the Stolper-

Samuelson theorem Milner and Kubota (2005) argue that democratization empowers the owners of

factors with which their country is abundantly endowed and therefore one should expect that trade

liberalization will ensue reflecting the median voterrsquos preferences Using MFN tariff rates averaged

across products they find that democratization in labor-abundant developing countries is associated

with lower trade barriers Other scholars have argued the reverse however suggesting that the need

to win elections makes democratic politicians sensitive to the demands of interest groups who offer

support in exchange for trade protection (Frieden and Rogowski 1996) Autocracies meanwhile

need appeal to only a very small segment of society to secure their power and therefore might be

less susceptible to broad societal pressures (Acemoglu and Robinson 2005 Henisz and Mansfield

2006)

We begin by examining whether trade policy varies between democracies and non-democracies

across HS2 industries Our industry-level analysis is motivated by the endogenous tariff literature in

which competing economic interests across sectors determine industry-level trade policy (eg Mayer

1984) In fact the Stolper-Samuelson theorem postulates that the distributional implications of

trade liberalization will be asymmetric in capital-abundant and labor-abundant industries resulting

in trade policy heterogeneity across industries Moreover as Grossman and Helpman (1994) show

political activities of industries such as lobbying interact with economic heterogeneity in import-

penetration and demand elasticity Consequently the canonical model of trade policy predicts

differences in trade policy across industries (see Proposition 2 in Grossman and Helpman 1994)

We will then gradually disaggregate our units of analysis to 4-digit (HS4) and 6-digit (HS6) levels

to illustrate the severity of ecological fallacy when researchers employ aggregated measures of trade

policies Our product-level analysis is also motivated by the possibility that underlying preferences

of various economic actors may interact differently with political institutions at the product-level

a prediction that is consistent with the firm-level theories in international political economy (Kim

2017) The hierarchical structure of the Harmonized System of product classification facilitates

aggregation and disaggregation at different levels of detail For example Brisket cuts a product

in the Animal sector can be classified very broadly by its HS 2-digit chapter or industry code

10

02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine

animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine

animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while

our full dataset captures trade policy for individual products at the tariff-line level we disaggregate

up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level

explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products

compared to HS2 or HS4 For example in the United States variance between HS2 groupings of

products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4

groupings of products explained 52 and between-HS6 variance explained 68 Second further

disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained

categorization that is internationally comparable as each country may set their own tariff lines using

8-digit or sometimes 10-digit codes independently4

311 Methodology

To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit

model of the observed MFN tariff rate τipt for importer i and product p in year t

τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt

τipt =

τ lowastipt if τ lowastipt ge 0

0 otherwise

(1)

where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero

otherwise We perform separate analyses aggregating products at either the HS2 industry level or

at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate

τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and

use logged values to address the high skewness of tariffs To compare our empirical findings against

existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV

score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner

and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries

although they tend to describe highly similar goods under the common HS6 category which is already highly specific

11

(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients

Zit represents a vector of country-level time-varying confounders of regime type and trade policy

log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and

an intercept5 To address missingness in covariate data we create multiple imputed datasets and

conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt

represents log import volume (at the same level of product aggregation as the tariff rate) All

covariates are lagged by 1 year

We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed

comparative advantages which captures technological differences across countries industries h (that

product p belongs to) and time t7 For example this allows us to account for the possibility

that developing and developed countries may use different production technologies even when they

produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and

εipt is idiosyncratic error assumed to be drawn from a Normal distribution

ηiiidsim N (0Ση) θt

iidsim N (0Σθ) εiptiidsim N (0 σ2

ε ) (2)

To be sure countries may have different domestic institutions that aggregate trade preferences

across various sectors meaning that trade policies of certain sectors tend to be highly correlated

For example the US Congress established the Agricultural Policy Advisory Committee (APAC)

and other advisory committees within the Department of Agriculture to provide advice on the

administration and implementation of US trade policy To account for heterogeneous political

processes across products we model the product-varying effects hierarchically Specifically we

allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg

Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use

ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log

trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade

volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally

intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total

exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2

level

12

02 (Meat and edible meat offal) more specifically by its HS 4-digit heading 0201 (Meat of bovine

animals fresh or chilled) or very specifically by its HS 6-digit subheading 020120 (Meat of bovine

animals cuts with bone in (excluding carcasses and half-carcasses) fresh or chilled) Note that while

our full dataset captures trade policy for individual products at the tariff-line level we disaggregate

up to but not beyond the HS6 level for two reasons First disaggregating tariffs to the HS6 level

explains the greatest amount of the variance in countriesrsquo MFN tariff profiles between products

compared to HS2 or HS4 For example in the United States variance between HS2 groupings of

products explained 38 of the variance in 2012 MFN applied tariffs while variance between HS4

groupings of products explained 52 and between-HS6 variance explained 68 Second further

disaggregation is ill-advised for comparisons between countries because HS6 is the most fine-grained

categorization that is internationally comparable as each country may set their own tariff lines using

8-digit or sometimes 10-digit codes independently4

311 Methodology

To estimate the effects of regime type on trade policy we introduce the following hierarchical Tobit

model of the observed MFN tariff rate τipt for importer i and product p in year t

τ lowastipt = βDit + γgtp Xit + δgtZit + λ1Vipt + λ2Wih[p]t + ηi + θt + εipt

τipt =

τ lowastipt if τ lowastipt ge 0

0 otherwise

(1)

where τ lowastipt is a latent tariff which we observe if it is greater than zero and is censored at zero

otherwise We perform separate analyses aggregating products at either the HS2 industry level or

at the HS4 or HS6 product levels In each case we compute the time-varying average tariff rate

τipt at the desired aggregation level from the tariff-line dataset that we compiled in Section 2 and

use logged values to address the high skewness of tariffs To compare our empirical findings against

existing studies we use a binary measure of democracy where Dit is unity if importer irsquos Polity IV

score is 6 or above in year t and zero otherwise (eg Mansfield Milner and Rosendorff 2000 Milner

and Kubota 2005 Persson and Tabellini 2005) Xit is a set of country-level covariatesmdashdemocracy4The same 10-digit tariff-line numbers do not necessarily correspond to the exactly same product across countries

although they tend to describe highly similar goods under the common HS6 category which is already highly specific

11

(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients

Zit represents a vector of country-level time-varying confounders of regime type and trade policy

log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and

an intercept5 To address missingness in covariate data we create multiple imputed datasets and

conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt

represents log import volume (at the same level of product aggregation as the tariff rate) All

covariates are lagged by 1 year

We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed

comparative advantages which captures technological differences across countries industries h (that

product p belongs to) and time t7 For example this allows us to account for the possibility

that developing and developed countries may use different production technologies even when they

produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and

εipt is idiosyncratic error assumed to be drawn from a Normal distribution

ηiiidsim N (0Ση) θt

iidsim N (0Σθ) εiptiidsim N (0 σ2

ε ) (2)

To be sure countries may have different domestic institutions that aggregate trade preferences

across various sectors meaning that trade policies of certain sectors tend to be highly correlated

For example the US Congress established the Agricultural Policy Advisory Committee (APAC)

and other advisory committees within the Department of Agriculture to provide advice on the

administration and implementation of US trade policy To account for heterogeneous political

processes across products we model the product-varying effects hierarchically Specifically we

allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg

Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use

ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log

trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade

volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally

intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total

exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2

level

12

(Dit) log GDP per capita and an interceptmdashfor which we estimate product-specific coefficients

Zit represents a vector of country-level time-varying confounders of regime type and trade policy

log GDP per capita (PPP basis) log population an indicator for GATTWTO membership and

an intercept5 To address missingness in covariate data we create multiple imputed datasets and

conduct estimation separately across them following Honaker King and Blackwell (2011)6 Vipt

represents log import volume (at the same level of product aggregation as the tariff rate) All

covariates are lagged by 1 year

We also include the continuous Balassa index Wih[p]t in order to control for countriesrsquo revealed

comparative advantages which captures technological differences across countries industries h (that

product p belongs to) and time t7 For example this allows us to account for the possibility

that developing and developed countries may use different production technologies even when they

produce similar goods Finally ηi and θt are importer- and year-varying intercepts respectively and

εipt is idiosyncratic error assumed to be drawn from a Normal distribution

ηiiidsim N (0Ση) θt

iidsim N (0Σθ) εiptiidsim N (0 σ2

ε ) (2)

To be sure countries may have different domestic institutions that aggregate trade preferences

across various sectors meaning that trade policies of certain sectors tend to be highly correlated

For example the US Congress established the Agricultural Policy Advisory Committee (APAC)

and other advisory committees within the Department of Agriculture to provide advice on the

administration and implementation of US trade policy To account for heterogeneous political

processes across products we model the product-varying effects hierarchically Specifically we

allow the effects to vary across products p (eg vegetables vs fish) but incorporate the complex5GDP and population figures come from the World Bank Open Data website httpdataworldbankorg

Trade volume data are sourced from the United Nations Comtrade Database In the exposition that follows we use

ldquonon-democracyrdquo as a shorthand to describe importer-years with Polity IV scores of less than 66The greater granularity at the HS4 and HS6 levels may results in a greater number of missing observations in log

trade volume for importers across all years In that case we performed linear extrapolation to impute missing trade

volume for a given importerrsquos products across years instead of using Amelia which is exceptionally computationally

intensive at such scale7The Balassa index of a given industry in a given country is the ratio of the industryrsquos share of the countryrsquos total

exports to the industryrsquos share of global exports The most granular level at which this can be measured is the HS2

level

12

correlations within a broader sector k (eg food sector) that operates differently from other sectors

(eg textile sector)

γp sim N (φk[p]Σγ) (3)

φk sim N (0Σφ) (4)

where the effect γp for product p belonging to sector k is drawn from a multivariate-Normal distribu-

tion with a mean vector φk[p] and covariance matrix Σγ and φk is drawn from a multivariate-Normal

distribution with mean 0 and covariance matrix Σφ This means that the product-specific coeffi-

cients vary based on the sector k to which the product belongs which increases the plausibility of

the exchangeability assumption for the product-specific effects

Our analysis requires substantial computational resources For example in the HS2 analysis we

examine 218903 MFN rates in total including 18199 duty-free rates (0) for 127 countries over 26

years (1990 to 2015) We overcome computational challenges by estimating the parameters of the

HS2 and HS4 models using the Hamiltonian Monte Carlo (HMC) method implemented in the Stan

program (Carpenter Gelman Hoffman et al 2016)8 For each of four imputed dataset we run

four separate Markov chains Our posterior sample combines the chains from the imputed datasets

While we focus specifically on the posterior means and credible intervals of our quantity of interest

when we present our findings below we also make the entire posterior samples publicly available

Finally for faster computation in the HS 6-digit case we use Variational Bayes (VB) instead of

HMC (Jordan Ghahramani Jaakkola et al 1999) We verify convergence of our models using the

Gelman-Rubin statistic To the best of our knowledge this is the first product-level study that

examines the relationships between regime type and MFN trade policy covering both developing

and developed nations

312 Empirical Results

Our quantity of interest is the product-specific effect of democracy on trade policy The model

given in equation (1) decomposes this quantity into two parts (1) the main effect β and (2) the8HMC is an appropriate tool to deal with the complexity of our model as the high dimensionality of the parameter

space might result in inefficient mixing and severe autocorrelation if we used a Markov Chain Monte Carlo (MCMC)

method (Betancourt 2017) HMC explores the parameter space efficiently making it possible to estimate parameter

values with accuracy within a reasonable length of time

13

01

02

03

04

05

06

07

08

09

10

11

12

13

14

15

16

17

18

1920

21

22

23

24

25

26

27

2829

30

31

32

3334

3536

373839

4041

42

43

44

45

46

47

48

49

50

5152

53

5455

5657

585960

6162

6364

65

66

67

68

6970

71

72

7374

75

76

78

79

80

81

82

8384

85

86

8788

89

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

minus2

minus1

0

1

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

All WTO

Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate

product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9

Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of

the posterior distribution of the main effect of democracy β (marked by the dotted horizontal

line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN

tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota

(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing

nations is associated with trade liberalization Notably this finding holds even though we include

a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy

variable Dit by γDEMp

14

existing studies do Thus the conclusion that on average democracies impose lower tariffs than

non-democracies appears to be robust

It is important to note however that the HS2 results reveal significant heterogeneity in the

effects of democracy across industries Visual inspection of Figure 3 showcases several industries

for which the effect of democracy differs notably from the posterior mean of the main effect β

(the dotted line) Animal Vegetable and Food products have positive estimates for the

majority of industries in these sectors the effects turn out to be statistically significant (marked

by black vertical lines) That is democracies are more protective of agricultural consumer goods

than non-democracies are This finding is consistent with agricultural protection in democracies

that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998

Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that

democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and

Metals industries have estimates that are significantly lower than the main effect suggesting that

democracies engage in deeper liberalization of these industries For comparability with Milner and

Kubota (2005) we conduct further analysis with only less-developed countries and we find similar

results for the average effect of democracy across all products see Figure C1 in Appendix C

Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-

anced product categories for example there are only two HS2 chapters for MachElec (ma-

chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that

when considering HS4 product groups as opposed to HS2 groups the estimated main effect of

democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially

increases We find stronger effects of democratic political institutions on agricultural protection at

the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or

chilled (0707) - stronger in fact than any other products We also find much stronger evidence

that democracies are more protective of textile products On the other hand intermediate goods

tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates

for the products that belong to Minerals Metals and MachElec This is consistent with

Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average

tariff rate of 1003

15

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

0102

03

04

05

06

07

08

09

10

11

12

13

14

15

1617

18

19

20

21

22

23

24

25 2627

28

29

30

31

32

33

34

353637

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

5455

56

57

5859

60

61 62

63

64

65

6667

6869

70

71

72

73

74

75

7678

79

80

81

8283

84

85

86

87

88

89

minus4

minus2

0

2

4

Harmonized System 4minusdigit

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box

intra-industry trade make it relatively easier for countries to liberalize intermediate goods than

finished products Our findings add nuance to this claim by showing that such effects are even

more pronounced among democracies holding economic size comparative advantages institutional

memberships and trade volumes constant

Finally we present our estimates from the HS 6-digit level across three distinct set of granular

products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-

tial variations in estimated effects even across similar products within the narrowly defined meat

industry We find that consumer products such as beef pork and lamb tend to be more protected in

democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that

there are industries such as cotton that exhibit relatively lower variation across products although

we see some differences between generic vs specialized cotton products

We conduct a more systematic analysis of the difference between consumption goods and in-

16

0201

1002

0120

0201

3002

0210

0202

2002

0230

0203

11 0203

1202

0319

0203

2102

0322

0203

2902

0410

0204

2102

0422

0204

2302

0430

0204

41 0204

4202

0443

0204

5002

0500

0206

1002

0621 02

0622

0206

2902

0630

0206

4102

0649

0206

80 0206

9002

0711

0207

1202

0713

0207

1402

0721

0207

2202

0723

0207

2402

0725

0207

2602

0727

0207

3202

0733

0207

3402

0735

0207

3602

0739

0207

4102

0742

0207

4302

0750

0208

1002

0820

0208

3002

0840

0208

5002

0890

0209

00 0210

1102

1012

0210

1902

1020

0210

9002

1091

0210

9302

1099

beef

frozen beef

pork

sheep

horse

edible animal organs

poultry

fatty livers

fatty livers

other meat

seasoned meat

minus4

minus2

0

2

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(a) HS2 chapter 02 Meat products

5201

0052

0210

5202

9152

0299

5203

0052

0411

5204

1952

0420

5205

1152

0512

5205

1352

0514

5205

1552

0521

5205

2252

0523

5205

2452

0525

5205

2652

0527

5205

2852

0531

5205

3252

0533

5205

3452

0535

5205

4152

0542

5205

4352

0544

5205

4552

0546

5205

4752

0548

5206

1152

0612

5206

1352

0614

5206

1552

0621 52

0622

5206

2352

0624

5206

2552

0631

5206

32 5206

3352

0634

5206

3552

0641

5206

42 5206

4352

0644

5206

4552

0710

5207

9052

0811

5208

1252

0813

5208

1952

0821

5208

2252

0823

5208

2952

0831

5208

3252

0833

5208

3952

0841

5208

4252

0843

5208

4952

0851

5208

5252

0853 52

0859

5209

1152

0912

5209

1952

0921

5209

2252

0929

5209

3152

0932

5209

3952

0941

5209

4252

0943

5209

4952

0951

5209

5252

0959

5210

1152

1012

5210

1952

1021

5210

2252

1029

5210

3152

1032

5210

3952

1041

5210

42 5210

4952

1051

5210

52 5210

5952

1111

5211

1252

1119

5211

2152

1122

5211

2952

1131

5211

3252

1139

5211

4152

1142

5211

4352

1149

5211

5152

1152

5211

5952

1211

5212

1252

1213

5212

1452

1215

5212

2152

1222

5212

2352

1224

5212

25

generic cotton

specialized cotton

minus2

minus1

0

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(b) HS2 chapter 52 Cotton products

Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products

7301

1073

0120

7302

1073

0220

7302

3073

0240

7302

9073

0300

7304

1073

0420

7304

2173

0429

7304

31 7304

3973

0441 73

0449 73

0451

7304

5973

0490

7305

1173

0512

7305

1973

0520

7305

3173

0539 73

0590

7306

1073

0620

7306

3073

0640

7306

5073

0660

7306

9073

0711

7307

1973

0721

7307

2273

0723

7307

2973

0791

7307

9273

0793

7307

9973

0810

7308

20 7308

30 7308

4073

0890

7309

0073

1010

7310

2173

1029

7311

0073

1210

7312

9073

1300

7314

1173

1412

7314

1373

1414

7314

1973

1420

7314

3073

1431

7314

3973

1441

7314

4273

1449

7314

5073

1511 73

1512

7315

1973

1520

7315

8173

1582

7315

8973

1590

7316

0073

1700

7318

1173

1812

7318

1373

1814

7318

1573

1816

7318

1973

1821

7318

2273

1823

7318

2473

1829

7319

1073

1920

7319

3073

1990

7320

1073

2020

7320

9073

2111

7321

1273

2113

7321

8173

2182

7321

83 7321

9073

2211

7322

1973

2290

7323

1073

2391

7323

92 7323

9373

2394

7323

9973

2410 73

2421

7324

2973

2490

7325

1073

2591

7325

9973

2611

7326

1973

2620

7326

90

sheetpiling

railwayrails

switch blades

tubespipestubepipefittings

buildingstructures

tankscasks

containers

wire

steelminuscontainingcloths

chainsanchors

nailstacks

screwsbolts

washersrivets

sewingneedles

springs

steel appliances

other

Trump 20182020 tariffs

Intermediate goods

Consumption goods

minus2

minus1

0

1

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)

17

termediate goods based on the differences across steel products To do this we create a mapping

from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based

on main end use (eg consumption vs intermediate) We find that democracies are more likely to

liberalize intermediate goods within the steel industry (shaded in blue box) However we find that

consumption goods such as steel appliances (shaded in red box) tend to receive higher protection

Interestingly we find that the set of steel products on which the Trump administration imposed

high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to

liberalize

Our findings are consistent with Betz and Pond (2019) who show that democracies are more

likely to protect consumer goods than non-democracies While it is well understood that consumers

incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-

ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer

interests do not get translated into trade policy-making in democracies especially when it comes to

consumer goods More generally our findings call into question the validity of the key assumptions

made in the literature that individual preferences (given by their factoral or sectoral interests) are

key determinants of actual trade policy-making Taken together the significant variability across

products suggest that the current empirical understanding of how political institutions interact with

underlying set of preferences of actors in economy is incomplete at best

32 Dyadic Analysis

Does the interaction of regime type between trading partners affect the depth of trade liberalization

We make three contributions to the analysis of this question First we directly analyze trade policies

between country-pairs rather than using a proxy measure such as trade volume The standard

gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade

which includes barriers to market access between the trading partners By using applied tariffs

as the dependent variable therefore our analysis returns more direct estimates of the relationship

between regime type and the choice of trade policy Second we distinguish the direction of trade

policy between importing and exporting countries A direct test of the hypothesis that pairs of

democracies are more likely to engage in liberalization requires researchers to examine the interactive

effect in two directions (1) whether a democratic importer is more likely to liberalize when its export

18

partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can

achieve freer market access when its negotiating import partner is a democracy instead of a non-

democracy Finally we investigate heterogeneity across industries The findings from the monadic

analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of

political institutions as well as by political pressures that vary across interest groups Consequently

we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific

political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine

the complexity of bilateral trade policy outcomes across industries

321 Methodology

We employ a difference-in-differences identification strategy Specifically we examine the industry-

specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral

Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after

FTAs between dyads with different regime types

Our proposed hierarchical linear model for the change in trade policy before and after an FTA

between importer i and exporter j is given by

∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt

+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt

+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt

+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)

where p indexes product at the chosen level of disaggregation The proposed model in equation (5)

distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity

IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt

and DNONDEMNONDEMijt are defined similarly

For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff

reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-

tively This accounts for the possibility of anticipation effects as well as phase-in periods that are

prevalent in trade agreements To minimize excessive extrapolation into the future we focus on

the comparison of tariff rates immediately before and after each trade agreement by setting L = 1

19

and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t

ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year

tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log

population and log GDP in year t Zijt represents dyad-level covariates including logged total trade

volume between the two countries log of the partner-specific mean tariff imposed by the importer

across all industries whether at least one of the pair is a major power whether both parties were

GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-

thermore to account for the fact that democracies might have lower underlying tariff rates to begin

with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each

product p Finally ξp is an product-specific intercept As in the monadic analysis we model the

prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp

]to be Normally distributed

γp sim N (φk[p]Σγ)

φk sim N (0Σφ)(6)

The quantities of interest are the differences in the degree of trade liberalization between demo-

cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is

a democracy and the other is not)

E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM

ijt ] = βDEMNONDEM + γDEMNONDEMp (7)

E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM

ijt ] = βNONDEMDEM + γNONDEMDEMp (8)

where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is

a democracy and the partner is not and equation (8) compares a dyad with two democracies to a

mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-

tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently

fit the HS4-level model12

11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check

the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is

significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots

20

322 Empirical Results

We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements

Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two

parties to the agreement and in which both parties are sovereign states to have a conceptually

cleaner analysis of interactions between regime type We therefore include agreements such as

the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA

for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44

are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties

are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies

and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in

Appendix B13

Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-

stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade

negotiations are not a random sample from the population of all possible dyads and therefore

we emphasize that our estimand is not the difference in tariff reduction between the population

of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff

reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This

interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the

formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes

in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade

agreements

In order to make a direct comparison between our analysis and prior research we estimate an

undirected version of equation (5) of the form

∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt

+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt

+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)

13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes

revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We

periodically check the underlying databases for changes and will update our analysis as the data are refreshed

21

HS4HS2

minus10

minus05

00

05

larrD

eepe

r re

duct

ions

S

hallo

wer

red

uctio

nsrarr

Undirected

HS4

HS2

HS4

HS2

Democratic ImporterNonminusDemocratic Exporter

NonminusDemocratic ImporterDemocratic Exporter

Directed

Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would

which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM

ijt ] = βMIXED +

γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic

pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents

our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find

any difference in the applied tariff rates of mixed dyads and democratic dyads14

To shed light on this finding we decompose the direction of trade liberalization among FTA

partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of

the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4

levels data respectively First we examine whether democratic importers are able to engage in

deeper trade liberalization when their counterpart is a democracy rather than a non-democracy

This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former

engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given

the small number of FTAs involving non-democracy pairs in our data

22

Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff

reductions when their export partner is a non-democracy rather than a democracy the magnitude is

substantially larger when disaggregating to the HS4 level Second we consider whether democratic

exporters can achieve better market access when their import partner is a democracy or a non-

democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)

in the panel we find that tariffs reductions are smaller when non-democratic importers are paired

with democratic exporters than when democratic importers are paired with democratic exporters

Again the magnitude is larger with more granular tariff data These results suggest that the finding

in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions

than mixed dyads might be due to the fact that non-democratic importers give shallower concessions

to democratic exporters than democratic importers give to democratic exporters

Finally we explore the complex bilateral strategic incentives among FTA partners at granular

levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in

panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)

of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions

than pairs of democracies The estimated effects are significant for agriculture and metals industries

with very small overall differences across industries This pattern is even more prominent with HS4-

level results whereby most HS4 estimates are negative and statistically significant Similarly panel

(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff

reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4

industries Importantly unlike our monadic analysis we do not find significant heterogeneities when

we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively

constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This

suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level

interests rather than highly heterogeneous interest of individual producers15

It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South

Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation

Korean government formed several task forces organized around broad industry groupings within relevant agencies

such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare

23

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243 44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 6364656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 8889 90 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 26

27

28 29 3031 32 3334353637 38

39 40

41 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 7172 73 74 75 76 787980 81 82 83

84 85

86 878889

90 91 92939495 96 97

minus25

minus20

minus15

minus10

minus05

00

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(a) Democratic Importer Non-Democratic Exporter

0102

03 0405

06 07 08 09 10 11 121314 15

161718192021222324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 888990 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

05

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 0405

06 07 08

09

10 11

12

1314

15

161718

19

2021

22

23

24

2526

27

28

2930

31

32

33

3435

36

37 38

3940

41 42

43

44

4546

4748

49

50

5152

53 5455

56

57

58

59

60 6162

63

6465

6667

68

69

7071

72

7374

75

76

78

79

80

81

82

8384 85

86

87

888990 91 929394

95

96

97

minus05

00

05

10

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(b) Non-Democratic Importer Democratic Exporter

Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)

24

causal effects of democratization on bilateral trade liberalization across industries We leave for

future research this challenging task of investigating how exactly changes in political institutions

translate into trade policy outcomes

4 Concluding Remarks

In this paper we present a novel dataset with nearly 6 billion observations of product-level applied

tariff rates that countries apply to their trading partners incorporating the universe of preferential

rates and the Generalized System of Preferences To do so we combine and augment existing

datasets available from the WTO and UNCTAD and we resolve conflicting information between

the two Our dataset lays an important empirical foundation for investigating trade politics at a

much more granular level than has previously been done

To illustrate the importance of product-specific tariff measurement we examine an enduring

question in international political economy whether there are systematic differences in trade policy

between countries with different regime types Consistent with prior work we find that democra-

cies have lower tariff rates than non-democracies on average However focusing on the country-

or sector-level average elides substantial heterogeneity We document that democracies are more

protective than non-democracies for many industries such as agriculture and textile industries

Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-

ization We find significant variations across products even within the same industry Specifically

we show that consumer products are more likely to be protected in democracies whereas intermedi-

ate goods receive much deeper liberalization compared to the average level across other products

Our findings demonstrate various instances of aggregation bias with the use of increasingly granular

units of analysis

Our data also allows researchers in the fields of international and comparative political economy

to track fine-grained temporal changes in partner-specific trade policy In particular we examine

whether interactions between regime types at the dyad level results in differences in the degree of

bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences

design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a

mixed pair with a democracy and a non-democracy However by disentangling who is the recipient

of concessions we show that the difference between democratic pairs and mixed pairs is due in

25

large part to shallower concessions granted by non-democratic importers to democratic partners

but not vice-versa Put another way democratic importers grant trade concessions to democratic

exporters and non-democratic exporters but democratic exporters win more trade concessions

from democratic importers than from non-democratic importers Future studies would benefit from

investigating these empirical findings with a theoretical focus on the direction of trade liberalization

Our dataset can be combined with industry-level covariates such as import and export concen-

tration as well as country-specific industry structures to further explore linkages between political

institutions and industry-level trade policies It can also be used to study the increasing complex-

ity of product-level trade policy that affects the deepening of global supply chain and production

networks In addition as other scholars have pointed out there exists significant variation in insti-

tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and

Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to

assemble are likely to result in different configurations of demands for trade protection Research

into the relationships between political institutions and trade policies continues to be relevant as

policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-

ing trade agreements in response to pressures from their constituents The question is not so much

whether there will be more or less liberalization but rather which products and industries will be

most exposed to a review of trade policies This article presents findings that usefully contribute

to this research agenda

26

ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press

Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank

Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340

Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289

Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323

Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt

Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600

mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http

peopletamuedu~apondresearchhtml

Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press

Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37

Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309

Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress

27

01

02

03

04

05

06

07

08

09

10

11

12

13

14

15

16

17

18

1920

21

22

23

24

25

26

27

2829

30

31

32

3334

3536

373839

4041

42

43

44

45

46

47

48

49

50

5152

53

5455

5657

585960

6162

6364

65

66

67

68

6970

71

72

7374

75

76

78

79

80

81

82

8384

85

86

8788

89

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

minus2

minus1

0

1

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

All WTO

Figure 3 Effect of Democracy on Log Tariffs This plot presents posterior means and 95 credibleintervals for the estimated effects of democracy on tariff rates for each HS2 chapter Across all industries MFNtariffs are about 31(asymp exp(027)minus1) lower on average for democracies than non-democracies (the dotted horizontalline) However there exists significant heterogeneity in the effect of democracy across industries Democracies tendto have relatively higher tariffs in agricultural sectors Industries with black lines are those in which the difference inMFN tariffs between democracies and non-democracies are statistically different from zero The HS2 industry codesare given at the bottom of each estimate

product-specific partial effect of democracy γDEMp at the HS2 HS4 or HS6 level9

Figure 3 reports the posterior distribution of our quantity of interest β + γDEMp The mean of

the posterior distribution of the main effect of democracy β (marked by the dotted horizontal

line) shows that across all industries democracies impose about 31(asymp exp(027)minus 1) lower MFN

tariffs on average compared to non-democracies This finding is consistent with Milner and Kubota

(2005) and Chaudoin Milner and Pang (2015) who find that the democratization of developing

nations is associated with trade liberalization Notably this finding holds even though we include

a large number of developed countries and use a more fine-grained industry-level dataset than the9Note that γp is a vector of product-varying effects and we denote the element corresponding to the democracy

variable Dit by γDEMp

14

existing studies do Thus the conclusion that on average democracies impose lower tariffs than

non-democracies appears to be robust

It is important to note however that the HS2 results reveal significant heterogeneity in the

effects of democracy across industries Visual inspection of Figure 3 showcases several industries

for which the effect of democracy differs notably from the posterior mean of the main effect β

(the dotted line) Animal Vegetable and Food products have positive estimates for the

majority of industries in these sectors the effects turn out to be statistically significant (marked

by black vertical lines) That is democracies are more protective of agricultural consumer goods

than non-democracies are This finding is consistent with agricultural protection in democracies

that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998

Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that

democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and

Metals industries have estimates that are significantly lower than the main effect suggesting that

democracies engage in deeper liberalization of these industries For comparability with Milner and

Kubota (2005) we conduct further analysis with only less-developed countries and we find similar

results for the average effect of democracy across all products see Figure C1 in Appendix C

Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-

anced product categories for example there are only two HS2 chapters for MachElec (ma-

chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that

when considering HS4 product groups as opposed to HS2 groups the estimated main effect of

democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially

increases We find stronger effects of democratic political institutions on agricultural protection at

the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or

chilled (0707) - stronger in fact than any other products We also find much stronger evidence

that democracies are more protective of textile products On the other hand intermediate goods

tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates

for the products that belong to Minerals Metals and MachElec This is consistent with

Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average

tariff rate of 1003

15

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

0102

03

04

05

06

07

08

09

10

11

12

13

14

15

1617

18

19

20

21

22

23

24

25 2627

28

29

30

31

32

33

34

353637

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

5455

56

57

5859

60

61 62

63

64

65

6667

6869

70

71

72

73

74

75

7678

79

80

81

8283

84

85

86

87

88

89

minus4

minus2

0

2

4

Harmonized System 4minusdigit

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box

intra-industry trade make it relatively easier for countries to liberalize intermediate goods than

finished products Our findings add nuance to this claim by showing that such effects are even

more pronounced among democracies holding economic size comparative advantages institutional

memberships and trade volumes constant

Finally we present our estimates from the HS 6-digit level across three distinct set of granular

products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-

tial variations in estimated effects even across similar products within the narrowly defined meat

industry We find that consumer products such as beef pork and lamb tend to be more protected in

democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that

there are industries such as cotton that exhibit relatively lower variation across products although

we see some differences between generic vs specialized cotton products

We conduct a more systematic analysis of the difference between consumption goods and in-

16

0201

1002

0120

0201

3002

0210

0202

2002

0230

0203

11 0203

1202

0319

0203

2102

0322

0203

2902

0410

0204

2102

0422

0204

2302

0430

0204

41 0204

4202

0443

0204

5002

0500

0206

1002

0621 02

0622

0206

2902

0630

0206

4102

0649

0206

80 0206

9002

0711

0207

1202

0713

0207

1402

0721

0207

2202

0723

0207

2402

0725

0207

2602

0727

0207

3202

0733

0207

3402

0735

0207

3602

0739

0207

4102

0742

0207

4302

0750

0208

1002

0820

0208

3002

0840

0208

5002

0890

0209

00 0210

1102

1012

0210

1902

1020

0210

9002

1091

0210

9302

1099

beef

frozen beef

pork

sheep

horse

edible animal organs

poultry

fatty livers

fatty livers

other meat

seasoned meat

minus4

minus2

0

2

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(a) HS2 chapter 02 Meat products

5201

0052

0210

5202

9152

0299

5203

0052

0411

5204

1952

0420

5205

1152

0512

5205

1352

0514

5205

1552

0521

5205

2252

0523

5205

2452

0525

5205

2652

0527

5205

2852

0531

5205

3252

0533

5205

3452

0535

5205

4152

0542

5205

4352

0544

5205

4552

0546

5205

4752

0548

5206

1152

0612

5206

1352

0614

5206

1552

0621 52

0622

5206

2352

0624

5206

2552

0631

5206

32 5206

3352

0634

5206

3552

0641

5206

42 5206

4352

0644

5206

4552

0710

5207

9052

0811

5208

1252

0813

5208

1952

0821

5208

2252

0823

5208

2952

0831

5208

3252

0833

5208

3952

0841

5208

4252

0843

5208

4952

0851

5208

5252

0853 52

0859

5209

1152

0912

5209

1952

0921

5209

2252

0929

5209

3152

0932

5209

3952

0941

5209

4252

0943

5209

4952

0951

5209

5252

0959

5210

1152

1012

5210

1952

1021

5210

2252

1029

5210

3152

1032

5210

3952

1041

5210

42 5210

4952

1051

5210

52 5210

5952

1111

5211

1252

1119

5211

2152

1122

5211

2952

1131

5211

3252

1139

5211

4152

1142

5211

4352

1149

5211

5152

1152

5211

5952

1211

5212

1252

1213

5212

1452

1215

5212

2152

1222

5212

2352

1224

5212

25

generic cotton

specialized cotton

minus2

minus1

0

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(b) HS2 chapter 52 Cotton products

Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products

7301

1073

0120

7302

1073

0220

7302

3073

0240

7302

9073

0300

7304

1073

0420

7304

2173

0429

7304

31 7304

3973

0441 73

0449 73

0451

7304

5973

0490

7305

1173

0512

7305

1973

0520

7305

3173

0539 73

0590

7306

1073

0620

7306

3073

0640

7306

5073

0660

7306

9073

0711

7307

1973

0721

7307

2273

0723

7307

2973

0791

7307

9273

0793

7307

9973

0810

7308

20 7308

30 7308

4073

0890

7309

0073

1010

7310

2173

1029

7311

0073

1210

7312

9073

1300

7314

1173

1412

7314

1373

1414

7314

1973

1420

7314

3073

1431

7314

3973

1441

7314

4273

1449

7314

5073

1511 73

1512

7315

1973

1520

7315

8173

1582

7315

8973

1590

7316

0073

1700

7318

1173

1812

7318

1373

1814

7318

1573

1816

7318

1973

1821

7318

2273

1823

7318

2473

1829

7319

1073

1920

7319

3073

1990

7320

1073

2020

7320

9073

2111

7321

1273

2113

7321

8173

2182

7321

83 7321

9073

2211

7322

1973

2290

7323

1073

2391

7323

92 7323

9373

2394

7323

9973

2410 73

2421

7324

2973

2490

7325

1073

2591

7325

9973

2611

7326

1973

2620

7326

90

sheetpiling

railwayrails

switch blades

tubespipestubepipefittings

buildingstructures

tankscasks

containers

wire

steelminuscontainingcloths

chainsanchors

nailstacks

screwsbolts

washersrivets

sewingneedles

springs

steel appliances

other

Trump 20182020 tariffs

Intermediate goods

Consumption goods

minus2

minus1

0

1

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)

17

termediate goods based on the differences across steel products To do this we create a mapping

from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based

on main end use (eg consumption vs intermediate) We find that democracies are more likely to

liberalize intermediate goods within the steel industry (shaded in blue box) However we find that

consumption goods such as steel appliances (shaded in red box) tend to receive higher protection

Interestingly we find that the set of steel products on which the Trump administration imposed

high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to

liberalize

Our findings are consistent with Betz and Pond (2019) who show that democracies are more

likely to protect consumer goods than non-democracies While it is well understood that consumers

incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-

ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer

interests do not get translated into trade policy-making in democracies especially when it comes to

consumer goods More generally our findings call into question the validity of the key assumptions

made in the literature that individual preferences (given by their factoral or sectoral interests) are

key determinants of actual trade policy-making Taken together the significant variability across

products suggest that the current empirical understanding of how political institutions interact with

underlying set of preferences of actors in economy is incomplete at best

32 Dyadic Analysis

Does the interaction of regime type between trading partners affect the depth of trade liberalization

We make three contributions to the analysis of this question First we directly analyze trade policies

between country-pairs rather than using a proxy measure such as trade volume The standard

gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade

which includes barriers to market access between the trading partners By using applied tariffs

as the dependent variable therefore our analysis returns more direct estimates of the relationship

between regime type and the choice of trade policy Second we distinguish the direction of trade

policy between importing and exporting countries A direct test of the hypothesis that pairs of

democracies are more likely to engage in liberalization requires researchers to examine the interactive

effect in two directions (1) whether a democratic importer is more likely to liberalize when its export

18

partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can

achieve freer market access when its negotiating import partner is a democracy instead of a non-

democracy Finally we investigate heterogeneity across industries The findings from the monadic

analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of

political institutions as well as by political pressures that vary across interest groups Consequently

we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific

political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine

the complexity of bilateral trade policy outcomes across industries

321 Methodology

We employ a difference-in-differences identification strategy Specifically we examine the industry-

specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral

Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after

FTAs between dyads with different regime types

Our proposed hierarchical linear model for the change in trade policy before and after an FTA

between importer i and exporter j is given by

∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt

+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt

+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt

+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)

where p indexes product at the chosen level of disaggregation The proposed model in equation (5)

distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity

IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt

and DNONDEMNONDEMijt are defined similarly

For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff

reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-

tively This accounts for the possibility of anticipation effects as well as phase-in periods that are

prevalent in trade agreements To minimize excessive extrapolation into the future we focus on

the comparison of tariff rates immediately before and after each trade agreement by setting L = 1

19

and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t

ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year

tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log

population and log GDP in year t Zijt represents dyad-level covariates including logged total trade

volume between the two countries log of the partner-specific mean tariff imposed by the importer

across all industries whether at least one of the pair is a major power whether both parties were

GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-

thermore to account for the fact that democracies might have lower underlying tariff rates to begin

with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each

product p Finally ξp is an product-specific intercept As in the monadic analysis we model the

prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp

]to be Normally distributed

γp sim N (φk[p]Σγ)

φk sim N (0Σφ)(6)

The quantities of interest are the differences in the degree of trade liberalization between demo-

cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is

a democracy and the other is not)

E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM

ijt ] = βDEMNONDEM + γDEMNONDEMp (7)

E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM

ijt ] = βNONDEMDEM + γNONDEMDEMp (8)

where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is

a democracy and the partner is not and equation (8) compares a dyad with two democracies to a

mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-

tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently

fit the HS4-level model12

11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check

the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is

significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots

20

322 Empirical Results

We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements

Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two

parties to the agreement and in which both parties are sovereign states to have a conceptually

cleaner analysis of interactions between regime type We therefore include agreements such as

the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA

for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44

are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties

are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies

and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in

Appendix B13

Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-

stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade

negotiations are not a random sample from the population of all possible dyads and therefore

we emphasize that our estimand is not the difference in tariff reduction between the population

of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff

reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This

interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the

formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes

in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade

agreements

In order to make a direct comparison between our analysis and prior research we estimate an

undirected version of equation (5) of the form

∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt

+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt

+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)

13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes

revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We

periodically check the underlying databases for changes and will update our analysis as the data are refreshed

21

HS4HS2

minus10

minus05

00

05

larrD

eepe

r re

duct

ions

S

hallo

wer

red

uctio

nsrarr

Undirected

HS4

HS2

HS4

HS2

Democratic ImporterNonminusDemocratic Exporter

NonminusDemocratic ImporterDemocratic Exporter

Directed

Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would

which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM

ijt ] = βMIXED +

γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic

pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents

our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find

any difference in the applied tariff rates of mixed dyads and democratic dyads14

To shed light on this finding we decompose the direction of trade liberalization among FTA

partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of

the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4

levels data respectively First we examine whether democratic importers are able to engage in

deeper trade liberalization when their counterpart is a democracy rather than a non-democracy

This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former

engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given

the small number of FTAs involving non-democracy pairs in our data

22

Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff

reductions when their export partner is a non-democracy rather than a democracy the magnitude is

substantially larger when disaggregating to the HS4 level Second we consider whether democratic

exporters can achieve better market access when their import partner is a democracy or a non-

democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)

in the panel we find that tariffs reductions are smaller when non-democratic importers are paired

with democratic exporters than when democratic importers are paired with democratic exporters

Again the magnitude is larger with more granular tariff data These results suggest that the finding

in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions

than mixed dyads might be due to the fact that non-democratic importers give shallower concessions

to democratic exporters than democratic importers give to democratic exporters

Finally we explore the complex bilateral strategic incentives among FTA partners at granular

levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in

panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)

of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions

than pairs of democracies The estimated effects are significant for agriculture and metals industries

with very small overall differences across industries This pattern is even more prominent with HS4-

level results whereby most HS4 estimates are negative and statistically significant Similarly panel

(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff

reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4

industries Importantly unlike our monadic analysis we do not find significant heterogeneities when

we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively

constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This

suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level

interests rather than highly heterogeneous interest of individual producers15

It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South

Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation

Korean government formed several task forces organized around broad industry groupings within relevant agencies

such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare

23

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243 44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 6364656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 8889 90 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 26

27

28 29 3031 32 3334353637 38

39 40

41 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 7172 73 74 75 76 787980 81 82 83

84 85

86 878889

90 91 92939495 96 97

minus25

minus20

minus15

minus10

minus05

00

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(a) Democratic Importer Non-Democratic Exporter

0102

03 0405

06 07 08 09 10 11 121314 15

161718192021222324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 888990 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

05

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 0405

06 07 08

09

10 11

12

1314

15

161718

19

2021

22

23

24

2526

27

28

2930

31

32

33

3435

36

37 38

3940

41 42

43

44

4546

4748

49

50

5152

53 5455

56

57

58

59

60 6162

63

6465

6667

68

69

7071

72

7374

75

76

78

79

80

81

82

8384 85

86

87

888990 91 929394

95

96

97

minus05

00

05

10

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(b) Non-Democratic Importer Democratic Exporter

Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)

24

causal effects of democratization on bilateral trade liberalization across industries We leave for

future research this challenging task of investigating how exactly changes in political institutions

translate into trade policy outcomes

4 Concluding Remarks

In this paper we present a novel dataset with nearly 6 billion observations of product-level applied

tariff rates that countries apply to their trading partners incorporating the universe of preferential

rates and the Generalized System of Preferences To do so we combine and augment existing

datasets available from the WTO and UNCTAD and we resolve conflicting information between

the two Our dataset lays an important empirical foundation for investigating trade politics at a

much more granular level than has previously been done

To illustrate the importance of product-specific tariff measurement we examine an enduring

question in international political economy whether there are systematic differences in trade policy

between countries with different regime types Consistent with prior work we find that democra-

cies have lower tariff rates than non-democracies on average However focusing on the country-

or sector-level average elides substantial heterogeneity We document that democracies are more

protective than non-democracies for many industries such as agriculture and textile industries

Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-

ization We find significant variations across products even within the same industry Specifically

we show that consumer products are more likely to be protected in democracies whereas intermedi-

ate goods receive much deeper liberalization compared to the average level across other products

Our findings demonstrate various instances of aggregation bias with the use of increasingly granular

units of analysis

Our data also allows researchers in the fields of international and comparative political economy

to track fine-grained temporal changes in partner-specific trade policy In particular we examine

whether interactions between regime types at the dyad level results in differences in the degree of

bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences

design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a

mixed pair with a democracy and a non-democracy However by disentangling who is the recipient

of concessions we show that the difference between democratic pairs and mixed pairs is due in

25

large part to shallower concessions granted by non-democratic importers to democratic partners

but not vice-versa Put another way democratic importers grant trade concessions to democratic

exporters and non-democratic exporters but democratic exporters win more trade concessions

from democratic importers than from non-democratic importers Future studies would benefit from

investigating these empirical findings with a theoretical focus on the direction of trade liberalization

Our dataset can be combined with industry-level covariates such as import and export concen-

tration as well as country-specific industry structures to further explore linkages between political

institutions and industry-level trade policies It can also be used to study the increasing complex-

ity of product-level trade policy that affects the deepening of global supply chain and production

networks In addition as other scholars have pointed out there exists significant variation in insti-

tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and

Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to

assemble are likely to result in different configurations of demands for trade protection Research

into the relationships between political institutions and trade policies continues to be relevant as

policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-

ing trade agreements in response to pressures from their constituents The question is not so much

whether there will be more or less liberalization but rather which products and industries will be

most exposed to a review of trade policies This article presents findings that usefully contribute

to this research agenda

26

ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press

Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank

Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340

Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289

Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323

Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt

Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600

mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http

peopletamuedu~apondresearchhtml

Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press

Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37

Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309

Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress

27

existing studies do Thus the conclusion that on average democracies impose lower tariffs than

non-democracies appears to be robust

It is important to note however that the HS2 results reveal significant heterogeneity in the

effects of democracy across industries Visual inspection of Figure 3 showcases several industries

for which the effect of democracy differs notably from the posterior mean of the main effect β

(the dotted line) Animal Vegetable and Food products have positive estimates for the

majority of industries in these sectors the effects turn out to be statistically significant (marked

by black vertical lines) That is democracies are more protective of agricultural consumer goods

than non-democracies are This finding is consistent with agricultural protection in democracies

that has been widely documented in the literature (Runge and von Witzke 1987 Varshney 1998

Anderson and Martin 2005 Bates and Block 2011 Park and Jensen 2007)10 We also find that

democracies tend to protect TextCloth (textile) industries Conversely Minerals Wood and

Metals industries have estimates that are significantly lower than the main effect suggesting that

democracies engage in deeper liberalization of these industries For comparability with Milner and

Kubota (2005) we conduct further analysis with only less-developed countries and we find similar

results for the average effect of democracy across all products see Figure C1 in Appendix C

Next moving from HS2 to HS4 reveals further heterogeneities as we account for more nu-

anced product categories for example there are only two HS2 chapters for MachElec (ma-

chineelectrical) goods but they consist of 133 unique HS4 subheadings Figure 4 shows that

when considering HS4 product groups as opposed to HS2 groups the estimated main effect of

democracy (dotted line) becomes smaller (8) while product-specific heterogeneity substantially

increases We find stronger effects of democratic political institutions on agricultural protection at

the HS4 level The effects are strongest for Meat of bovine animals (0201) and Cucumbers fresh or

chilled (0707) - stronger in fact than any other products We also find much stronger evidence

that democracies are more protective of textile products On the other hand intermediate goods

tend to have lower tariffs in democracies than non-democracies as shown by the smaller estimates

for the products that belong to Minerals Metals and MachElec This is consistent with

Baccini Duumlr and Elsig (2018) who show that increasing importance of global value chains and10The tariff rate on agriculture products is 2256 on average while non-agriculture products have an average

tariff rate of 1003

15

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

0102

03

04

05

06

07

08

09

10

11

12

13

14

15

1617

18

19

20

21

22

23

24

25 2627

28

29

30

31

32

33

34

353637

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

5455

56

57

5859

60

61 62

63

64

65

6667

6869

70

71

72

73

74

75

7678

79

80

81

8283

84

85

86

87

88

89

minus4

minus2

0

2

4

Harmonized System 4minusdigit

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box

intra-industry trade make it relatively easier for countries to liberalize intermediate goods than

finished products Our findings add nuance to this claim by showing that such effects are even

more pronounced among democracies holding economic size comparative advantages institutional

memberships and trade volumes constant

Finally we present our estimates from the HS 6-digit level across three distinct set of granular

products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-

tial variations in estimated effects even across similar products within the narrowly defined meat

industry We find that consumer products such as beef pork and lamb tend to be more protected in

democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that

there are industries such as cotton that exhibit relatively lower variation across products although

we see some differences between generic vs specialized cotton products

We conduct a more systematic analysis of the difference between consumption goods and in-

16

0201

1002

0120

0201

3002

0210

0202

2002

0230

0203

11 0203

1202

0319

0203

2102

0322

0203

2902

0410

0204

2102

0422

0204

2302

0430

0204

41 0204

4202

0443

0204

5002

0500

0206

1002

0621 02

0622

0206

2902

0630

0206

4102

0649

0206

80 0206

9002

0711

0207

1202

0713

0207

1402

0721

0207

2202

0723

0207

2402

0725

0207

2602

0727

0207

3202

0733

0207

3402

0735

0207

3602

0739

0207

4102

0742

0207

4302

0750

0208

1002

0820

0208

3002

0840

0208

5002

0890

0209

00 0210

1102

1012

0210

1902

1020

0210

9002

1091

0210

9302

1099

beef

frozen beef

pork

sheep

horse

edible animal organs

poultry

fatty livers

fatty livers

other meat

seasoned meat

minus4

minus2

0

2

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(a) HS2 chapter 02 Meat products

5201

0052

0210

5202

9152

0299

5203

0052

0411

5204

1952

0420

5205

1152

0512

5205

1352

0514

5205

1552

0521

5205

2252

0523

5205

2452

0525

5205

2652

0527

5205

2852

0531

5205

3252

0533

5205

3452

0535

5205

4152

0542

5205

4352

0544

5205

4552

0546

5205

4752

0548

5206

1152

0612

5206

1352

0614

5206

1552

0621 52

0622

5206

2352

0624

5206

2552

0631

5206

32 5206

3352

0634

5206

3552

0641

5206

42 5206

4352

0644

5206

4552

0710

5207

9052

0811

5208

1252

0813

5208

1952

0821

5208

2252

0823

5208

2952

0831

5208

3252

0833

5208

3952

0841

5208

4252

0843

5208

4952

0851

5208

5252

0853 52

0859

5209

1152

0912

5209

1952

0921

5209

2252

0929

5209

3152

0932

5209

3952

0941

5209

4252

0943

5209

4952

0951

5209

5252

0959

5210

1152

1012

5210

1952

1021

5210

2252

1029

5210

3152

1032

5210

3952

1041

5210

42 5210

4952

1051

5210

52 5210

5952

1111

5211

1252

1119

5211

2152

1122

5211

2952

1131

5211

3252

1139

5211

4152

1142

5211

4352

1149

5211

5152

1152

5211

5952

1211

5212

1252

1213

5212

1452

1215

5212

2152

1222

5212

2352

1224

5212

25

generic cotton

specialized cotton

minus2

minus1

0

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(b) HS2 chapter 52 Cotton products

Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products

7301

1073

0120

7302

1073

0220

7302

3073

0240

7302

9073

0300

7304

1073

0420

7304

2173

0429

7304

31 7304

3973

0441 73

0449 73

0451

7304

5973

0490

7305

1173

0512

7305

1973

0520

7305

3173

0539 73

0590

7306

1073

0620

7306

3073

0640

7306

5073

0660

7306

9073

0711

7307

1973

0721

7307

2273

0723

7307

2973

0791

7307

9273

0793

7307

9973

0810

7308

20 7308

30 7308

4073

0890

7309

0073

1010

7310

2173

1029

7311

0073

1210

7312

9073

1300

7314

1173

1412

7314

1373

1414

7314

1973

1420

7314

3073

1431

7314

3973

1441

7314

4273

1449

7314

5073

1511 73

1512

7315

1973

1520

7315

8173

1582

7315

8973

1590

7316

0073

1700

7318

1173

1812

7318

1373

1814

7318

1573

1816

7318

1973

1821

7318

2273

1823

7318

2473

1829

7319

1073

1920

7319

3073

1990

7320

1073

2020

7320

9073

2111

7321

1273

2113

7321

8173

2182

7321

83 7321

9073

2211

7322

1973

2290

7323

1073

2391

7323

92 7323

9373

2394

7323

9973

2410 73

2421

7324

2973

2490

7325

1073

2591

7325

9973

2611

7326

1973

2620

7326

90

sheetpiling

railwayrails

switch blades

tubespipestubepipefittings

buildingstructures

tankscasks

containers

wire

steelminuscontainingcloths

chainsanchors

nailstacks

screwsbolts

washersrivets

sewingneedles

springs

steel appliances

other

Trump 20182020 tariffs

Intermediate goods

Consumption goods

minus2

minus1

0

1

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)

17

termediate goods based on the differences across steel products To do this we create a mapping

from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based

on main end use (eg consumption vs intermediate) We find that democracies are more likely to

liberalize intermediate goods within the steel industry (shaded in blue box) However we find that

consumption goods such as steel appliances (shaded in red box) tend to receive higher protection

Interestingly we find that the set of steel products on which the Trump administration imposed

high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to

liberalize

Our findings are consistent with Betz and Pond (2019) who show that democracies are more

likely to protect consumer goods than non-democracies While it is well understood that consumers

incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-

ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer

interests do not get translated into trade policy-making in democracies especially when it comes to

consumer goods More generally our findings call into question the validity of the key assumptions

made in the literature that individual preferences (given by their factoral or sectoral interests) are

key determinants of actual trade policy-making Taken together the significant variability across

products suggest that the current empirical understanding of how political institutions interact with

underlying set of preferences of actors in economy is incomplete at best

32 Dyadic Analysis

Does the interaction of regime type between trading partners affect the depth of trade liberalization

We make three contributions to the analysis of this question First we directly analyze trade policies

between country-pairs rather than using a proxy measure such as trade volume The standard

gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade

which includes barriers to market access between the trading partners By using applied tariffs

as the dependent variable therefore our analysis returns more direct estimates of the relationship

between regime type and the choice of trade policy Second we distinguish the direction of trade

policy between importing and exporting countries A direct test of the hypothesis that pairs of

democracies are more likely to engage in liberalization requires researchers to examine the interactive

effect in two directions (1) whether a democratic importer is more likely to liberalize when its export

18

partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can

achieve freer market access when its negotiating import partner is a democracy instead of a non-

democracy Finally we investigate heterogeneity across industries The findings from the monadic

analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of

political institutions as well as by political pressures that vary across interest groups Consequently

we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific

political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine

the complexity of bilateral trade policy outcomes across industries

321 Methodology

We employ a difference-in-differences identification strategy Specifically we examine the industry-

specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral

Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after

FTAs between dyads with different regime types

Our proposed hierarchical linear model for the change in trade policy before and after an FTA

between importer i and exporter j is given by

∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt

+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt

+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt

+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)

where p indexes product at the chosen level of disaggregation The proposed model in equation (5)

distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity

IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt

and DNONDEMNONDEMijt are defined similarly

For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff

reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-

tively This accounts for the possibility of anticipation effects as well as phase-in periods that are

prevalent in trade agreements To minimize excessive extrapolation into the future we focus on

the comparison of tariff rates immediately before and after each trade agreement by setting L = 1

19

and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t

ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year

tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log

population and log GDP in year t Zijt represents dyad-level covariates including logged total trade

volume between the two countries log of the partner-specific mean tariff imposed by the importer

across all industries whether at least one of the pair is a major power whether both parties were

GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-

thermore to account for the fact that democracies might have lower underlying tariff rates to begin

with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each

product p Finally ξp is an product-specific intercept As in the monadic analysis we model the

prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp

]to be Normally distributed

γp sim N (φk[p]Σγ)

φk sim N (0Σφ)(6)

The quantities of interest are the differences in the degree of trade liberalization between demo-

cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is

a democracy and the other is not)

E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM

ijt ] = βDEMNONDEM + γDEMNONDEMp (7)

E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM

ijt ] = βNONDEMDEM + γNONDEMDEMp (8)

where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is

a democracy and the partner is not and equation (8) compares a dyad with two democracies to a

mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-

tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently

fit the HS4-level model12

11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check

the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is

significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots

20

322 Empirical Results

We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements

Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two

parties to the agreement and in which both parties are sovereign states to have a conceptually

cleaner analysis of interactions between regime type We therefore include agreements such as

the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA

for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44

are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties

are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies

and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in

Appendix B13

Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-

stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade

negotiations are not a random sample from the population of all possible dyads and therefore

we emphasize that our estimand is not the difference in tariff reduction between the population

of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff

reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This

interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the

formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes

in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade

agreements

In order to make a direct comparison between our analysis and prior research we estimate an

undirected version of equation (5) of the form

∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt

+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt

+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)

13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes

revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We

periodically check the underlying databases for changes and will update our analysis as the data are refreshed

21

HS4HS2

minus10

minus05

00

05

larrD

eepe

r re

duct

ions

S

hallo

wer

red

uctio

nsrarr

Undirected

HS4

HS2

HS4

HS2

Democratic ImporterNonminusDemocratic Exporter

NonminusDemocratic ImporterDemocratic Exporter

Directed

Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would

which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM

ijt ] = βMIXED +

γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic

pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents

our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find

any difference in the applied tariff rates of mixed dyads and democratic dyads14

To shed light on this finding we decompose the direction of trade liberalization among FTA

partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of

the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4

levels data respectively First we examine whether democratic importers are able to engage in

deeper trade liberalization when their counterpart is a democracy rather than a non-democracy

This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former

engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given

the small number of FTAs involving non-democracy pairs in our data

22

Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff

reductions when their export partner is a non-democracy rather than a democracy the magnitude is

substantially larger when disaggregating to the HS4 level Second we consider whether democratic

exporters can achieve better market access when their import partner is a democracy or a non-

democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)

in the panel we find that tariffs reductions are smaller when non-democratic importers are paired

with democratic exporters than when democratic importers are paired with democratic exporters

Again the magnitude is larger with more granular tariff data These results suggest that the finding

in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions

than mixed dyads might be due to the fact that non-democratic importers give shallower concessions

to democratic exporters than democratic importers give to democratic exporters

Finally we explore the complex bilateral strategic incentives among FTA partners at granular

levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in

panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)

of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions

than pairs of democracies The estimated effects are significant for agriculture and metals industries

with very small overall differences across industries This pattern is even more prominent with HS4-

level results whereby most HS4 estimates are negative and statistically significant Similarly panel

(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff

reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4

industries Importantly unlike our monadic analysis we do not find significant heterogeneities when

we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively

constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This

suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level

interests rather than highly heterogeneous interest of individual producers15

It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South

Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation

Korean government formed several task forces organized around broad industry groupings within relevant agencies

such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare

23

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243 44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 6364656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 8889 90 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 26

27

28 29 3031 32 3334353637 38

39 40

41 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 7172 73 74 75 76 787980 81 82 83

84 85

86 878889

90 91 92939495 96 97

minus25

minus20

minus15

minus10

minus05

00

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(a) Democratic Importer Non-Democratic Exporter

0102

03 0405

06 07 08 09 10 11 121314 15

161718192021222324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 888990 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

05

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 0405

06 07 08

09

10 11

12

1314

15

161718

19

2021

22

23

24

2526

27

28

2930

31

32

33

3435

36

37 38

3940

41 42

43

44

4546

4748

49

50

5152

53 5455

56

57

58

59

60 6162

63

6465

6667

68

69

7071

72

7374

75

76

78

79

80

81

82

8384 85

86

87

888990 91 929394

95

96

97

minus05

00

05

10

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(b) Non-Democratic Importer Democratic Exporter

Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)

24

causal effects of democratization on bilateral trade liberalization across industries We leave for

future research this challenging task of investigating how exactly changes in political institutions

translate into trade policy outcomes

4 Concluding Remarks

In this paper we present a novel dataset with nearly 6 billion observations of product-level applied

tariff rates that countries apply to their trading partners incorporating the universe of preferential

rates and the Generalized System of Preferences To do so we combine and augment existing

datasets available from the WTO and UNCTAD and we resolve conflicting information between

the two Our dataset lays an important empirical foundation for investigating trade politics at a

much more granular level than has previously been done

To illustrate the importance of product-specific tariff measurement we examine an enduring

question in international political economy whether there are systematic differences in trade policy

between countries with different regime types Consistent with prior work we find that democra-

cies have lower tariff rates than non-democracies on average However focusing on the country-

or sector-level average elides substantial heterogeneity We document that democracies are more

protective than non-democracies for many industries such as agriculture and textile industries

Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-

ization We find significant variations across products even within the same industry Specifically

we show that consumer products are more likely to be protected in democracies whereas intermedi-

ate goods receive much deeper liberalization compared to the average level across other products

Our findings demonstrate various instances of aggregation bias with the use of increasingly granular

units of analysis

Our data also allows researchers in the fields of international and comparative political economy

to track fine-grained temporal changes in partner-specific trade policy In particular we examine

whether interactions between regime types at the dyad level results in differences in the degree of

bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences

design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a

mixed pair with a democracy and a non-democracy However by disentangling who is the recipient

of concessions we show that the difference between democratic pairs and mixed pairs is due in

25

large part to shallower concessions granted by non-democratic importers to democratic partners

but not vice-versa Put another way democratic importers grant trade concessions to democratic

exporters and non-democratic exporters but democratic exporters win more trade concessions

from democratic importers than from non-democratic importers Future studies would benefit from

investigating these empirical findings with a theoretical focus on the direction of trade liberalization

Our dataset can be combined with industry-level covariates such as import and export concen-

tration as well as country-specific industry structures to further explore linkages between political

institutions and industry-level trade policies It can also be used to study the increasing complex-

ity of product-level trade policy that affects the deepening of global supply chain and production

networks In addition as other scholars have pointed out there exists significant variation in insti-

tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and

Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to

assemble are likely to result in different configurations of demands for trade protection Research

into the relationships between political institutions and trade policies continues to be relevant as

policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-

ing trade agreements in response to pressures from their constituents The question is not so much

whether there will be more or less liberalization but rather which products and industries will be

most exposed to a review of trade policies This article presents findings that usefully contribute

to this research agenda

26

ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press

Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank

Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340

Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289

Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323

Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt

Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600

mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http

peopletamuedu~apondresearchhtml

Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press

Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37

Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309

Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress

27

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

0102

03

04

05

06

07

08

09

10

11

12

13

14

15

1617

18

19

20

21

22

23

24

25 2627

28

29

30

31

32

33

34

353637

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

5455

56

57

5859

60

61 62

63

64

65

6667

6869

70

71

72

73

74

75

7678

79

80

81

8283

84

85

86

87

88

89

minus4

minus2

0

2

4

Harmonized System 4minusdigit

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

Figure 4 Effect of Democracy on HS 4-Digit Level Log Tariffs This plot presents posteriormeans and 95 credible intervals for the estimated effects of democracy on tariff rates for each HS4 product groupAcross all industries MFN tariffs are about 8(asymp exp(minus008) minus 1) lower on average for democracies than non-democracies (the dotted horizontal line) Boxes group together products belonging to a common HS2 industry withthe chapter code given at the bottom of each box

intra-industry trade make it relatively easier for countries to liberalize intermediate goods than

finished products Our findings add nuance to this claim by showing that such effects are even

more pronounced among democracies holding economic size comparative advantages institutional

memberships and trade volumes constant

Finally we present our estimates from the HS 6-digit level across three distinct set of granular

products (1) meat (2) cotton and (3) steel Panel (a) of Figure 5 shows that there exist substan-

tial variations in estimated effects even across similar products within the narrowly defined meat

industry We find that consumer products such as beef pork and lamb tend to be more protected in

democracies whereas animal organs are likely to have lower tariffs Conversely panel (b) shows that

there are industries such as cotton that exhibit relatively lower variation across products although

we see some differences between generic vs specialized cotton products

We conduct a more systematic analysis of the difference between consumption goods and in-

16

0201

1002

0120

0201

3002

0210

0202

2002

0230

0203

11 0203

1202

0319

0203

2102

0322

0203

2902

0410

0204

2102

0422

0204

2302

0430

0204

41 0204

4202

0443

0204

5002

0500

0206

1002

0621 02

0622

0206

2902

0630

0206

4102

0649

0206

80 0206

9002

0711

0207

1202

0713

0207

1402

0721

0207

2202

0723

0207

2402

0725

0207

2602

0727

0207

3202

0733

0207

3402

0735

0207

3602

0739

0207

4102

0742

0207

4302

0750

0208

1002

0820

0208

3002

0840

0208

5002

0890

0209

00 0210

1102

1012

0210

1902

1020

0210

9002

1091

0210

9302

1099

beef

frozen beef

pork

sheep

horse

edible animal organs

poultry

fatty livers

fatty livers

other meat

seasoned meat

minus4

minus2

0

2

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(a) HS2 chapter 02 Meat products

5201

0052

0210

5202

9152

0299

5203

0052

0411

5204

1952

0420

5205

1152

0512

5205

1352

0514

5205

1552

0521

5205

2252

0523

5205

2452

0525

5205

2652

0527

5205

2852

0531

5205

3252

0533

5205

3452

0535

5205

4152

0542

5205

4352

0544

5205

4552

0546

5205

4752

0548

5206

1152

0612

5206

1352

0614

5206

1552

0621 52

0622

5206

2352

0624

5206

2552

0631

5206

32 5206

3352

0634

5206

3552

0641

5206

42 5206

4352

0644

5206

4552

0710

5207

9052

0811

5208

1252

0813

5208

1952

0821

5208

2252

0823

5208

2952

0831

5208

3252

0833

5208

3952

0841

5208

4252

0843

5208

4952

0851

5208

5252

0853 52

0859

5209

1152

0912

5209

1952

0921

5209

2252

0929

5209

3152

0932

5209

3952

0941

5209

4252

0943

5209

4952

0951

5209

5252

0959

5210

1152

1012

5210

1952

1021

5210

2252

1029

5210

3152

1032

5210

3952

1041

5210

42 5210

4952

1051

5210

52 5210

5952

1111

5211

1252

1119

5211

2152

1122

5211

2952

1131

5211

3252

1139

5211

4152

1142

5211

4352

1149

5211

5152

1152

5211

5952

1211

5212

1252

1213

5212

1452

1215

5212

2152

1222

5212

2352

1224

5212

25

generic cotton

specialized cotton

minus2

minus1

0

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(b) HS2 chapter 52 Cotton products

Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products

7301

1073

0120

7302

1073

0220

7302

3073

0240

7302

9073

0300

7304

1073

0420

7304

2173

0429

7304

31 7304

3973

0441 73

0449 73

0451

7304

5973

0490

7305

1173

0512

7305

1973

0520

7305

3173

0539 73

0590

7306

1073

0620

7306

3073

0640

7306

5073

0660

7306

9073

0711

7307

1973

0721

7307

2273

0723

7307

2973

0791

7307

9273

0793

7307

9973

0810

7308

20 7308

30 7308

4073

0890

7309

0073

1010

7310

2173

1029

7311

0073

1210

7312

9073

1300

7314

1173

1412

7314

1373

1414

7314

1973

1420

7314

3073

1431

7314

3973

1441

7314

4273

1449

7314

5073

1511 73

1512

7315

1973

1520

7315

8173

1582

7315

8973

1590

7316

0073

1700

7318

1173

1812

7318

1373

1814

7318

1573

1816

7318

1973

1821

7318

2273

1823

7318

2473

1829

7319

1073

1920

7319

3073

1990

7320

1073

2020

7320

9073

2111

7321

1273

2113

7321

8173

2182

7321

83 7321

9073

2211

7322

1973

2290

7323

1073

2391

7323

92 7323

9373

2394

7323

9973

2410 73

2421

7324

2973

2490

7325

1073

2591

7325

9973

2611

7326

1973

2620

7326

90

sheetpiling

railwayrails

switch blades

tubespipestubepipefittings

buildingstructures

tankscasks

containers

wire

steelminuscontainingcloths

chainsanchors

nailstacks

screwsbolts

washersrivets

sewingneedles

springs

steel appliances

other

Trump 20182020 tariffs

Intermediate goods

Consumption goods

minus2

minus1

0

1

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)

17

termediate goods based on the differences across steel products To do this we create a mapping

from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based

on main end use (eg consumption vs intermediate) We find that democracies are more likely to

liberalize intermediate goods within the steel industry (shaded in blue box) However we find that

consumption goods such as steel appliances (shaded in red box) tend to receive higher protection

Interestingly we find that the set of steel products on which the Trump administration imposed

high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to

liberalize

Our findings are consistent with Betz and Pond (2019) who show that democracies are more

likely to protect consumer goods than non-democracies While it is well understood that consumers

incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-

ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer

interests do not get translated into trade policy-making in democracies especially when it comes to

consumer goods More generally our findings call into question the validity of the key assumptions

made in the literature that individual preferences (given by their factoral or sectoral interests) are

key determinants of actual trade policy-making Taken together the significant variability across

products suggest that the current empirical understanding of how political institutions interact with

underlying set of preferences of actors in economy is incomplete at best

32 Dyadic Analysis

Does the interaction of regime type between trading partners affect the depth of trade liberalization

We make three contributions to the analysis of this question First we directly analyze trade policies

between country-pairs rather than using a proxy measure such as trade volume The standard

gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade

which includes barriers to market access between the trading partners By using applied tariffs

as the dependent variable therefore our analysis returns more direct estimates of the relationship

between regime type and the choice of trade policy Second we distinguish the direction of trade

policy between importing and exporting countries A direct test of the hypothesis that pairs of

democracies are more likely to engage in liberalization requires researchers to examine the interactive

effect in two directions (1) whether a democratic importer is more likely to liberalize when its export

18

partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can

achieve freer market access when its negotiating import partner is a democracy instead of a non-

democracy Finally we investigate heterogeneity across industries The findings from the monadic

analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of

political institutions as well as by political pressures that vary across interest groups Consequently

we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific

political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine

the complexity of bilateral trade policy outcomes across industries

321 Methodology

We employ a difference-in-differences identification strategy Specifically we examine the industry-

specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral

Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after

FTAs between dyads with different regime types

Our proposed hierarchical linear model for the change in trade policy before and after an FTA

between importer i and exporter j is given by

∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt

+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt

+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt

+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)

where p indexes product at the chosen level of disaggregation The proposed model in equation (5)

distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity

IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt

and DNONDEMNONDEMijt are defined similarly

For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff

reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-

tively This accounts for the possibility of anticipation effects as well as phase-in periods that are

prevalent in trade agreements To minimize excessive extrapolation into the future we focus on

the comparison of tariff rates immediately before and after each trade agreement by setting L = 1

19

and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t

ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year

tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log

population and log GDP in year t Zijt represents dyad-level covariates including logged total trade

volume between the two countries log of the partner-specific mean tariff imposed by the importer

across all industries whether at least one of the pair is a major power whether both parties were

GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-

thermore to account for the fact that democracies might have lower underlying tariff rates to begin

with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each

product p Finally ξp is an product-specific intercept As in the monadic analysis we model the

prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp

]to be Normally distributed

γp sim N (φk[p]Σγ)

φk sim N (0Σφ)(6)

The quantities of interest are the differences in the degree of trade liberalization between demo-

cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is

a democracy and the other is not)

E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM

ijt ] = βDEMNONDEM + γDEMNONDEMp (7)

E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM

ijt ] = βNONDEMDEM + γNONDEMDEMp (8)

where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is

a democracy and the partner is not and equation (8) compares a dyad with two democracies to a

mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-

tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently

fit the HS4-level model12

11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check

the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is

significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots

20

322 Empirical Results

We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements

Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two

parties to the agreement and in which both parties are sovereign states to have a conceptually

cleaner analysis of interactions between regime type We therefore include agreements such as

the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA

for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44

are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties

are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies

and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in

Appendix B13

Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-

stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade

negotiations are not a random sample from the population of all possible dyads and therefore

we emphasize that our estimand is not the difference in tariff reduction between the population

of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff

reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This

interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the

formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes

in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade

agreements

In order to make a direct comparison between our analysis and prior research we estimate an

undirected version of equation (5) of the form

∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt

+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt

+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)

13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes

revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We

periodically check the underlying databases for changes and will update our analysis as the data are refreshed

21

HS4HS2

minus10

minus05

00

05

larrD

eepe

r re

duct

ions

S

hallo

wer

red

uctio

nsrarr

Undirected

HS4

HS2

HS4

HS2

Democratic ImporterNonminusDemocratic Exporter

NonminusDemocratic ImporterDemocratic Exporter

Directed

Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would

which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM

ijt ] = βMIXED +

γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic

pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents

our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find

any difference in the applied tariff rates of mixed dyads and democratic dyads14

To shed light on this finding we decompose the direction of trade liberalization among FTA

partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of

the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4

levels data respectively First we examine whether democratic importers are able to engage in

deeper trade liberalization when their counterpart is a democracy rather than a non-democracy

This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former

engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given

the small number of FTAs involving non-democracy pairs in our data

22

Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff

reductions when their export partner is a non-democracy rather than a democracy the magnitude is

substantially larger when disaggregating to the HS4 level Second we consider whether democratic

exporters can achieve better market access when their import partner is a democracy or a non-

democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)

in the panel we find that tariffs reductions are smaller when non-democratic importers are paired

with democratic exporters than when democratic importers are paired with democratic exporters

Again the magnitude is larger with more granular tariff data These results suggest that the finding

in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions

than mixed dyads might be due to the fact that non-democratic importers give shallower concessions

to democratic exporters than democratic importers give to democratic exporters

Finally we explore the complex bilateral strategic incentives among FTA partners at granular

levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in

panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)

of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions

than pairs of democracies The estimated effects are significant for agriculture and metals industries

with very small overall differences across industries This pattern is even more prominent with HS4-

level results whereby most HS4 estimates are negative and statistically significant Similarly panel

(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff

reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4

industries Importantly unlike our monadic analysis we do not find significant heterogeneities when

we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively

constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This

suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level

interests rather than highly heterogeneous interest of individual producers15

It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South

Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation

Korean government formed several task forces organized around broad industry groupings within relevant agencies

such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare

23

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243 44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 6364656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 8889 90 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 26

27

28 29 3031 32 3334353637 38

39 40

41 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 7172 73 74 75 76 787980 81 82 83

84 85

86 878889

90 91 92939495 96 97

minus25

minus20

minus15

minus10

minus05

00

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(a) Democratic Importer Non-Democratic Exporter

0102

03 0405

06 07 08 09 10 11 121314 15

161718192021222324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 888990 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

05

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 0405

06 07 08

09

10 11

12

1314

15

161718

19

2021

22

23

24

2526

27

28

2930

31

32

33

3435

36

37 38

3940

41 42

43

44

4546

4748

49

50

5152

53 5455

56

57

58

59

60 6162

63

6465

6667

68

69

7071

72

7374

75

76

78

79

80

81

82

8384 85

86

87

888990 91 929394

95

96

97

minus05

00

05

10

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(b) Non-Democratic Importer Democratic Exporter

Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)

24

causal effects of democratization on bilateral trade liberalization across industries We leave for

future research this challenging task of investigating how exactly changes in political institutions

translate into trade policy outcomes

4 Concluding Remarks

In this paper we present a novel dataset with nearly 6 billion observations of product-level applied

tariff rates that countries apply to their trading partners incorporating the universe of preferential

rates and the Generalized System of Preferences To do so we combine and augment existing

datasets available from the WTO and UNCTAD and we resolve conflicting information between

the two Our dataset lays an important empirical foundation for investigating trade politics at a

much more granular level than has previously been done

To illustrate the importance of product-specific tariff measurement we examine an enduring

question in international political economy whether there are systematic differences in trade policy

between countries with different regime types Consistent with prior work we find that democra-

cies have lower tariff rates than non-democracies on average However focusing on the country-

or sector-level average elides substantial heterogeneity We document that democracies are more

protective than non-democracies for many industries such as agriculture and textile industries

Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-

ization We find significant variations across products even within the same industry Specifically

we show that consumer products are more likely to be protected in democracies whereas intermedi-

ate goods receive much deeper liberalization compared to the average level across other products

Our findings demonstrate various instances of aggregation bias with the use of increasingly granular

units of analysis

Our data also allows researchers in the fields of international and comparative political economy

to track fine-grained temporal changes in partner-specific trade policy In particular we examine

whether interactions between regime types at the dyad level results in differences in the degree of

bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences

design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a

mixed pair with a democracy and a non-democracy However by disentangling who is the recipient

of concessions we show that the difference between democratic pairs and mixed pairs is due in

25

large part to shallower concessions granted by non-democratic importers to democratic partners

but not vice-versa Put another way democratic importers grant trade concessions to democratic

exporters and non-democratic exporters but democratic exporters win more trade concessions

from democratic importers than from non-democratic importers Future studies would benefit from

investigating these empirical findings with a theoretical focus on the direction of trade liberalization

Our dataset can be combined with industry-level covariates such as import and export concen-

tration as well as country-specific industry structures to further explore linkages between political

institutions and industry-level trade policies It can also be used to study the increasing complex-

ity of product-level trade policy that affects the deepening of global supply chain and production

networks In addition as other scholars have pointed out there exists significant variation in insti-

tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and

Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to

assemble are likely to result in different configurations of demands for trade protection Research

into the relationships between political institutions and trade policies continues to be relevant as

policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-

ing trade agreements in response to pressures from their constituents The question is not so much

whether there will be more or less liberalization but rather which products and industries will be

most exposed to a review of trade policies This article presents findings that usefully contribute

to this research agenda

26

ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press

Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank

Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340

Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289

Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323

Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt

Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600

mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http

peopletamuedu~apondresearchhtml

Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press

Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37

Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309

Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress

27

0201

1002

0120

0201

3002

0210

0202

2002

0230

0203

11 0203

1202

0319

0203

2102

0322

0203

2902

0410

0204

2102

0422

0204

2302

0430

0204

41 0204

4202

0443

0204

5002

0500

0206

1002

0621 02

0622

0206

2902

0630

0206

4102

0649

0206

80 0206

9002

0711

0207

1202

0713

0207

1402

0721

0207

2202

0723

0207

2402

0725

0207

2602

0727

0207

3202

0733

0207

3402

0735

0207

3602

0739

0207

4102

0742

0207

4302

0750

0208

1002

0820

0208

3002

0840

0208

5002

0890

0209

00 0210

1102

1012

0210

1902

1020

0210

9002

1091

0210

9302

1099

beef

frozen beef

pork

sheep

horse

edible animal organs

poultry

fatty livers

fatty livers

other meat

seasoned meat

minus4

minus2

0

2

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(a) HS2 chapter 02 Meat products

5201

0052

0210

5202

9152

0299

5203

0052

0411

5204

1952

0420

5205

1152

0512

5205

1352

0514

5205

1552

0521

5205

2252

0523

5205

2452

0525

5205

2652

0527

5205

2852

0531

5205

3252

0533

5205

3452

0535

5205

4152

0542

5205

4352

0544

5205

4552

0546

5205

4752

0548

5206

1152

0612

5206

1352

0614

5206

1552

0621 52

0622

5206

2352

0624

5206

2552

0631

5206

32 5206

3352

0634

5206

3552

0641

5206

42 5206

4352

0644

5206

4552

0710

5207

9052

0811

5208

1252

0813

5208

1952

0821

5208

2252

0823

5208

2952

0831

5208

3252

0833

5208

3952

0841

5208

4252

0843

5208

4952

0851

5208

5252

0853 52

0859

5209

1152

0912

5209

1952

0921

5209

2252

0929

5209

3152

0932

5209

3952

0941

5209

4252

0943

5209

4952

0951

5209

5252

0959

5210

1152

1012

5210

1952

1021

5210

2252

1029

5210

3152

1032

5210

3952

1041

5210

42 5210

4952

1051

5210

52 5210

5952

1111

5211

1252

1119

5211

2152

1122

5211

2952

1131

5211

3252

1139

5211

4152

1142

5211

4352

1149

5211

5152

1152

5211

5952

1211

5212

1252

1213

5212

1452

1215

5212

2152

1222

5212

2352

1224

5212

25

generic cotton

specialized cotton

minus2

minus1

0

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(b) HS2 chapter 52 Cotton products

Figure 5 Effect of Democracy on HS 6-Digit Level Log Tariffs Meat and Cotton Prod-ucts This plot presents posterior means and 95 credible intervals for the estimated effects of democracy tariffrates for HS6 products in the meat and cotton industries respectively Boxes indicate distinct product categoriesMFN tariffs are about 43(asymp exp(036)minus 1) higher for democracies than non-democracies across all meat productsand 42(asymp exp(minus055)minus 1) lower democracies than non-democracies across all cotton products

7301

1073

0120

7302

1073

0220

7302

3073

0240

7302

9073

0300

7304

1073

0420

7304

2173

0429

7304

31 7304

3973

0441 73

0449 73

0451

7304

5973

0490

7305

1173

0512

7305

1973

0520

7305

3173

0539 73

0590

7306

1073

0620

7306

3073

0640

7306

5073

0660

7306

9073

0711

7307

1973

0721

7307

2273

0723

7307

2973

0791

7307

9273

0793

7307

9973

0810

7308

20 7308

30 7308

4073

0890

7309

0073

1010

7310

2173

1029

7311

0073

1210

7312

9073

1300

7314

1173

1412

7314

1373

1414

7314

1973

1420

7314

3073

1431

7314

3973

1441

7314

4273

1449

7314

5073

1511 73

1512

7315

1973

1520

7315

8173

1582

7315

8973

1590

7316

0073

1700

7318

1173

1812

7318

1373

1814

7318

1573

1816

7318

1973

1821

7318

2273

1823

7318

2473

1829

7319

1073

1920

7319

3073

1990

7320

1073

2020

7320

9073

2111

7321

1273

2113

7321

8173

2182

7321

83 7321

9073

2211

7322

1973

2290

7323

1073

2391

7323

92 7323

9373

2394

7323

9973

2410 73

2421

7324

2973

2490

7325

1073

2591

7325

9973

2611

7326

1973

2620

7326

90

sheetpiling

railwayrails

switch blades

tubespipestubepipefittings

buildingstructures

tankscasks

containers

wire

steelminuscontainingcloths

chainsanchors

nailstacks

screwsbolts

washersrivets

sewingneedles

springs

steel appliances

other

Trump 20182020 tariffs

Intermediate goods

Consumption goods

minus2

minus1

0

1

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

Figure 6 Effect of Democracy on HS 6-Digit Level Log Tariffs Steel Products This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for HS6goods classified as Steel products (HS2 chapter 73) Boxes indicate distinct product categories colored by BroadEconomy Category (BEC) Boxes outlined in black indicate steel products targeted for protection in the UnitedStates by the Trump administration in 2018 and 2020 Across all products in this industry MFN tariffs are about47(asymp exp(039)minus 1) higher on average for democracies than non-democracies (the dotted horizontal line)

17

termediate goods based on the differences across steel products To do this we create a mapping

from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based

on main end use (eg consumption vs intermediate) We find that democracies are more likely to

liberalize intermediate goods within the steel industry (shaded in blue box) However we find that

consumption goods such as steel appliances (shaded in red box) tend to receive higher protection

Interestingly we find that the set of steel products on which the Trump administration imposed

high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to

liberalize

Our findings are consistent with Betz and Pond (2019) who show that democracies are more

likely to protect consumer goods than non-democracies While it is well understood that consumers

incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-

ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer

interests do not get translated into trade policy-making in democracies especially when it comes to

consumer goods More generally our findings call into question the validity of the key assumptions

made in the literature that individual preferences (given by their factoral or sectoral interests) are

key determinants of actual trade policy-making Taken together the significant variability across

products suggest that the current empirical understanding of how political institutions interact with

underlying set of preferences of actors in economy is incomplete at best

32 Dyadic Analysis

Does the interaction of regime type between trading partners affect the depth of trade liberalization

We make three contributions to the analysis of this question First we directly analyze trade policies

between country-pairs rather than using a proxy measure such as trade volume The standard

gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade

which includes barriers to market access between the trading partners By using applied tariffs

as the dependent variable therefore our analysis returns more direct estimates of the relationship

between regime type and the choice of trade policy Second we distinguish the direction of trade

policy between importing and exporting countries A direct test of the hypothesis that pairs of

democracies are more likely to engage in liberalization requires researchers to examine the interactive

effect in two directions (1) whether a democratic importer is more likely to liberalize when its export

18

partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can

achieve freer market access when its negotiating import partner is a democracy instead of a non-

democracy Finally we investigate heterogeneity across industries The findings from the monadic

analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of

political institutions as well as by political pressures that vary across interest groups Consequently

we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific

political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine

the complexity of bilateral trade policy outcomes across industries

321 Methodology

We employ a difference-in-differences identification strategy Specifically we examine the industry-

specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral

Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after

FTAs between dyads with different regime types

Our proposed hierarchical linear model for the change in trade policy before and after an FTA

between importer i and exporter j is given by

∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt

+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt

+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt

+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)

where p indexes product at the chosen level of disaggregation The proposed model in equation (5)

distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity

IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt

and DNONDEMNONDEMijt are defined similarly

For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff

reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-

tively This accounts for the possibility of anticipation effects as well as phase-in periods that are

prevalent in trade agreements To minimize excessive extrapolation into the future we focus on

the comparison of tariff rates immediately before and after each trade agreement by setting L = 1

19

and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t

ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year

tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log

population and log GDP in year t Zijt represents dyad-level covariates including logged total trade

volume between the two countries log of the partner-specific mean tariff imposed by the importer

across all industries whether at least one of the pair is a major power whether both parties were

GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-

thermore to account for the fact that democracies might have lower underlying tariff rates to begin

with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each

product p Finally ξp is an product-specific intercept As in the monadic analysis we model the

prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp

]to be Normally distributed

γp sim N (φk[p]Σγ)

φk sim N (0Σφ)(6)

The quantities of interest are the differences in the degree of trade liberalization between demo-

cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is

a democracy and the other is not)

E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM

ijt ] = βDEMNONDEM + γDEMNONDEMp (7)

E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM

ijt ] = βNONDEMDEM + γNONDEMDEMp (8)

where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is

a democracy and the partner is not and equation (8) compares a dyad with two democracies to a

mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-

tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently

fit the HS4-level model12

11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check

the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is

significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots

20

322 Empirical Results

We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements

Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two

parties to the agreement and in which both parties are sovereign states to have a conceptually

cleaner analysis of interactions between regime type We therefore include agreements such as

the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA

for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44

are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties

are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies

and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in

Appendix B13

Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-

stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade

negotiations are not a random sample from the population of all possible dyads and therefore

we emphasize that our estimand is not the difference in tariff reduction between the population

of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff

reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This

interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the

formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes

in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade

agreements

In order to make a direct comparison between our analysis and prior research we estimate an

undirected version of equation (5) of the form

∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt

+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt

+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)

13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes

revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We

periodically check the underlying databases for changes and will update our analysis as the data are refreshed

21

HS4HS2

minus10

minus05

00

05

larrD

eepe

r re

duct

ions

S

hallo

wer

red

uctio

nsrarr

Undirected

HS4

HS2

HS4

HS2

Democratic ImporterNonminusDemocratic Exporter

NonminusDemocratic ImporterDemocratic Exporter

Directed

Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would

which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM

ijt ] = βMIXED +

γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic

pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents

our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find

any difference in the applied tariff rates of mixed dyads and democratic dyads14

To shed light on this finding we decompose the direction of trade liberalization among FTA

partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of

the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4

levels data respectively First we examine whether democratic importers are able to engage in

deeper trade liberalization when their counterpart is a democracy rather than a non-democracy

This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former

engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given

the small number of FTAs involving non-democracy pairs in our data

22

Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff

reductions when their export partner is a non-democracy rather than a democracy the magnitude is

substantially larger when disaggregating to the HS4 level Second we consider whether democratic

exporters can achieve better market access when their import partner is a democracy or a non-

democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)

in the panel we find that tariffs reductions are smaller when non-democratic importers are paired

with democratic exporters than when democratic importers are paired with democratic exporters

Again the magnitude is larger with more granular tariff data These results suggest that the finding

in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions

than mixed dyads might be due to the fact that non-democratic importers give shallower concessions

to democratic exporters than democratic importers give to democratic exporters

Finally we explore the complex bilateral strategic incentives among FTA partners at granular

levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in

panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)

of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions

than pairs of democracies The estimated effects are significant for agriculture and metals industries

with very small overall differences across industries This pattern is even more prominent with HS4-

level results whereby most HS4 estimates are negative and statistically significant Similarly panel

(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff

reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4

industries Importantly unlike our monadic analysis we do not find significant heterogeneities when

we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively

constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This

suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level

interests rather than highly heterogeneous interest of individual producers15

It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South

Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation

Korean government formed several task forces organized around broad industry groupings within relevant agencies

such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare

23

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243 44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 6364656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 8889 90 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 26

27

28 29 3031 32 3334353637 38

39 40

41 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 7172 73 74 75 76 787980 81 82 83

84 85

86 878889

90 91 92939495 96 97

minus25

minus20

minus15

minus10

minus05

00

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(a) Democratic Importer Non-Democratic Exporter

0102

03 0405

06 07 08 09 10 11 121314 15

161718192021222324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 888990 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

05

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 0405

06 07 08

09

10 11

12

1314

15

161718

19

2021

22

23

24

2526

27

28

2930

31

32

33

3435

36

37 38

3940

41 42

43

44

4546

4748

49

50

5152

53 5455

56

57

58

59

60 6162

63

6465

6667

68

69

7071

72

7374

75

76

78

79

80

81

82

8384 85

86

87

888990 91 929394

95

96

97

minus05

00

05

10

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(b) Non-Democratic Importer Democratic Exporter

Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)

24

causal effects of democratization on bilateral trade liberalization across industries We leave for

future research this challenging task of investigating how exactly changes in political institutions

translate into trade policy outcomes

4 Concluding Remarks

In this paper we present a novel dataset with nearly 6 billion observations of product-level applied

tariff rates that countries apply to their trading partners incorporating the universe of preferential

rates and the Generalized System of Preferences To do so we combine and augment existing

datasets available from the WTO and UNCTAD and we resolve conflicting information between

the two Our dataset lays an important empirical foundation for investigating trade politics at a

much more granular level than has previously been done

To illustrate the importance of product-specific tariff measurement we examine an enduring

question in international political economy whether there are systematic differences in trade policy

between countries with different regime types Consistent with prior work we find that democra-

cies have lower tariff rates than non-democracies on average However focusing on the country-

or sector-level average elides substantial heterogeneity We document that democracies are more

protective than non-democracies for many industries such as agriculture and textile industries

Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-

ization We find significant variations across products even within the same industry Specifically

we show that consumer products are more likely to be protected in democracies whereas intermedi-

ate goods receive much deeper liberalization compared to the average level across other products

Our findings demonstrate various instances of aggregation bias with the use of increasingly granular

units of analysis

Our data also allows researchers in the fields of international and comparative political economy

to track fine-grained temporal changes in partner-specific trade policy In particular we examine

whether interactions between regime types at the dyad level results in differences in the degree of

bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences

design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a

mixed pair with a democracy and a non-democracy However by disentangling who is the recipient

of concessions we show that the difference between democratic pairs and mixed pairs is due in

25

large part to shallower concessions granted by non-democratic importers to democratic partners

but not vice-versa Put another way democratic importers grant trade concessions to democratic

exporters and non-democratic exporters but democratic exporters win more trade concessions

from democratic importers than from non-democratic importers Future studies would benefit from

investigating these empirical findings with a theoretical focus on the direction of trade liberalization

Our dataset can be combined with industry-level covariates such as import and export concen-

tration as well as country-specific industry structures to further explore linkages between political

institutions and industry-level trade policies It can also be used to study the increasing complex-

ity of product-level trade policy that affects the deepening of global supply chain and production

networks In addition as other scholars have pointed out there exists significant variation in insti-

tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and

Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to

assemble are likely to result in different configurations of demands for trade protection Research

into the relationships between political institutions and trade policies continues to be relevant as

policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-

ing trade agreements in response to pressures from their constituents The question is not so much

whether there will be more or less liberalization but rather which products and industries will be

most exposed to a review of trade policies This article presents findings that usefully contribute

to this research agenda

26

ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press

Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank

Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340

Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289

Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323

Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt

Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600

mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http

peopletamuedu~apondresearchhtml

Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press

Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37

Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309

Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress

27

termediate goods based on the differences across steel products To do this we create a mapping

from each HS6 product to a Broad Economic Category (BEC) which distinguishes products based

on main end use (eg consumption vs intermediate) We find that democracies are more likely to

liberalize intermediate goods within the steel industry (shaded in blue box) However we find that

consumption goods such as steel appliances (shaded in red box) tend to receive higher protection

Interestingly we find that the set of steel products on which the Trump administration imposed

high tariffs (blue box with dark boundary) are tubes pipes and wires that democracies tend to

liberalize

Our findings are consistent with Betz and Pond (2019) who show that democracies are more

likely to protect consumer goods than non-democracies While it is well understood that consumers

incur dispersed costs of protection in contrast to concentrated benefits that import-competing pro-

ducers may enjoy our findings raise an important question for IPE scholarship as to why consumer

interests do not get translated into trade policy-making in democracies especially when it comes to

consumer goods More generally our findings call into question the validity of the key assumptions

made in the literature that individual preferences (given by their factoral or sectoral interests) are

key determinants of actual trade policy-making Taken together the significant variability across

products suggest that the current empirical understanding of how political institutions interact with

underlying set of preferences of actors in economy is incomplete at best

32 Dyadic Analysis

Does the interaction of regime type between trading partners affect the depth of trade liberalization

We make three contributions to the analysis of this question First we directly analyze trade policies

between country-pairs rather than using a proxy measure such as trade volume The standard

gravity model of trade predicts that bilateral trade volume depends directly on the costs of trade

which includes barriers to market access between the trading partners By using applied tariffs

as the dependent variable therefore our analysis returns more direct estimates of the relationship

between regime type and the choice of trade policy Second we distinguish the direction of trade

policy between importing and exporting countries A direct test of the hypothesis that pairs of

democracies are more likely to engage in liberalization requires researchers to examine the interactive

effect in two directions (1) whether a democratic importer is more likely to liberalize when its export

18

partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can

achieve freer market access when its negotiating import partner is a democracy instead of a non-

democracy Finally we investigate heterogeneity across industries The findings from the monadic

analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of

political institutions as well as by political pressures that vary across interest groups Consequently

we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific

political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine

the complexity of bilateral trade policy outcomes across industries

321 Methodology

We employ a difference-in-differences identification strategy Specifically we examine the industry-

specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral

Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after

FTAs between dyads with different regime types

Our proposed hierarchical linear model for the change in trade policy before and after an FTA

between importer i and exporter j is given by

∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt

+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt

+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt

+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)

where p indexes product at the chosen level of disaggregation The proposed model in equation (5)

distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity

IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt

and DNONDEMNONDEMijt are defined similarly

For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff

reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-

tively This accounts for the possibility of anticipation effects as well as phase-in periods that are

prevalent in trade agreements To minimize excessive extrapolation into the future we focus on

the comparison of tariff rates immediately before and after each trade agreement by setting L = 1

19

and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t

ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year

tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log

population and log GDP in year t Zijt represents dyad-level covariates including logged total trade

volume between the two countries log of the partner-specific mean tariff imposed by the importer

across all industries whether at least one of the pair is a major power whether both parties were

GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-

thermore to account for the fact that democracies might have lower underlying tariff rates to begin

with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each

product p Finally ξp is an product-specific intercept As in the monadic analysis we model the

prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp

]to be Normally distributed

γp sim N (φk[p]Σγ)

φk sim N (0Σφ)(6)

The quantities of interest are the differences in the degree of trade liberalization between demo-

cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is

a democracy and the other is not)

E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM

ijt ] = βDEMNONDEM + γDEMNONDEMp (7)

E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM

ijt ] = βNONDEMDEM + γNONDEMDEMp (8)

where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is

a democracy and the partner is not and equation (8) compares a dyad with two democracies to a

mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-

tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently

fit the HS4-level model12

11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check

the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is

significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots

20

322 Empirical Results

We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements

Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two

parties to the agreement and in which both parties are sovereign states to have a conceptually

cleaner analysis of interactions between regime type We therefore include agreements such as

the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA

for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44

are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties

are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies

and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in

Appendix B13

Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-

stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade

negotiations are not a random sample from the population of all possible dyads and therefore

we emphasize that our estimand is not the difference in tariff reduction between the population

of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff

reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This

interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the

formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes

in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade

agreements

In order to make a direct comparison between our analysis and prior research we estimate an

undirected version of equation (5) of the form

∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt

+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt

+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)

13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes

revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We

periodically check the underlying databases for changes and will update our analysis as the data are refreshed

21

HS4HS2

minus10

minus05

00

05

larrD

eepe

r re

duct

ions

S

hallo

wer

red

uctio

nsrarr

Undirected

HS4

HS2

HS4

HS2

Democratic ImporterNonminusDemocratic Exporter

NonminusDemocratic ImporterDemocratic Exporter

Directed

Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would

which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM

ijt ] = βMIXED +

γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic

pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents

our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find

any difference in the applied tariff rates of mixed dyads and democratic dyads14

To shed light on this finding we decompose the direction of trade liberalization among FTA

partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of

the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4

levels data respectively First we examine whether democratic importers are able to engage in

deeper trade liberalization when their counterpart is a democracy rather than a non-democracy

This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former

engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given

the small number of FTAs involving non-democracy pairs in our data

22

Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff

reductions when their export partner is a non-democracy rather than a democracy the magnitude is

substantially larger when disaggregating to the HS4 level Second we consider whether democratic

exporters can achieve better market access when their import partner is a democracy or a non-

democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)

in the panel we find that tariffs reductions are smaller when non-democratic importers are paired

with democratic exporters than when democratic importers are paired with democratic exporters

Again the magnitude is larger with more granular tariff data These results suggest that the finding

in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions

than mixed dyads might be due to the fact that non-democratic importers give shallower concessions

to democratic exporters than democratic importers give to democratic exporters

Finally we explore the complex bilateral strategic incentives among FTA partners at granular

levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in

panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)

of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions

than pairs of democracies The estimated effects are significant for agriculture and metals industries

with very small overall differences across industries This pattern is even more prominent with HS4-

level results whereby most HS4 estimates are negative and statistically significant Similarly panel

(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff

reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4

industries Importantly unlike our monadic analysis we do not find significant heterogeneities when

we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively

constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This

suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level

interests rather than highly heterogeneous interest of individual producers15

It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South

Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation

Korean government formed several task forces organized around broad industry groupings within relevant agencies

such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare

23

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243 44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 6364656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 8889 90 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 26

27

28 29 3031 32 3334353637 38

39 40

41 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 7172 73 74 75 76 787980 81 82 83

84 85

86 878889

90 91 92939495 96 97

minus25

minus20

minus15

minus10

minus05

00

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(a) Democratic Importer Non-Democratic Exporter

0102

03 0405

06 07 08 09 10 11 121314 15

161718192021222324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 888990 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

05

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 0405

06 07 08

09

10 11

12

1314

15

161718

19

2021

22

23

24

2526

27

28

2930

31

32

33

3435

36

37 38

3940

41 42

43

44

4546

4748

49

50

5152

53 5455

56

57

58

59

60 6162

63

6465

6667

68

69

7071

72

7374

75

76

78

79

80

81

82

8384 85

86

87

888990 91 929394

95

96

97

minus05

00

05

10

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(b) Non-Democratic Importer Democratic Exporter

Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)

24

causal effects of democratization on bilateral trade liberalization across industries We leave for

future research this challenging task of investigating how exactly changes in political institutions

translate into trade policy outcomes

4 Concluding Remarks

In this paper we present a novel dataset with nearly 6 billion observations of product-level applied

tariff rates that countries apply to their trading partners incorporating the universe of preferential

rates and the Generalized System of Preferences To do so we combine and augment existing

datasets available from the WTO and UNCTAD and we resolve conflicting information between

the two Our dataset lays an important empirical foundation for investigating trade politics at a

much more granular level than has previously been done

To illustrate the importance of product-specific tariff measurement we examine an enduring

question in international political economy whether there are systematic differences in trade policy

between countries with different regime types Consistent with prior work we find that democra-

cies have lower tariff rates than non-democracies on average However focusing on the country-

or sector-level average elides substantial heterogeneity We document that democracies are more

protective than non-democracies for many industries such as agriculture and textile industries

Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-

ization We find significant variations across products even within the same industry Specifically

we show that consumer products are more likely to be protected in democracies whereas intermedi-

ate goods receive much deeper liberalization compared to the average level across other products

Our findings demonstrate various instances of aggregation bias with the use of increasingly granular

units of analysis

Our data also allows researchers in the fields of international and comparative political economy

to track fine-grained temporal changes in partner-specific trade policy In particular we examine

whether interactions between regime types at the dyad level results in differences in the degree of

bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences

design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a

mixed pair with a democracy and a non-democracy However by disentangling who is the recipient

of concessions we show that the difference between democratic pairs and mixed pairs is due in

25

large part to shallower concessions granted by non-democratic importers to democratic partners

but not vice-versa Put another way democratic importers grant trade concessions to democratic

exporters and non-democratic exporters but democratic exporters win more trade concessions

from democratic importers than from non-democratic importers Future studies would benefit from

investigating these empirical findings with a theoretical focus on the direction of trade liberalization

Our dataset can be combined with industry-level covariates such as import and export concen-

tration as well as country-specific industry structures to further explore linkages between political

institutions and industry-level trade policies It can also be used to study the increasing complex-

ity of product-level trade policy that affects the deepening of global supply chain and production

networks In addition as other scholars have pointed out there exists significant variation in insti-

tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and

Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to

assemble are likely to result in different configurations of demands for trade protection Research

into the relationships between political institutions and trade policies continues to be relevant as

policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-

ing trade agreements in response to pressures from their constituents The question is not so much

whether there will be more or less liberalization but rather which products and industries will be

most exposed to a review of trade policies This article presents findings that usefully contribute

to this research agenda

26

ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press

Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank

Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340

Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289

Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323

Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt

Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600

mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http

peopletamuedu~apondresearchhtml

Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press

Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37

Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309

Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress

27

partner is a democracy rather than a non- democracy and (2) whether a democratic exporter can

achieve freer market access when its negotiating import partner is a democracy instead of a non-

democracy Finally we investigate heterogeneity across industries The findings from the monadic

analysis in Section 31 confirm that unilateral incentives to liberalize are affected by the structure of

political institutions as well as by political pressures that vary across interest groups Consequently

we expect that bilateral trade negotiations will also be affected by trading partnersrsquo industry-specific

political constraints The bilateral tariff data that we introduced in Section 2 enables us to examine

the complexity of bilateral trade policy outcomes across industries

321 Methodology

We employ a difference-in-differences identification strategy Specifically we examine the industry-

specific interactive effects of regime type on the degree of trade liberalization as a result of bilateral

Free Trade Agreements (FTAs) We compare the magnitudes of tariff reductions before and after

FTAs between dyads with different regime types

Our proposed hierarchical linear model for the change in trade policy before and after an FTA

between importer i and exporter j is given by

∆τijpt = α + (βDEMNONDEM + γDEMNONDEMp )DDEMNONDEMijt

+ (βNONDEMDEM + γNONDEMDEMp )DNONDEMDEMijt

+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt

+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (5)

where p indexes product at the chosen level of disaggregation The proposed model in equation (5)

distinguishes the direction of trade liberalization DDEMNONDEMijt is an indicator equal to 1 if the Polity

IV score for importer i is 6 or above and the score for exporting partner j is below 6 DNONDEMDEMijt

and DNONDEMNONDEMijt are defined similarly

For an FTA between i and j that goes into effect in year tlowast we compare the degree of tariff

reduction between tlowast minus L and tlowast + F where L and F denote the length of lags and leads respec-

tively This accounts for the possibility of anticipation effects as well as phase-in periods that are

prevalent in trade agreements To minimize excessive extrapolation into the future we focus on

the comparison of tariff rates immediately before and after each trade agreement by setting L = 1

19

and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t

ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year

tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log

population and log GDP in year t Zijt represents dyad-level covariates including logged total trade

volume between the two countries log of the partner-specific mean tariff imposed by the importer

across all industries whether at least one of the pair is a major power whether both parties were

GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-

thermore to account for the fact that democracies might have lower underlying tariff rates to begin

with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each

product p Finally ξp is an product-specific intercept As in the monadic analysis we model the

prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp

]to be Normally distributed

γp sim N (φk[p]Σγ)

φk sim N (0Σφ)(6)

The quantities of interest are the differences in the degree of trade liberalization between demo-

cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is

a democracy and the other is not)

E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM

ijt ] = βDEMNONDEM + γDEMNONDEMp (7)

E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM

ijt ] = βNONDEMDEM + γNONDEMDEMp (8)

where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is

a democracy and the partner is not and equation (8) compares a dyad with two democracies to a

mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-

tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently

fit the HS4-level model12

11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check

the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is

significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots

20

322 Empirical Results

We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements

Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two

parties to the agreement and in which both parties are sovereign states to have a conceptually

cleaner analysis of interactions between regime type We therefore include agreements such as

the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA

for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44

are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties

are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies

and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in

Appendix B13

Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-

stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade

negotiations are not a random sample from the population of all possible dyads and therefore

we emphasize that our estimand is not the difference in tariff reduction between the population

of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff

reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This

interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the

formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes

in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade

agreements

In order to make a direct comparison between our analysis and prior research we estimate an

undirected version of equation (5) of the form

∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt

+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt

+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)

13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes

revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We

periodically check the underlying databases for changes and will update our analysis as the data are refreshed

21

HS4HS2

minus10

minus05

00

05

larrD

eepe

r re

duct

ions

S

hallo

wer

red

uctio

nsrarr

Undirected

HS4

HS2

HS4

HS2

Democratic ImporterNonminusDemocratic Exporter

NonminusDemocratic ImporterDemocratic Exporter

Directed

Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would

which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM

ijt ] = βMIXED +

γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic

pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents

our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find

any difference in the applied tariff rates of mixed dyads and democratic dyads14

To shed light on this finding we decompose the direction of trade liberalization among FTA

partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of

the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4

levels data respectively First we examine whether democratic importers are able to engage in

deeper trade liberalization when their counterpart is a democracy rather than a non-democracy

This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former

engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given

the small number of FTAs involving non-democracy pairs in our data

22

Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff

reductions when their export partner is a non-democracy rather than a democracy the magnitude is

substantially larger when disaggregating to the HS4 level Second we consider whether democratic

exporters can achieve better market access when their import partner is a democracy or a non-

democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)

in the panel we find that tariffs reductions are smaller when non-democratic importers are paired

with democratic exporters than when democratic importers are paired with democratic exporters

Again the magnitude is larger with more granular tariff data These results suggest that the finding

in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions

than mixed dyads might be due to the fact that non-democratic importers give shallower concessions

to democratic exporters than democratic importers give to democratic exporters

Finally we explore the complex bilateral strategic incentives among FTA partners at granular

levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in

panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)

of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions

than pairs of democracies The estimated effects are significant for agriculture and metals industries

with very small overall differences across industries This pattern is even more prominent with HS4-

level results whereby most HS4 estimates are negative and statistically significant Similarly panel

(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff

reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4

industries Importantly unlike our monadic analysis we do not find significant heterogeneities when

we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively

constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This

suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level

interests rather than highly heterogeneous interest of individual producers15

It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South

Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation

Korean government formed several task forces organized around broad industry groupings within relevant agencies

such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare

23

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243 44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 6364656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 8889 90 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 26

27

28 29 3031 32 3334353637 38

39 40

41 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 7172 73 74 75 76 787980 81 82 83

84 85

86 878889

90 91 92939495 96 97

minus25

minus20

minus15

minus10

minus05

00

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(a) Democratic Importer Non-Democratic Exporter

0102

03 0405

06 07 08 09 10 11 121314 15

161718192021222324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 888990 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

05

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 0405

06 07 08

09

10 11

12

1314

15

161718

19

2021

22

23

24

2526

27

28

2930

31

32

33

3435

36

37 38

3940

41 42

43

44

4546

4748

49

50

5152

53 5455

56

57

58

59

60 6162

63

6465

6667

68

69

7071

72

7374

75

76

78

79

80

81

82

8384 85

86

87

888990 91 929394

95

96

97

minus05

00

05

10

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(b) Non-Democratic Importer Democratic Exporter

Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)

24

causal effects of democratization on bilateral trade liberalization across industries We leave for

future research this challenging task of investigating how exactly changes in political institutions

translate into trade policy outcomes

4 Concluding Remarks

In this paper we present a novel dataset with nearly 6 billion observations of product-level applied

tariff rates that countries apply to their trading partners incorporating the universe of preferential

rates and the Generalized System of Preferences To do so we combine and augment existing

datasets available from the WTO and UNCTAD and we resolve conflicting information between

the two Our dataset lays an important empirical foundation for investigating trade politics at a

much more granular level than has previously been done

To illustrate the importance of product-specific tariff measurement we examine an enduring

question in international political economy whether there are systematic differences in trade policy

between countries with different regime types Consistent with prior work we find that democra-

cies have lower tariff rates than non-democracies on average However focusing on the country-

or sector-level average elides substantial heterogeneity We document that democracies are more

protective than non-democracies for many industries such as agriculture and textile industries

Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-

ization We find significant variations across products even within the same industry Specifically

we show that consumer products are more likely to be protected in democracies whereas intermedi-

ate goods receive much deeper liberalization compared to the average level across other products

Our findings demonstrate various instances of aggregation bias with the use of increasingly granular

units of analysis

Our data also allows researchers in the fields of international and comparative political economy

to track fine-grained temporal changes in partner-specific trade policy In particular we examine

whether interactions between regime types at the dyad level results in differences in the degree of

bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences

design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a

mixed pair with a democracy and a non-democracy However by disentangling who is the recipient

of concessions we show that the difference between democratic pairs and mixed pairs is due in

25

large part to shallower concessions granted by non-democratic importers to democratic partners

but not vice-versa Put another way democratic importers grant trade concessions to democratic

exporters and non-democratic exporters but democratic exporters win more trade concessions

from democratic importers than from non-democratic importers Future studies would benefit from

investigating these empirical findings with a theoretical focus on the direction of trade liberalization

Our dataset can be combined with industry-level covariates such as import and export concen-

tration as well as country-specific industry structures to further explore linkages between political

institutions and industry-level trade policies It can also be used to study the increasing complex-

ity of product-level trade policy that affects the deepening of global supply chain and production

networks In addition as other scholars have pointed out there exists significant variation in insti-

tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and

Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to

assemble are likely to result in different configurations of demands for trade protection Research

into the relationships between political institutions and trade policies continues to be relevant as

policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-

ing trade agreements in response to pressures from their constituents The question is not so much

whether there will be more or less liberalization but rather which products and industries will be

most exposed to a review of trade policies This article presents findings that usefully contribute

to this research agenda

26

ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press

Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank

Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340

Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289

Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323

Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt

Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600

mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http

peopletamuedu~apondresearchhtml

Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press

Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37

Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309

Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress

27

and F = 111 To simplify the notation we denote the year prior to the FTA taking effect by t

ie t = tlowast minus L Then ∆τijpt represents a change in tariffs (logged) for product p between year

tlowast minus L and tlowast + F Zit and Zjt represent covariates for the importer and partner and include log

population and log GDP in year t Zijt represents dyad-level covariates including logged total trade

volume between the two countries log of the partner-specific mean tariff imposed by the importer

across all industries whether at least one of the pair is a major power whether both parties were

GATTWTO members as well as logged distance (in kilometers) between the two countries Fur-

thermore to account for the fact that democracies might have lower underlying tariff rates to begin

with we control for pre-existing tariff levels by including the pre-FTA MFN rates Mipt for each

product p Finally ξp is an product-specific intercept As in the monadic analysis we model the

prior distribution of the product-varying coefficient γp =[ξp γDEMNONDEMp γNONDEMDEMp γNONDEMNONDEMp

]to be Normally distributed

γp sim N (φk[p]Σγ)

φk sim N (0Σφ)(6)

The quantities of interest are the differences in the degree of trade liberalization between demo-

cratic pairs (ie dyads in which both parties are democracies) and mixed dyads (ie one party is

a democracy and the other is not)

E[∆τijpt | DDEMNONDEMijt ]minus E[∆τijpt | DDEMDEM

ijt ] = βDEMNONDEM + γDEMNONDEMp (7)

E[∆τijpt | DNONDEMDEMijt ]minus E[∆τijpt | DDEMDEM

ijt ] = βNONDEMDEM + γNONDEMDEMp (8)

where equation (7) compares a dyad with two democracies to a mixed dyad where the importer is

a democracy and the partner is not and equation (8) compares a dyad with two democracies to a

mixed dyad where the exporter is a democracy and the importing partner is not We estimate equa-

tion (5) at HS2 and HS4 levels with Stan using a variational approximation method to efficiently

fit the HS4-level model12

11To account for more extensive phase-in periods as well as anticipation effects in trade agreements we also check

the robustness of our findings by setting L = F = 3 We find that the direction of bilateral trade liberalization is

significant in this analysis as well12We perform various diagnostics to check the convergence See Appendix for the traceplots

20

322 Empirical Results

We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements

Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two

parties to the agreement and in which both parties are sovereign states to have a conceptually

cleaner analysis of interactions between regime type We therefore include agreements such as

the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA

for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44

are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties

are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies

and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in

Appendix B13

Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-

stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade

negotiations are not a random sample from the population of all possible dyads and therefore

we emphasize that our estimand is not the difference in tariff reduction between the population

of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff

reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This

interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the

formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes

in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade

agreements

In order to make a direct comparison between our analysis and prior research we estimate an

undirected version of equation (5) of the form

∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt

+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt

+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)

13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes

revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We

periodically check the underlying databases for changes and will update our analysis as the data are refreshed

21

HS4HS2

minus10

minus05

00

05

larrD

eepe

r re

duct

ions

S

hallo

wer

red

uctio

nsrarr

Undirected

HS4

HS2

HS4

HS2

Democratic ImporterNonminusDemocratic Exporter

NonminusDemocratic ImporterDemocratic Exporter

Directed

Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would

which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM

ijt ] = βMIXED +

γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic

pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents

our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find

any difference in the applied tariff rates of mixed dyads and democratic dyads14

To shed light on this finding we decompose the direction of trade liberalization among FTA

partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of

the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4

levels data respectively First we examine whether democratic importers are able to engage in

deeper trade liberalization when their counterpart is a democracy rather than a non-democracy

This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former

engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given

the small number of FTAs involving non-democracy pairs in our data

22

Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff

reductions when their export partner is a non-democracy rather than a democracy the magnitude is

substantially larger when disaggregating to the HS4 level Second we consider whether democratic

exporters can achieve better market access when their import partner is a democracy or a non-

democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)

in the panel we find that tariffs reductions are smaller when non-democratic importers are paired

with democratic exporters than when democratic importers are paired with democratic exporters

Again the magnitude is larger with more granular tariff data These results suggest that the finding

in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions

than mixed dyads might be due to the fact that non-democratic importers give shallower concessions

to democratic exporters than democratic importers give to democratic exporters

Finally we explore the complex bilateral strategic incentives among FTA partners at granular

levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in

panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)

of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions

than pairs of democracies The estimated effects are significant for agriculture and metals industries

with very small overall differences across industries This pattern is even more prominent with HS4-

level results whereby most HS4 estimates are negative and statistically significant Similarly panel

(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff

reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4

industries Importantly unlike our monadic analysis we do not find significant heterogeneities when

we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively

constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This

suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level

interests rather than highly heterogeneous interest of individual producers15

It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South

Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation

Korean government formed several task forces organized around broad industry groupings within relevant agencies

such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare

23

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243 44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 6364656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 8889 90 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 26

27

28 29 3031 32 3334353637 38

39 40

41 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 7172 73 74 75 76 787980 81 82 83

84 85

86 878889

90 91 92939495 96 97

minus25

minus20

minus15

minus10

minus05

00

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(a) Democratic Importer Non-Democratic Exporter

0102

03 0405

06 07 08 09 10 11 121314 15

161718192021222324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 888990 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

05

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 0405

06 07 08

09

10 11

12

1314

15

161718

19

2021

22

23

24

2526

27

28

2930

31

32

33

3435

36

37 38

3940

41 42

43

44

4546

4748

49

50

5152

53 5455

56

57

58

59

60 6162

63

6465

6667

68

69

7071

72

7374

75

76

78

79

80

81

82

8384 85

86

87

888990 91 929394

95

96

97

minus05

00

05

10

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(b) Non-Democratic Importer Democratic Exporter

Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)

24

causal effects of democratization on bilateral trade liberalization across industries We leave for

future research this challenging task of investigating how exactly changes in political institutions

translate into trade policy outcomes

4 Concluding Remarks

In this paper we present a novel dataset with nearly 6 billion observations of product-level applied

tariff rates that countries apply to their trading partners incorporating the universe of preferential

rates and the Generalized System of Preferences To do so we combine and augment existing

datasets available from the WTO and UNCTAD and we resolve conflicting information between

the two Our dataset lays an important empirical foundation for investigating trade politics at a

much more granular level than has previously been done

To illustrate the importance of product-specific tariff measurement we examine an enduring

question in international political economy whether there are systematic differences in trade policy

between countries with different regime types Consistent with prior work we find that democra-

cies have lower tariff rates than non-democracies on average However focusing on the country-

or sector-level average elides substantial heterogeneity We document that democracies are more

protective than non-democracies for many industries such as agriculture and textile industries

Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-

ization We find significant variations across products even within the same industry Specifically

we show that consumer products are more likely to be protected in democracies whereas intermedi-

ate goods receive much deeper liberalization compared to the average level across other products

Our findings demonstrate various instances of aggregation bias with the use of increasingly granular

units of analysis

Our data also allows researchers in the fields of international and comparative political economy

to track fine-grained temporal changes in partner-specific trade policy In particular we examine

whether interactions between regime types at the dyad level results in differences in the degree of

bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences

design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a

mixed pair with a democracy and a non-democracy However by disentangling who is the recipient

of concessions we show that the difference between democratic pairs and mixed pairs is due in

25

large part to shallower concessions granted by non-democratic importers to democratic partners

but not vice-versa Put another way democratic importers grant trade concessions to democratic

exporters and non-democratic exporters but democratic exporters win more trade concessions

from democratic importers than from non-democratic importers Future studies would benefit from

investigating these empirical findings with a theoretical focus on the direction of trade liberalization

Our dataset can be combined with industry-level covariates such as import and export concen-

tration as well as country-specific industry structures to further explore linkages between political

institutions and industry-level trade policies It can also be used to study the increasing complex-

ity of product-level trade policy that affects the deepening of global supply chain and production

networks In addition as other scholars have pointed out there exists significant variation in insti-

tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and

Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to

assemble are likely to result in different configurations of demands for trade protection Research

into the relationships between political institutions and trade policies continues to be relevant as

policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-

ing trade agreements in response to pressures from their constituents The question is not so much

whether there will be more or less liberalization but rather which products and industries will be

most exposed to a review of trade policies This article presents findings that usefully contribute

to this research agenda

26

ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press

Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank

Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340

Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289

Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323

Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt

Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600

mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http

peopletamuedu~apondresearchhtml

Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press

Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37

Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309

Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress

27

322 Empirical Results

We obtain data on preferential trade agreements from the WTOrsquos Regional Trade Agreements

Information System (RTA-IS) database We focus on bilateral FTAs in which there are only two

parties to the agreement and in which both parties are sovereign states to have a conceptually

cleaner analysis of interactions between regime type We therefore include agreements such as

the USA-Australia FTA but exclude NAFTA the EU-Canada FTA and the EFTA-SACU FTA

for example Our dataset consists of 90 unique bilateral FTAs Of these 90 bilateral FTAs 44

are signed between democratic dyads 38 are mixed dyads and 8 are dyads in which both parties

are non-democracies There are 36 unique parties to these 90 FTAs of which 26 are democracies

and 10 are non-democracies The full list of bilateral FTAs included in our analysis is given in

Appendix B13

Our emphasis on bilateral FTAs arises from our interest in understanding how democratic in-

stitutions relate to the outcomes of trade negotiations Certainly countries that enter into trade

negotiations are not a random sample from the population of all possible dyads and therefore

we emphasize that our estimand is not the difference in tariff reduction between the population

of democratic pairs and mixed pairs in general Rather we are interested in differences in tariff

reductions between dyad types among those dyads that successfully negotiate bilateral FTAs This

interest in the ldquointensive marginrdquo of negotiated outcomes is the same premise that motivates the

formal model developed by Mansfield Milner and Rosendorff (2000) who compare the changes

in trade volumes that result from democratic dyad trade agreements to those of mixed dyad trade

agreements

In order to make a direct comparison between our analysis and prior research we estimate an

undirected version of equation (5) of the form

∆τijpt = α + (βMIXED + γMIXEDp )DMIXEDijt

+ (βNONDEMNONDEM + γNONDEMNONDEMp )DNONDEMNONDEMijt

+ δgt0 Zit + δgt1 Zjt + δgt2 Zijt + λMipt + ξp + εijpt (9)

13As Table B1 shows 19 of the bilateral FTAs are fairly recent taking effect on or after 2010 Importers sometimes

revise the data they previously reported to the WTO and UNCTAD including revisions to tariff schedules We

periodically check the underlying databases for changes and will update our analysis as the data are refreshed

21

HS4HS2

minus10

minus05

00

05

larrD

eepe

r re

duct

ions

S

hallo

wer

red

uctio

nsrarr

Undirected

HS4

HS2

HS4

HS2

Democratic ImporterNonminusDemocratic Exporter

NonminusDemocratic ImporterDemocratic Exporter

Directed

Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would

which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM

ijt ] = βMIXED +

γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic

pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents

our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find

any difference in the applied tariff rates of mixed dyads and democratic dyads14

To shed light on this finding we decompose the direction of trade liberalization among FTA

partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of

the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4

levels data respectively First we examine whether democratic importers are able to engage in

deeper trade liberalization when their counterpart is a democracy rather than a non-democracy

This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former

engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given

the small number of FTAs involving non-democracy pairs in our data

22

Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff

reductions when their export partner is a non-democracy rather than a democracy the magnitude is

substantially larger when disaggregating to the HS4 level Second we consider whether democratic

exporters can achieve better market access when their import partner is a democracy or a non-

democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)

in the panel we find that tariffs reductions are smaller when non-democratic importers are paired

with democratic exporters than when democratic importers are paired with democratic exporters

Again the magnitude is larger with more granular tariff data These results suggest that the finding

in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions

than mixed dyads might be due to the fact that non-democratic importers give shallower concessions

to democratic exporters than democratic importers give to democratic exporters

Finally we explore the complex bilateral strategic incentives among FTA partners at granular

levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in

panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)

of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions

than pairs of democracies The estimated effects are significant for agriculture and metals industries

with very small overall differences across industries This pattern is even more prominent with HS4-

level results whereby most HS4 estimates are negative and statistically significant Similarly panel

(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff

reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4

industries Importantly unlike our monadic analysis we do not find significant heterogeneities when

we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively

constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This

suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level

interests rather than highly heterogeneous interest of individual producers15

It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South

Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation

Korean government formed several task forces organized around broad industry groupings within relevant agencies

such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare

23

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243 44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 6364656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 8889 90 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 26

27

28 29 3031 32 3334353637 38

39 40

41 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 7172 73 74 75 76 787980 81 82 83

84 85

86 878889

90 91 92939495 96 97

minus25

minus20

minus15

minus10

minus05

00

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(a) Democratic Importer Non-Democratic Exporter

0102

03 0405

06 07 08 09 10 11 121314 15

161718192021222324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 888990 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

05

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 0405

06 07 08

09

10 11

12

1314

15

161718

19

2021

22

23

24

2526

27

28

2930

31

32

33

3435

36

37 38

3940

41 42

43

44

4546

4748

49

50

5152

53 5455

56

57

58

59

60 6162

63

6465

6667

68

69

7071

72

7374

75

76

78

79

80

81

82

8384 85

86

87

888990 91 929394

95

96

97

minus05

00

05

10

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(b) Non-Democratic Importer Democratic Exporter

Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)

24

causal effects of democratization on bilateral trade liberalization across industries We leave for

future research this challenging task of investigating how exactly changes in political institutions

translate into trade policy outcomes

4 Concluding Remarks

In this paper we present a novel dataset with nearly 6 billion observations of product-level applied

tariff rates that countries apply to their trading partners incorporating the universe of preferential

rates and the Generalized System of Preferences To do so we combine and augment existing

datasets available from the WTO and UNCTAD and we resolve conflicting information between

the two Our dataset lays an important empirical foundation for investigating trade politics at a

much more granular level than has previously been done

To illustrate the importance of product-specific tariff measurement we examine an enduring

question in international political economy whether there are systematic differences in trade policy

between countries with different regime types Consistent with prior work we find that democra-

cies have lower tariff rates than non-democracies on average However focusing on the country-

or sector-level average elides substantial heterogeneity We document that democracies are more

protective than non-democracies for many industries such as agriculture and textile industries

Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-

ization We find significant variations across products even within the same industry Specifically

we show that consumer products are more likely to be protected in democracies whereas intermedi-

ate goods receive much deeper liberalization compared to the average level across other products

Our findings demonstrate various instances of aggregation bias with the use of increasingly granular

units of analysis

Our data also allows researchers in the fields of international and comparative political economy

to track fine-grained temporal changes in partner-specific trade policy In particular we examine

whether interactions between regime types at the dyad level results in differences in the degree of

bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences

design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a

mixed pair with a democracy and a non-democracy However by disentangling who is the recipient

of concessions we show that the difference between democratic pairs and mixed pairs is due in

25

large part to shallower concessions granted by non-democratic importers to democratic partners

but not vice-versa Put another way democratic importers grant trade concessions to democratic

exporters and non-democratic exporters but democratic exporters win more trade concessions

from democratic importers than from non-democratic importers Future studies would benefit from

investigating these empirical findings with a theoretical focus on the direction of trade liberalization

Our dataset can be combined with industry-level covariates such as import and export concen-

tration as well as country-specific industry structures to further explore linkages between political

institutions and industry-level trade policies It can also be used to study the increasing complex-

ity of product-level trade policy that affects the deepening of global supply chain and production

networks In addition as other scholars have pointed out there exists significant variation in insti-

tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and

Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to

assemble are likely to result in different configurations of demands for trade protection Research

into the relationships between political institutions and trade policies continues to be relevant as

policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-

ing trade agreements in response to pressures from their constituents The question is not so much

whether there will be more or less liberalization but rather which products and industries will be

most exposed to a review of trade policies This article presents findings that usefully contribute

to this research agenda

26

ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press

Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank

Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340

Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289

Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323

Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt

Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600

mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http

peopletamuedu~apondresearchhtml

Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press

Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37

Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309

Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress

27

HS4HS2

minus10

minus05

00

05

larrD

eepe

r re

duct

ions

S

hallo

wer

red

uctio

nsrarr

Undirected

HS4

HS2

HS4

HS2

Democratic ImporterNonminusDemocratic Exporter

NonminusDemocratic ImporterDemocratic Exporter

Directed

Figure 7 Tariff Reductions by Dyad Type The left panel shows the difference in tariff reductionsbetween mixed dyads (where one party to the FTA is a democracy and the other is a non-democracy) and democraticdyads On average we do not find a statistical difference in reduction The right panel disaggregates mixed dyads intotwo types one in which the importer is the democracy and one in which the exporter is the democracy Black dotsindicate tariff reductions using HS2-level tariff and volume measures (with 95 confidence interval) blue trianglesindicate tariff reductions using HS4-level tariff and volume measures The middle panel shows that a democraticimporter gives deeper concessions to a non-democratic partner than it would to a democratic partner In contrastthe right panel shows that a non-democratic importer secures shallower reductions from a democratic partner thana democratic import partner would

which produces a comparable quantity of interest E[∆τijpt | DMIXEDijt ]minusE[∆τijpt | DDEMDEM

ijt ] = βMIXED +

γMIXEDp that is the tariff reduction for product p by mixed pairs compared to that between democratic

pairs without distinguishing the direction of trade liberalization The left panel in Figure 7 presents

our estimates for the main effect βMIXED using HS2 and HS4 level data On average we do not find

any difference in the applied tariff rates of mixed dyads and democratic dyads14

To shed light on this finding we decompose the direction of trade liberalization among FTA

partners The right panel in Figure 7 reports the posterior mean and 95 credible intervals of

the main effects βDEMNONDEM and βNONDEMDEM given in equations (7) and (8) using the HS2 and HS4

levels data respectively First we examine whether democratic importers are able to engage in

deeper trade liberalization when their counterpart is a democracy rather than a non-democracy

This corresponds to the estimates on the left-hand side (ldquoDemocratic Importer Non-Democratic14Our model also allows us to compare pairs of non-democracies to pairs of democracies We find that the former

engages in deeper liberalization than the latter (-034 log points) although this estimate is likely to be noisy given

the small number of FTAs involving non-democracy pairs in our data

22

Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff

reductions when their export partner is a non-democracy rather than a democracy the magnitude is

substantially larger when disaggregating to the HS4 level Second we consider whether democratic

exporters can achieve better market access when their import partner is a democracy or a non-

democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)

in the panel we find that tariffs reductions are smaller when non-democratic importers are paired

with democratic exporters than when democratic importers are paired with democratic exporters

Again the magnitude is larger with more granular tariff data These results suggest that the finding

in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions

than mixed dyads might be due to the fact that non-democratic importers give shallower concessions

to democratic exporters than democratic importers give to democratic exporters

Finally we explore the complex bilateral strategic incentives among FTA partners at granular

levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in

panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)

of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions

than pairs of democracies The estimated effects are significant for agriculture and metals industries

with very small overall differences across industries This pattern is even more prominent with HS4-

level results whereby most HS4 estimates are negative and statistically significant Similarly panel

(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff

reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4

industries Importantly unlike our monadic analysis we do not find significant heterogeneities when

we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively

constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This

suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level

interests rather than highly heterogeneous interest of individual producers15

It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South

Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation

Korean government formed several task forces organized around broad industry groupings within relevant agencies

such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare

23

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243 44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 6364656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 8889 90 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 26

27

28 29 3031 32 3334353637 38

39 40

41 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 7172 73 74 75 76 787980 81 82 83

84 85

86 878889

90 91 92939495 96 97

minus25

minus20

minus15

minus10

minus05

00

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(a) Democratic Importer Non-Democratic Exporter

0102

03 0405

06 07 08 09 10 11 121314 15

161718192021222324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 888990 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

05

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 0405

06 07 08

09

10 11

12

1314

15

161718

19

2021

22

23

24

2526

27

28

2930

31

32

33

3435

36

37 38

3940

41 42

43

44

4546

4748

49

50

5152

53 5455

56

57

58

59

60 6162

63

6465

6667

68

69

7071

72

7374

75

76

78

79

80

81

82

8384 85

86

87

888990 91 929394

95

96

97

minus05

00

05

10

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(b) Non-Democratic Importer Democratic Exporter

Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)

24

causal effects of democratization on bilateral trade liberalization across industries We leave for

future research this challenging task of investigating how exactly changes in political institutions

translate into trade policy outcomes

4 Concluding Remarks

In this paper we present a novel dataset with nearly 6 billion observations of product-level applied

tariff rates that countries apply to their trading partners incorporating the universe of preferential

rates and the Generalized System of Preferences To do so we combine and augment existing

datasets available from the WTO and UNCTAD and we resolve conflicting information between

the two Our dataset lays an important empirical foundation for investigating trade politics at a

much more granular level than has previously been done

To illustrate the importance of product-specific tariff measurement we examine an enduring

question in international political economy whether there are systematic differences in trade policy

between countries with different regime types Consistent with prior work we find that democra-

cies have lower tariff rates than non-democracies on average However focusing on the country-

or sector-level average elides substantial heterogeneity We document that democracies are more

protective than non-democracies for many industries such as agriculture and textile industries

Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-

ization We find significant variations across products even within the same industry Specifically

we show that consumer products are more likely to be protected in democracies whereas intermedi-

ate goods receive much deeper liberalization compared to the average level across other products

Our findings demonstrate various instances of aggregation bias with the use of increasingly granular

units of analysis

Our data also allows researchers in the fields of international and comparative political economy

to track fine-grained temporal changes in partner-specific trade policy In particular we examine

whether interactions between regime types at the dyad level results in differences in the degree of

bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences

design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a

mixed pair with a democracy and a non-democracy However by disentangling who is the recipient

of concessions we show that the difference between democratic pairs and mixed pairs is due in

25

large part to shallower concessions granted by non-democratic importers to democratic partners

but not vice-versa Put another way democratic importers grant trade concessions to democratic

exporters and non-democratic exporters but democratic exporters win more trade concessions

from democratic importers than from non-democratic importers Future studies would benefit from

investigating these empirical findings with a theoretical focus on the direction of trade liberalization

Our dataset can be combined with industry-level covariates such as import and export concen-

tration as well as country-specific industry structures to further explore linkages between political

institutions and industry-level trade policies It can also be used to study the increasing complex-

ity of product-level trade policy that affects the deepening of global supply chain and production

networks In addition as other scholars have pointed out there exists significant variation in insti-

tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and

Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to

assemble are likely to result in different configurations of demands for trade protection Research

into the relationships between political institutions and trade policies continues to be relevant as

policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-

ing trade agreements in response to pressures from their constituents The question is not so much

whether there will be more or less liberalization but rather which products and industries will be

most exposed to a review of trade policies This article presents findings that usefully contribute

to this research agenda

26

ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press

Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank

Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340

Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289

Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323

Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt

Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600

mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http

peopletamuedu~apondresearchhtml

Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press

Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37

Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309

Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress

27

Exporterrdquo) in the panel We find that in fact democratic importers tend to engage in deeper tariff

reductions when their export partner is a non-democracy rather than a democracy the magnitude is

substantially larger when disaggregating to the HS4 level Second we consider whether democratic

exporters can achieve better market access when their import partner is a democracy or a non-

democracy As shown in the right-hand side (ldquoNon-Democratic Importer Democratic Exporterrdquo)

in the panel we find that tariffs reductions are smaller when non-democratic importers are paired

with democratic exporters than when democratic importers are paired with democratic exporters

Again the magnitude is larger with more granular tariff data These results suggest that the finding

in Mansfield Milner and Rosendorff (2000) that democratic dyads achieve greater tariff reductions

than mixed dyads might be due to the fact that non-democratic importers give shallower concessions

to democratic exporters than democratic importers give to democratic exporters

Finally we explore the complex bilateral strategic incentives among FTA partners at granular

levels Figure C3 presents our estimates of the industry-varying effects βDEMNONDEM + γDEMNONDEMp in

panel (a) and βNONDEMDEM + γNONDEMDEMp in panel (b) respectively Consistent with Figure 7 panel (a)

of Figure C3 shows that mixed pairs with democratic importer engage in deeper tariff reductions

than pairs of democracies The estimated effects are significant for agriculture and metals industries

with very small overall differences across industries This pattern is even more prominent with HS4-

level results whereby most HS4 estimates are negative and statistically significant Similarly panel

(b) at the bottom shows that mixed pairs with non-democratic importer engage in shallower tariff

reductions than pairs of democracies Again our findings are consistent across various HS2 and HS4

industries Importantly unlike our monadic analysis we do not find significant heterogeneities when

we disaggregate the analysis at the HS4 level Rather we find that the point estimates are relatively

constant across HS4 goods that belong to the same sector (divided by solid vertical lines) This

suggests in bilateral FTAs reciprocal concessions are made reflecting more broad industry-level

interests rather than highly heterogeneous interest of individual producers15

It is important to note that the results presented in this section should not be interpreted as15The authorrsquos interview with a Korean diplomat who participated in the KORUS negotiation (FTA between South

Korea and the US) confirms that this was the case in the agreement Upon the announcement of the negotiation

Korean government formed several task forces organized around broad industry groupings within relevant agencies

such as the Ministry of Agriculture and Ministry of Trade Industry and Energy and Ministry of Health and Welfare

23

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243 44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 6364656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 8889 90 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 26

27

28 29 3031 32 3334353637 38

39 40

41 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 7172 73 74 75 76 787980 81 82 83

84 85

86 878889

90 91 92939495 96 97

minus25

minus20

minus15

minus10

minus05

00

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(a) Democratic Importer Non-Democratic Exporter

0102

03 0405

06 07 08 09 10 11 121314 15

161718192021222324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 888990 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

05

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 0405

06 07 08

09

10 11

12

1314

15

161718

19

2021

22

23

24

2526

27

28

2930

31

32

33

3435

36

37 38

3940

41 42

43

44

4546

4748

49

50

5152

53 5455

56

57

58

59

60 6162

63

6465

6667

68

69

7071

72

7374

75

76

78

79

80

81

82

8384 85

86

87

888990 91 929394

95

96

97

minus05

00

05

10

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(b) Non-Democratic Importer Democratic Exporter

Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)

24

causal effects of democratization on bilateral trade liberalization across industries We leave for

future research this challenging task of investigating how exactly changes in political institutions

translate into trade policy outcomes

4 Concluding Remarks

In this paper we present a novel dataset with nearly 6 billion observations of product-level applied

tariff rates that countries apply to their trading partners incorporating the universe of preferential

rates and the Generalized System of Preferences To do so we combine and augment existing

datasets available from the WTO and UNCTAD and we resolve conflicting information between

the two Our dataset lays an important empirical foundation for investigating trade politics at a

much more granular level than has previously been done

To illustrate the importance of product-specific tariff measurement we examine an enduring

question in international political economy whether there are systematic differences in trade policy

between countries with different regime types Consistent with prior work we find that democra-

cies have lower tariff rates than non-democracies on average However focusing on the country-

or sector-level average elides substantial heterogeneity We document that democracies are more

protective than non-democracies for many industries such as agriculture and textile industries

Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-

ization We find significant variations across products even within the same industry Specifically

we show that consumer products are more likely to be protected in democracies whereas intermedi-

ate goods receive much deeper liberalization compared to the average level across other products

Our findings demonstrate various instances of aggregation bias with the use of increasingly granular

units of analysis

Our data also allows researchers in the fields of international and comparative political economy

to track fine-grained temporal changes in partner-specific trade policy In particular we examine

whether interactions between regime types at the dyad level results in differences in the degree of

bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences

design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a

mixed pair with a democracy and a non-democracy However by disentangling who is the recipient

of concessions we show that the difference between democratic pairs and mixed pairs is due in

25

large part to shallower concessions granted by non-democratic importers to democratic partners

but not vice-versa Put another way democratic importers grant trade concessions to democratic

exporters and non-democratic exporters but democratic exporters win more trade concessions

from democratic importers than from non-democratic importers Future studies would benefit from

investigating these empirical findings with a theoretical focus on the direction of trade liberalization

Our dataset can be combined with industry-level covariates such as import and export concen-

tration as well as country-specific industry structures to further explore linkages between political

institutions and industry-level trade policies It can also be used to study the increasing complex-

ity of product-level trade policy that affects the deepening of global supply chain and production

networks In addition as other scholars have pointed out there exists significant variation in insti-

tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and

Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to

assemble are likely to result in different configurations of demands for trade protection Research

into the relationships between political institutions and trade policies continues to be relevant as

policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-

ing trade agreements in response to pressures from their constituents The question is not so much

whether there will be more or less liberalization but rather which products and industries will be

most exposed to a review of trade policies This article presents findings that usefully contribute

to this research agenda

26

ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press

Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank

Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340

Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289

Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323

Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt

Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600

mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http

peopletamuedu~apondresearchhtml

Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press

Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37

Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309

Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress

27

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243 44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 6364656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 8889 90 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 04 05

06 07 08 09 10 11 121314 15

16171819202122 2324

25 26

27

28 29 3031 32 3334353637 38

39 40

41 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 7172 73 74 75 76 787980 81 82 83

84 85

86 878889

90 91 92939495 96 97

minus25

minus20

minus15

minus10

minus05

00

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(a) Democratic Importer Non-Democratic Exporter

0102

03 0405

06 07 08 09 10 11 121314 15

161718192021222324

25 2627

28 29 3031 32 3334353637 38

39 4041 4243

44 454647 48 49

50 51 52 53 54 55 565758 59 60 61 62 63

64656667

68 69 70 71 72 73 74 75 76 787980 81 82 83 84 85

86 87 888990 91 92939495 96 97

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

TransportMiscellan

minus05

00

05

Harmonized System 2minusdigit

larrD

eepe

r

Sha

llow

errarr

0102 03 0405

06 07 08

09

10 11

12

1314

15

161718

19

2021

22

23

24

2526

27

28

2930

31

32

33

3435

36

37 38

3940

41 42

43

44

4546

4748

49

50

5152

53 5455

56

57

58

59

60 6162

63

6465

6667

68

69

7071

72

7374

75

76

78

79

80

81

82

8384 85

86

87

888990 91 929394

95

96

97

minus05

00

05

10

Harmonized System 4minusdigit

larrD

eepe

r ta

riff r

educ

tions

Sha

llow

er ta

riff r

educ

tions

rarr

(b) Non-Democratic Importer Democratic Exporter

Figure 8 Mixed Dyads Compared to Democratic Dyads Panel (a) shows that a democraticimporter tends to give deeper tariff reductions to their non-democratic partners than they do to their democraticpartners We find significant effects from agriculture and metals industries at the HS2 level while most estimatesachieve statistical significance at the HS4 level Conversely Panel (b) shows that a non-democratic importer tends togive shallower tariff reductions compared to a democratic importer when the FTA partner is a democracy (Equiva-lently a democratic importer gives deeper tariff reductions to a democratic export partner than to a non-democraticexporter)

24

causal effects of democratization on bilateral trade liberalization across industries We leave for

future research this challenging task of investigating how exactly changes in political institutions

translate into trade policy outcomes

4 Concluding Remarks

In this paper we present a novel dataset with nearly 6 billion observations of product-level applied

tariff rates that countries apply to their trading partners incorporating the universe of preferential

rates and the Generalized System of Preferences To do so we combine and augment existing

datasets available from the WTO and UNCTAD and we resolve conflicting information between

the two Our dataset lays an important empirical foundation for investigating trade politics at a

much more granular level than has previously been done

To illustrate the importance of product-specific tariff measurement we examine an enduring

question in international political economy whether there are systematic differences in trade policy

between countries with different regime types Consistent with prior work we find that democra-

cies have lower tariff rates than non-democracies on average However focusing on the country-

or sector-level average elides substantial heterogeneity We document that democracies are more

protective than non-democracies for many industries such as agriculture and textile industries

Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-

ization We find significant variations across products even within the same industry Specifically

we show that consumer products are more likely to be protected in democracies whereas intermedi-

ate goods receive much deeper liberalization compared to the average level across other products

Our findings demonstrate various instances of aggregation bias with the use of increasingly granular

units of analysis

Our data also allows researchers in the fields of international and comparative political economy

to track fine-grained temporal changes in partner-specific trade policy In particular we examine

whether interactions between regime types at the dyad level results in differences in the degree of

bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences

design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a

mixed pair with a democracy and a non-democracy However by disentangling who is the recipient

of concessions we show that the difference between democratic pairs and mixed pairs is due in

25

large part to shallower concessions granted by non-democratic importers to democratic partners

but not vice-versa Put another way democratic importers grant trade concessions to democratic

exporters and non-democratic exporters but democratic exporters win more trade concessions

from democratic importers than from non-democratic importers Future studies would benefit from

investigating these empirical findings with a theoretical focus on the direction of trade liberalization

Our dataset can be combined with industry-level covariates such as import and export concen-

tration as well as country-specific industry structures to further explore linkages between political

institutions and industry-level trade policies It can also be used to study the increasing complex-

ity of product-level trade policy that affects the deepening of global supply chain and production

networks In addition as other scholars have pointed out there exists significant variation in insti-

tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and

Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to

assemble are likely to result in different configurations of demands for trade protection Research

into the relationships between political institutions and trade policies continues to be relevant as

policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-

ing trade agreements in response to pressures from their constituents The question is not so much

whether there will be more or less liberalization but rather which products and industries will be

most exposed to a review of trade policies This article presents findings that usefully contribute

to this research agenda

26

ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press

Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank

Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340

Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289

Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323

Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt

Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600

mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http

peopletamuedu~apondresearchhtml

Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press

Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37

Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309

Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress

27

causal effects of democratization on bilateral trade liberalization across industries We leave for

future research this challenging task of investigating how exactly changes in political institutions

translate into trade policy outcomes

4 Concluding Remarks

In this paper we present a novel dataset with nearly 6 billion observations of product-level applied

tariff rates that countries apply to their trading partners incorporating the universe of preferential

rates and the Generalized System of Preferences To do so we combine and augment existing

datasets available from the WTO and UNCTAD and we resolve conflicting information between

the two Our dataset lays an important empirical foundation for investigating trade politics at a

much more granular level than has previously been done

To illustrate the importance of product-specific tariff measurement we examine an enduring

question in international political economy whether there are systematic differences in trade policy

between countries with different regime types Consistent with prior work we find that democra-

cies have lower tariff rates than non-democracies on average However focusing on the country-

or sector-level average elides substantial heterogeneity We document that democracies are more

protective than non-democracies for many industries such as agriculture and textile industries

Furthermore we offer the first product-level estimates of how political regimes affect trade liberal-

ization We find significant variations across products even within the same industry Specifically

we show that consumer products are more likely to be protected in democracies whereas intermedi-

ate goods receive much deeper liberalization compared to the average level across other products

Our findings demonstrate various instances of aggregation bias with the use of increasingly granular

units of analysis

Our data also allows researchers in the fields of international and comparative political economy

to track fine-grained temporal changes in partner-specific trade policy In particular we examine

whether interactions between regime types at the dyad level results in differences in the degree of

bilateral trade liberalization Our analysis of 90 bilateral FTAs based on a difference-in-differences

design partially confirms prior findings that democratic pairs achieve greater tariff reductions than a

mixed pair with a democracy and a non-democracy However by disentangling who is the recipient

of concessions we show that the difference between democratic pairs and mixed pairs is due in

25

large part to shallower concessions granted by non-democratic importers to democratic partners

but not vice-versa Put another way democratic importers grant trade concessions to democratic

exporters and non-democratic exporters but democratic exporters win more trade concessions

from democratic importers than from non-democratic importers Future studies would benefit from

investigating these empirical findings with a theoretical focus on the direction of trade liberalization

Our dataset can be combined with industry-level covariates such as import and export concen-

tration as well as country-specific industry structures to further explore linkages between political

institutions and industry-level trade policies It can also be used to study the increasing complex-

ity of product-level trade policy that affects the deepening of global supply chain and production

networks In addition as other scholars have pointed out there exists significant variation in insti-

tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and

Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to

assemble are likely to result in different configurations of demands for trade protection Research

into the relationships between political institutions and trade policies continues to be relevant as

policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-

ing trade agreements in response to pressures from their constituents The question is not so much

whether there will be more or less liberalization but rather which products and industries will be

most exposed to a review of trade policies This article presents findings that usefully contribute

to this research agenda

26

ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press

Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank

Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340

Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289

Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323

Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt

Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600

mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http

peopletamuedu~apondresearchhtml

Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press

Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37

Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309

Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress

27

large part to shallower concessions granted by non-democratic importers to democratic partners

but not vice-versa Put another way democratic importers grant trade concessions to democratic

exporters and non-democratic exporters but democratic exporters win more trade concessions

from democratic importers than from non-democratic importers Future studies would benefit from

investigating these empirical findings with a theoretical focus on the direction of trade liberalization

Our dataset can be combined with industry-level covariates such as import and export concen-

tration as well as country-specific industry structures to further explore linkages between political

institutions and industry-level trade policies It can also be used to study the increasing complex-

ity of product-level trade policy that affects the deepening of global supply chain and production

networks In addition as other scholars have pointed out there exists significant variation in insti-

tutional structures within democracies and non-democracies (Rickard 2015 Geddes Wright and

Frantz 2014) Differences in the scale and scope of support coalitions that a government needs to

assemble are likely to result in different configurations of demands for trade protection Research

into the relationships between political institutions and trade policies continues to be relevant as

policymakers around the world re-evaluate the merits of trade liberalization and re-negotiate exist-

ing trade agreements in response to pressures from their constituents The question is not so much

whether there will be more or less liberalization but rather which products and industries will be

most exposed to a review of trade policies This article presents findings that usefully contribute

to this research agenda

26

ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press

Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank

Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340

Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289

Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323

Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt

Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600

mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http

peopletamuedu~apondresearchhtml

Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press

Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37

Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309

Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress

27

ReferencesAcemoglu Daron and James A Robinson 2005 Economic Origins of Dictatorship and DemocracyCambridge UK Cambridge University Press

Anderson Kym and Will Martin 2005 Agricultural Trade Reform and the Doha DevelopmentAgenda Washington DC The World Bank

Baccini Leonardo Andreas Duumlr and Manfred Elsig 2018 Intra-Industry Trade Global ValueChains and Preferential Tariff Liberalization International Studies Quarterly 62 (2)329ndash340

Bastiaens Ida and Nita Rudra 2016 Trade Liberalization and the Challenges of Revenue Mobi-lization Can International Financial Institutions Make a Difference Review of InternationalPolitical Economy 23 (2)261ndash289

Bates Robert H and Steven Block 2011 Political Institutions and Agricultural Trade Interven-tions in Africa American Journal of Agricultural Economics 93 (2)317ndash323

Betancourt Michael 2017 A Conceptual Introduction to Hamiltonian Monte Carlo Unpublishedmanuscript available at lthttpsarxivorgabs170102434gt

Betz Timm and Amy Pond 2019 The Absence of Consumer Interests in Trade Policy TheJournal of Politics 81 (2)585ndash600

mdashmdashmdash 2020 Governments as Borrowers and Regulators Working Paper available from http

peopletamuedu~apondresearchhtml

Carnegie Allison 2015 Power Plays How International Institutions Reshape Coercive DiplomacyCambridge UK Cambridge University Press

Carpenter Bob Andrew Gelman Matt Hoffman Daniel Lee Ben Goodrich Michael BetancourtMichael A Brubaker Jiqiang Guo Peter Li and Allen Riddell 2016 Stan A ProbabilisticProgramming Language Journal of Statistical Software 201ndash37

Chaudoin Stephen Helen V Milner and Xun Pang 2015 International Systems and DomesticPolitics Linking Complex Interactions with Empirical Models in International Relations Inter-national Organization 69 (2)275ndash309

Frieden Jeffry A and Ronald Rogowski 1996 The Impact of the International Economy onNational Policies An Analytical Overview In Internationalization and Domestic Politics editedby Robert O Keohane and Helen V Milner chap 2 25ndash47 Cambridge Cambridge UniversityPress

27

Gawande Kishore and Wendy L Hansen 1999 Retaliation Bargaining and the Pursuit of Freeand Fair Trade International Organization 53 (1)117ndash159

Geddes Barbara Joseph Wright and Erica Frantz 2014 Autocratic Breakdown and RegimeTransitions A New Data Set Perspectives on Politics 12 (2)313ndash331

Grossman Gene M and Elhanan Helpman 1994 Protection for Sale The American EconomicReview 84 (4)833ndash850

Hansen John Mark 1990 Taxation and the Political Economy of the Tariff International Orga-nization 44 (4)527ndash551

Henisz Witold J and Edward D Mansfield 2006 Votes and Vetoes The Political Determinantsof Commercial Openness International Studies Quarterly 50 (1)189ndash212

Honaker James Gary King and Matthew Blackwell 2011 Amelia II A Program for Missing DataJournal of Statistical Software 45 (7)1ndash47

Jordan Michael Zoubin Ghahramani Tommi Jaakkola and Lawrence Saul 1999 An Introductionto Variational Methods for Graphical Models Machine learning 37 (2)183ndash233

Kim In Song 2017 Political Cleavages within Industry Firm-level Lobbying for Trade Liberaliza-tion American Political Science Review 111 (1)1ndash20

Kim In Song and Iain Osgood 2019 Firms in Trade and Trade Politics Annual Review of PoliticalScience Forthcoming

Kono Daniel Y 2006 Optimal Obfuscation Democracy and Trade Policy Transparency AmericanPolitical Science Review 100 (3)369ndash384

Lake David A 2009 Open Economy Politics A critical Review The Review of InternationalOrganizations 4 (3)219ndash244

Luuml Xiaobo Kenneth F Scheve and Matthew J Slaughter 2012 Inequity Aversion and theInternational Distribution of Trade Protection American Journal of Political Science 56 (3)638ndash654

Maggi Giovanni and Andres Rodriguez-Clare 2007 A Political-Economy Theory of Trade Agree-ments American Economic Review 97 (4)1374ndash1406

Mansfield Edward D and Marc L Busch 1995 The Political Economy of Nontariff Barriers ACross-national Analysis International Organization 49 (4)723ndash749

28

Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321

mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513

Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430

Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985

Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143

Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661

Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329

Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102

Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press

Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297

Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier

Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222

Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292

Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press

29

Appendix A Tariff-line Dataset

A1 Bilateral Tariff Data Collection and Processing

A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System

There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present

To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2

Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000

A11 WTO IDB Duty Collection and Processing

We perform the following steps to collect and process IDB duties

Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg

We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013

16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg

1

Mansfield Edward D Helen V Milner and B Peter Rosendorff 2000 Free to Trade DemocraciesAutocracies and International Trade The American Political Science Review 94 (2)305ndash321

mdashmdashmdash 2002 Why Democracies Cooperate More Electoral Control and International Trade Agree-ments International Organization 56 (3)477ndash513

Mayda Anna Maria and Dani Rodrik 2005 Why Are Some People (and Countries) More Protec-tionist than Others European Economic Review 49 (6)1393ndash1430

Mayer Wolfgang 1984 Endogenous Tariff Formation The American Economic Review 74 (5)970ndash985

Milner Helen V and Keiko Kubota 2005 Why the Move to Free Trade Democracy and TradePolicy in the Developing Countries International Organization 59 (1)107ndash143

Morrow James D Randolph M Siverson and Tressa E Tabares 1998 The Political Determi-nants of International Trade The Major Powers 1907ndash1990 American Political Science Review92 (3)649ndash661

Park Jong Hee and Nathan Jensen 2007 Electoral Competition and Agricultural Support inOECD Countries American Journal of Political Science 51 (2)314ndash329

Pelc Krzysztof J 2013 The Cost of Wiggle-Room Looking at the Welfare Effects of Flexibility inTariff Rates at the WTO International Studies Quarterly 57 (1)91ndash102

Persson Torsten and Guido Enrico Tabellini 2005 The Economic Effects of Constitutions Cam-bridge MA MIT press

Rickard Stephanie J 2015 Electoral Systems and Trade Oxford Handbook of Political Economyof International Trade 15280ndash297

Rodrik Dani 1995 Chapter 28 Political Economy of Trade Policy In Handbook of InternationalEconomics vol 3 edited by Gene M Grossman and Kenneth Rogoff 1457 ndash 1494 Elsevier

Runge Carlisle Ford and Harald von Witzke 1987 Institutional Change in the Common Agri-cultural Policy of the European Community American Journal of Agricultural Economics69 (2)213ndash222

Scheve Kenneth F and Matthew J Slaughter 2001 What Determines Individual Trade-policyPreferences Journal of International Economics 54 (2)267ndash292

Varshney Ashutosh 1998 Democracy Development and the Countryside Urban-Rural Strugglesin India Cambridge UK Cambridge University Press

29

Appendix A Tariff-line Dataset

A1 Bilateral Tariff Data Collection and Processing

A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System

There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present

To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2

Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000

A11 WTO IDB Duty Collection and Processing

We perform the following steps to collect and process IDB duties

Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg

We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013

16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg

1

Appendix A Tariff-line Dataset

A1 Bilateral Tariff Data Collection and Processing

A tariff-line is a numeric code that each importer uses to identify a unique product For a givenproduct tariff-lines can differ from country to country however the first six digits of the tariff-lineare internationally standardized under the Harmonized System

There are two existing sources of tariff-line data the WTOrsquos Integrated Database (IDB) pub-licly accessible at the WTOrsquos public Tariff Analysis Online (TAO) facility and UNCTADrsquos TradeAnalysis Information System (TRAINS) publicly accessible at the World Bankrsquos World IntegratedTrade Solution (WITS) website16 Together they form a comprehensive collection of ad valoremand non-ad valorem tariff rates across all WTO countries and Harmonized System products from1988 to the present

To compile this universe of tariffs we first web-scrape tariff-lines for all available importers andyears An observation in this dataset is a tariff imposed in a given year by an importer on a productimported from a country (eg Republic of Korea) or a group of countries (eg NAFTA MercosurWTO members) Where the tariff affects a group of countries we identify the members of the groupand expand the observation so that each new observation is a dyad with an importer and exporterFinally for each resulting (year importer exporter tariff-line) we compare duties from IDB andTRAINS to select the most likely applied duty using the algorithm detailed in Appendix A2

Figure 2 illustrates the data collection processing and merging steps in our tariff datasetcreation using an example United States tariff-line The next sections detail each of these steps forIDB and TRAINS respectively To further clarify each step we use a recurring example tariff-lineThe United Statersquos (USA) 2013 tariff on HS product 62011330 (Overcoats raincoats car-coatscapes cloaks and similar articles) from South Africa (ZAF) Notably this particular tariff-line is abeneficiary of the African Growth and Opportunity Act (AGOA) enacted by the US in 2000

A11 WTO IDB Duty Collection and Processing

We perform the following steps to collect and process IDB duties

Step 1 (Web scrape product-level duties) For each year and importer we scrape allIDB product-level applied tariffs available through WTOrsquos public Tariff Analysis Online(TAO) facility Each duty is identified by its year importer and Harmonized Systemproduct code and contains information on its specific beneficiary group as well as the rateapplied Eg

We acquire two different reported duties from IDB for American imports of overcoat-likeapparels from WTO countries (including South Africa) in 2013

16TAOrsquos URL is httptaowtoorg and WITSrsquos URL is httpwitsworldbankorg

1

Year Imp Code Full description Type Reported rate

2013 USA 62011330ldquoMFN appliedduty raterdquo

02 497 centskg + 197

2013 USA 62011330ldquoGeneralduty raterdquo

80 529 centskg + 585

Step 2 (Parse compound and mixed tariff rates) In IDB all tariff-lines withcompound or mixed rates (rates that have both an ad valorem and non ad valoremcomponent) have a NULL in the field for the numerical duty rate Rather than discardingthese complex tariffs we parse the ad valorem component from the reported rate text anduse it as a approximation of the full duty rate Eg

Year Imp Code Full description Type Reported rate (asymp imputed AVE)

2013 USA 62011330ldquoMFN appliedduty raterdquo

02 497 centskg + 197 (asymp 197)

2013 USA 62011330ldquoGeneralduty raterdquo

80 529 centskg + 585 (asymp 585)

We now have an approximate lsquoad valorem equivalentrsquo rate imputed for these and all otherIDB mixedcompound duty rates

Step 3 (Disaggregate duty beneficiaries to countries) Each duty has a type field anddescription field that uniquely indicates its specific beneficiary which may be a country (egPreferential rate for Canada) members of an agreement (eg North-American FreeTrade Agreement) or a group of countries (eg G16) We use a mix of hand-coding fromofficial materials and string matching with country names and regional trade agreementtitles in order to map each duty type appearing in IDB data to its respective set ofcountries 17 Eg

Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)

2013 USA ZAF 62011330ldquoMFN appliedduty raterdquo

02 497 centskg + 197 (asymp 197)

2013 USA ZAF 62011330ldquoGeneralduty raterdquo

80 529 centskg + 585 (asymp 585)

We find that both IDB duty types stipulate South Africa as a beneficiary

A12 UNCTAD TRAINS Duty Collection and Processing

Likewise we perform the following corresponding steps for TRAINS tariffs

Step 1 (Web scrape product-level duties) For each year we scrape all TRAINSproduct-level tariffs available through the WITS web site Eg

17We use official preference beneficiaries for many tariff measures from httpswitsworldbankorgWITSWITSSupport+MaterialsTrfMeasuresaspxPage=TfMeasures We map beneficiaries of regional trade agreements fromthe Regional Trade Agreements Information System (RTA-IS) publicly accessible at httpsrtaiswtoorg

2

Year Imp Code Full description Type Reported rate (asymp UNCTAD AVE)

2013 USA 62011330ldquoMost FavouredNation dutyrate treatmentrdquo

002 497 centskg + 197 (asymp 2122)

2013 USA 62011330

ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo

051 00

We find two different duties applicable to 2013 American imports of HS product 62011330from South Africa Unlike IDB however TRAINS reports a preferential rate (AGOA) Alsounlike IDB TRAINS provides its own ad valorem equivalent (2122) for the compoundMFN tariff (497 centskg + 197)

Step 2 (Disaggregate duty beneficiaries to countries) Using a combination of aregion-to-countries mapping and a type-to-countries mapping both provided by the WorldBank we expand each beneficiary-level duty to its disaggregated partner-specific dutiesEg

Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)

2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo

002 497 centskg + 197 (asymp 2122)

2013 USA ZAF 62011330

ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo

051 00

Again we find that both the duties found in TRAINS stipulate South Africa as a beneficiary

Performing these procedures we acquire 41 billion IDB and 47 billion TRAINS product-levelpartner-specific duties However as noted in our example for each (year importer exporterproduct) we may have multiple conflicting duties of which only one is actually applied In the nextsection we describe the merging algorithm used to solve this problem

A2 Tariff Merging Algorithm

A given (year importer exporter industry) query may return multiple possible duties from theWTO IDB database and the UNCTAD TRAINS database In some cases both sources agreeon an ad valorem rate but TRAINS provides a more informative specific duty rate In othercases TRAINS correctly accounts for a compound or mixed rate while IDB does not Moreoverfor some years one source correctly retrieves a newly enforced preferential rate while the othermistakenly reports previous yearsrsquo Most Favored Nation (MFN) duty rate Finally for all non-ad

3

valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers

The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps

Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg

Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)

2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo

02 497 centskg + 197 (asymp 197)

2013 USA ZAF 62011330ldquoGeneralduty raterdquo

80 529 centskg + 585 (asymp 585)

In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values

Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg

Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)

2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo

002 497 centskg + 197 (asymp 2122)

2013 USA ZAF 62011330

ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo

051 00

Since there is only a single preferential duty we select it as the best TRAINS candidate

4

Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg

Year Imp Exp Code Original description Final applied rate Source

2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo

497 centskg + 197 (asymp 197) IDB

2013 USA ZAF 62011330

ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo

00 TRAINS

Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line

The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year

18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size

5

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

Year

JPNUSABRAPHLARGCHECANKORCOLNZL

NORAUSMEXZAFIDN

CHLBOLEUNURYIND

EGYPAKNIC

ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI

MDGISL

PERLKA

GTMTURBGDDOM

ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV

PNGJORBFAMACHKGSENBENZMBNERATG

MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ

MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI

KNAZWEGABGUYVCTUKR

FJIRUSAREMDV

HTICPVAGOKHMGHAVUTSUR

DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE

WSMYEMSYCAFGLBRLIE

PRT

Impo

rter

WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)

Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail

6

Appendix B List of Bilateral FTAs

Table B1 List of Bilateral Free Trade Agreements

Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009

Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008

Table B1 Continued on next page

7

Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012

Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009

Table B1 Continued on next page

8

Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012

9

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Animal

dem demdem nondemnondem dem

Vegetable FoodProd Minerals

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Fuels Chemicals PlastiRub HidesSkin

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Wood TextCloth Footwear StoneGlas

10 0 10years from agreement

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Metals

10 0 10years from agreement

MachElec

10 0 10years from agreement

Transport

10 0 10years from agreement

Miscellan

Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure

10

Appendix C Alternative Specifications and Diagnostics

01

02

03

04

05

0607

0809

1011

12

13

14

1516

17

18

1920

2122

23

24

25

26272829

30

31

32

33

3435

36

37

3839

40

41

42

43

44

45

46

47

48

4950

5152

53

5455

56

57

58

59

60

6162

63

64

65

6667

68

6970

71

72

7374

75

76

78

79

80

81

82

8384

85

86

87

88

89

Animal

Vegetable

FoodProd

Minerals

Fuels

Chemicals

PlastiRub

HidesSkin

Wood

TextCloth

Footwear

StoneGlas

Metals

MachElec

Transport

minus10

minus05

00

05

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

LDC only

Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate

11

01

02

03

04

05

06

07

08

09

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

2829

30

31

3233

34

3536

37

38

39

4041

42

43

44

45

46

47

48

4950

5152

53

545556

57

5859

60

6162

63

64

65

6667

68

6970

71

72

73

74

75

76

78

79

80

81

8283

84

85

86

87

88

89

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

minus3

minus2

minus1

0

1

2

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

All WTO

(a) HS 2-digit level

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

01

02

03

04

05

0607

08

09

10

11

12

13

14

15

1617

18

19

20

21

22

23

24

2526

27

28

29

30

31

32

33

34

3536

37

38

39

40

41

42

4344

45

46

47

48

49

50

51

52

53

54

55

56

57

5859

6061 62

63

64

65

66

67

68

69

70

71

72

73

74

75

7678

79

80

81

8283

84 8586

87

88

89

minus4

minus2

0

2

4

Harmonized System 4minusdigit Product

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(b) HS 4-digit level

Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does

12

02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus03

00

03

minus05

00

05

10

00

05

10

15

iteration

(a) HS 2-digit (HMC)

0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)

500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000

00

04

08

12

00

05

10

minus10

minus05

00

iteration

(b) HS 4-digit (HMC)

020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus01

00

01

02

minus06

minus03

00

minus15

minus10

minus05

00

iteration

(c) HS 6-digit (VB)

Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification

13

• Title Page
• 1 Introduction
• 2 New Database Bilateral Product-level Applied Tariffs
• 3 The Effects of Regime Type on Trade Policy
• • 31 Monadic Analysis
  • • 311 Methodology
    • 312 Empirical Results
    • • 32 Dyadic Analysis
      • • 321 Methodology
        • 322 Empirical Results
        • • 4 Concluding Remarks
          • References
          • A Tariff-line Dataset
          • • A1 Bilateral Tariff Data Collection and Processing
            • • A11 WTO IDB Duty Collection and Processing
              • A12 UNCTAD TRAINS Duty Collection and Processing
              • • A2 Tariff Merging Algorithm
                • • B List of Bilateral FTAs
                  • C Alternative Specifications and Diagnostics

valorem tariffs TRAINS provides an ad valorem equivalent (AVE) rate using a custom statisticalmethod that allows comparisons to be made between products with ad valorem and non-ad valoremrates For such tariffs IDB only provides the original non-ad valorem rate which is typically lessinformative for trade researchers

The goal of the merging algorithm is to account for all of these cases in order to select the singlemost accurate and informative duty that an importer applies to a industry and partner in a givenyear We illustrate how this is done using the previous example of United Statesrsquo 2013 tariff onHS product 62011330 from South Africa In this case it is clear that United States in practiceapplies the preferential AGOA duty rate over the Most Favored Nation duty rate Our algorithmcorrectly picks this rate in three steps

Step 1 (Pick IDB candidate) If there are any preferential IDB duties for the giventariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (or the imputed AVE in the case of mixedcompound tariffs) if noduties in the set have an ad valorem component sort using the parsed specific rate Eg

Year Imp Exp Code Full description Type Reported rate (asymp imputed AVE)

2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo

02 497 centskg + 197 (asymp 197)

2013 USA ZAF 62011330ldquoGeneralduty raterdquo

80 529 centskg + 585 (asymp 585)

In this case since there are no preferential duties reported by IDB we pick the lower of thenon-preferential duties using the imputed AVE values

Step 2 (Pick TRAINS candidate) If there are any preferential TRAINS duties for thegiven tariff-line pick the preferential duty with the lowest rate Otherwise pick thenon-preferential duty with the lowest rate When picking from either set sort duties usingthe ad valorem rate (either the reported ad valorem rate or the AVE imputed by UNCTAD)Eg

Year Imp Exp Code Full description Type Reported rate (asymp UNCTAD AVE)

2013 USA ZAF 62011330ldquoMost FavouredNation dutyrate treatmentrdquo

002 497 centskg + 197 (asymp 2122)

2013 USA ZAF 62011330

ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo

051 00

Since there is only a single preferential duty we select it as the best TRAINS candidate

4

Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg

Year Imp Exp Code Original description Final applied rate Source

2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo

497 centskg + 197 (asymp 197) IDB

2013 USA ZAF 62011330

ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo

00 TRAINS

Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line

The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year

18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size

5

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

Year

JPNUSABRAPHLARGCHECANKORCOLNZL

NORAUSMEXZAFIDN

CHLBOLEUNURYIND

EGYPAKNIC

ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI

MDGISL

PERLKA

GTMTURBGDDOM

ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV

PNGJORBFAMACHKGSENBENZMBNERATG

MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ

MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI

KNAZWEGABGUYVCTUKR

FJIRUSAREMDV

HTICPVAGOKHMGHAVUTSUR

DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE

WSMYEMSYCAFGLBRLIE

PRT

Impo

rter

WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)

Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail

6

Appendix B List of Bilateral FTAs

Table B1 List of Bilateral Free Trade Agreements

Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009

Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008

Table B1 Continued on next page

7

Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012

Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009

Table B1 Continued on next page

8

Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012

9

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Animal

dem demdem nondemnondem dem

Vegetable FoodProd Minerals

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Fuels Chemicals PlastiRub HidesSkin

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Wood TextCloth Footwear StoneGlas

10 0 10years from agreement

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Metals

10 0 10years from agreement

MachElec

10 0 10years from agreement

Transport

10 0 10years from agreement

Miscellan

Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure

10

Appendix C Alternative Specifications and Diagnostics

01

02

03

04

05

0607

0809

1011

12

13

14

1516

17

18

1920

2122

23

24

25

26272829

30

31

32

33

3435

36

37

3839

40

41

42

43

44

45

46

47

48

4950

5152

53

5455

56

57

58

59

60

6162

63

64

65

6667

68

6970

71

72

7374

75

76

78

79

80

81

82

8384

85

86

87

88

89

Animal

Vegetable

FoodProd

Minerals

Fuels

Chemicals

PlastiRub

HidesSkin

Wood

TextCloth

Footwear

StoneGlas

Metals

MachElec

Transport

minus10

minus05

00

05

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

LDC only

Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate

11

01

02

03

04

05

06

07

08

09

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

2829

30

31

3233

34

3536

37

38

39

4041

42

43

44

45

46

47

48

4950

5152

53

545556

57

5859

60

6162

63

64

65

6667

68

6970

71

72

73

74

75

76

78

79

80

81

8283

84

85

86

87

88

89

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

minus3

minus2

minus1

0

1

2

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

All WTO

(a) HS 2-digit level

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

01

02

03

04

05

0607

08

09

10

11

12

13

14

15

1617

18

19

20

21

22

23

24

2526

27

28

29

30

31

32

33

34

3536

37

38

39

40

41

42

4344

45

46

47

48

49

50

51

52

53

54

55

56

57

5859

6061 62

63

64

65

66

67

68

69

70

71

72

73

74

75

7678

79

80

81

8283

84 8586

87

88

89

minus4

minus2

0

2

4

Harmonized System 4minusdigit Product

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(b) HS 4-digit level

Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does

12

02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus03

00

03

minus05

00

05

10

00

05

10

15

iteration

(a) HS 2-digit (HMC)

0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)

500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000

00

04

08

12

00

05

10

minus10

minus05

00

iteration

(b) HS 4-digit (HMC)

020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus01

00

01

02

minus06

minus03

00

minus15

minus10

minus05

00

iteration

(c) HS 6-digit (VB)

Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification

13

• Title Page
• 1 Introduction
• 2 New Database Bilateral Product-level Applied Tariffs
• 3 The Effects of Regime Type on Trade Policy
• • 31 Monadic Analysis
  • • 311 Methodology
    • 312 Empirical Results
    • • 32 Dyadic Analysis
      • • 321 Methodology
        • 322 Empirical Results
        • • 4 Concluding Remarks
          • References
          • A Tariff-line Dataset
          • • A1 Bilateral Tariff Data Collection and Processing
            • • A11 WTO IDB Duty Collection and Processing
              • A12 UNCTAD TRAINS Duty Collection and Processing
              • • A2 Tariff Merging Algorithm
                • • B List of Bilateral FTAs
                  • C Alternative Specifications and Diagnostics

Step 3 (Select between candidates) Given the best IDB and TRAINS candidateduties if one is preferential and the other is not select the duty that is preferential If bothare either non-preferential or preferential and the TRAINS candidate has an imputed AVEselect the TRAINS candidate Otherwise select the candidate with the lowest ad valoremrate If either a TRAINS or IDB candidate could not be found select the candidate that isavailable Eg

Year Imp Exp Code Original description Final applied rate Source

2013 USA ZAF 62011330ldquoMFN appliedduty ratesrdquo

497 centskg + 197 (asymp 197) IDB

2013 USA ZAF 62011330

ldquoAGOA preferenceon certain textilesand apparelfor eligible countriesrdquo

00 TRAINS

Since TRAINS provides a preferential rate and IDB does not we select the TRAINScandidate as the applied duty for this tariff-line

The result is a unique tariff for each (year importer exporter product) query In sum thisprocedure merges 41 billion IDB duties with 47 billion TRAINS duties to produce 57 billionlsquoresolvedrsquo bilateral tariffs18 Figure A1 summarizes the coverage of the resulting tariff-line datasetfor each WTO importer and year

18We implement this procedure as a distributed SQL operation on the Hadoop big data ecosystem Overall thisoperation takes more than 72 hours to complete on a 10 node computing cluster (256 GB RAM per node 24 CPUper node) and the resulting un-indexed dataset is more than 900 GB in size

5

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

Year

JPNUSABRAPHLARGCHECANKORCOLNZL

NORAUSMEXZAFIDN

CHLBOLEUNURYIND

EGYPAKNIC

ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI

MDGISL

PERLKA

GTMTURBGDDOM

ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV

PNGJORBFAMACHKGSENBENZMBNERATG

MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ

MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI

KNAZWEGABGUYVCTUKR

FJIRUSAREMDV

HTICPVAGOKHMGHAVUTSUR

DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE

WSMYEMSYCAFGLBRLIE

PRT

Impo

rter

WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)

Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail

6

Appendix B List of Bilateral FTAs

Table B1 List of Bilateral Free Trade Agreements

Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009

Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008

Table B1 Continued on next page

7

Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012

Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009

Table B1 Continued on next page

8

Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012

9

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Animal

dem demdem nondemnondem dem

Vegetable FoodProd Minerals

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Fuels Chemicals PlastiRub HidesSkin

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Wood TextCloth Footwear StoneGlas

10 0 10years from agreement

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Metals

10 0 10years from agreement

MachElec

10 0 10years from agreement

Transport

10 0 10years from agreement

Miscellan

Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure

10

Appendix C Alternative Specifications and Diagnostics

01

02

03

04

05

0607

0809

1011

12

13

14

1516

17

18

1920

2122

23

24

25

26272829

30

31

32

33

3435

36

37

3839

40

41

42

43

44

45

46

47

48

4950

5152

53

5455

56

57

58

59

60

6162

63

64

65

6667

68

6970

71

72

7374

75

76

78

79

80

81

82

8384

85

86

87

88

89

Animal

Vegetable

FoodProd

Minerals

Fuels

Chemicals

PlastiRub

HidesSkin

Wood

TextCloth

Footwear

StoneGlas

Metals

MachElec

Transport

minus10

minus05

00

05

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

LDC only

Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate

11

01

02

03

04

05

06

07

08

09

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

2829

30

31

3233

34

3536

37

38

39

4041

42

43

44

45

46

47

48

4950

5152

53

545556

57

5859

60

6162

63

64

65

6667

68

6970

71

72

73

74

75

76

78

79

80

81

8283

84

85

86

87

88

89

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

minus3

minus2

minus1

0

1

2

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

All WTO

(a) HS 2-digit level

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

01

02

03

04

05

0607

08

09

10

11

12

13

14

15

1617

18

19

20

21

22

23

24

2526

27

28

29

30

31

32

33

34

3536

37

38

39

40

41

42

4344

45

46

47

48

49

50

51

52

53

54

55

56

57

5859

6061 62

63

64

65

66

67

68

69

70

71

72

73

74

75

7678

79

80

81

8283

84 8586

87

88

89

minus4

minus2

0

2

4

Harmonized System 4minusdigit Product

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(b) HS 4-digit level

Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does

12

02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus03

00

03

minus05

00

05

10

00

05

10

15

iteration

(a) HS 2-digit (HMC)

0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)

500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000

00

04

08

12

00

05

10

minus10

minus05

00

iteration

(b) HS 4-digit (HMC)

020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus01

00

01

02

minus06

minus03

00

minus15

minus10

minus05

00

iteration

(c) HS 6-digit (VB)

Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification

13

• Title Page
• 1 Introduction
• 2 New Database Bilateral Product-level Applied Tariffs
• 3 The Effects of Regime Type on Trade Policy
• • 31 Monadic Analysis
  • • 311 Methodology
    • 312 Empirical Results
    • • 32 Dyadic Analysis
      • • 321 Methodology
        • 322 Empirical Results
        • • 4 Concluding Remarks
          • References
          • A Tariff-line Dataset
          • • A1 Bilateral Tariff Data Collection and Processing
            • • A11 WTO IDB Duty Collection and Processing
              • A12 UNCTAD TRAINS Duty Collection and Processing
              • • A2 Tariff Merging Algorithm
                • • B List of Bilateral FTAs
                  • C Alternative Specifications and Diagnostics

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

Year

JPNUSABRAPHLARGCHECANKORCOLNZL

NORAUSMEXZAFIDN

CHLBOLEUNURYIND

EGYPAKNIC

ECUPRYMYSMUSSGPCRISLVVENCUBTGOBRNCHNMLI

MDGISL

PERLKA

GTMTURBGDDOM

ISRUGATHAHNDKENBHRMARNPLSAUPANNGACIV

PNGJORBFAMACHKGSENBENZMBNERATG

MMRALBTTOGEOSWZTZAKGZVNMMKDBWALCAMDABLZ

MNGTUNDMAOMNKWTCMRQATMOZMWICOGNAMRWAGNBGRDJAMARMCAFLSOBRBTCDBDI

KNAZWEGABGUYVCTUKR

FJIRUSAREMDV

HTICPVAGOKHMGHAVUTSUR

DJISLBMRTGMBMNETONGINTJKKAZLAOCODSLE

WSMYEMSYCAFGLBRLIE

PRT

Impo

rter

WTO IDB only UNCTAD TRAINS only Both available (some rates disagree) Both available (all rates agree)

Figure A1 Data Availability across Importers and Years This figure summarizes the availabilityof our data for each WTO importer and year Although the large number of missing import-year observations fromboth primary sources (white cells) prevents our dataset from being fully comprehensive it shows that our datasetcovers tariff policies for all major participants of global trade (top 50 trading countries in volume) starting in 1995Moreover we make several improvements by combining data from the two available sources (red and blue cells)and resolving various discrepancies where the sources may conflict (black cells) Altogether we compile 2476 WTOimporter-year tariff profiles (3080 importer-year profiles overall) from the WTO Integrated Database (IDB) andthe UNCTAD Trade Analysis Information System (TRAINS) As illustrated in this figure less than 50 of theseobservations are available from both sources where the reported duty rates agree Appendix A1 explains datacollection and processing in detail

6

Appendix B List of Bilateral FTAs

Table B1 List of Bilateral Free Trade Agreements

Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009

Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008

Table B1 Continued on next page

7

Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012

Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009

Table B1 Continued on next page

8

Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012

9

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Animal

dem demdem nondemnondem dem

Vegetable FoodProd Minerals

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Fuels Chemicals PlastiRub HidesSkin

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Wood TextCloth Footwear StoneGlas

10 0 10years from agreement

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Metals

10 0 10years from agreement

MachElec

10 0 10years from agreement

Transport

10 0 10years from agreement

Miscellan

Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure

10

Appendix C Alternative Specifications and Diagnostics

01

02

03

04

05

0607

0809

1011

12

13

14

1516

17

18

1920

2122

23

24

25

26272829

30

31

32

33

3435

36

37

3839

40

41

42

43

44

45

46

47

48

4950

5152

53

5455

56

57

58

59

60

6162

63

64

65

6667

68

6970

71

72

7374

75

76

78

79

80

81

82

8384

85

86

87

88

89

Animal

Vegetable

FoodProd

Minerals

Fuels

Chemicals

PlastiRub

HidesSkin

Wood

TextCloth

Footwear

StoneGlas

Metals

MachElec

Transport

minus10

minus05

00

05

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

LDC only

Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate

11

01

02

03

04

05

06

07

08

09

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

2829

30

31

3233

34

3536

37

38

39

4041

42

43

44

45

46

47

48

4950

5152

53

545556

57

5859

60

6162

63

64

65

6667

68

6970

71

72

73

74

75

76

78

79

80

81

8283

84

85

86

87

88

89

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

minus3

minus2

minus1

0

1

2

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

All WTO

(a) HS 2-digit level

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

01

02

03

04

05

0607

08

09

10

11

12

13

14

15

1617

18

19

20

21

22

23

24

2526

27

28

29

30

31

32

33

34

3536

37

38

39

40

41

42

4344

45

46

47

48

49

50

51

52

53

54

55

56

57

5859

6061 62

63

64

65

66

67

68

69

70

71

72

73

74

75

7678

79

80

81

8283

84 8586

87

88

89

minus4

minus2

0

2

4

Harmonized System 4minusdigit Product

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(b) HS 4-digit level

Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does

12

02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus03

00

03

minus05

00

05

10

00

05

10

15

iteration

(a) HS 2-digit (HMC)

0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)

500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000

00

04

08

12

00

05

10

minus10

minus05

00

iteration

(b) HS 4-digit (HMC)

020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus01

00

01

02

minus06

minus03

00

minus15

minus10

minus05

00

iteration

(c) HS 6-digit (VB)

Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification

13

• Title Page
• 1 Introduction
• 2 New Database Bilateral Product-level Applied Tariffs
• 3 The Effects of Regime Type on Trade Policy
• • 31 Monadic Analysis
  • • 311 Methodology
    • 312 Empirical Results
    • • 32 Dyadic Analysis
      • • 321 Methodology
        • 322 Empirical Results
        • • 4 Concluding Remarks
          • References
          • A Tariff-line Dataset
          • • A1 Bilateral Tariff Data Collection and Processing
            • • A11 WTO IDB Duty Collection and Processing
              • A12 UNCTAD TRAINS Duty Collection and Processing
              • • A2 Tariff Merging Algorithm
                • • B List of Bilateral FTAs
                  • C Alternative Specifications and Diagnostics

Appendix B List of Bilateral FTAs

Table B1 List of Bilateral Free Trade Agreements

Panel A Non-Democratic PairsArmenia Ukraine 1994Azerbaijan Ukraine 1994Ukraine Uzbekistan 1994Jordan Singapore 2003Morocco Turkey 2004Egypt Turkey 2005China Singapore 2007Jordan Turkey 2009

Panel B Mixed PairsGeorgia Ukraine 1994Israel Turkey 1995Georgia Turkmenistan 1998Macedonia Turkey 1998Jordan United States 1999New Zealand Singapore 1999Japan Singapore 2000Australia Singapore 2001Singapore United States 2002Australia Thailand 2003Moldova Ukraine 2003New Zealand Thailand 2003Tunisia Turkey 2003Bahrain United States 2004Chile China 2004Japan Malaysia 2004South Korea Singapore 2004Morocco United States 2004Panama Singapore 2004China Pakistan 2005Japan Thailand 2005Albania Turkey 2006China New Zealand 2006Georgia Turkey 2006Malaysia Pakistan 2006Peru Singapore 2007Oman United States 2007China Peru 2008

Table B1 Continued on next page

7

Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012

Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009

Table B1 Continued on next page

8

Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012

9

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Animal

dem demdem nondemnondem dem

Vegetable FoodProd Minerals

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Fuels Chemicals PlastiRub HidesSkin

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Wood TextCloth Footwear StoneGlas

10 0 10years from agreement

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Metals

10 0 10years from agreement

MachElec

10 0 10years from agreement

Transport

10 0 10years from agreement

Miscellan

Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure

10

Appendix C Alternative Specifications and Diagnostics

01

02

03

04

05

0607

0809

1011

12

13

14

1516

17

18

1920

2122

23

24

25

26272829

30

31

32

33

3435

36

37

3839

40

41

42

43

44

45

46

47

48

4950

5152

53

5455

56

57

58

59

60

6162

63

64

65

6667

68

6970

71

72

7374

75

76

78

79

80

81

82

8384

85

86

87

88

89

Animal

Vegetable

FoodProd

Minerals

Fuels

Chemicals

PlastiRub

HidesSkin

Wood

TextCloth

Footwear

StoneGlas

Metals

MachElec

Transport

minus10

minus05

00

05

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

LDC only

Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate

11

01

02

03

04

05

06

07

08

09

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

2829

30

31

3233

34

3536

37

38

39

4041

42

43

44

45

46

47

48

4950

5152

53

545556

57

5859

60

6162

63

64

65

6667

68

6970

71

72

73

74

75

76

78

79

80

81

8283

84

85

86

87

88

89

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

minus3

minus2

minus1

0

1

2

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

All WTO

(a) HS 2-digit level

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

01

02

03

04

05

0607

08

09

10

11

12

13

14

15

1617

18

19

20

21

22

23

24

2526

27

28

29

30

31

32

33

34

3536

37

38

39

40

41

42

4344

45

46

47

48

49

50

51

52

53

54

55

56

57

5859

6061 62

63

64

65

66

67

68

69

70

71

72

73

74

75

7678

79

80

81

8283

84 8586

87

88

89

minus4

minus2

0

2

4

Harmonized System 4minusdigit Product

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(b) HS 4-digit level

Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does

12

02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus03

00

03

minus05

00

05

10

00

05

10

15

iteration

(a) HS 2-digit (HMC)

0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)

500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000

00

04

08

12

00

05

10

minus10

minus05

00

iteration

(b) HS 4-digit (HMC)

020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus01

00

01

02

minus06

minus03

00

minus15

minus10

minus05

00

iteration

(c) HS 6-digit (VB)

Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification

13

• Title Page
• 1 Introduction
• 2 New Database Bilateral Product-level Applied Tariffs
• 3 The Effects of Regime Type on Trade Policy
• • 31 Monadic Analysis
  • • 311 Methodology
    • 312 Empirical Results
    • • 32 Dyadic Analysis
      • • 321 Methodology
        • 322 Empirical Results
        • • 4 Concluding Remarks
          • References
          • A Tariff-line Dataset
          • • A1 Bilateral Tariff Data Collection and Processing
            • • A11 WTO IDB Duty Collection and Processing
              • A12 UNCTAD TRAINS Duty Collection and Processing
              • • A2 Tariff Merging Algorithm
                • • B List of Bilateral FTAs
                  • C Alternative Specifications and Diagnostics

Table B1 ndash Continued from previous pageMontenegro Turkey 2008China Costa Rica 2009Canada Jordan 2010Chile Malaysia 2010Australia Malaysia 2011Costa Rica Singapore 2011South Korea Turkey 2011Montenegro Ukraine 2011Mauritius Turkey 2011Switzerland China 2012

Panel C Democratic PairsColombia Mexico 1993Canada Chile 1995Canada Israel 1995Chile Mexico 1997Israel Mexico 1998Canada Costa Rica 2000Chile Costa Rica 2000Chile El Salvador 2000Panama El Salvador 2001Chile South Korea 2002Mexico Uruguay 2002Chile United States 2002Australia United States 2003Japan Mexico 2003Sri Lanka Pakistan 2003Chile Japan 2005Mauritius Pakistan 2005Costa Rica Panama 2006Indonesia Japan 2006Japan Philippines 2006Chile Panama 2006Australia Chile 2007Canada Peru 2007Switzerland Japan 2007Chile Colombia 2007Guatemala Panama 2007Honduras Panama 2007Nicaragua Panama 2007Chile Peru 2007Peru United States 2007Chile Guatemala 2008Canada Colombia 2009

Table B1 Continued on next page

8

Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012

9

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Animal

dem demdem nondemnondem dem

Vegetable FoodProd Minerals

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Fuels Chemicals PlastiRub HidesSkin

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Wood TextCloth Footwear StoneGlas

10 0 10years from agreement

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Metals

10 0 10years from agreement

MachElec

10 0 10years from agreement

Transport

10 0 10years from agreement

Miscellan

Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure

10

Appendix C Alternative Specifications and Diagnostics

01

02

03

04

05

0607

0809

1011

12

13

14

1516

17

18

1920

2122

23

24

25

26272829

30

31

32

33

3435

36

37

3839

40

41

42

43

44

45

46

47

48

4950

5152

53

5455

56

57

58

59

60

6162

63

64

65

6667

68

6970

71

72

7374

75

76

78

79

80

81

82

8384

85

86

87

88

89

Animal

Vegetable

FoodProd

Minerals

Fuels

Chemicals

PlastiRub

HidesSkin

Wood

TextCloth

Footwear

StoneGlas

Metals

MachElec

Transport

minus10

minus05

00

05

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

LDC only

Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate

11

01

02

03

04

05

06

07

08

09

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

2829

30

31

3233

34

3536

37

38

39

4041

42

43

44

45

46

47

48

4950

5152

53

545556

57

5859

60

6162

63

64

65

6667

68

6970

71

72

73

74

75

76

78

79

80

81

8283

84

85

86

87

88

89

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

minus3

minus2

minus1

0

1

2

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

All WTO

(a) HS 2-digit level

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

01

02

03

04

05

0607

08

09

10

11

12

13

14

15

1617

18

19

20

21

22

23

24

2526

27

28

29

30

31

32

33

34

3536

37

38

39

40

41

42

4344

45

46

47

48

49

50

51

52

53

54

55

56

57

5859

6061 62

63

64

65

66

67

68

69

70

71

72

73

74

75

7678

79

80

81

8283

84 8586

87

88

89

minus4

minus2

0

2

4

Harmonized System 4minusdigit Product

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(b) HS 4-digit level

Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does

12

02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus03

00

03

minus05

00

05

10

00

05

10

15

iteration

(a) HS 2-digit (HMC)

0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)

500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000

00

04

08

12

00

05

10

minus10

minus05

00

iteration

(b) HS 4-digit (HMC)

020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus01

00

01

02

minus06

minus03

00

minus15

minus10

minus05

00

iteration

(c) HS 6-digit (VB)

Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification

13

• Title Page
• 1 Introduction
• 2 New Database Bilateral Product-level Applied Tariffs
• 3 The Effects of Regime Type on Trade Policy
• • 31 Monadic Analysis
  • • 311 Methodology
    • 312 Empirical Results
    • • 32 Dyadic Analysis
      • • 321 Methodology
        • 322 Empirical Results
        • • 4 Concluding Remarks
          • References
          • A Tariff-line Dataset
          • • A1 Bilateral Tariff Data Collection and Processing
            • • A11 WTO IDB Duty Collection and Processing
              • A12 UNCTAD TRAINS Duty Collection and Processing
              • • A2 Tariff Merging Algorithm
                • • B List of Bilateral FTAs
                  • C Alternative Specifications and Diagnostics

Table B1 ndash Continued from previous pageSouth Korea Peru 2009Colombia United States 2010Japan Peru 2010South Korea United States 2010Mexico Peru 2010Chile Nicaragua 2010Panama Peru 2010Panama United States 2010Canada Panama 2011Costa Rica Peru 2011Canada Honduras 2012Australia South Korea 2012

9

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Animal

dem demdem nondemnondem dem

Vegetable FoodProd Minerals

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Fuels Chemicals PlastiRub HidesSkin

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Wood TextCloth Footwear StoneGlas

10 0 10years from agreement

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Metals

10 0 10years from agreement

MachElec

10 0 10years from agreement

Transport

10 0 10years from agreement

Miscellan

Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure

10

Appendix C Alternative Specifications and Diagnostics

01

02

03

04

05

0607

0809

1011

12

13

14

1516

17

18

1920

2122

23

24

25

26272829

30

31

32

33

3435

36

37

3839

40

41

42

43

44

45

46

47

48

4950

5152

53

5455

56

57

58

59

60

6162

63

64

65

6667

68

6970

71

72

7374

75

76

78

79

80

81

82

8384

85

86

87

88

89

Animal

Vegetable

FoodProd

Minerals

Fuels

Chemicals

PlastiRub

HidesSkin

Wood

TextCloth

Footwear

StoneGlas

Metals

MachElec

Transport

minus10

minus05

00

05

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

LDC only

Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate

11

01

02

03

04

05

06

07

08

09

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

2829

30

31

3233

34

3536

37

38

39

4041

42

43

44

45

46

47

48

4950

5152

53

545556

57

5859

60

6162

63

64

65

6667

68

6970

71

72

73

74

75

76

78

79

80

81

8283

84

85

86

87

88

89

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

minus3

minus2

minus1

0

1

2

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

All WTO

(a) HS 2-digit level

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

01

02

03

04

05

0607

08

09

10

11

12

13

14

15

1617

18

19

20

21

22

23

24

2526

27

28

29

30

31

32

33

34

3536

37

38

39

40

41

42

4344

45

46

47

48

49

50

51

52

53

54

55

56

57

5859

6061 62

63

64

65

66

67

68

69

70

71

72

73

74

75

7678

79

80

81

8283

84 8586

87

88

89

minus4

minus2

0

2

4

Harmonized System 4minusdigit Product

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(b) HS 4-digit level

Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does

12

02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus03

00

03

minus05

00

05

10

00

05

10

15

iteration

(a) HS 2-digit (HMC)

0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)

500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000

00

04

08

12

00

05

10

minus10

minus05

00

iteration

(b) HS 4-digit (HMC)

020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus01

00

01

02

minus06

minus03

00

minus15

minus10

minus05

00

iteration

(c) HS 6-digit (VB)

Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification

13

• Title Page
• 1 Introduction
• 2 New Database Bilateral Product-level Applied Tariffs
• 3 The Effects of Regime Type on Trade Policy
• • 31 Monadic Analysis
  • • 311 Methodology
    • 312 Empirical Results
    • • 32 Dyadic Analysis
      • • 321 Methodology
        • 322 Empirical Results
        • • 4 Concluding Remarks
          • References
          • A Tariff-line Dataset
          • • A1 Bilateral Tariff Data Collection and Processing
            • • A11 WTO IDB Duty Collection and Processing
              • A12 UNCTAD TRAINS Duty Collection and Processing
              • • A2 Tariff Merging Algorithm
                • • B List of Bilateral FTAs
                  • C Alternative Specifications and Diagnostics

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Animal

dem demdem nondemnondem dem

Vegetable FoodProd Minerals

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Fuels Chemicals PlastiRub HidesSkin

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Wood TextCloth Footwear StoneGlas

10 0 10years from agreement

00

05

10

15

20

25

30

aver

age

duty

acro

ss p

rodu

cts

(log

)

Metals

10 0 10years from agreement

MachElec

10 0 10years from agreement

Transport

10 0 10years from agreement

Miscellan

Figure B1 Tariff Reduction Trajectories by Industry and Dyad Type This plot showsthe average logged HS 4-digit level tariffs between FTA partners over time across different industries (HS 2-digit)and dyad type For instance the solid red line indicates average industry-specific tariffs before and after an FTAwhen both FTA partners are democracies while the blue dashed line summarizes agreements where the importer isa democracy and the exporter is a non-democracy Due to the relative sparsity of free trade agreements betweennon-democracy pairs we exclude them from the set of dyads in this figure

10

Appendix C Alternative Specifications and Diagnostics

01

02

03

04

05

0607

0809

1011

12

13

14

1516

17

18

1920

2122

23

24

25

26272829

30

31

32

33

3435

36

37

3839

40

41

42

43

44

45

46

47

48

4950

5152

53

5455

56

57

58

59

60

6162

63

64

65

6667

68

6970

71

72

7374

75

76

78

79

80

81

82

8384

85

86

87

88

89

Animal

Vegetable

FoodProd

Minerals

Fuels

Chemicals

PlastiRub

HidesSkin

Wood

TextCloth

Footwear

StoneGlas

Metals

MachElec

Transport

minus10

minus05

00

05

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

LDC only

Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate

11

01

02

03

04

05

06

07

08

09

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

2829

30

31

3233

34

3536

37

38

39

4041

42

43

44

45

46

47

48

4950

5152

53

545556

57

5859

60

6162

63

64

65

6667

68

6970

71

72

73

74

75

76

78

79

80

81

8283

84

85

86

87

88

89

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

minus3

minus2

minus1

0

1

2

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

All WTO

(a) HS 2-digit level

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

01

02

03

04

05

0607

08

09

10

11

12

13

14

15

1617

18

19

20

21

22

23

24

2526

27

28

29

30

31

32

33

34

3536

37

38

39

40

41

42

4344

45

46

47

48

49

50

51

52

53

54

55

56

57

5859

6061 62

63

64

65

66

67

68

69

70

71

72

73

74

75

7678

79

80

81

8283

84 8586

87

88

89

minus4

minus2

0

2

4

Harmonized System 4minusdigit Product

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(b) HS 4-digit level

Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does

12

02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus03

00

03

minus05

00

05

10

00

05

10

15

iteration

(a) HS 2-digit (HMC)

0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)

500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000

00

04

08

12

00

05

10

minus10

minus05

00

iteration

(b) HS 4-digit (HMC)

020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus01

00

01

02

minus06

minus03

00

minus15

minus10

minus05

00

iteration

(c) HS 6-digit (VB)

Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification

13

• Title Page
• 1 Introduction
• 2 New Database Bilateral Product-level Applied Tariffs
• 3 The Effects of Regime Type on Trade Policy
• • 31 Monadic Analysis
  • • 311 Methodology
    • 312 Empirical Results
    • • 32 Dyadic Analysis
      • • 321 Methodology
        • 322 Empirical Results
        • • 4 Concluding Remarks
          • References
          • A Tariff-line Dataset
          • • A1 Bilateral Tariff Data Collection and Processing
            • • A11 WTO IDB Duty Collection and Processing
              • A12 UNCTAD TRAINS Duty Collection and Processing
              • • A2 Tariff Merging Algorithm
                • • B List of Bilateral FTAs
                  • C Alternative Specifications and Diagnostics

Appendix C Alternative Specifications and Diagnostics

01

02

03

04

05

0607

0809

1011

12

13

14

1516

17

18

1920

2122

23

24

25

26272829

30

31

32

33

3435

36

37

3839

40

41

42

43

44

45

46

47

48

4950

5152

53

5455

56

57

58

59

60

6162

63

64

65

6667

68

6970

71

72

7374

75

76

78

79

80

81

82

8384

85

86

87

88

89

Animal

Vegetable

FoodProd

Minerals

Fuels

Chemicals

PlastiRub

HidesSkin

Wood

TextCloth

Footwear

StoneGlas

Metals

MachElec

Transport

minus10

minus05

00

05

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

LDC only

Figure C1 Effect of Democracy on Log Tariffs Less-Developed Countries This plotpresents posterior means and 95 credible intervals for the estimated effects of democracy on trade policy for eachHS2 industry The Harmonized System 2-digit chapter codes are given at the bottom of each estimate

11

01

02

03

04

05

06

07

08

09

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

2829

30

31

3233

34

3536

37

38

39

4041

42

43

44

45

46

47

48

4950

5152

53

545556

57

5859

60

6162

63

64

65

6667

68

6970

71

72

73

74

75

76

78

79

80

81

8283

84

85

86

87

88

89

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

minus3

minus2

minus1

0

1

2

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

All WTO

(a) HS 2-digit level

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

01

02

03

04

05

0607

08

09

10

11

12

13

14

15

1617

18

19

20

21

22

23

24

2526

27

28

29

30

31

32

33

34

3536

37

38

39

40

41

42

4344

45

46

47

48

49

50

51

52

53

54

55

56

57

5859

6061 62

63

64

65

66

67

68

69

70

71

72

73

74

75

7678

79

80

81

8283

84 8586

87

88

89

minus4

minus2

0

2

4

Harmonized System 4minusdigit Product

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(b) HS 4-digit level

Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does

12

02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus03

00

03

minus05

00

05

10

00

05

10

15

iteration

(a) HS 2-digit (HMC)

0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)

500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000

00

04

08

12

00

05

10

minus10

minus05

00

iteration

(b) HS 4-digit (HMC)

020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus01

00

01

02

minus06

minus03

00

minus15

minus10

minus05

00

iteration

(c) HS 6-digit (VB)

Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification

13

• Title Page
• 1 Introduction
• 2 New Database Bilateral Product-level Applied Tariffs
• 3 The Effects of Regime Type on Trade Policy
• • 31 Monadic Analysis
  • • 311 Methodology
    • 312 Empirical Results
    • • 32 Dyadic Analysis
      • • 321 Methodology
        • 322 Empirical Results
        • • 4 Concluding Remarks
          • References
          • A Tariff-line Dataset
          • • A1 Bilateral Tariff Data Collection and Processing
            • • A11 WTO IDB Duty Collection and Processing
              • A12 UNCTAD TRAINS Duty Collection and Processing
              • • A2 Tariff Merging Algorithm
                • • B List of Bilateral FTAs
                  • C Alternative Specifications and Diagnostics

01

02

03

04

05

06

07

08

09

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

2829

30

31

3233

34

3536

37

38

39

4041

42

43

44

45

46

47

48

4950

5152

53

545556

57

5859

60

6162

63

64

65

6667

68

6970

71

72

73

74

75

76

78

79

80

81

8283

84

85

86

87

88

89

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

minus3

minus2

minus1

0

1

2

Harmonized System 2minusdigit Industry

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

All WTO

(a) HS 2-digit level

AnimalVegetable

FoodProdMinerals

FuelsChemicals

PlastiRubHidesSkin

WoodTextCloth

FootwearStoneGlas

MetalsMachElec

Transport

01

02

03

04

05

0607

08

09

10

11

12

13

14

15

1617

18

19

20

21

22

23

24

2526

27

28

29

30

31

32

33

34

3536

37

38

39

40

41

42

4344

45

46

47

48

49

50

51

52

53

54

55

56

57

5859

6061 62

63

64

65

66

67

68

69

70

71

72

73

74

75

7678

79

80

81

8283

84 8586

87

88

89

minus4

minus2

0

2

4

Harmonized System 4minusdigit Product

larrD

emoc

raci

es h

ave

low

er ta

riffs

Dem

ocra

cies

hav

e hi

gher

tarif

fsrarr

(b) HS 4-digit level

Figure C2 Effect of Democracy on Log Tariffs Variational Bayes Results This plot presentsposterior means and 95 credible intervals for the estimated effects of democracy on tariff rates for all HS2 chaptersand HS4 headings using Variational Bayes (VB) rather than Hamiltonian Monte Carlo (HMC) The advantage of VBover HMC is a significant gain in computational speed which facilitates easier replicability VB takes approximately20 minutes to converge for both the HS2 and HS4 monadic models while HMC takes 1 day and approximately3 weeks for the HS2 and HS4 monadic models respectively The disadvantage is that there no exact convergenceguarantees A comparison with Figures 3 and 4 shows that our substantive findings do not significantly changehowever the magnitude of effect sizes does

12

02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus03

00

03

minus05

00

05

10

00

05

10

15

iteration

(a) HS 2-digit (HMC)

0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)

500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000

00

04

08

12

00

05

10

minus10

minus05

00

iteration

(b) HS 4-digit (HMC)

020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus01

00

01

02

minus06

minus03

00

minus15

minus10

minus05

00

iteration

(c) HS 6-digit (VB)

Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification

13

• Title Page
• 1 Introduction
• 2 New Database Bilateral Product-level Applied Tariffs
• 3 The Effects of Regime Type on Trade Policy
• • 31 Monadic Analysis
  • • 311 Methodology
    • 312 Empirical Results
    • • 32 Dyadic Analysis
      • • 321 Methodology
        • 322 Empirical Results
        • • 4 Concluding Remarks
          • References
          • A Tariff-line Dataset
          • • A1 Bilateral Tariff Data Collection and Processing
            • • A11 WTO IDB Duty Collection and Processing
              • A12 UNCTAD TRAINS Duty Collection and Processing
              • • A2 Tariff Merging Algorithm
                • • B List of Bilateral FTAs
                  • C Alternative Specifications and Diagnostics

02 (Meat)Intercept 52 (Cotton)Polity 73 (Steel)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus03

00

03

minus05

00

05

10

00

05

10

15

iteration

(a) HS 2-digit (HMC)

0201 (beef)Intercept 5204 (sewing thread)Polity 7302 (railway tracks)log(GDPPC)

500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000

00

04

08

12

00

05

10

minus10

minus05

00

iteration

(b) HS 4-digit (HMC)

020130 (boneless beef)Intercept 520420 (retail sewing thread)Polity 730210 (rails)log(GDPPC)

200 400 600 800 1000 200 400 600 800 1000 200 400 600 800 1000

minus01

00

01

02

minus06

minus03

00

minus15

minus10

minus05

00

iteration

(c) HS 6-digit (VB)

Figure C3 Effect of Democracy on Log Tariffs Model Traceplots Panels show the fourMarkov chain traces of all three product-specific coefficients estimated for a set of representative products p (γp inequation 1) Panel (a) corresponds to a subset of model results in Figure 3 panel (b) corresponds to Figure 4 andpanel (c) corresponds to Figures 5 and 6 Thus each row shows the same product-specific coefficient at varying levelsof Harmonized System classification

13

• Title Page
• 1 Introduction
• 2 New Database Bilateral Product-level Applied Tariffs
• 3 The Effects of Regime Type on Trade Policy
• • 31 Monadic Analysis
  • • 311 Methodology
    • 312 Empirical Results
    • • 32 Dyadic Analysis
      • • 321 Methodology
        • 322 Empirical Results
        • • 4 Concluding Remarks
          • References
          • A Tariff-line Dataset
          • • A1 Bilateral Tariff Data Collection and Processing
            • • A11 WTO IDB Duty Collection and Processing
              • A12 UNCTAD TRAINS Duty Collection and Processing
              • • A2 Tariff Merging Algorithm
                • • B List of Bilateral FTAs
                  • C Alternative Specifications and Diagnostics