The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
The Employment
Consequences of
Anti-Dumping Tariffs:
Lessons from Brazil
Gustavo de Souza and Haishi Li
October 5, 2022
WP 2022-46
https://doi.org/10.21033/wp-2022-46
*Working papers are not edited, and all opinions are the responsibility
of the author(s). The views expressed do not necessarily reflect
the views of the Federal Reserve Bank of Chicago or the Federal
Reserve System.
The Employment Consequences of Anti-Dumping
Tariffs: Lessons from Brazil*
Gustavo de Souzaand Haishi Li
October 5, 2022
Abstract
Can anti-dumping tariffs increase employment?
We compile data on all anti-
dumping (AD) investigations in Brazil matching it to firm-level administrative employment information. Using difference-in-differences, we find that, an AD tariff decreases
imports and increases employment in the protected sector. Moreover, downstream
firms decrease employment, while upstream ones are not affected. To quantify the aggregate effect of these tariffs, we build a model with international trade, input-output
linkages, and labor force participation. We show that the Brazilian AD policy increased
employment by 0.06%, but they decreased welfare by 2.4%. Using tariffs, the government can increase employment by as much as 2.8%.
Keywords: employment, tariffs, anti-dumping, international trade
JEL Codes: F13, F16
* We thank for their valuable comments Rodrigo Adao, Tibor Besedeš, Chad Bown, Jonathan Dingel,
**
Jingting Fan, Armen Khederlarian, Bingjing Li, Justin Pierce, Giovanni Peri, Heiwai Tang, Chang Sun, Felix
Tintelnot, Yoto Yotov, David Nagy, Giacomo Ponzetto, Greg Kaplan, Andrés Rodrı́guez-Clare, and Esteban
Rossi-Hansberg. And seminar participants at the China Trade Research Group, Latin America Meeting
of Econometric Society, Midwest Trade Group, University of California San Diego, Shanghai University of
Finance and Economics, Virtual International Trade and Macro Seminar, Urban Economics Association,
Chicago Economics Trade Group, IIES Labor Group, Centre de Recerca en Economia Internacional, Federal
Reserve Bank of Chicago and Katholieke Universiteit Leuven. All remaining errors are ours.
Federal Reserve Bank of Chicago. E-mail: gustavo@microtomacro.net
University of California, San Diego and University of Hong Kong. E-mail: hal101@ucsd.edu
1
1
Introduction
With the promise of ‘bringing jobs back’, tariffs are usually advocated as a tool to increase
local employment. Despite tariffs’ relevance to policy and their prominence in political
debates, economists still do not know how tariffs in general and on particular products affect
firms, wages, and employment.
On the one hand, tariffs shift demand for foreign products to those produced in the home
market. Therefore, through this shift in demand, the national producer and sectors upstream to it (that is, the sectors that provide inputs to the national producer) could increase
production and employment. On the other hand, downstream sectors (the ones that use the
tariffed good as an input) face higher costs, which could lead to lower employment among
them. Therefore, the final effect of anti-dumping (AD) tariffs on employment will depend
on the employment elasticity of the national producer, upstream sectors, and downstream
sectors.
In this paper, we ask: What is the effect of tariffs on aggregate employment and wages?
And how does the effect of tariffs propagate through the value chain? We use data and
a model to show that import tariffs might increase employment, depending on the tariffed
product’s position along the value chain.
To answer our questions, we start by collecting information on all anti-dumping investigations initiated by Brazil.1 Next, we link each investigation to a national producer, an
upstream sector, and a downstream sector. For a subset of the investigations, we also get
the name of the firm that filed the complaint. This information is then matched to an
employer-employee dataset that contains details on wages and employment at the firm level.
To identify the causal effect of tariffs, we implement a difference-in-differences strategy.
The treatment group is the set of products whose anti-dumping investigations led to a tariff
increase. The control group is the set of products whose anti-dumping investigations did not
result in a tariff change.
As usual in difference-in-differences, the identifying assumption is of parallel trends be1
Among global economies, Brazil ranks the sixth in terms of the number of anti-dumping (AD) investigations launched, only after the United States, India, European Union, Canada, and Argentina (Bown 2005).
Also see https://www.reuters.com/article/us-brazil-china-wto-idUSKCN11F2MS.
2
tween the treatment and control groups. In fact, in our setting, the assumption of parallel
trends is supported by anti-dumping regulations. According to World Trade Organization
(WTO) regulations, the decision to impose an AD tariff depends on international prices of
the investigated product before the investigation. Therefore, conditional on a product being
investigated, the decision to impose a tariff and its size are made based on pre-determined
variables, which can be teased out with fixed effects. Importantly, AD tariffs should not
depend on labor market trends or political connections.
Supporting our identification strategy, we show that political connections cannot predict
AD tariffs but international prices can, as stipulated by WTO regulations. We show that
treatment and control groups are equally likely to make campaign contributions or to receive
procurement contracts, subsidies, tax breaks, or subsidized loans from the government, which
shows that AD tariffs are not targeted to protect politically connected sectors. We also show
that AD tariffs do not correlate with preferential trade agreements or MFN tariffs. Moreover,
we can predict Brazilian AD tariffs with an R-squared above 0.95 using only international
prices, which is expected because AD tariffs should be a non-linear function of international
prices. Therefore, the data strongly indicates that AD tariffs in Brazil are imposed according
to WTO regulations, which supports our identifying assumption.
To further validate our identification strategy, we implement a battery of exercises and
robustness checks. First, we show that pre-period parallel trends hold for all the variables we
consider. Second, we also show that our results cannot be explained by other major shocks
hitting the Brazilian economy, such as the Brazilian trade liberalization or fluctuations in the
exchange rate. Third, we implement two placebo tests showing that our results are not driven
by sectoral or labor market trends. Fourth, we also found that adding or removing controls
from our main specification does not change the results. Fifth, we show that running our
main regressions at the region level, instead of the firm level, delivers similar results. Finally,
we also find the same results using an instrumented diff-in-diff with pre-period international
prices as instrument.
We show that anti-dumping tariffs decrease imports and increase employment in the
protected sector without affecting upstream firms. Yet, employment and wages in the downstream firms decrease. When anti-dumping tariffs are imposed on a product, imports of that
3
product fall. A 100% ad valorem anti-dumping tariff generates a 25% drop in imports. Moreover, in contrast to Flaaen et al. (2019), we do not find any corresponding increase in imports
from other locations. Employment increased at the national producer when anti-dumping
tariffs are imposed against its foreign competitors. A 100% ad valorem tariff generates a
1.8% employment increase among firms shielded from international competition. Upstream
firms are not affected by anti-dumping tariffs. This phenomenon can be explained by an
increase in imports of inputs by the national producer. A 100% ad valorem tariff increases
imports of inputs used by the national producer by 2.8%. Therefore, local firms do not
benefit because the national producer is sourcing its inputs internationally. Downstream
firms, as expected, are negatively affected. A 100% ad valorem tariff on all inputs of a firm
decreases employment by 3.8%.
The empirical results are informative about firm-level responses to AD tariffs but are
silent about aggregate effects. To make aggregate quantitative predictions, we build a small
open economy model that features international trade, input-output linkages, and labor
force participation. We show that the key parameters from the model can be identified
from the estimated reduced-form elasticities. In the model, workers choose to work between
different sectors or to stay outside the labor force. To produce, firms use labor and input
from all sectors. The sectoral input is supplied by imperfectly substitutable domestic and
foreign producers. The Frisch elasticity, the Armington trade elasticity2 and the elasticity
of substitution across product lines are estimated from the effect that anti-dumping tariffs
have on employment and international trade.
From the quantitative model, we conclude that the Brazilian anti-dumping policy increased employment and GDP, but the effect of tariffs on employment depends on the position
of the tariffed product along the value chain. We find that the Brazilian anti-dumping policy
increased employment and gross domestic product (GDP) by 0.06% and 0.05%, respectively,
with a decrease in consumption by 2.43%. Input-output linkages affect these aggregate predictions. A model without the input-output linkages predicts employment and GDP to each
rise by 0.15%–these jumps constitute overestimations of 100%. Moreover, the aggregate effect of a tariff depends on the position of the product along the value chain. Imposing tariffs
2
The Armington trade elasticity refers to the elasticity of substitution across countries.
4
that protect computer, electrical and machinery sectors (which are further down the value
chain) increases aggregate employment. However, imposing tariffs that protect agriculture
and mining sectors (which are relatively upstream in the value chain) decreases aggregate
employment. These results indicate that, if the goal of tariffs is to increase employment,
they should be targeted at items produced by sectors at the end of the value chain that use
inputs from a wide range of sectors.
We show that, if the goal is to maximize employment, tariffs should be higher on products made by downstream sectors than on products made by upstream sectors; i.e., they
should follow “tariff-escalation”.3 These tariffs increase aggregate employment by 2.8% but
substantially decrease welfare by 15.9%. With this exercise, we highlight the important
trade-off between promoting employment and raising consumer welfare that policymakers
face when setting tariffs.
This paper contributes to the literature that studies the effect of AD tariffs.4 The literature has shown that AD tariffs reduce imports; i.e., they cause trade depression.5 But
the evidence on trade diversion–the impact of tariffs on the imports of other products and
countries–is mixed. Prusa (1997), Bown and Crowley (2006), Bown and Crowley (2007),
Baylis and Perloff (2010), and Flaaen et al. (2019) find that AD tariffs increase imports from
non-targeted countries, while Konings et al. (2001) and Durling and Prusa (2006) do not
find a significant third-country effect. This literature has also found that AD tariffs affect
firm performance in the protected sector (Konings and Vandenbussche 2008, Pierce 2011,
and Jabbour et al. 2019), as well as employment (Trimarchi 2020, Barattieri and Cacciatore
2020, and Bown et al. 2021).
We make two contributions to this literature on AD tariff effects—with our empirical
strategy and with our quantitative model. First, we propose a new identification strategy
3
Tariff escalation refers to the fact that tariffs on intermediate goods are generally lower than those on
final goods. Travis et al. (1964), Balassa (1965), Bown and Crowley (2016), and Shapiro (2021) document
tariff-escalation for most favored nation (MFN) tariffs. Antràs et al. (2022) and Caliendo et al. (2021)
explain this pattern with a government that maximizes consumer welfare and the free entry-exit of firms
into upstream and downstream sectors. We provide an alternative explanation to it with a government
interested in maximizing employment.
4
For a review of the literature, see Blonigen and Prusa (2016)
5
See Staiger and Wolak (1994), Lloyd et al. (1998), Prusa (1997), Vandenbussche and Zanardi (2010),
Irwin (2014), Besedeš and Prusa (2017), and Sandkamp (2020).
5
exploiting the institutions of AD regulation. Unlike previous research, this strategy allows
us to tease out the effects of uncertainty caused by AD investigations.6 Second, we examine
more than just the effect of tariffs on trade. To the best of our knowledge, we provide the
first general equilibrium analysis of the aggregate employment effect of AD tariffs that takes
into account all midstream, upstream, and downstream impacts.7
Our paper also contributes to the literature that studies the labor market consequences
of international trade. Several papers within this literature have found that import competition leads to a decrease in employment and wages (Trefler 2004, Autor et al. 2013, Autor
et al. 2014, Pierce and Schott 2016, Dix-Carneiro and Kovak 2015, Dix-Carneiro and Kovak 2017, and Devlin et al. 2021), and affects employment in upstream and downstream
sectors (Acemoglu et al. 2014 and Pierce and Schott 2016). While these empirical works
largely agree that tariffs cause employment declines in downstream sectors, they disagree
about the impact on the protected sector and its propagation upstream, casting uncertainty
on the aggregate effect of tariffs.8 Quantitative works, including Caliendo et al. (2019) and
Rodrı́guez-Clare et al. (2020), predict that tariffs can increase total employment, whereas
Barattieri et al. (2021) predicts otherwise.
We contribute to this literature studying international trade’s impact on the domestic
labor market in the following three ways. First, we develop a new empirical strategy and
provide new evidence for the employment effects of trade policies along the supply chain.
Second, basing our model on empirical estimates, we find moderate aggregate employment
gains from AD tariffs, yet we highlight the importance of the input-output linkages to such
6
For example, Staiger and Wolak (1994) and Prusa (1997), among others, find that AD investigations
can reduce trade even if they do not conclude with tariffs.
7
Previous works that study the general equilibrium effect of AD tariffs focus on welfare. Using a small
open economy model with firm dynamics, Ruhl (2014) finds significant U.S. welfare loss due to AD tariffs.
Gallaway et al. (1999) also evaluates the welfare loss from AD tariffs with a computable general equilibrium
model.
8
Flaaen and Pierce (2019) and Trimarchi (2020) find that 2018-19 U.S. tariffs weakly increase employment
in the protected sectors, but Barattieri and Cacciatore (2020) and Bown et al. (2021) find that anti-dumping
tariffs have insignificant employment effect on the protected sectors. Flaaen and Pierce (2019), Barattieri
and Cacciatore (2020), and Bown et al. (2021) find significant declines in employment in the downstream
sectors. Blonigen (2016) studies the impact of industrial policy on downstream firms’ exports. Huang et al.
(2019) studies how 2018-19 U.S. tariffs affect the financial market performance of American and Chinese
firms in upstream and downstream sectors. Conconi et al. (2018) studies how input requirements by a free
trade agreement divert international trade. Handley et al. (2020) finds that 2018-19 U.S. tariffs significantly
reduce exports by downstream firms.
6
an aggregate employment effect. Third, we find that to maximize employment, tariffs should
be set higher on goods produced by downstream sectors than on goods produced by upstream
sectors and that these tariffs cause a significant loss in consumer welfare.
The rest of the paper proceeds as follows. In Section 2 we go over the WTO AD rules, the
practice of AD investigations in Brazil, and the data used in this paper. Then, in Section 3
we explain our empirical strategy. In Section 4 we present the main empirical results. Next,
in Section 5 we introduce the model. In Section 6 we describe the procedure to estimate
the model. In Section 7 we show the quantitative results. Finally, in Section 8 we state our
conclusions.
2
Institutions and Data
2.1
Anti-Dumping Investigations
Dumping is defined as an international price discrimination where the exporter charges
a lower price in the destination market that in their home market. According to WTO
regulations, the destination market harmed by dumping is allowed to set an AD tariff to
exactly offset this price difference. The WTO AD regulations, which Brazil follows, define
three steps for the creation of an AD tariff: (1) firms harmed by dumping file a complaint
to the Ministry of Economy, (2) the government opens an investigation into whether the
foreign competitor engaged in dumping, and (3) the government decides whether to impose
the AD tariff and, if so, its size.9
The process starts with a domestic firm or a group of domestic firms filing a complaint
with the Ministry of Economy. The complaint must show that the sector is harmed by
foreign dumping practices. Firms must present evidence that they experience a decrease in
profits, sales, or wages, and link this to increased import competition from an international
competitor. Moreover, the international competitor must have both an increasing volume
of the good it’s exporting to Brazil and a decreasing price on that good in Brazil. This
suggests that the sales and price of investigated and non-investigated products may have
9
See the Agreement on Implementation of Article VI of the GATT 1994 (The Anti-Dumping Agreement,
https://www.wto.org/english/docs_e/legal_e/19-adp_01_e.htm).
7
different trends. We discuss this further in Section A.3.
The government, upon receiving the complaint, determines whether it should open an
investigation or dismiss the case. This decision is made based on whether there is enough
proof that links the national supplier’s decline in economic performance to increased imports
from the international competitor. We only consider the cases in which an investigation is
opened.
After the government opens an investigation, it identifies the price of the imported product before the investigation in its home market (called the “normal value”) and in Brazil.
If the imported product comes from a non-market economy, the normal value is calculated
using the pre-investigation price in a third market.101112
If the government finds that the foreign competitor is involved in unfair trade practices
by charging a lower price in Brazil than its normal value, the government will create an AD
tariff to equate the Brazilian post-tariff price to the normal value. Therefore, the AD tariff
is set based on pre-determined price differences. The AD tariff, once imposed, lasts for five
years and is then reevaluated.
2.2
Data
We use four datasets. They contain information on AD tariffs, product-level imports, firmlevel employment, and firm-level imports. For information on AD tariffs and investigations,
we use the Global Anti-dumping Database (Bown 2005). For each AD investigation in
Brazil, the Global Anti-dumping Database contains the product name and classification
investigated, the country of origin, the start and conclusion dates of each investigation, and
the measures taken. Section A.1 presents a set of summary statistics of AD investigations
in Brazil.
Data on imports comes from the Secretary of International Trade of the Ministry of
Economy in Brazil. It provides monthly statistics on imports and exports for Brazil at the
10
The Brazilian government considers only China and Vietnam as non-market economies. In those cases,
the third country chosen as a reference for the normal value will depend on data availability.
11
In general, the normal value in a non-market economy can also be estimated using estimates of the
production cost, but this method is not used in Brazil.
12
See WTO’s Technical Information on Anti-dumping (https://www.wto.org/english/tratop_e/adp_
e/adp_info_e.htm).
8
product level. This is used to understand the effect of tariffs on trade. The third database,
RAIS, covers employment information of Brazilian firms. It is a yearly employer-employee
matched dataset containing information on wages, hours, occupation, and demographics of
workers. It also contains data on the sector and location of the firm. Using a concordance
table provided by the Brazilian Secretary of International Trade, we link each AD investigation to its sector. This allows us to study how tariff increases affect employment in domestic
sectors.
Throughout the paper, we constrain the analysis to firms with more than one worker that
have been active for more than 10 years. The goal is to prevent changes in the composition
of firms from driving the results.13 We constrain our analysis from 1995 to 2016. We also
drop from the empirical analysis the service sector and the government sector.
3
Empirical Strategy
We use difference-in-differences to identify the effect of AD tariffs on trade and employment.
The control group is the set of products with AD investigations that did not result in tariff
changes. The treatment group is the set of products with AD investigations that led to tariff
increases.14
3.1
Validation
In this section, we show that parallel trends are supported by institutional facts and exogeneity tests. Moreover, AD tariffs do not correlate with other policies implemented in the
period, political connections, or sectoral shocks.
13
One could be worried that this choice could lead to sample selection. Indeed, that would be the case if
AD tariffs could lead firms to enter or exit the market. On section A.5.3, we show that results are the same
if we keep all the firms on the sample. We also show that AD tariffs did not led to firm entry or exit.
14
Notice that to identify the effect of tariffs, one cannot compare products with AD investigations against
products that do not have AD investigations. This happens because of two factors–selection and the effect
of tariff uncertainty. The first factor stems from the fact that AD investigations are not random. As Section
A.3 shows, investigated products have a lower price and higher volume than non-investigated ones. They
are also in a decreasing price and increasing volume trend at the time of the investigation. Furthermore, AD
complaints are filed by expanding firms. Therefore, products and sectors undergoing an AD investigation
are in a special trend. The second factor is the trade policy uncertainty created by AD investigations. As
highlighted by Staiger and Wolak (1994), Prusa (2001), Lu et al. (2013), Besedeš and Prusa (2017), among
others, the investigation itself might have effects on trade and employment.
9
Institutions As discussed in Section 2, conditional on an AD investigation, the decision
to impose a tariff is a function of pre-determined characteristics outside of Brazil. Therefore,
conditional on an investigation, the probability to be treated should be fully captured by
product/firm-level fixed effects. Therefore, sectors that produce goods under AD investigations should have similar labor market trends.
Exogeneity Test In Section A.4.1, as the WTO regulations suggest, we show that prices
outside of Brazil can predict tariffs. We can predict Brazilian AD tariffs with an R-squared
above 0.95 using the distribution of international prices and AD tariffs imposed by other
countries.15 These results suggest that it is very unlikely that labor trends are affecting the
AD policy of the Brazilian government, which supports our assumption of parallel trends.
Placebo Tests To further guarantee that treatment and control groups do not differ in
underlying shocks or trends, we implement two placebo tests, as shown in Section A.4.2.
First, we show that tariffs do not correlate with employment changes in sectors that are
not subject to AD tariffs but have similar employment trends. This placebo test indicates
that the results are not driven by sectoral shocks to sectors following a certain employment
trend. In our second placebo test, we show that there is no correlation of AD tariffs with
employment changes 5 years before the AD tariff is implemented, supporting the notion that
the identifying effect is not coming from labor market trends.
Other Policies and Political Connection We also show, in Section A.4.3, that the AD
policy does not correlate with political connections, public procurement, subsidies from the
government, tax breaks, or other tariffs.
15
As discussed in Section A.4.1, AD tariffs should be a function of the import’s price in Brazil and in the
home country during the pre-period. But because the price in the home country are not observed, we proxy
for this price with the AD tariff of other countries and the distribution of prices of the investigated product.
The details of our test can be found in Section A.4.1.
10
3.2
3.2.1
Empirical Model
Imports
We use the following empirical model to identify the effect of imposing an AD tariff τp,c on
imports of product p from country c:16
yp,c,q = θτp,c,q + βIp,c,q {After AD} + ηp,c + ηq,c + p,c,q ,
(1)
where yp,c,q is the log of total imports of product p from country c in quarter q, Ip,c,q {After AD}
is a dummy taking 1 after the beginning of the first investigation, τp,c is the ad valorem AD
tariff imposed (for the control group, this variable takes the value of zero), ηp,c is a productcountry fixed effect removing any level differences between treatment and control, and ηq,c
is a quarter-country fixed effect. The parameter of interest is θ, which captures the effect
of tariff τp,c . β, which is common to the treatment and control groups, captures the effect
of being exposed to an AD investigation. In this regression we constrain the sample to
product-destination pairs that have at least one AD investigations.
To test for parallel trends in the pre-period, we use the following specification:17
yp,c,q =
X
j
θj τp,c,first Ip,c,q {j Qrt. to AD} +
X
βj Ip,c,q {j Qrt. to AD} + ηp,c + ηq,c + p,c,q ,
j
(2)
where Ip,c,q {j Qrt. to AD} is a dummy which takes one if quarter q is j quarters to the
beginning of the first AD investigation, τp,c,first is the first AD tariff imposed on product p
from country c, and θj captures the dynamic effects of the first AD tariff. Parallel trends in
the pre-period imply that θj = 0 for all j < 0.
16
In our analysis each product refers to an 8-digit Nomenclatura Comum do Mercosul (NCM) code. The
first 6 digits of the NCM code are the same as those of a Harmonized System (HS) code. The Brazilian
government adds two additional digits to improve granularity.
17
In these tests, we consider only the first AD investigation, as is standard in the literature, to ensure that
there is no confounding investigation in the pre-period.
11
3.3
Midstream Firms
We use difference-in-differences to identify the effect of AD tariffs on the national producer,
which we also refer to as a midstream firm. The treatment group is the set of firms whose
products faced AD investigations that led to tariff increases. The control group is the set of
firms whose products never got protection with tariff changes despite being investigated.
The main specification is:
mid
0
yi,s,t = θτs,t
+ βIs,t {After AD} + Xi,s,t
κ + ηi + ηt + i,t ,
(3)
mid
is the average of
where yi,s,t is a labor outcome of firm i in sector s in year t18 and where τs,t
ad-valorem tariffs imposed on products produced by sector s in year t. Firms in the control
mid
increases from zero
group have zero tariffs, but for the treatment group the variable τs,t
after the decision of the first investigation and keeps changing as AD tariffs are imposed
or removed. The variable Is,t {After AD} is a dummy that takes one after the first AD
investigation. It captures the effect of being exposed to an AD investigation and any other
trend that leads to the investigation. Finally, ηi is a firm fixed effect and ηt is a time fixed
effect.
To test for pre-period parallel trends, we use the following model:
yi,s,t =
X
mid
θj τs,f
irst × Is,t {j Yrs. to AD} +
j
q
X
βj Is,t {j Yrs. to AD} + ηi + ηt + i,t , (4)
j=−q
mid
where τs,f
irst is the first AD tariff imposed on products of sector s and where Is,t {j Yrs. to AD}
is a dummy that takes one if year t is j years before the beginning of the investigation that
results in the first AD tariff.
3.3.1
Main Downstream Firms
To identify the effect of tariffs on downstream firms, we link each sector protected by an
AD tariff to its main consumer. The downstream sector is the one that buys the largest
18
Each sector in Brazil has a 4-digit sector code. There are a total of 297 non-service sectors in Brazil
that could be protected against dumping by foreign competitors with tariff increases. See Section A.2 for
the detailed definition.
12
share of the reference sector’s production.19 By linking each sector to one downstream, we
are able to reproduce the same clean identification strategy we use for the midstream sector.
The treatment group consists of the firms in the sector whose main supplier had an AD
investigation for one of its products that led to a tariff increase, while the control group
consists of the firms in the sector whose main supplier never got protection with an AD
tariff despite being investigated.20
We chose not to implement a difference-in-differences with all firms downstream to an
investigated firm because we would not be able to test for pre-period parallel trends. According to our input-output table, almost all sectors buy inputs from all other sectors. Therefore,
every firm is downstream to a treatment firm and downstream to a control firm. In this case,
all the firms would be both in the treatment group and in the control group as soon as the
first AD tariff is imposed.
For each sector, we create an exposure measure based on the input cost increase caused
by the AD tariff. The increase in input costs due to tariff τs,t on sector d(s), which is
downstream to sector s, is:
down
τ̃d(s),t
=
Input Demand of Sector d(s) from Sector s
× τs,t ,
Aggregate Input Demand of Sector d(s)
where τs,t is the average AD tariff that affects sector s in year t. The main downstream
specification is:
down
0
yi,d(s),t = θτ̃d(s),t
+ βIs,t {After AD} + Xi,d(s),t
κ + ηi + ηt + i,t ,
(5)
where yi,d(s),t is a labor outcome of firm i in sector d(s) (which buys inputs from sector s),
19
Section A.2 describes how we construct an input-output table at the level of the 4-digit sector code
for Brazil and formally defines main downstream (and upstream) sectors. A similar empirical strategy is
also employed by Feng and Li (2021) who examine how the impact of climate disasters propagates to main
upstream and main downstream countries.
20
It is important to notice that, as is usual in difference-in-differences, we are recovering the relative effect
of the tariff. All firms are affected through the input-output connections and other general equilibrium
effects. Still, firms that are the main consumers of a product with a tariff hike, should be relatively more
affected than others. In the sections below, we use a model to match these estimated relative effects by
running the same regressions in the model. In this way, we recover the aggregate effect of tariffs common to
all firms. In Section A.5.2, we consider a regression that includes all midstream, weighted average upstream,
and weighted average downstream tariffs.
13
0
τ̃d(s),t is the change in input cost caused by an AD tariff in sector s, Xi,d(s),t
is a set of controls,
ηi is a firm fixed effect, and ηt is a year fixed effect.
To test for parallel trends between the control and the treatment group in the pre-period,
we use the following specification:
yi,d(s),t =
X
down
θj τ̃d(s)
× Is,t {j Yrs. to AD} +
q
X
βj Is,t {j Yrs. to AD} + ηi + ηt + i,t , (6)
j=−q
j
down
where τ̃d(s)
is the increase in input cost caused by the first AD tariff imposed on a product
produced by sector s, and where Is,t {j Yrs. to AD} is a dummy that takes one if year t is j
years before the beginning that results in the investigation of the first AD tariff.
3.3.2
Main Upstream Firms
We implement the same identification strategy at the main supplier of firms with an AD
investigation. We calculate the exposure measure at upstream firms as
up
τ̃u(s),t
=
Sales to Sector s from Sector u(s)
× τs,t ,
Production of Sector u(s)
where τs,t is the average AD tariff affecting sector s in year t. The main model is
up
0
yi,u(s),t = θτ̃u(s),t
+ βIs,t {After AD} + Xi,u(s),t
κ + ηi + ηt + i,t ,
(7)
where yi,u(s),t is a labor outcome of firm i in sector u(s) and year t. The other variables are
as described earlier. To test for pre-period parallel trends, we use an equation similar to 6.
4
4.1
Results
Imports
According to Figure 1, which shows the dynamic effects of the initial AD tariff on log imports
in dollars, AD tariffs cause a drop in imports. In the quarters before the announcement of the
tariff increase, the control and treatment groups had a similar import trend. This abruptly
14
changed when the investigation began; 10 quarters later, a 100% marginal tariff led to a
decrease of approximately 50% in imports.
Tariffs cause a drop in the quantity imported but but do not affect prices, according to
Table 1. Using variation from all AD tariffs, Column 1 of Table 1 shows that a 100% increase
in tariffs leads to a drop of 25% in imports. According to columns 2 and 3, which show the
effect of tariffs on the quantity imported and on the price of imported goods, the drop can be
explained by a decrease in quantity imported. The lack of price effect, displayed in column
3 of Table 1, indicates that Brazilian demand for international goods is too small to have a
significant effect on international prices.21
Figure 1: Effect of AD Tariffs on Imports
Description: This figure contains the coefficients of the effect of AD tariff on imports using the dynamic model 2 plotted
against quarters to the beginning of the investigation in the x-axis. Imports are measured in freight on board (FOB) current
dollars at the NCM product code level. Import data is from the Secretary of International Trade of the Ministry of Economy,
and AD data is from the Global Anti-dumping Database. The sample is composed of product-origin that had at least one AD
investigation. The shaded area contains the 95% confidence interval. Standard errors are clustered at the product-origin level.
We find limited evidence for trade diversion; i.e., firms do not shift imports of the tariffed
goods to another country. We run a regression of imports of the tariffed products from
countries that are not affected by AD investigations on product-level AD tariff. Table 2
21
This indicates that we find a complete pass-through of AD tariffs in Brazil.This finding is consistent
with Blonigen and Haynes (2002), Sandkamp (2020), and the recent works that study 2018-19 U.S. tariffs
(Amiti (2019), Amiti et al. (2020), Fajgelbaum et al. (2020), and Cavallo et al. (2021)).
15
Table 1: Effect of AD Tariffs on International Trade
τp,c,t
(1)
log(Value Imports)
-0.259***
(0.0811)
(2)
log(Quantity Imports)
-0.273***
(0.0999)
(3)
log(Price)
0.0157
(0.0428)
20803
0.635
12.703
.18
20733
0.652
11.654
.18
20732
0.787
1.077
.18
N
R2
Mean Dep. Var
Mean Ind. Var
Description: This table presents the estimated parameters of model 1. The sample is composed of
product-origin that had at least one AD investigation. log(Value Imports) is the log of FOB current
dollars imports at the NCM level. log(Quantity Imports) is the log of quantity imported, and log(Price)
is the log of price per-unit. Import data is from the Secretary of International Trade of the Ministry of
Economy, and AD data is from the Global Anti-dumping Database. The sample runs from 1995 to 2016.
Standard errors are clustered at the product-origin level.
shows that AD tariffs do not affect imports from other countries.
We also investigate whether firms shift from importing the tariffed products to importing
other products. We run a regression of imports of the products that do not face AD investigations but fall within the same 4-digit sector impacted by the product-level AD tariff. Table
2 shows that imports of similar products from the taxed country are negatively affected by
AD tariffs. This is explained by the model in Section 5: tariffs reduce the production of
downstream firms by increasing their marginal cost. Due to the production reduction, firms
reduce the demand for all the inputs they use, including employment and other imported
inputs.
4.2
Midstream Firms
This section shows that an AD tariff increases the wage bill, employment, exports, and
imports by the national producer. The effect of AD tariffs on the wage bill is presented in
Figure 2. It shows that before the introduction of the tariff, treatment and control groups
followed similar trends. But, the introduction of the tariff led to a relative increase in the
wage bill of firms protected by the AD tariff. Five years after its introduction, a 100% AD
16
Table 2: Effect of AD Tariffs on Trade Diversion
AD Tariff
(1)
log(Value Imports)
-0.0269
(0.0357)
N
R2
Mean Dep. Var
Mean Ind. Var
Product X Orig. FE
Time X Country FE
Specification
Same
(2)
log(Quantity Imports)
-0.0575
(0.0489)
60327
59792
0.659
0.707
10.832
8.661
.32
.32
Yes
Yes
Yes
Yes
Product, Other Countries
(3)
log(Value Imports)
-0.0871**
(0.0443)
(4)
log(Quantity Imports)
-0.0926*
(0.0536)
120603
118222
0.694
0.746
10.993
8.952
.09
.09
Yes
Yes
Yes
Yes
Same Country, Other Products
Description: This table shows the effect of the AD tariff on imports of other countries and other products. In columns 1 and 2 we show the coefficient
of a regression of average AD tariff at the product level on imports of countries not exposed to AD investigations. Import data is from the Secretary of
International Trade of the Ministry of Economy, and AD data is from the Global Anti-dumping Database. In columns 3 and 4 we show the coefficient
of a regression of AD tariffs on the imports of other products at the same 4-digit HS code from the same country. Standard errors are clustered at the
product-origin level.
tariff increased the wage bill of the national producer by about 3%.
Figure 2: Midstream Wage-Bill
Description: This figure contains the coefficients of the effect of an AD tariff on the log wage bill using the dynamic model
4. The x-axis contains the number of years to the first AD investigation. Wage bill data is from RAIS, and AD data is from
the Global Anti-dumping Database. The sample is composed of firms in sectors producing the product under AD investigation.
We constrain the sample to the set of firms observed 5 years around the AD investigation. These sample restrictions are made
to avoid compositional change. The shaded area contains the 95% confidence interval. Standard errors are clustered at the firm
level.
17
According to results in Table 3, AD tariffs increase employment, wage bill, exports and
imports in the national producer. Columns 1 through 2 of Table 3 show the effect of tariffs
on employment, and the wage bill. A 100% AD tariff increases employment by 1.8% and
the wage bill by 1.9%. Columns 3 and 4 show how the probability of becoming an exporter
or importer is affected by tariffs. Column 3 shows that a 100% AD tariff would increase
by 0.4% the probability of the national producer of becoming an exporter, while column 4
shows that the same tariff would increase by 0.3% the probability that the same domestic
firm would become an importer. Columns 5 and 6 show the effects of AD tariffs on the
intensive margin of exporting and importing. Therefore, tariffs affect employment at the
national producer, as well as its international trade.
Our estimates are within range of what the literature has found. Flaaen and Pierce (2019)
finds that a 100% increase in tariffs leads to 0.8% increase in employment in the protected
sector. Bown et al. (2021) finds that tariffs does not increase employment in the protected
sector, instead, they significantly decrease employment in downstream sectors.
Table 3: Effect of AD Tariffs on the National Producer
mid
τs,t
N
R2
# Firms
Mean Dep. Var
Mean Ind. Var
(1)
log(# Workers)
0.0184***
(0.00359)
(2)
log(Wage Bill)
0.0186***
(0.00390)
(3)
I(Exporter)
0.00421***
(0.00114)
(4)
I(Importer)
0.00330***
(0.00119)
(5)
log(Exports)
0.0133
(0.0107)
(6)
log(Imports)
0.0286***
(0.00937)
119368
0.829
6277
2.68
1.19
119368
0.863
6277
10.069
1.19
132816
0.613
6277
.165
1.07
132816
0.635
6277
.189
1.07
17057
0.832
1635
12.988
1.07
24052
0.798
2087
12.806
1.07
Description: This table presents the estimated parameters of model 3. The sample is composed of firms in sectors that produce the product under
an AD investigation. We limit the sample to the set of firms observed 5-years around the AD investigation. log(Wage Bill) is the log of the firm’s total
labor expenditure. log(Number Workers) is the log of the total number of workers in the firm. I{exporter} is a dummy that takes one if the protected
firm exports any product that year, I{importer} is a dummy taking one if the protected firm imports any product that year, log(Imports) is the log of
expected imports of the firm, and log(Exports) is the log of expected exports. Expected exports and imports are calculated following de Souza (2021),
mid is the average AD tariff imposed on products produced by the
who describes how expected export and imports are calculated at the firm level. τs,t
sector of each firm. Standard errors are clustered at the firm level.
18
4.3
Downstream Firms
We now show that the effects of tariffs propagate downstream, decreasing employment and
the wage bill. Figure 3 traces the dynamic effects of AD tariffs on downstream firms. Once
again, treatment and control firms show similar trends prior to the introduction of the AD
tariff and they diverge only after the beginning of the investigation. When tariffs are imposed
on the inputs of these firms, employment decreases. A 100% AD tariff on all the inputs of
a firm would lead to a drop of 10% in the wage bill 5 years after the beginning of the
investigation.
Figure 3: Downstream Wage-Bill
Description: This figure contains the coefficients of the effect of an AD tariff on the log wage bill of firms downstream to a AD
tariff using the dynamic model 6. The x-axis contains the number of years to the first AD investigation. Wage bill data is from
RAIS, and AD data is from the Global Anti-dumping Database. The sample is composed of firms whose main input is under
AD investigation. We constrain the sample to the set of firms observed 5 years around the AD investigation. These sample
restrictions are imposed to avoid compositional change. The shaded area contains the 95% confidence interval. Standard errors
are clustered at the firm level.
Columns 1 through 2 of Table 4 shows the effect of AD tariffs on the main downstream
firms.22 A 100% AD tariff on all the inputs of these firms would lead to a 3.8% drop in
employment and a 8.5% drop in wage bill.
The downstream effect of AD tariffs is not limited to the main buyer of an input, according
22
As Figure 3 demonstrates, these firms follow the same trend before the introduction of the tariff.
19
to results in Table 4. Columns 3 through 4 show the effect of tariffs on all downstream firms.
AD tariffs have no significant effect on employment downstream, but they do have an impact
on wage bill.
Table 4: Effect of AD Tariffs on Downstream Firms
(1)
log(# Workers)
-0.0383*
(0.0221)
down
τ̃d(s),t
Sample
Main Downstream
N
182790
2
R
0.812
# Firms
8686
Mean Dep. Var
2.412
Mean Ind. Var
.07
(2)
log(Wage Bill)
-0.0857***
(0.0244)
(3)
log(# Workers)
0.000765
(0.0173)
(4)
log(Wage Bill)
-0.0430**
(0.0191)
Main Downstream
182790
0.833
8686
9.599
.07
All Downstream
969619
0.806
55505
2.147
.05
All Downstream
969619
0.834
55505
9.313
.05
Description: This table presents the estimated parameters of model 5. The sample is composed of firms in sectors downstream to
the product under AD investigation. We limit the sample to the set of firms observed 5-years around the AD investigation. Columns
1 through 3 limit the sample to the main downstream firms. Columns 4 through 6 contain all downstream firms. log(Wage Bill) is
down is the
the log of total labor expenditure of the firm. log(Number Workers) is the log of the total number of workers in the firm. τ̃d(s),t
average AD tariff imposed on the inputs used by the sector of each downstream firm. Standard errors are clustered at the firm level.
4.4
Upstream Firms
This section shows that AD tariffs do not affect firms upstream, i.e., the firms that sell
inputs to the midstream firms, despite increasing employment midstream and the tariffs’
effects propagating downstream. Figure 4, which traces the dynamic effects of an AD tariff
on the wage bill of the main input provider of the national supplier, indicates that there
is no difference between the treatment and control groups before and after the AD tariff is
implemented.
Table 5 shows that even using variation from all the AD investigations, we do not find any
effect of AD tariffs on employment or the wage bill. Columns 1 through 2 of Table 5 show the
effect of AD tariffs on numbers of workers, and the wage bill in the main suppliers of sectors
protected by the AD tariff. The estimates identified are not statistically nor economically
significant. Columns 3 through 4 show the effect of AD tariffs on all upstream sectors. We
20
Figure 4: Upstream Wage-Bill
Description: This figure contains the coefficients of the effect of an AD tariff on the log wage bill of firms upstream to a AD
tariff using the dynamic model. The x-axis contains the number of years to the first AD investigation. Wage bill data is from
RAIS, and AD data is from the Global Anti-dumping Database. The sample is composed of firms whose main input is under
AD investigation. We constrain the sample to the set of firms observed 5 years around the AD investigation. These sample
restrictions are made to avoid compositional change. The shaded area contains the 95% confidence interval. Standard errors
are clustered at the firm level.
do not find any significant effect.
4.5
Robustness
We find that AD tariffs increase employment of the national producer and the effects of these
tariffs propagate to downstream firms, thereby lowering their employment, but they do not
significantly affect upstream firms. In this section we show that these results are robust to
the addition of controls, to different specifications, and to the use of alternative identification
strategies.
Controls. Tables A.11 through A.13 in Section A.5.1 show that the effect of AD tariffs on
the wage bill is stable across specifications. We show that by adding as a control a 1-digit
sector-year fixed effect, a 2-digit sector-year fixed effect, dummies for the number of products
investigated, dummies for the number of products with AD tariffs or by controlling for AD
21
Table 5: Effect of AD Tariffs on Upstream Firms
up
τ̃u(s),t
(1)
log(# Workers)
0.00321
(0.00792)
(2)
log(Wage Bill)
-0.000384
(0.00809)
(3)
log(# Workers)
0.00680*
(0.00379)
(4)
log(Wage Bill)
0.00637
(0.00401)
Main Upstream
74735
0.816
3694
2.55
.29
Main Upstream
74735
0.840
3694
9.8
.29
All Upstream
3238468
0.807
185354
2.144
.05
All Upstream
3238468
0.835
185354
9.313
.05
Sample
N
R2
# Firms
Mean Dep. Var
Mean Ind. Var
Description: This table presents the estimated parameters of model 7. The sample is composed of firms in sectors
upstream to the product under AD investigation. We limit the sample to the set of firms observed 5 years around the AD
investigation. Columns 1 through 3 limit the sample to the main upstream firm. Columns 4 through 6 contain all upstream
firms. log(Wage Bill) is the log of total labor expenditure of the firm. log(Number Workers) is the log of the total number
up
of workers in the firm. τ̃u(s),t
is the average AD tariff imposed on the sectors that each firm sells to. Standard errors are
clustered at the firm level.
tariffs upstream and downstream does not change the conclusion that AD tariffs increase
the wage bill midstream, decrease it downstream, and has no effect upstream.
All Sectors. Following Acemoglu et al. (2014) and Bown et al. (2021), in Section A.5.2 we
run a specification with the exposure of each firm to midstream, downstream, and upstream
tariffs. Under this specification, the results are still the same–i.e., AD tariffs increase employment midstream, and their effects propagate downstream, and do not affect upstream
firms.
Sectoral Regressions. In Section A.5.3, we study the effect of AD tariffs on sectoral
aggregates. We show that the results remain the same: AD tariffs boost employment midstream, and their effects propagate to downstream firms, and do not affect upstream firms.
Instrumental Variable. In Section A.4.1, we show that AD tariffs can be predicted with
high accuracy using international prices and the AD policy of other countries. Exploiting this
result, we use the AD policy of other countries as an instrument for AD tariffs in Brazil. In
22
Section A.5.4 we show that tariffs increase employment midstream, propagate downstream,
and have no effect upstream.
Regional Variation. In Section A.5.5, we also identify the effect of AD tariffs in local
labor markets. Leveraging heterogeneous exposure to the tariffs’ effects on account of the
heterogeneous sectoral composition of regional labor markets, we find that tariffs increase
employment midstream, decrease employment downstream (in particular, by educational
attainment group), and have no impact upstream.
Other Shocks In section A.4.4, we show that heterogeneous exposure to aggregate shocks
cannot explain our results. We add as controls to our main specifications terms capturing
heterogeneous exposure to exchange rate fluctuation and the Brazilian trade liberalization.
We still find that tariffs midstream increase employment, while they decrease employment
downstream, and have no effect upstream.
5
Model
We have found that AD tariffs increase employment in the protected sector, decrease employment downstream, and do not have an impact upstream. To study the aggregate employment
consequence, we build a quantitative model of the Brazilian economy. The model translates
the relative employment effects that we identified to aggregate effects, taking into account
the general equilibrium forces.
The model has households, firms, and a government. Households supply labor to different
sectors and receive an income transfer if they stay outside the labor force. Firms produce
using labor and inputs. Inputs come from national producers and from outside of Brazil.
The government collects taxes, imposes tariffs, and receives transfers from abroad to finance
the payment to non-working households. When the government imposes a tariff, it increases
the price of the international good, shifts demand from overseas to the national market,
increases production costs downstream, and increases demand for national inputs upstream.
In the next section, we use the elasticities we identified on the data to calibrate the important
parameters of the model.
23
5.1
Environment
The model is static. There are i ∈ {0, 1, ...N } countries; i = 0 denotes Brazil. Brazil has S
production sectors and a population with measure L. Households optimally choose to work
in one of the S sectors and supply labor to the sector in which they work, or stay out of the
labor force. If they work, they earn a sector-specific wage, decide how much labor to supply,
and receive a disutility from working. If they do not work, they receive social insurance from
the government.
To finance the social insurance, the government generates revenue from three sources: it
imposes an income tax on all households, borrows from the rest of the world, and collects
tariff revenues.
The product market is competitive.23 Each production sector has a representative firm.
The firm’s problem builds on Caliendo and Parro (2015)–it produces tradable output with
labor and non-tradeable input from all sectors according to a constant-return-to-scale technology. To make the non-tradable input, the firm sources tradable output from all sectors
and countries. Brazilian tradable output is used domestically and exported.
We assume Brazil is a small open economy in the sense that Brazilian AD tariffs do not
affect the ex-tariff foreign price faced by Brazilian importers, an assumption supported by
the empirical results (see Table 1).
5.2
Workers
Brazil is endowed with a population of fixed measure L. A representative household ω
chooses which sector to work in, the amount of labor to supply in this sector, receive income,
and choose their consumption bundle. Labor force in all sectors and those that do not work
add up to the total population. Households are heterogeneous in their disutility to work in
each sector.
23
We follow Ramondo and Rodrı́guez-Clare (2013), Caliendo and Parro (2015), and Caliendo et al. (2019)
on this assumption. As long as mark-ups are not affected by tariffs, a model with monopolistic competition
would deliver the same results.
24
Consumption Household ω chooses final goods consumption of sector r, cr (ω). The preference of households across different sectoral goods is given by
C(ω) =
S
X
θ
! θ−1
1
(dr ) θ (cr (ω))
θ−1
θ
,
(8)
r=1
where dr is a taste parameter and θ is the consumption elasticity of substitution across
sectors.
Income Households working ls (ω) at sector s receive wage ws . If the household chooses to
stay out of the labor force, ω receives welfare transfer b from the government. In any case,
households pay a fraction δ of their income in taxes.24 The budget constraint is
s
X
P r cr (ω) =
r=1
(1 − δ)ws ls (ω)
, s>0
(1 − δ)b
, s=0
,
(9)
where P r is the price index of the final good produced by sector r and ls (ω) is the labor
supply of household ω to sector s.
Labor Supply Conditional on working in sector s, households decide how much labor to
supply. The utility of worker ω, supplying ls (ω) to sector s and consuming C(ω), is given by
U s (ω) =
C s (ω) −
ψs s
l (ω)
ψ s +1
C s (ω)
ψ s +1
ψs
, s>0
,
(10)
, s=0
where ψ s , the Frisch labor supply elasticity within sector s, governs the elasticity of substitution between labor and leisure. In some sectors, ψ s is low and it is costly for workers to
adjust their labor supply. In sectors in which ψ s is higher, the disutility of working does not
increase much with the labor supplied. There is no disutility from working for those who do
24
This assumption ensures that fiscal policies do not distort the impact of trade shocks on a household’s
sector choice.
25
not work: ψ 0 = 0.25
Households have to choose between the different sectors. They receive the idiosyncratic
preference shock z s (ω) for working in sector s. z s (ω) follows a Frechet distribution with
shape parameter µ: F (z s (ω)) = exp(−(z s (ω))−µ ). Households also have an exogenous taste
preference for working in sector s given by as . In the end, the utility of a sector s household
is the product of the utility from consumption and leisure, sector taste, and the preference
shock:
U s (ω)as z s (ω),
(11)
where U s (ω) is given by 10.
Households’ Problem Households maximize utility 11, subject to the consumption bundle 8, the budget constraint 9, and the endogenous utility 10.
max
S
s,{crs (ω)}S
r=1 ,{ls (ω)}s=1
U s (ω)as z s (ω)
(12)
s.t. 8, 9, and 10
Heterogeneous Labor Supply Elasticity The solution to the households’ problem implies that the sectoral labor supply, Ls , is heterogeneous across sectors. To demonstrate
that, notice that the labor supplied to sector s equals:26
Ls =
s λs
ãs ( wC )
P
PS
µL
s λs
ãs w
+ã0 ( bC )
P
s=1 ( P C )
, s>0
PS
, s=0
s=1
µ
b
PC
s
ws λ
+ã0
PC
)
ã0 (
ãs
(
)
25
µ
( PbC )
L
,
(13)
We assume that a household’s labor supply problem is governed by the Greenwood et al. (1988) (GHH)
preference for tractability. The GHH preference focuses on the substitution between leisure and consumption
as it mutes the income effect. Cravino and Levchenko (2017) and Bonadio et al. (2021) also assume that a
representative household supplies labor to each sector following GHH preferences. However, they abstract
from the household’s sector choice problem and their self-selection into the non-working sector.
26
See Section B.1 for the proof.
26
where λs = µ(1+ψ s )+ψ s denotes the Frisch elasticity. An increase in a sector’s s wage causes
an increase in the sector’s labor supply through two channels. First, more households choose
this sector to work in (governed by µ). Second, each household in this sector supplies more
labor (governed by ψ s ). With ψ s > 0, the second channel creates heterogeneous sectoral
labor supply elasticities. If ψ s = 0, ∀s, then the labor supply problem will be reduced to a
discrete choice problem where households only choose sectors, and the labor supply elasticity
will be the same (and equal to the Frechet shape parameter µ) for all sectors.
5.3
Government
The social insurance system is financed by three sources of government revenue. The first is
the income tax δ. The second is borrowing from the rest of the world through a Trade Deficit
(T D).27 The third is the Tariff Revenue (T R).28 With the fiscal revenues, the government
pays each non-working household a fixed social insurance income b. The government budget
constraint is given by:
S
X
0
bL = δ
!
s
s
w L + bL
0
+ T D + T R,
(14)
s=1
where bL0 is the total government transfer to households outside the labor force.
5.4
Firms
Intermediate Goods Each sector s contains a representative competitive firm. Firms
use labor and a composite bundle from other sectors to produce. The production function
is given by:
Y s = As
1
ρ
(es ) (Ls )
ρ−1
ρ
+
S
X
(f
ss0
1
ρ
) (M
ss0
)
ρ−1
ρ
ρ
! ρ−1
,
s0 =1
27
This is a common property of static models of international trade: the value of foreign borrowing equals
the trade deficit.
28
The tariff revenue is a function of import values and the tariffs imposed on these imports, which we
specify below.
27
where As is the total factor productivity (TFP), Ls is the labor demand by sector s and M ss
0
is the quantity of sector s0 output used by sector s. ρ denotes the elasticity of substitution
0
across inputs. es and f ss are labor- and input-augmenting technology parameters.29
A firm’s profit maximization problem implies that Brazilian firms set prices, P0s , that are
equal to the marginal cost:
P0s =
1
As
es (ws )1−ρ +
S
X
1
! 1−ρ
0
0
f ss (P s )1−ρ
,
(15)
s0 =1
0
where ws denotes the sector s wage and where P s denotes the price of input from sector
s0 .30
Sector s producer’s expenditure share on input from sector s0 equals the following:
0
0
sss
M
0
f ss (P s )1−ρ
=
.
P
es (ws )1−ρ + Ss0 =1 f ss0 (P s0 )1−ρ
And sector s producer’s expenditure share on labor equals the following: ssL = 1 −
PS
s0 =1
0
sss
M.
Composite Intermediate Goods Firms use inputs from different countries. Inputs are
aggregated at the sector level according to a Dixit-Stiglitz style technology (Dixit and Stiglitz
1977). Therefore,
Qs =
N
X
1
(gis ) σs (Yis )
σ s −1
σs
s
! σsσ−1
,
i=0
where Qs denotes the quantity of the non-tradable input bundle, gis is a preference shifter for
inputs from sector s and country i, and σ s is the Armington trade elasticity (the elasticity of
substitution across countries).31 Yis denotes the quantity of sector s tradable goods imported
from foreign country i. Y0s is the quantity of Brazilian tradable output used in Brazil.32
29
These parameters allow us to match the factor shares observed in the data.
Following Caliendo and Parro (2015), we assume that the non-tradable input is used as a production
input and is also consumed. Thus, the sector s consumption goods price faced by consumers equals the
sector s input price faced by firms, and both are equal to P s .
31
Like many works in the trade literature, including Broda and Weinstein (2006) and Caliendo and Parro
(2015), we let the trade elasticity differ across sectors.
32
Therefore, the rest of Brazilian sector s output, Y s − Y0s , is exported.
30
28
Profit maximization and competitive markets imply that Brazilian sector s has the following expenditure function for country i:
s
xsi
g s (P s )1−σ
= i si 1−σs X s
(P )
(16)
where Pis is the price of a composite good of sector s from country i, xsi = Pis Yis denotes
the expenditure by sector s on country i, and X s = P s Qs denotes the total expenditure by
sector s. We further denote the country-sector level expenditure share: ssi =
Pis Yis 33
.
P s Qs
The
relationship between the sectoral input price and the sector-origin-level output price can be
established as follows:
s
(P s )1−σ =
N
X
s
gis (Pis )1−σ .
(17)
i=0
We assume that Brazilian exporters face the same Armington trade elasticity σ s . Foreign
demand for Brazilian sector s output can be written as:
s
YFs0 = (P0s )−σ EFs ,
where EFs is a function of foreign income and price. Because we assume that Brazil is a small
open economy, EFs can be treated as being exogenous to Brazilian AD tariffs. We denote
the Foreign expenditure on Brazilian output: EFs 0 = P0s YFs0 .
Product Lines. Sector-origin-level import, Yis , which is produced by combining different
product lines, is denoted by the following:34
s
ζsζ−1
Yis =
X
1
(hsil ) ζs (yils )
ζ s −1
ζs
,
l∈Ωsi
where Ωsi denotes the set of product lines that Brazil imports from country i in sector s,
yils denotes the quantity of imports in product line l of sector s from country i, hsil is a
(P s )−σ
33
s
The demand function for sectoral tradable output from individual countries is Yis = (Pis )−σs Qs .
34
Fajgelbaum et al. (2020) uses a similar technology that aggregates products to sectors.
29
preference parameter for products, and ζ s is the elasticity of substitution across product
lines. We allow the product-line-level elasticity of substitution to be heterogeneous across
sectors and to differ from country-level substitution.
Brazil imposes tariffs τlis on the product lines. The ex-tariff import price of product line
l is denoted with psli . As mentioned before, we assume that Brazil is a small open economy.
Therefore, psli can be treated as exogenous to Brazilian AD tariffs.35 The competitive market
and the profit maximization assumption imply the following expenditure function on product
line l of sector s from country i:
s
xsil
hs (ps tsil )1−ζ s
= il ils 1−ζ
xi ,
(Pi ) s
(18)
where tsil = 1 + τlis . We denote the product l’s share in the expenditure on sector s of country
i: ssil =
s
psil tsil yil
.36
s
s
P i Yi
The sector-origin-level output price, Pis , can be written as a function of
product-line-level prices and tariffs:
s
(Pis )1−ζ =
X
s
hsil (psil tsil )1−ζ .
(19)
l∈Ωsi
Market Clearing The market clearing condition for Brazilian sector s output is:
s
Y =
s
(P0s )−σ
1
Qs + EFs
s
(P )−σs
.
(20)
s
On the right-hand side,
(P0s )−σ
Qs
(P s )−σs
denotes the domestic demand for Brazilian output. The
s
rest, YFs0 = (P0s )−σ EFs , denotes the foreign demand.
The sectoral input, Qs , is used for both consumption and the production of tradable
output. Thus, the market clearing condition is:
s
Q =
S
X
0
M s s + C s,
(21)
s0 =1
35
We do not find that the AD tariff affects import prices, despite the fact that there is a large drop in
demand for the imported good. This supports our assumption that Brazil is a small open economy.
36
s
The demand for product line l of sector s imports from country i is denoted by the following: yil
=
(psil tsil )−ζ
(Pis )−ζ s
s
Yis .
30
0
where M s s is the quantity of composite goods from sector s and used by sector s0 and C s
refers to total consumption by all households of sector s composite good:
s
s
s
P C = α (1 − δ)
S
X
!
s
s
0
w L + bL
,
s=1
Labor is hired to produce the tradable output. The market clearing condition for labor
equates the labor supply to labor demand in each production sector:
Ls =
1 s s s
s P Y .
ws L 0
(22)
We finally relate the trade deficit and tariff revenue. Trade deficit equals total imports
minus total exports:
TD =
S X
N X
X
psil yils
−
s=1 i=1 l∈Ωsi
S
X
s
(P0s )1−σ EFs .
s=1
And the tariff revenue equals tariff import values multiplied by tariffs:
TR =
S X
N X
X
psil yils τils .
s=1 i=1 l∈Ωsi
Equilibrium Given the government’s fiscal and tariff policy, {δ, b, {τils }i,l,s } and foreign
prices and demand, {{psil }i,l,s , {EFs }s }, the equilibrium is defined as a set of sectoral input
prices, {P s }s , and sectoral wages, {ws }s such that the following hold:37
1. Firms maximize profit (equation 15);
2. The price index satisfies equations 17 and 19;
3. The goods markets clear, satisfying equations 20 and 21;
4. The labor market clears, satisfying equation 22;
5. Government budget constraint (equation 14) holds.
37
0
The equilibrium also depends on fundamentals, {{ãs }s , {ds }s , {As }s , {es }s , {f ss }s,s0 , {gis }i,s , {hsil }i,l,s }.
31
To compute counterfactuals, we rewrite the model in changes, which we present in Section
B.2.38 We also present in Section B.2 the equilibrium definition for the model in changes.
6
Identification of Model Parameters
In this section we identify, in five sequential steps, the parameters of the model using the
elasticities we identified in Section 3. First, we calibrate a set of parameters targeting moments in the Brazilian economy. Second, we estimate the elasticity of substitution across
product lines, taking into account the effect of anti-dumping tariffs on product-level imports. Third, we estimate the elasticity of substitution across countries, taking into account
the effect of anti-dumping tariffs on country-level imports. Fourth, we estimate the labor
supply elasticity, taking into account the effect of anti-dumping tariffs on employment and
wages. And finally, we identify the elasticities of substitution across inputs (ρ) and the consumption elasticity across sectors (θ) from the effect of tariffs on midstream and downstream
employment.
6.1
Calibration
The baseline economy is calibrated to Brazil in 1995, which is the initial year of our database.
We let each sector s ∈ {1, 2, ..., S − 1} refer to a Classificação Nacional de Atividades
Econômicas (CNAE) 2.0 4-digit goods sector in agriculture, livestock, extractive industry,
and manufacturing. s = S represents the combined service sector. The input-output co0
efficient, sss
M , is taken from the input-output table. We let each product line l represent
a Harmonized System (HS) 6-digit product. With a concordance table between HS codes
and CNAE 2.0 4-digit sectors from the IBGE (the Brazilian Institute of Geography and
s
Statistics), we calculate the sector-level exports E0F
. We get the Brazilian population and
the share of population that is not working from the IPEA database—a macroeconomic,
social, and regional database maintained by the Brazilian government.39 We compute the
sector population share κs with RAIS and total population. We further compute both
38
Doing so we eliminate the economic fundamentals that are exogenous to tariff changes and are difficult
to calibrate or estimate.
39
The link to the IPEA database can be found here.
32
the sector-level consumption expenditure share αs and the labor and input shares in gross
output, ssL and ssM , from the estimated input-output table. We calibrate the expenditure
shares on countries and products, ssi and ssil , by merging the estimated input-output table
with sector- and product-level imports data. We calibrate the social insurance tax rate to
the variable “government transfer rate” (“Renda de transferências governamentais”) in the
IPEA’s database, which equaled 10.3% in 1995. Using the government budget constraint (as
denoted by equation 14), we calibrate social insurance b to be 668.54 (Brazilian Real).40 We
calibrate the elasticity of the non-working population with respect to the social insurance
to the estimated value in the literature that studies the cost of public funds (Kleven and
Kreiner 2006) and set it to 0.2.
6.2
Elasticity of Substitution across Product Lines
To estimate the elasticity of substitution across product lines, ζ s , we study the effect of
anti-dumping tariffs on the import of products from a particular destination. ζ s captures
how easily the importer can switch product lines within sector-origin-level imports, and it
governs the impact of AD tariffs on sector-origin-level prices.
Table 6: Concordance between Broad Sectors and CNAE 2.0 2-digit sectors
No.
1
2
3
4
5
6
7
Broad Sector Name
Agriculture, Mining, Food and Textile
Leather, Wood and Paper
Petrochemicals
Mineral and Metal products
Computer, Electrical and Machinery Equipment
Automobiles and Transportation Equipment
Service
2 Digit CNAE 2.0 Sectors
1-14
15-18
19-21
22-25
26-28
29-33
35-97
Description: This table presents the concordance between (a) the broad sectors on which level we estimate the trade
and labor supply elasticities and (b) the CNAE 2.0 2-digit sectors.
We show that the elasticity of substitution across product lines, ζ s , which captures how
easily the importer can switch product lines, can be identified from the effect of anti-dumping
tariffs on imports. Taking the log of equation 18 and adding controls as in our specification
40
More specifically, the unit of value for this amount is 1995 Brazilian Real per annum.
33
in equation 1, we have:
s
log(xsi,l,t ) = (1 − ζ s ) log(tsi,l,t ) + β2s Isi,l,t {After AD} + β3s Nsi,l,t {No. of AD} + Φsi,t + ηi,l
+ si,l,t ,
where xsi,l,t are the imports of product l from country i in quarter t; 1 − ζ s is the effect of AD
tariffs on imports; Φsi,t summarizes the sector-origin-quarter-level price index, the sectororigin-level expenditure, and other factors that are common to all products in the same
s
sector from the same origin (see equation 18); and ηi,l
denotes the origin-product-level fixed
effect. To address the potential correlation between the error term and tariffs, we implement
a difference-in-differences, as before, adding Isi,l,t {After AD} (a dummy that takes value one
after the first AD investigation) and Nsi,l,t {No. of AD} (the number of AD investigations)
as the control. We constrain our sample to the set of products under investigation. The
identification assumption is that conditional on AD investigations, shocks to the originproduct-level consumer preference and the international price (including non-tariff trade
barriers) are not correlated with contemporaneous AD tariff changes.
Although we allow ζ s to vary across sectors, to ensure that there are sufficient variations
in AD tariffs, we classify CNAE 2.0 sectors into 6 broad sectors based on their definitions,
and we estimate ζ s for each broad sector.41 Table 6 presents the concordance between the
broad sectors and CNAE 2.0 2-digit sectors.
Table 7: Elasticity of substitution across product lines
ζs
Standard Err.
Sector name
8.005
(2.514)
Agriculture, Mining, Food and Textile
Wood and Paper
2.185
(0.801)
Petrochemicals
1.547
(0.435)
Minerals and Metals
1.152
(0.451)
Computer, Electrical and Machinery Equipment 5.062
(1.714)
Automobiles and Transportation Equipment
1.808
(0.601)
1.633
(0.338)
All Sectors
Description: This table presents the elasticity of substitution across product lines for CNAE 2.0 4digit sectors. The elasticities are assumed to be the same within each broad sector but to vary across
broad sectors. Standard errors are clustered at the product-origin-level.
Table 7 shows the estimates of ζ s . The implied elasticities of substitution across product
41
That is, we assume that ζ s is heterogeneous across the broad sectors but remains the same within each
broad sector.
34
lines range from 1.152 for mineral and metal sectors to 8.005 for agriculture, mining, food,
and textile sectors. These results are consistent with our intuition that products in primary
sectors (harvesting and extracting natural resources) are more substitutable than those in
secondary sectors (manufacturing and processing). The cross-sector average elasticity of
substitution across product lines equals 1.633. This low estimate is consistent with the
insignificant trade diversion to other products that we discovered, as noted in Section 4.1.
6.3
Elasticity of Substitution across Countries
In this section, we estimate the Armington trade elasticity, σ s , which captures how easily
sector-level imports can be substituted across different countries. We show that σ s can be
identified from the effect of anti-dumping tariffs on imports at the country level.
Taking logs of equation 16 and adding controls, we have:
log(xsi,t ) = (1 − σ s ) log(tsi,t ) + β2s Isi,t {After AD} + β3s Nsi,t {No. of AD} + Φst + ηis + si,t ,
where xsi,t are imports of sector s from country i in quarter t; tsi,t is the average AD tariffs
at the country-sector-quarter level;42 1 − σ s captures the effect of AD tariffs on country
level imports; Φst is a sector-quarter fixed effect capturing the sectoral import price index,
expenditure, and other factors that are common to all origin countries (see equation 16);
ηis is an origin-sector fixed effect; Isi,t {After AD} is a dummy taking the value of one after
the first AD investigation happens in sector s and targets country i; and Nsi,t {No. of AD}
counts the number of AD investigations that target country i and sector s in quarter t.
The identification assumption is that conditional on AD investigations, shocks to the originlevel consumer preference and international price (including non-tariff trade barriers) are not
correlated with contemporaneous AD tariff changes.
Table 8 shows the estimates of the elasticity of substitution across countries, σ s . Estimates range from 1.339 for the petrochemical sector to 5.158 for the computer, electrical,
and machinery sectors. For all sectors except the agriculture, mining, food, and textile
42
P
Formally, tsi,t = l∈Ωs ssi,l,t−1 tsi,l,t , where ssi,l,t−1 denotes the share of product l in sector s imports from
i
country i in year t − 1.
35
sector and the petrochemicals sector, the cross-country elasticity is higher than the crossproduct elasticity. This suggests that within each non-primary 4-digit sector, imports are
more homogeneous across countries than across products. The cross-sector average elasticity of substitution across countries equals 1.633, which is consistent with the limited trade
diversion to other countries that we reported in Section 4.1.
For most sectors, our estimates are much lower than what Caliendo and Parro (2015) find.
For example, their estimates are all above 10 for the computer, electrical, and machinery
equipment sector. A notable difference in their estimation procedure from ours is that
Caliendo and Parro (2015) do not account for the impact of trade policy uncertainties as we
do. These findings are hence consistent with the predictions in Handley and Limão (2017)
that accounting for these uncertainties can significantly reduce trade elasticity estimates.
Table 8: Elasticity of substitution across countries
σs
Standard Err.
Sector name
2.044
(0.260)
Agriculture, Mining, Food and Textile
Wood and Paper
3.060
(0.414)
Petrochemicals
1.339
(0.176)
Minerals and Metals
2.338
(0.171)
Computer, Electrical, and Machinery Equipment 5.158
(1.147)
Automobiles and Transportation Equipment
2.248
(0.350)
2.054
(0.091)
All Sectors
Description: This table presents the elasticity of substitution across countries for CNAE 2.0 4-digit
sectors. The elasticities are assumed to be the same within each broad sector but to vary across broad
sectors. Standard errors are clustered at the CNAE 2.0 4-digit sector level.
6.4
Labor Supply Elasticity
The labor supply elasticity, λs , governs the effects of tariffs on employment. We show that
λs can be identified from the effects of anti-dumping tariffs on wages and employment.
Equation 13 shows the relation between sectoral employment and wages. Taking the log
of that equation and adding controls, we get:
s
log(wi,t
)=
1
log(Lsi,t ) + β2s Isi,t {After AD} + β3s Nsi,t {No. of AD} + ηi + Ψst + si,t ,
s
λ
s
where wi,t
is wages at firm i in sector s in year t; Lsi,t is employment at firm i in sector s in
36
year t; ηi is a firm fixed effect; and Ψst is a year fixed effect. We again implement the same
identification strategy and control for exposure to an AD investigation with Ist {After AD}.
We instrument employment at the firm, Lsi,t , with AD tariffs in sector s, tst .43 Employment
is an endogenous choice of firms and correlates with other shocks affecting the firm. To deal
with that, we instrument Lsi,t with AD tariffs imposed on products made by sector s.44 . As
we discussed before, AD tariffs affect employment at the firm level, satisfying the relevance
condition, and are unlikely to correlate with firm-level shocks, satisfying the exogeneity
condition.
We present our estimates in Table 9. Labor supply elasticities are heterogeneous across
sectors, ranging from 0.678 to 1.666. Our estimates are higher than the micro estimates (see
Chetty et al. 2011), but lower than the macro elasticities. Our numbers are close to Eckert
(2019), who studies the elasticity of workers’ sector choice to sector income and finds an
elasticity of around 1.1 to 1.5.45
Table 9: Labor Supply Elasticity
Sector name
Agriculture, Mining, Food and Textile
Wood and Paper
Petroleum and Chemicals
Minerals and Metals
Computer, Electrical and Machinery Equipment
Automobiles and Transportation Equipment
All Non-service Sectors
Service
Implied λs
1.009
0.678
0.771
1.666
1.592
0.943
1.115
0.431
Standard Err.
(0.199)
(0.354)
(0.572)
(0.242)
(0.251)
(0.123)
(0.083)
(0.038)
Description: This table presents the labor supply elasticity for CNAE 2.0 4-digit sectors. The elasticities
are assumed to be the same within each broad sector but to vary across broad sectors. Standard errors are
clustered on the firm level.
We show how the estimated product, trade, and labor supply elasticities are associated
with the position of the sector in the value chain. We measure how upstream a sector is
by taking advantage of the procedure in Fally (2011), Antràs et al. (2012), and Antràs and
Chor (2013). The upstreamness measure computes the average number of sectors that one
PN
Formally, tst = i=1 ssi,t−1 tsi,t , where ssi,t−1 denotes the share of country i in sector s imports in year
t − 1 and tsi,t denotes country-sector level tariffs.
44
Because the service sector has no product that is subject to AD tariff, we instrument service sector
employment with upstream tariffs, as discussed in Section 3
45
Eckert (2019) assumes labor supply elasticity is the same across all sectors but is heterogeneous across
worker skill groups.
43
37
dollar of a sector’s output passes through to arrive at final demand (we present more details
in Section B.5). Figure B.1 shows that relatively downstream sectors have larger elasticity of
substitution across products and across countries, but they have weakly lower labor supply
elasticities.
6.5
Indirect Inference
We estimate ρ, the elasticity of substitution across labor and materials, and θ, the demand
elasticity across sectoral products, by indirect inference. ρ governs how much demand for
workers increases out of an increase in the demand for the final product. Because of that,
we choose a value for ρ to approximate the effect of AD tariffs on midstream employment.
θ governs how much the demand of the final consumer changes from a change in prices.
Because of that, we choose a value for θ to approximate the effect of AD tariffs downstream.
The greater is the elasticity of substitution across sectoral final consumption, θ, the larger
the decline in demand downstream firms will face because of their increase in production
costs.
We run our estimation algorithm as follows. We guess a set of parameters, {ρ, θ}, and
s
we provide annual tariffs, τi,t
, to the model. For each year, we solve the counterfactual
equilibrium with the model in changes (see Section B.2). Then we run the same panel
regression in the model as in the data. We target the effects of anti-dumping tariffs on
employment at midstream firms and at the main downstream firms. We present the detailed
procedures in Section B.3.
Table 10: Estimated elasticities and targeted moments
Parameters
Elasticity of substitution
across inputs ρ
Elasticity of
final demand θ
Targeted moments
Elasticity of midstream employment
with respect to midstream tariffs
Elasticity of main downstream employment
with respect to midstream tariffs
Parameter Value
0.6694
(0.5960,0.7428)
4.4082
(4.3790,4.4374)
Description: This table presents the elasticities that are estimated in the model–i.e., the of substitution between labor and inputs
ρ and the elasticity of final demand θ. The values presented in parentheses are the lower and upper bounds of the 95% confidence
interval of the estimated parameters. Standard errors are calculated by bootstrapping.
Table 10 shows the estimated parameters and their confidence intervals. Labor and ma38
terials are complements, with an elasticity of substitution equal to 0.67. Final goods are
substitutes, with an elasticity of substitution of 4.41. These values fall within the estimated
range in Oberfield and Raval (2021), who find that across different specifications and industries, the elasticity of substitution across inputs falls between 0.6 and 1.0 and the final
demand elasticity varies between 3.0 and 7.0.46
6.6
Model Validation
In this section, we show that the model can approximate well both targeted and non-targeted
moments. Table 11 displays a series of elasticities that we identified with data in Section 3
alongside their model-generated counterparts. As expected, the model matches exactly the
targeted moments–i.e., the effects of tariffs on midstream employment and on downstream
employment at main downstream firms. The second panel shows the model performance on
a set of non-targeted moments. The model performs reasonably well.47 In particular, we
did not find any upstream effects in the model, as we have not found in the data. In Table
B.1, we show how the distribution of trade elasticities contribute to this null effect of tariffs
upstream.
6.7
Different Model Specifications
In this section we study how different model specifications impact their ability to match
non-targeted moments. In this way we understand how each model assumption helps the
model match to the data. We consider the following alternative models and present the
results in Table B.1.
Same Input and Final Demand Elasticity In Column 3 of Table B.1, we require that
the input elasticity equals the final demand elasticity (ρ = θ). We target midstream employment with both ρ and θ. We estimate that both parameters equal 2.01. This overestimates
the input elasticity and underestimates the final demand elasticity, and thus substantially
46
Su (2017) also finds that inputs are complements, while outputs are substitutes.
The model predicts a small negative effect of tariffs on midstream exports, while the data in Table 3
shows a positive effect; that said, the positive effect documented in the data is insignificant.
47
39
Table 11: Targeted and non-targeted moments, data and model
Data
Model
Targeted Moments
1
Midstream employment
0.0184
0.0184
2 Main downstream employment -0.0383
-0.0383
Non-targeted Moments
3
Main upstream employment
0.0032
0.0029
4
Midstream wage bill
0.0186
0.0218
5
Main downstream wage bill
-0.0857
-0.0769
6
Main upstream wage bill
-0.0003
0.0037
7
Exports by midstream firms
0.0133
-0.0061
8
Imports by midstream firms
0.0286
0.0224
Employment Elasticity with Respect to Average Tariffs
9
Midstream tariffs
0.009
0.0117
10
Upstream tariffs
-0.0158
-0.0276
11
Downstream tariffs
-0.009
-0.0021
Description: This table presents the targeted and non-targeted moments in
the data and in the model. Moments 1-8 refer to the elasticity of midstream,
main downstream, and main upstream employment and wage bill, as well as
exports and imports with respect to midstream tariffs. Moments 9-11 refer
to the joint impact of midstream, average downstream, and average upstream
tariffs (see Section A.5.2). The data moments (Column 1) refer to the corresponding estimated coefficients that are presented in the empirical section.
The model moments refer to those estimated with model-simulated data and
equations B.13 and B.14. The employment elasticity with respect to average
tariffs refers to the joint impact of own sector, average upstream, and average
downstream tariffs that we document in Table A.14.
understates the negative employment consequences for downstream sectors.
Heterogeneous Elasticity of Substitution across Inputs In Column 4 of Table B.1,
we consider a model where the elasticity of substitution across inputs differs across sectors.
To reduce the number of parameters to estimate, we assume that the input elasticity is log
linear in sector upstreamness: ρ = exp(β0 + β1 · U ), where U denotes sector upstreamness
(see Section B.5 for a detailed description of how we compute the sector upstreamness). We
target the employment response in midstream, main upstream, and downstream sectors to
midstream tariffs with β0 , β1 , and θ. We find that the elasticity of substitution across inputs
weakly decreases in sector upstreamness (labor and inputs are less substitutable with each
other in more upstream sectors). However, the estimated slope coefficient β1 (2.5e-3) is close
40
to zero and thus, compared with the baseline, this model does not significantly improve in
matching the upstream employment effect of tariffs and other non-targeted moments.
Homogeneous Labor Supply Elasticity In Column 5 of Table B.1, we assume that all
sectors have the same labor supply elasticity. We set them to their estimated economy-wide
values that we presented in Table 9. We find that this model overestimates the midstream
wage response. This suggests that it underestimates the labor supply elasticity associated
with the average midstream sector and overestimates the labor supply elasticity associated
with the average downstream sector; thus, a much larger wage increase midstream and a
smaller wage decrease downstream are required to achieve the same midstream employment
response.
Homogeneous Trade Elasticity In Column 6 of Table B.1, we assume that all sectors have the same trade elasticity. We set them to their estimated economy-wide values
that we presented in Tables 8. We find that the model with homogeneous trade elasticity
overestimates the upstream employment effect by about 100%, suggesting that this model
underestimates the trade elasticity for an average upstream sector such that foreign inputs
substitute less with domestic labor. This is also supported by the small increase in imports
by midstream firms in this model.
No Input-output Linkages In Column 7 of Table B.1, we present the model that does not
have input-output linkages. To estimate this model, we target the midstream employment
effect with θ. Similar to the model that sets input and final demand elasticity to be the
same, the model without input-output linkages underestimates the final demand elasticity
and the magnitude of the downstream employment effect.
Cobb-Douglas Technology and Preference In Column 8, we investigate the case in
which both the production function and final demand are Cobb-Douglas (ρ = θ = 1). The
Cobb-Douglas model has been widely used in the international trade literature that takes
into account input-output linkages (for instance, Caliendo and Parro 2015, Ossa 2015, and
Caliendo et al. 2019, among others). However, Column 8 shows that this model substantially
41
misrepresents the employment and trade consequences of tariffs. As this model ignores that
the increasing sector price due to tariffs can cause final demand to substitute away from
that sector, it understates the negative impact on downstream employment and exaggerates
the impact on midstream and upstream employment. Because this model does not consider
that tariffs lead to cross-sector substitution in inputs, it also underestimates the magnitude
of the tariff effects on both imports and exports.
7
Quantitative Results
7.1
Brazilian Anti-Dumping Policy
Table 12 shows the effect of Brazil’s anti-dumping policy on its economy.48 The Brazilian
anti-dumping policy increased employment by 0.06%. Tariffs shift the demand for protected
goods from overseas to the national market, which increase employment at the national
producer and decrease it at downstream firms. However, because the downstream effect is
not strong enough, aggregate employment and GDP both increase.
The Brazilian anti-dumping policy decreased real income by 1.3% and welfare by 2.4%.
To achieve a comparable metric of welfare, we calculate it in consumption-equivalent terms;
that is, we hold the labor-leisure decision constant and we determine the percent change
in consumption that makes households indifferent between a world with and without an
AD policy.49 Tariffs increase producers’ marginal cost and final prices to consumers. As a
consequence, the real income of workers and consumption both go down, even though there
are more workers. These two effects together—the decrease in consumption and the increase
in labor supply—contribute to a decrease in welfare by 2.4%.
It is important to consider the input-output relationship of firms. Column 3 of Table
48
We call the Brazilian anti-dumping policy the cumulative anti-dumping tariffs implemented by the
Brazilian government. In Section B.7, instead of looking at the overall anti-dumping policy, we study the
impact of AD tariffs that Brazil imposed each year.
49
In Section B.4 we define precisely the equation that calculates the welfare change in consumptionequivalent terms. Since Lucas (1987), consumption-equivalent terms often have been used to measure the
welfare change. Our equation transforms non-consumption terms that enter the utility, e.g., leisure, into
consumption, such that they can be compared to other real economic variables, e.g., GDP and employment
(Jones and Klenow 2016). Because our model has a household labor-leisure choice within sectors, we use the
change in consumption-equivalent terms as our preferred measure of the welfare change.
42
Table 12: Aggregate Effect of Brazilian AD Policy
Variable
Baseline Model
Employment
GDP
Real income
Welfare
0.06%
0.05%
-1.32%
-2.43%
No
Input-output
0.15%
0.14%
-0.75%
-1.53%
Description: This table shows the effect of Brazilian anti-dumping policy
on aggregate employment, GDP, real income, and welfare. For each of
the variables of interest, we calculate the percentage change between the
equilibrium with the Brazilian anti-dumping policy and the benchmark
equilibrium in which no anti-dumping tariff is imposed. These aggregate
variables are defined in Section B.4. The Brazilian AD policy refers to each
sector’s maximum AD tariff of all years. We show the effect predicted by
the baseline model and a model that does not have input-output linkages.
12 shows the aggregate effect of the Brazilian anti-dumping policy according to a model
without an input-output connection between firms. Without this connection, the predicted
effect on employment is almost twice as large, while the welfare cost is lower. The model
without input-output connections fails to take into account that anti-dumping tariffs decrease employment downstream and, consequently, overstates the positive impact on overall
employment from anti-dumping tariffs.
7.2
Propagation Through Input-Output Linkages
The aggregate effect of anti-dumping tariffs depends on the position of the protected sector
in the value chain. To show this, we plot on the y-axis of Figure 5 the effect of a 200% antidumping tariff on each CNAE 2.0 4-digit sector. We further take the average of all 4-digit
sectors within each broad sector (defined in Table 6). We plot on the x-axis the average
upstreamness of each broad sector, i.e., the average number of sectors that one dollar of a
sector’s output passes through to reach final demand.50
Figure 5a shows that the impact of sector-level tariffs on employment is strongly negatively correlated with sector upstreamness.51 There are two empirically relevant channels:
50
51
In Section B.5, our concept of upstreamness is more precisely defined.
The negative correlation is robust to sector characteristic controls, including broad sector fixed effects,
43
the direct effect on the national producer and the downstream propagation.52 When a sector
is downstream in the value chain (i.e., when it has low upstreamness), few sectors are downstream to it. Therefore, the negative employment effect in downstream sectors is smaller and
the positive employment effect on the national producer dominates. Yet, for a sector that
provides inputs to several other sectors, the tariff’s negative employment effect downstream
is larger and, as a consequence, the effect of the tariff on overall employment is weaker. For
example, Figure 5a indicates that imposing tariffs on products made by the petrochemicals
sector and the agriculture, mining, food, and textile sector decreases aggregate employment.
However, protecting products made by the computer, electrical and machinery equipment
sector can raise it.53
Figure 5b shows that the impact of sector-level tariffs on welfare is not correlated with
sector upstreamness and is negative for almost all sectors. The tariff affects economic welfare
through two main channels: prices and wages. Taxing downstream sectors substitutes more
imports with domestic labor, which increases domestic prices. Higher prices imply lower
real income and, therefore, welfare decreases. On the other hand, taxing upstream sectors
decreases employment in more downstream sectors by cutting their wages, and this, too,
leads to lower nominal income and welfare.54
7.3
Optimal Import Tariff Policy
If the goal of the government is to maximize employment, how should it choose tariffs? To
answer this question, we study the choice of tariffs that maximize employment conditional on
keeping the government’s budget constant.55 We formally present this optimal tariff problem
in Section B.6.
Figure 6a shows that the input-output linkages are an important factor for the choice of
value added shares in GDP, employment shares, import shares, and trade elasticities (see Table B.3).
52
As discussed in the empirical section and supported by the model in B.1, there is no propagation of the
tariff effects upstream.
53
In Section B.7.2 we show the aggregate consequences of imposing 200% sector-level tariffs on each 4-digit
sector, without taking their means for broad sectors.
54
Figure B.4 confirms this intuition, which shows that the impact of sector-level tariffs on both the nominal
income and consumer price are negatively correlated with sector upstreamness.
55
For tractability, we require tariffs to be bellow 1000%. We also experimented with setting the AD tariff
upper bound to 900%. Our findings are robust to the bounds.
44
Figure 5: Effect of a 200% Sectoral Import Tariff on Employment and Welfare
(a) Employment and Sector Upstreamness
(b) Welfare and Sector Upstreamness
Description: These figures show the effect of sectoral, 200% anti-dumping tariff on aggregate employment and welfare. For
each large sector, the x-axis plots the average upstreamness, which measures the average number of sectors that one dollar of
a sector’s output passes through to reach final demand. In Section B.5, we present more details about how we measure sector
upstreamness. The y-axis of Figure 5a plots the change in employment caused by a 200% anti-dumping tariff. The y-axis of
Figure 5b plots the effect of a 200% tariff on welfare. To avoid cluttering the figure, we average the effect within large sectors.
45
employment-maximizing tariffs.56 Figure 6a plots on the x-axis the average upstreamness
of each large sector and on the y-axis the average tariff that maximizes employment, conditional on keeping the government’s revenue constant. It shows that if the government’s goal
is to maximize employment, tariffs should be large (an average of 373%) and strongly negatively correlated with sector upstreamness. Given that the effects of tariffs only propagate
downstream, the government should minimize the negative employment effect downstream
by setting higher tariffs on sectors selling directly to the final consumer.
Figure 6b shows that the optimal tariffs that maximize welfare should be low (an average
of 7.8%) and positively correlated with sector upstreamness.57 Imposing higher tariffs on
goods produced by downstream sectors reallocates more production from abroad to domestic
labor than imposing them on goods produced by upstream sectors.That, in turn, contributes
to higher consumer prices, which further reduces welfare by more.
Antràs et al. (2022) and Caliendo et al. (2021) find that tariffs are larger in downstream
sectors, a phenomenon that they call “tariff escalation”. Our findings suggest a new mechanism for tariff escalation. We show that a policymaker whose goal is to maximize employment
will also increase tariffs less in upstream sectors because they negatively impact employment
in other sectors. This suggests that “bringing jobs back” may also be the motivation for
tariff escalation.58
According to Table 13, a government that uses tariffs can increase employment by at
most 2.8%, but it precipitates a 15.9% decrease in welfare.59 If, instead, the government
chooses tariffs to maximize welfare, employment will increase by 0.01% but welfare should
decrease only by 1.46%.60 These findings highlight the trade-offs that policymakers face: the
56
In Section B.7.3 we show the optimal tariffs that maximize employment, GDP, real income and welfare
on each 4-digit sector, without taking their means for broad sectors.
57
See Section B.6 for this government’s problem.
58
Antràs et al. (2022) and Caliendo et al. (2021) explain tariff escalation with a model in which a government chooses sectoral tariffs to maximize consumer welfare and firms freely enter and exit sectors. In
these models, protecting downstream sectors leads to more entries in upstream sectors, thus alleviating the
pressure of tariffs on downstream prices. Protecting upstream sectors, on the other hand, leads to more exit
in downstream sectors, which further increases the consumer price and causes more welfare loss.
59
The real income loss from optimal tariffs that maximize employment is comparable to the real income
loss from autarky for Brazil that is described in the literature. Using an input-output table that has 251
sectors,Ossa (2015) shows that the Brazilian gains from trade equal 9.8%. This translates into a 8.9% real
income loss.
60
Compared with the baseline equilibrium where no AD tariff is imposed, the optimal tariffs that maximize
real income and welfare still reduce real income and welfare because of the fiscal constraint: the government
46
tariffs that increase employment are likely to harm consumer welfare.
Table 13: Optimal Import Tariff Policy
Effect on
Employment
GDP
Real income
Welfare
Optimal tariffs that maximize
Employment
GDP
Real income
2.82%
2.46%
0.03%
2.32%
2.49%
0.07%
-7.97%
-6.64%
-0.87%
-15.85%
-14.63%
-1.77%
Welfare
0.01%
-0.10%
-1.02%
-1.46%
Baseline tariffs
0.06%
0.05%
-1.32%
-2.43%
Description: This table shows the optimal tariff according to different objectives of the government and its effect
on aggregate variables. The change in outcomes is made from the equilibrium without tariffs.
8
Conclusion
In this paper we study how the impacts of AD tariffs propagate along the value chain and
their aggregate consequences. We compile detailed data on AD investigations, trade, and
the input-output table, and we match them to firm-level administrative employment data
in Brazil. Using a difference-in-differences strategy, we find that AD tariffs reduce imports
but do not significantly divert trade to the imports of similar products from other foreign
countries. AD tariffs significantly increase employment in the protected sector and strongly
decrease employment in downstream sectors, but do not significantly increase employment
in upstream sectors.
To quantify the aggregate, general equilibrium effects of these tariffs, we build a small
open economy model of Brazil that takes into account international trade, input-output
linkages and labor force participation. The model can reproduce the micro-elasticities we
found and it matches the aggregate moments of the Brazilian economy.
We find that Brazilian AD policy weakly increases aggregate employment but decreases
consumer welfare. On average, protecting downstream sectors increases aggregate employment more than it protects upstream sectors. A government whose objective is to maximize
is required to collect the same AD tariff revenue as it collects from the benchmark tariffs. If tariffs are set
low for some sectors, they have to be high for other sectors to ensure the fiscal constraint holds. These tariffs
raise welfare relative to the incumbent Brazilian AD policy.
47
Figure 6: Tariffs to Maximize Employment and Welfare
(a) Tariffs that Maximize Employment
(b) Tariffs that Maximize Welfare
Description: These figures shows the sectoral optimal tariffs that maximize employment and welfare as well as the Brazilian
AD policy. The optimal tariffs solve a problem that maximize the respective aggregate variable, subject to the equilibrium
constraints and the additional constraint that the government collects the same tariff revenue as it collects from the benchmark
tariffs (see Section 7.3). The Brazilian AD policy refers to each sector’s maximum AD tariff in all years. The x-axis plots the
average upstreamness of each broad sector, which measures the average number of sectors that one dollar of a sector’s output
passes through to reach final demand. In Section B.5, we present more details about how we measure sector upstreamness. The
y-axis plots the tariff that maximizes employment (Figure 6a) and welfare (Figure 6b).
48
employment has strong incentives to increase tariffs, especially for downstream sectors. This
moderately increases aggregate employment but substantially undermines consumer welfare.
From our exercise, policymakers can learn that when setting tariffs, they face an important trade-off between increasing employment and promoting domestic welfare. Strong
WTO rules, trade agreements, and domestic political institutions could prevent the policy
makers from creating jobs by raising tariffs without limits–an act that imposes excessive
harm on consumer welfare.
49
References
Acemoglu, D., D. Autor, D. Dorn, G. H. Hanson, and B. Price (2014): “Import Competition and the Great US Employment Sag of the 2000s,” Journal of Labor
Economics, 34(S1), 141–198.
Amiti, M., S. J. Redding, and D. E. Weinstein (2020): “Who’s paying for the US
tariffs? A longer-term perspective,” in AEA Papers and Proceedings, vol. 110, 541–46.
Amiti, Mary & Redding, S. J. . W. D. (2019): “The Impact of the 2018 Trade War on
U.S. Prices and Welfare,” NBER Working Papers 25672.
Antràs, P. and D. Chor (2013): “Organizing the global value chain,” Econometrica, 81,
2127–2204.
Antràs, P., D. Chor, T. Fally, and R. Hillberry (2012): “Measuring the upstreamness of production and trade flows,” American Economic Review, 102, 412–16.
Antràs, P., A. Gutiérrez, T. C. Fort, and F. Tintelnot (2022): “Trade Policy
and Global Sourcing: A Rationale for Tariff Escalation,” .
Autor, D. H., D. Dorn, and G. H. Hanson (2013): “The China Syndrome: Local
Labor Market Effects of Import Competition in the United States,” American Economic
Review, 103(6).
Autor, D. H., D. Dorn, G. H. Hanson, and J. Song (2014): “Trade adjustment:
Worker-level evidence,” The Quarterly Journal of Economics, 129, 1799–1860.
Balassa, B. (1965): “Tariff protection in industrial countries: an evaluation,” Journal of
Political Economy, 73, 573–594.
Barattieri, A. and M. Cacciatore (2020): “Self-harming trade policy? protectionism
and production networks,” Tech. rep., National Bureau of Economic Research.
Barattieri, A., M. Cacciatore, and F. Ghironi (2021): “Protectionism and the
business cycle,” Journal of International Economics, 129, 103417.
50
Baylis, K. and J. M. Perloff (2010): “Trade diversion from tomato suspension agreements,” Canadian Journal of Economics/Revue canadienne d’économique, 43, 127–151.
Besedeš, T. and T. J. Prusa (2017): “The hazardous effects of antidumping,” Economic
Inquiry, 55, 9–30.
Blonigen, B. and T. Prusa (2016): “Chapter 3 - Dumping and Antidumping Duties,”
North-Holland, vol. 1 of Handbook of Commercial Policy, 107 – 159.
Blonigen, B. A. (2016): “Industrial policy and downstream export performance,” The
Economic Journal, 126, 1635–1659.
Blonigen, B. A. and S. E. Haynes (2002): “Antidumping investigations and the passthrough of antidumping duties and exchange rates,” American Economic Review, 92,
1044–1061.
Blume, L., S. Durlauf, and L. E. Blume (2008): The new Palgrave dictionary of
economics, Palgram Macmillan.
Bonadio, B., Z. Huo, A. A. Levchenko, and N. Pandalai-Nayar (2021): “Globalization, Structural Change and International Comovement,” Tech. rep., Mimeo, Michigan,
Yale, and UT Austin.
Bown, C. P. (2005): “Global antidumping database version 1.0,” World Bank Policy
Research Working Paper.
Bown, C. P., P. Conconi, A. Erbahar, and L. Trimarchi (2021): “Trade protection
along supply chains,” .
Bown, C. P. and M. A. Crowley (2006): “Policy externalities: How US antidumping
affects Japanese exports to the EU,” European Journal of Political Economy, 22, 696–714.
——— (2007): “Trade deflection and trade depression,” Journal of International Economics,
72, 176–201.
——— (2016): “The empirical landscape of trade policy,” Handbook of commercial policy,
1, 3–108.
51
Broda, C. and D. E. Weinstein (2006): “Globalization and the Gains from Variety,”
The Quarterly journal of economics, 121, 541–585.
Caliendo, L., M. Dvorkin, and F. Parro (2019): “Trade and labor market dynamics:
General equilibrium analysis of the china trade shock,” Econometrica, 87, 741–835.
Caliendo, L., R. C. Feenstra, J. Romalis, and A. M. Taylor (2021): “A secondbest argument for low optimal tariffs,” Tech. rep., National Bureau of Economic Research.
Caliendo, L. and F. Parro (2015): “Estimates of the Trade and Welfare Effects of
NAFTA,” The Review of Economic Studies, 82, 1–44.
Cavallo, A., G. Gopinath, B. Neiman, and J. Tang (2021): “Tariff pass-through at
the border and at the store: Evidence from us trade policy,” American Economic Review:
Insights, 3, 19–34.
Chetty, R., A. Guren, D. Manoli, and A. Weber (2011): “Are micro and macro
labor supply elasticities consistent? A review of evidence on the intensive and extensive
margins,” American Economic Review, 101, 471–75.
Conconi, P., M. Garcı́a-Santana, L. Puccio, and R. Venturini (2018): “From final
goods to inputs: the protectionist effect of rules of origin,” American Economic Review,
108, 2335–65.
Cravino, J. and A. A. Levchenko (2017): “Multinational firms and international business cycle transmission,” The Quarterly Journal of Economics, 132, 921–962.
de Souza, G. (2021): “The Labor Market Consequences of Appropriate Technology,” Working Paper.
Devlin, A., B. K. Kovak, and P. M. Morrow (2021): “The Long-Run Labor Market
Effects of the Canada-US Free Trade Agreement,” University of Toronto, mimeo.
Dix-Carneiro, R. and B. K. Kovak (2015): “Trade liberalization and the skill premium:
A local labor markets approach,” American Economic Review, 105, 551–57.
52
——— (2017): “Trade liberalization and regional dynamics,” American Economic Review,
107, 2908–46.
Dixit, A. K. and J. E. Stiglitz (1977): “Monopolistic competition and optimum product
diversity,” The American economic review, 67, 297–308.
Durling, J. P. and T. J. Prusa (2006): “The trade effects associated with an antidumping epidemic: The hot-rolled steel market, 1996–2001,” European Journal of Political
Economy, 22, 675–695.
Eckert, F. (2019): “Growing Apart: Tradable Services and the Fragmentation of the U.S.
Economy,” Mimeograph, Yale University.
Fajgelbaum, P. D., P. K. Goldberg, P. J. Kennedy, and A. K. Khandelwal
(2020): “The return to protectionism,” The Quarterly Journal of Economics, 135, 1–55.
Fally, T. (2011): “On the Fragmentation of Production in the US,” University of Colorado
mimeo.
Feng, A. and H. Li (2021): “We Are All in the Same Boat: Cross-Border Spillovers of
Climate Risk through International Trade and Supply Chain,” Tech. rep., International
Monetary Fund.
Flaaen, A., A. Hortaçsu, and F. Tintelnot (2019): “The Production Relocation and
Price Effects of U.S. Trade Policy: The Case of Washing Machines,” Working Paper.
Flaaen, A. and J. R. Pierce (2019): “Disentangling the effects of the 2018-2019 tariffs
on a globally connected US manufacturing sector,” .
Gallaway, M. P., B. A. Blonigen, and J. E. Flynn (1999): “Welfare costs of the
US antidumping and countervailing duty laws,” Journal of International Economics, 49,
211–244.
Greenwood, J., Z. Hercowitz, and G. W. Huffman (1988): “Investment, capacity
utilization, and the real business cycle,” The American Economic Review, 402–417.
53
Guilhoto, J., C. R. Azzoni, S. M. Ichihara, D. K. Kadota, and E. A. Haddad
(2010a): “Matriz de Insumo-Produto do Nordeste e Estados: Metodologia e Resultados
(Input-Output Matrix of the Brazilian Northeast Region: Methodology and Results),”
Available at SSRN 1853629.
Guilhoto, J. et al. (2010b): “Estimação da Matriz Insumo-Produto Utilizando Dados
Preliminares das Contas Nacionais: Aplicação e Análise de Indicadores Econômicos para
o Brasil em 2005 (Using Data from the System of National Accounts to Estimate InputOutput Matrices: An Application Using Brazilian Data for 2005),” Available at SSRN
1836495.
Günlük-Şenesen, G. and J. M. Bates (1988): “Some experiments with methods of
adjusting unbalanced data matrices,” Journal of the Royal Statistical Society: Series A
(Statistics in Society), 151, 473–490.
Handley, K., F. Kamal, and R. Monarch (2020): “Rising import tariffs, falling export
growth: When modern supply chains meet old-style protectionism,” Tech. rep., National
Bureau of Economic Research.
Handley, K. and N. Limão (2017): “Policy uncertainty, trade, and welfare: Theory and
evidence for China and the United States,” American Economic Review, 107, 2731–83.
Huang, Y., C. Lin, S. Liu, and H. Tang (2019): “Trade networks and firm value:
Evidence from the US-China trade war,” Available at SSRN 3504602.
Irwin, D. A. (2014): “Tariff Incidence: Evidence from U.S. Sugar Duties, 1890-1930,”
NBER Working Papers 20635.
Jabbour, L., Z. Tao, E. Vanino, and Y. Zhang (2019): “The good, the bad and the
ugly: Chinese imports, European Union anti-dumping measures and firm performance,”
Journal of International Economics, 117, 1–20.
Jones, C. I. and P. J. Klenow (2016): “Beyond GDP? Welfare across countries and
time,” American Economic Review, 106, 2426–57.
54
Junius, T. and J. Oosterhaven (2003): “The solution of updating or regionalizing a
matrix with both positive and negative entries,” Economic Systems Research, 15, 87–96.
Kleven, H. J. and C. T. Kreiner (2006): “The marginal cost of public funds: Hours of
work versus labor force participation,” Journal of Public Economics, 90, 1955–1973.
Konings, J. and H. Vandenbussche (2008): “Heterogeneous responses of firms to trade
protection,” Journal of International Economics, 76, 371–383.
Konings, J., H. Vandenbussche, and L. Springael (2001): “Import diversion under
European antidumping policy,” Journal of Industry, Competition and Trade, 1, 283–299.
Leontief, W. (1949): “Recent developments in the study of interindustrial relationships,”
The American Economic Review, 39, 211–225.
Lloyd, T., O. Morrissey, and G. Reed (1998): “Estimating the impact of antidumping and anti-cartel actions using intervention analysis,” The Economic Journal, 108,
458–476.
Lu, Y., Z. Tao, and Y. Zhang (2013): “How do exporters respond to antidumping
investigations?” Journal of International Economics, 91, 290–300.
Lucas, R. E. (1987): Models of business cycles, vol. 26, Basil Blackwell Oxford.
Miller, R. E. and P. D. Blair (2009): Input-output analysis: foundations and extensions, Cambridge university press.
Muendler, M.-A. (2002): “Definitions of Brazilian mining and manufacturing sectors and
their conversion,” University of California, San Diego, Mimeograph.
Oberfield, E. and D. Raval (2021): “Micro data and macro technology,” Econometrica,
89, 703–732.
Ossa, R. (2015): “Why trade matters after all,” Journal of International Economics, 97,
266–277.
55
Pierce, J. R. (2011): “Plant-level responses to antidumping duties: Evidence from US
manufacturers,” Journal of International Economics, 85, 222–233.
Pierce, J. R. and P. K. Schott (2016): “The surprisingly swift decline of US manufacturing employment,” American Economic Review, 106, 1632–62.
Prusa, T. J. (1997): “The Trade Effects of US Antidumping Actions,” in The Effects of
US Trade Protection and Promotion Policies, University of Chicago Press, 191–214.
——— (2001): “On the Spread and Impact of Anti-Dumping,” Canadian Journal of Economics, 591–611.
Ramondo, N. and A. Rodrı́guez-Clare (2013): “Trade, multinational production, and
the gains from openness,” Journal of Political Economy, 121, 273–322.
Rodrı́guez-Clare, A., M. Ulate, and J. P. Vásquez (2020): “New-Keynesian Trade:
Understanding the Employment and Welfare Effects of Trade Shocks,” Tech. rep., National
Bureau of Economic Research.
Ruhl, K. J. (2014): “The aggregate impact of antidumping policies,” New York University.
Sandkamp, A. (2020): “The trade effects of antidumping duties: Evidence from the 2004
EU enlargement,” Journal of International Economics, 123, 103307.
Shapiro, J. S. (2021): “The environmental bias of trade policy,” The Quarterly Journal of
Economics, 136, 831–886.
Staiger, R. W. and F. A. Wolak (1994): “Measuring Industry Specific Protection:
Antidumping in the United States,” NBER Working Papers, 4696.
Stone, R. (1961): Input-output and national accounts, Organisation for European Economic
Co-operation.
Su, Y. (2017): “The Reflection Channel of Shock Transmission in Production Networks,”
Available at SSRN 3060965.
56
Temursho, U., J. Oosterhaven, and M. A. Cardenete (2021): “A multi-regional
generalized RAS updating technique,” Spatial Economic Analysis, 16, 271–286.
Travis, W. P. et al. (1964): The theory of trade and protection, 121, Harvard University
Press.
Trefler, D. (2004): “The long and short of the Canada-US free trade agreement,” American Economic Review, 94, 870–895.
Trimarchi, L. (2020): “Trade policy and the China syndrome,” .
Vandenbussche, H. and M. Zanardi (2010): “The chilling trade effects of antidumping
proliferation,” European Economic Review, 54, 760–777.
57
A
Empirical Evidence
A.1
Statistics of Anti-Dumping Investigation in Brazil
In this section we discuss the anti-dumping measures and investigations taken by Brazil
between 1989 and 2017. Table A.1 has the number of investigations, different products and
countries. Treated are the product-country pair that had an AD tariff applied or price adjustment. Figure A.1a show the AD investigations by year. In 1994 the Brazilian government
Table A.1: Statistics of Brazilian AD Investigations
# Investigations
# Products
# Countries
Avg. Tariff
Treated
Control
All
393
155
50
0.35
183
108
45
0
576
227
65
0.24
Notes: This table presents the statistics of Brazilian antidumping investigations between 1989 and 2017. Each investigation is a product-country pair. The average tariff is calculated using the imposed ad-valorem tariff. In case the tariff is
per-unit, we calculate the corresponding ad-valorem value using trade data of the preceding year.
filed a broad complaint that covered 124 types of artificial and synthetic fabric from South
Korea. Because we count investigations on the product level and not on the complaint level,
we observe the large spike that year. The complaint was rejected in all products. Except for
this spike, the treatment and control groups are evenly distributed over time.
Table A.2 show the top 5 countries with the most investigations. China is the leader, and
80% of the investigations on China end with a tariff increase or price adjustment. Moreover,
there is large variation on the tariff imposed.
While AD tariffs target specific products and countries, they can lead to significant price
changes on the sector level. Figure A.2 and Table A.3 show the average AD tariff that each
CNAE 2.0 4-digit sector faces. Figure A.2a shows, for each CNAE 2.0 4-digit sector, the
AD tariff faced by the average product-country pair that received an AD tariff. Figure A.2b
shows, for each CNAE 2.0 4-digit sector, the AD tariff faced by the average product-country
58
Figure A.1: Brazilian Anti-Dumping Policy Over-Time
(a)
(b)
Investigations by Year
Average AD Tariff by Year
Description: Figure A.1a shows the average number of AD investigations per year at the product level. Figure A.1b shows
the average AD tariff at the product level conditional on an AD being imposed.
Table A.2: Statistics of Brazilian AD Investigations
Country
China
South Korea
United States
India
Taiwan
Germany
# Investigations
113
63
58
34
25
22
% Treated
0.850
0.317
0.638
0.588
0.800
0.773
Avg. Tariff
0.782
0.336
0.581
0.324
0.445
0.388
Notes: This table presents the number of products investigated for dumping between 1989
and 2017 in Brazil against different countries. The data source is the Global Antidumping
Database.
pair taking into account the product-country pairs that never faced AD tariffs.61 Even if
we include into the average the product-country pairs that never had AD tariff changes, for
some sectors, the average AD tariff is as high as 30%. These figures show that AD tariffs
can lead to substantial price variations across sectors, even if only a subset of products and
countries in each sector were hit by AD tariffs.
Table A.3 shows the summary statistics of AD tariffs by broad sectors. In Section 6,
we estimate cross-product and cross-country elasticities of substitution for the same set of
61
In Figure A.2, we only show the 4-digit sectors that received AD tariffs. To compute the average AD
tariff for each sector, first, we compute the imports in each product from each country in an average year
during the sample period. Then we compute the weighted average of the maximum of product-country pair
specific AD tariff during the sample period, using these product-country level imports as weights.
59
broad sectors. Within each broad sector, some 4-digit sectors received an AD tariff. In
the aggregate, about 20% of all 4-digit sectors (53 out of 297) were protected. Taking into
account the 4-digit sectors that never received an AD tariff, the average AD tariff per 4digit sector is 1.94%, is the highest for wood and paper sector (5.4%) and is the lowest for
computer, electrical and machinery equipment sector (0.1%). Among the 4-digit sectors that
received AD tariffs, the average AD tariff per 4-digit sector is 10.5%, is the highest for wood
and paper sector (29.0%) and is the lowest for computer, electrical and machinery equipment
sector (0.77%). The percentiles of the tariffs also show that within each broad sector, some
4-digit sectors face large tariffs.
Figure A.2: Anti-Dumping Tariff by Sector
(b)
Product-countries with Positive Tariffs
100
Avg. Tariff Rate (%)
150
200
All Product-countries
0892
1031
1052
1312
1313
1323
1330
1531
1532
1533
1539
1721
2012
2019
2022
2029
2031
2033
2040
2099
2121
2122
2211
2219
2221
2229
2311
2312
2319
2320
2341
2349
2399
2422
2423
2424
2431
2439
2443
2449
2541
2543
2592
2599
2640
2759
2790
2812
2822
3092
3250
3291
3299
0
50
800
600
400
0892
1031
1052
1312
1313
1323
1330
1531
1532
1533
1539
1721
2012
2019
2022
2029
2031
2033
2040
2099
2121
2122
2211
2219
2221
2229
2311
2312
2319
2320
2341
2349
2399
2422
2423
2424
2431
2439
2443
2449
2541
2543
2592
2599
2640
2759
2790
2812
2822
3092
3250
3291
3299
0
200
Avg. Tariff Rate (%)
1,000
(a)
Description: Figure A.2a shows, for each CNAE 4-digit sector, the average AD tariff of all product-countries that face positive
AD tariffs. Figure A.2b shows, for each CNAE 4-digit sector, the average AD tariff of all product-countries that face positive AD
tariffs. To compute the average AD tariff for each sector, first, we compute the imports in each product from each country in an
average year during the sample period. Then we compute the weighted average of the maximum of product-country pair specific
AD tariff during the sample period, using these product-country level imports as weights. When computing the average tariff,
Figure A.2a only includes the product-countries that had positive AD tariffs, and Figure A.2b includes all product-countries.
Table A.3: Statistics of AD Tariff by Sector
Sector Name
Agriculture, Mining, Food and Textile
Wood and Paper
Petrochemicals
Minerals and Metals
Computer, Electrical and Machinery Equipment
Automobiles and Transportation Equipment
All Sectors
No. of 4-digit Sectors
100
37
41
40
47
32
297
No. with Positive AD
7
5
14
18
5
4
53
Uncond. Mean (%)
0.49
5.36
4.89
2.82
0.08
1.47
1.94
Cond. Mean (%)
6.96
28.95
14.32
6.27
0.77
11.74
10.52
Cond. p50
1.37
31.77
0.86
0.39
0.78
10.43
1.27
Cond. p95
40.31
44.37
175.73
48.34
1.90
21.78
44.37
Notes: This table presents summary statistics of AD tariff by broad sectors. The same set of broad sectors is used in Section 6, where we estimate the cross-product and cross-country elasticities of substitution. The Uncond.
Mean refers to, for each sector, the AD tariff faced by an average 4-digit sector when we account for the 4-digit sectors that never received an AD tariff. Cond. Mean refers to, for each sector, the AD tariff faced by an average
4-digit sector that received an AD tariff. Cond. p50 and Cond. p95 shows the 50th percentile and 95th percentile of the AD tariffs faced by 4-digit sectors within each broad sector.
60
A.2
A.2.1
Input-Output Table
Estimating an Input-Output Table
We base our sectoral findings on CNAE 2.0 4-digit level (CNAE4 level) sectors.62 There
are a total of 297 goods (agriculture, mining, and manufacturing) sectors and 375 service
sectors. As there is no AD tariff variation on service sectors, we combine all service sectors
into one single sector. In order to identify a sector’s main upstream and downstream sectors,
and to compute the weighted average upstream and downstream tariffs, we need to know
a input-output table for Brazil that has information about a sector’s input expenditure on
all sectors and from both domestic and foreign sources. We call such a table the complete
input-output table. However, the most disaggregated complete input-output table for
Brazil that is readily available is tabulated on a different sector classification–Niv, which has
only 67 broad sectors (among the Niv sectors just 36 are goods and the rest are service).63
To acquire complete input-output information on CNAE4 level, we take advantage of the
following datasets: a CNAE4 level imports table (details described below), CNAE4 level
gross output and expenditure on input, as well as Niv level complete input-output table.
We then apply a generalized-RAS (GRAS) estimation algorithm (Temursho et al. 2021) on
these databases to estimate the desirable input-output matrix.
We proceed in the following steps. We start with a unique database that we acquire
from the Secretary of International Trade of the Ministry of Economy on sector-product
level imports–the value of each HS 6-digit product that is imported by a domestic sector.
Using a concordance table between HS 6-digit products and CNAE4 sectors from Secretary
of International Trade, we construct the input-output table for imports, i.e., CNAE4 level
imports by each domestic CNAE4 sector. We call it the imports table. A few works in this
literature, for example, Flaaen and Pierce (2019), Handley et al. (2020), use the imports table
directly to compute domestic sectors’ exposure to upstream tariffs. However, due to the home
bias, the IO coefficients calculated with the imports table may not equal to those calculated
62
See https://cnae.ibge.gov.br/ for the background information about this sector classification.
Muendler (2002) discusses the relationship between CNAE sectors and Niv sectors. The Niv level
complete input-output table is available from IBGE (the Brazilian Institute of Geography and Statistics).
As the Niv level input-output table is only available for 2015, we fix all other datasets to the same year.
63
61
with the complete input-output table, which includes both domestic sales and imports.64
Using only the imports table may miss the sectors in which the domestic producers are main
upstream and downstream to the protected sectors but do not import or export extensively.
Therefore, we need to update the imports table with domestic input-output information.
In the third step, we apply the GRAS estimation algorithm on these datasets to estimate a CNAE4 level complete input-output table. RAS (Leontief 1949, Stone 1961) is an
estimation algorithm that has widely been used to estimate input-output tables. The algorithm minimizes the weighted distance between the unknown matrix and an initial guess
of it, subject to constraints on the row- and column-sums of the unknown matrix (sectoral
gross output and total input expenditure in our setting). The GRAS algorithm (GünlükŞenesen and Bates 1988, Junius and Oosterhaven 2003) extends RAS. It imposes additional
constraints such that the unknown matrix, once aggregated to a set of broad sectors, is
consistent with a known input-output matrix at the same broad sectoral. The GRAS algorithm is recommended by the Brazilian government to estimate the national input-output
table when such a table is not available for the current year (Guilhoto et al. 2010b), and
to estimate the regional input-output table using region-sectoral gross output, total input,
and the national level input-output table (Guilhoto et al. 2010a). In our setting, the initial
guess is the “normalized” imports table, where we multiply each entry in the imports table
with the ratio of total input expenditure (the sum of all entries in the Niv level input-output
table) to total imported intermediate input (the sum of all entries in the imports table).65
Our constraints are the data on CNAE4 level gross output, input expenditure, and the Niv
level input-output table. Following Temursho et al. (2021), we set up the problem as:
64
For example, an upstream sector’s share in other sectors’ domestic expenditure can be different from its
share on other sectors’ imports. Similarly, a downstream sector’s share in the sales of other sectors’ domestic
producers may not equal to its share in other sectors’ foreign producers.
65
Consequently, total expenditure in the “normalized” imports table, as we add up all elements in the
matrix, equals total input expenditure in the Brazilian economy.
62
S+F
X
(A.1)
xO
ij zij = ui , ∀i ∈ {1, ..., S}
(A.2)
xO
ij zij = vj , ∀j ∈ {1, ..., S}
(A.3)
{zij }
s.t.
S
X
|xO
ij |zij log(zij )
minf (Z) =
i=1 j=1
S
X
j=1
S+F
X
i=1
XX
xO
ij zij = wIJ , ∀I ∈ {1, ..., M } , J ∈ {1, ..., M } ,
(A.4)
i∈ΩI i∈ΩJ
where {i}Si=1 , {j}Si=1 denotes CNAE4 sectors. i = F denotes the final sector (final demand
and exports). I, J denotes Niv sectors, ΩI , ΩJ denotes the CNAE4 sectors in Niv sectors.
O i=S,j=S
Niv sectors do not overlap–ΩI ∩ ΩJ = ∅, ∀I 6= J, and ∪M
xij i=1,j=1 –the
I=1 ΩI = {1, ..., S}.
initial guess–represents the normalized imports table–imports of sector j output by sector i.
66
zij represents the distance between the
xO
F j denotes sales by sector j to the final sector.
normalized imports table and the unknown, complete input-output table. ui denotes sector
i total input expenditure. vj denotes sector j supplies (imports and domestic products). wIJ
denotes Niv sector J total output used by Niv sector I. The objective function minimizes
the weighted distance between the imports table and the complete input-output table. The
complete I-O table is consistent with the following information in the data: CNAE4 level
total input expenditure according to constraint A.2, CNAE4 level gross output according
to constraint A.3, and the cross-sector flows in the complete input-output table on the Niv
level according to constraint A.4. Junius and Oosterhaven (2003), Miller and Blair (2009)
and Temursho et al. (2021) show that the solution to this problem is unique, and Temursho
et al. (2021) provides an iterative solver that can give the solution.
Armed with the solved {zij }, we recover the complete input output table with xij = zij xO
ij .
We can then get the input-output coefficients. We define the sector expenditure share,
γij =
xij
,
ui
as the share of input that sector i spends on sector j. The numerator denotes
the input demand of sector i from sector j and the denominator denotes aggregate input
66
This initial guess is set to sector j’s gross output plus this sector’s imports minus
63
PJ
i=1
xO
ij .
demand of sector i. We define the sector output share, Sij =
xij
,
vj
as the share of output
that sector j sells to sector i. The numerator denotes the sales to sector i from sector j and
the denominator denotes the production of sector j. With these two sets of market shares,
we can construct the main upstream and downstream sectors as well as average downstream
and upstream tariffs.
A.2.2
Main upstream and downstream
For each sector i, we define its main upstream sector j(i) as the sector that sells the largest
share of output to sector i:
j(i) = arg max Sij .
j
For each sector j, we define its main downstream sector i(j) as the sector that spends the
largest share of input on sector j:
i(j) = arg max γij .
i
In the event studies we focus on non-service main upstream and downstream sectors.
A.3
Endogeneity of AD Tariffs
In this section, we show that products and sectors that are targeted by an AD investigation
are not similar to the ones that are not investigated. Products investigated have higher trade
volume and lower prices. Moreover, they have an increasing trend in trade volume and a
decreasing trend in prices. Sectors investigated have higher employment and wage, and have
a decreasing trend in wage. These findings suggest that one cannot compare investigated
products to non-investigated ones, because one cannot tease apart the effect of AD tariffs
from product level trends.
64
We use the following model to calculate the probability of investigation:
Ip,o,t {Investigation} = β0 log(Importsp,o,t−1 ) + β1 log(Pricep,o,t−1 ) + µp,o + µt,o + p,o,t ,
(A.5)
where Ip,o,t {Investigation} is a dummy taking one if there is an AD investigation against
product p, from destination o, in year t; Importsp,o,t−1 are imports of product p, from origin
o, in year t − 1; µp,o is a production-origin fixed effect, and µt,o is a time-origin fixed effect.
Column 1 and 2 of Table A.4 show that AD investigations are more likely to target higher
volume and lower price product-destinations. Column 3 and 4 show that investigations are
more likely to target product-destinations in a increasing volume and decreasing price trend.
Columns 5 and 6 show that AD tariffs are also more likely to be implemented on higher
volume and lower price product-destinations, and Columns 7 and 8 show that they are also
implemented on products in a decreasing price and increasing volume trend. Given that
AD tariffs should be implemented on lower price producers, it is expected that they are in
an increasing volume and decreasing prices trend. As that investigated products are not in
the same trend as non-investigated ones, a comparison between them would deliver a biased
estimate–one cannot tease apart the effect of an AD tariff from a pre-existing trend.
Table A.4: Probability of Dumping Investigation and Anti-Dumping Tariff
log(Importst−1 )
log(P ricet−1 )
N
R2
Year FE
Product-Destination FE
Year-Country FE
(2)
(1)
I {Investigation} I {Investigation}
0.000121***
0.000120***
(0.000)
(0.000)
-0.0000746***
-0.0000750***
(0.000)
(0.000)
1542747
1542747
0.001
0.001
X
(3)
(4)
(5)
I {Investigation} I {Investigation} I {AD Tariff}
0.000108***
0.000111***
0.0000983***
(0.000)
(0.000)
(0.000)
-0.0000168**
-0.0000181**
-0.0000575***
(0.033)
(0.023)
(0.000)
1509536
1508929
1542747
0.087
0.090
0.001
X
X
X
X
(6)
I {AD Tariff}
0.0000972***
(0.000)
-0.0000589***
(0.000)
1542747
0.001
X
(7)
I {AD Tariff}
0.0000924***
(0.000)
-0.0000201***
(0.003)
1509536
0.090
X
X
(8)
I {AD Tariff}
0.0000941***
(0.000)
-0.0000207***
(0.002)
1508929
0.093
X
X
Notes: This table shows the estimated parameters of model A.5. I {Investigation} is a dummy taking one if that product-destination has an AD investigation starting at that year. I {AD Tariff} is a dummy taking one if a product had
a AD investigation starting that year. log(Importst−1 ) the lagged FOB imports in dollars, and log(P ricet−1 ) is lagged prices. Trade data is from the the Secretary of International Trade of the Ministry of Economy in Brazil and AD data
is from the Global Anti-dumping database. Standard errors, cluster at the origin-product level, are in parenthesis.
Table A.5 studies the relationship between AD investigations and firm-level labor market
outcomes. Column 1 and 2 show that investigations are more likely to start on sectors
that have higher wage, higher employment, and smaller number of establishments. Column
3 shows that investigations are more likely to start in sectors that have increasing wage,
increasing number of workers, and decreasing number of establishments trends. Column
65
4-6 show that the same relationship holds between AD tariffs and firm-level labor market
outcomes.
Table A.5: Probability of Dumping Investigation and AD Tariff
log(Avg. Waget−1 )
log(N. Workerst−1 )
log(N. Establishmentst−1 )
N
Year FE
Sector FE
(1)
I {Investigation}
0.000102***
(0.000)
0.000358***
(0.000)
-0.000503***
(0.000)
36677266
(2)
(3)
I {Investigation} I {Investigation}
0.000335***
0.0000146
(0.000)
(0.191)
0.000333***
-0.00000987**
(0.000)
(0.027)
-0.000529***
-0.0000349**
(0.000)
(0.034)
36677266
33294706
X
X
X
(4)
I {AD Tariff}
0.0000770***
(0.000)
0.000343***
(0.000)
-0.000489***
(0.000)
36677266
(5)
I {AD Tariff}
0.000294***
(0.000)
0.000319***
(0.000)
-0.000516***
(0.000)
36677266
X
(6)
I {AD Tariff}
0.0000130
(0.236)
-0.00000676+
(0.122)
-0.0000363**
(0.025)
33294706
X
X
Notes: This table shows the estimated parameters of a regression of AD policy and firm level characteristics. I {Investigation} is a dummy taking one if that product-destination has
an AD investigation starting at that year. I {AD Tariff} is a dummy taking one if a product had a AD investigation starting that year. log(N. Workerst−1 ) is lagged employment and
log(N. Establishmentst−1 ) is lagged number of establishments. Trade data is from the the Secretary of International Trade of the Ministry of Economy in Brazil and AD data is from the
Global Anti-dumping database. Standard errors, cluster at the origin-product level, are in parenthesis.
A.4
A.4.1
Validation
Predicting Tariffs
According to the WTO regulation, AD tariffs should be equal to the price exporters charged
in their home country minus the price they charged in Brazil. Therefore, if we had international data in prices we could test if WTO regulations are being followed with
τp,c = β
πp,c,c − πp,c,BR
+ p,c ,
πp,c,BR
where τp,c is the AD tariff imposed against product p from country c, πp,c,c is the price
charged by the exporter of product p, from country c, in country c, πp,c,BR is the price of
product p, from country c, in Brazil. If WTO regulations are being followed, β = 1.
But, life is not so easy. We do not observe the price charged by the exporter in their
home market. Instead, we approximate that with the distribution of prices and the AD
policy of other countries. The idea is that the distribution of prices of good p from country
c and the AD tariffs imposed against product p from country c contain indirect information
66
on the price charged in country c. We use the specification:
0
min
max
p75
p25
median
avg
θ + ,
+ Xp,c
+ β2 τp,c
+ β2 τp,c
+ β2 τp,c
+ β2 τp,c
+ β2 τp,c
τp,c = β1 τp,c
avg
where τp,c
is the AD tariff if the price charged in country c and product p was the average
price charged from imports of product p from country c across all countries in the world
min
max
p75
p25
median
uses the tariff that would
, and τp,c
, τp,c
, τp,c
, τp,c
except for Brazil. Similarly, τp,c
have been implemented if the price charged by the supplier in its home country were the
median, the 25th percentile, the 75th percentile, the maximum price, or the minimum price,
respectively. Xp,c is a set of fixed effects for the number of countries imposing AD tariff
against pair (p, c) or imposing AD an investigation against (p, c).
A.4.2
Placebo Tests
In this section, we discuss the results of two placebo tests. First, we evaluate if the results
are driven by sectoral shocks. To do that we match each sector that received an AD tariff,
to another sector that did not receive an AD tariff but was in a similar trend before the
introduction of the tariff. We take these matches as the fake treatment group and compare
it to the same control group as in the main specification. Second, we evaluate if the results
are driven by sectoral trends. To do that we implement the difference-in-differences strategy
pretending that the AD tariff was implemented 5 years before its de-facto implementation.
These placebo tests support that results are not driven by sectoral shocks or trends.
To test if results are driven by sectoral shocks, we match each sector that faces an AD
investigation to a sector that belongs to the same large sector group, had similar employment
and international trade trends but did not face an AD investigation. More specifically, for
each 4-digit sector i that has an AD investigation, we match it to sector q that is in the
same 1-digit sector and had a similar level of employment and wage bill in the three years
before the beginning of the AD investigation. Then, we treat each firm at sector q as if they
had been affected by the investigation and reproduce regression 4. If a sectoral shock that
affected sectors in a particular trend is behind the results identified, AD tariffs should also
correlate with employment movements at sector q.
67
Table A.6: AD Tariffs and the Distribution of Prices
avg
τp,c
median
τp,c
p25
τp,c
p75
τp,c
max
τp,c
min
τp,c
Level
Sample
N
R2
adj. R2
(1)
AD tariff
0.731***
(0.0511)
-0.237***
(0.0622)
3.676***
(0.407)
-0.752***
(0.0647)
-0.000000196***
(5.01e-08)
-5.787***
(0.639)
Product X Origin
All
129
0.800
0.680
(2)
(3)
(4)
AD tariff
AD tariff
AD tariff
0.550***
0.101*
-0.0400
(0.0593)
(0.0554)
(0.0992)
-1.095***
0.0316
0.112
(0.153)
(0.117)
(0.0931)
2.732***
-0.183
-0.0272
(0.442)
(0.190)
(0.232)
0.0735
0.0000453
0.00163
(0.164)
(0.00457)
(0.00319)
0.000000238
2.98e-08 0.00000379***
(0.000000241)
(5.61e-08) (0.000000562)
-4.072***
-0.0230
-0.282
(0.722)
(0.219)
(0.483)
Product X Origin
Sector
Sector
Positive Tariff
All
Positive Tariff
100
62
49
0.904
0.830
0.972
0.828
0.390
0.853
Notes: This table shows the estimated parameters of a regression of AD policy on different values of predicted
πp,avg −πp,c,BR
where πp,c,avg is the average price charged by country c for good p to all other
πp,c,BR
median = πp,median −πp,c,BR where π
countries except Brazil , τp,c
p,median is the median price charged by
πp,c,BR
median = πp,median −πp,c,BR where π
country c for good p to all other countries except Brazil, τp,c
p,median is
πp,c,BR
π
−πp,c,BR
p25
the median price charged by country c for good p to all other countries except Brazil, τp,c
= p,p25
πp,c,BR
avg
tariffs. τp,c
=
where πp,p25 is the 25th percentile of prices charged by country c for good p to all other countries except
p75
=
Brazil, τp,c
πp,p75 −πp,c,BR
πp,c,BR
where πp,p75 is the 75th percentile of prices charged by country c for good p
max =
to all other countries except Brazil, τp,c
πp,max −πp,c,BR
πp,c,BR
where πp,max is the maximum price charged by
min =
country c for good p to all other countries except Brazil, and τp,c
πp,min −πp,c,BR
πp,c,BR
where πp,min is the
minimum price charged by country c for good p to all other countries except Brazil. ADtarif f is the AD tariff
imposed at the product level. Trade data is from the the Secretary of International Trade of the Ministry
of Economy in Brazil and the United Nations Comtrade, and AD data is from the Global Anti-dumping
database. Standard errors, cluster at the origin-product level, are in parenthesis.
68
The results of the placebo test are presented in Figure A.3a. It indicates that there is
no correlation between employment in sectors that did not receive an AD tariff but had a
similar trend in employment and the wage bill, and AD tariffs. We conclude that results are
not driven by sectoral shocks affecting sectors with similar employment characteristics.
We also test if results are driven by sectoral trends. To do so we implement regression
4 but we pretend that the investigation started 5 years before its de-facto implementation.
Figure A.3b shows that, as expected, we don’t find any difference in the wage bill between
treatment and control five years before the introduction of the tariff.
A.4.3
Political Connection and Other Policies
We show that AD tariffs are not correlated with political engagement or other policies. If
firms protected by a tariff are also targeted by other policies, we will not be able to tease
apart the effect of tariffs from the effect of these other policies. Table A.7 tests that for
a series of prominent policies in Brazil. It shows that AD tariffs do not correlate with
signing a procurement contract with the federal government (Column 2) nor with receiving
a subsidized loan (Column 3).
During the 2000s, the Brazilian government implemented policies facilitating access to
the stock market, reducing taxes, and privatizing state-owned firms. Columns 4 to 7 show
that these policies do not correlate with AD tariffs.
There is no correlation between tariffs and campaign contributions, according to A.7.
Therefore, this result indicates that it is unlikely that firms targeted by AD tariff are systematically lobbying for other blessings from the government.
AD tariffs do not correlate with other international trade policies. Columns 8 and 9
show that treatment and control groups are equally exposed to changes in MFN tariffs and
preferential tariffs.
A.4.4
Other Shocks
In this section, we show that heterogeneous exposure to aggregate shocks cannot explain our
results. In particular, we focus on important shocks to the Brazilian economy in the past
years – the exchange rate fluctuation and trade liberalization, discussed in Dix-Carneiro and
69
(a) Placebo Test with Fake Investigated Sectors
(b) Placebo Test with Fake Investigation Year
Description: Figure A.3a shows the coefficients of regression 4 but using placebo firms. For each sector with an AD
investigation, we match it to a sector in the same 4-digit classification that had similar employment and wage-bill in the 3 years
before the investigation. Then, we treat the matched sector as if it was subject to the AD investigation and tariff. Figure A.3a
shows the coefficients of regression 4 pretending that the AD investigation started 5 years before it actually did.
70
Table A.7: AD Tariffs are not Correlated with Political Connection and Other
Policies
(2)
(1)
I{Campaign Contribution} I{Gov. Demand}
0.000937*
-0.000709
(0.00101)
(0.000507)
τs,t
N
R2
# Firms
Mean Dep. Var
Mean Ind. Var
Firm FE
Year FE
20857
0.501
7108
.013
.96
Yes
Yes
(3)
I{Subsidize Loan}
-0.0000218
(0.000140)
(4)
I{Publicly Traded}
-0.000324
(0.000306)
154641
0.174
7111
.001
.96
Yes
Yes
154591
0.536
7111
.006
.96
Yes
Yes
81134
0.516
7109
.019
.96
Yes
Yes
(5)
(6)
I{Public} I{Multinational}
-0.0000105
-0.00000956
(0.000181)
(0.0000106)
154636
0.193
7111
.004
.96
Yes
Yes
154639
0.085
7111
0
.96
Yes
Yes
(7)
(8)
(9)
I{Simples} MFN Tariff Tariffs
-0.000561
0.0871
0.0435
(0.00130)
(0.104)
(0.0940)
108408
0.751
7111
.525
.96
Yes
Yes
154641
0.947
36
17.65
.96
Yes
Yes
142808
0.945
36
17.919
.96
Yes
Yes
This table presents the estimated parameters of model 3. The sample is composed of firms in sectors that produce the product under an AD investigation. We limit the sample to the set of firms observed 5-years around the AD investigation.
I{Campaign Contribution} is a dummy taking one if the firm has made a campaign contribution in the past election, I{Gov. Demand} is a dummy taking one if the firm has won a government procurement, I{Subsidize Loan} is a dummy if the firm
has collected a subsidized loan from the government, I{Publicly Traded} is a dummy if the firm is publicly traded, I{Public} is a dummy if the firm is owned by the government, I{Multinational} is a dummy if the firm is part of a multinational
corporation, I{Simples} is a dummy if the firm is part of the Simples plan, which is a plan with lower taxes and simplified tax filling, MFN Tariff is the most favored nation tariff, i.e., the tariff imposed by Brazil to other WTO member, and Tariffs
mid is the average AD tariff imposed on products produced by the sector of each firm. Standard errors are clustered at the firm level for columns 1 to 7 and at the sector level for columns 8 and 9. We
is the tariff imposed by Brazil excluding AD. τs,t
cluster tariffs at the sector level because they do not vary at the firm level.
Kovak 2015 and Dix-Carneiro and Kovak 2017.
To control for heterogeneous exposure to exchange rate fluctuation, we use the following
model:
mid
0
yi,s,t = θτs,t
+ βIs,t {After AD} + αs Et + Xi,s,t
κ + ηi + ηt + i,t
(A.6)
where αs is a parameter capturing the correlation of exchange rate fluctuation Et and sector
s labor outcomes. Equivalently, we write similar specification for the effect of AD tariffs
upstream and downstream. Tables A.8, A.9, and A.10 shows that AD tariffs increase employment at midstream firms, decreases it downstream, and has no effect upstream, as we
have found on the main specification.
One could be worried that we are capturing reminiscences of the Brazilian trade liberalization experience. To test if this is the case, we use the following functional form
0
mid
yi,s,t = θτs,t
+ βIs,t {After AD} + αt Libs + Xi,s,t
κ + ηi + ηt + i,t
(A.7)
where Libs is the tariff change between 1995 and 1990 calculated by Dix-Carneiro and Kovak
2017. αt is an year-specific parameter. Tables A.8, A.9, and A.10 shows that AD tariffs increase employment at midstream firms, decreases it downstream, and has no effect upstream,
as we have found on the main specification.
71
Table A.8: Effect of AD Tariffs on Midstream Firms Controlling for Shocks
mid
τs,t
Control:
N
R2
# Firms
Mean Dep. Var
Mean Ind. Var
Firm FE
Year FE
(1)
log(# Workers)
0.0215***
(0.00378)
(2)
log(Wage Bill)
0.0225***
(0.00412)
(3)
log(# Workers)
0.0227***
(0.00557)
(4)
log(Wage Bill)
0.0232***
(0.00617)
Exchange Rate
132816
0.811
6277
2.684
1.07
Yes
Yes
Exchange Rate
132816
0.846
6277
10.062
1.07
Yes
Yes
Trade Liberalization
128745
0.808
6098
2.684
1.07
Yes
Yes
Trade Liberalization
128745
0.843
6098
10.062
1.07
Yes
Yes
This table presents the estimated parameters of model 3. The sample is composed of firms in sectors that produce the product under
an AD investigation. We limit the sample to the set of firms observed 5-years around the AD investigation. I{Campaign Contribution}
is a dummy taking one if the firm has made a campaign contribution in the past election, I{Gov. Demand} is a dummy taking one if the
firm has won a government procurement, I{Subsidize Loan} is a dummy if the firm has collected a subsidized loan from the government,
mid is
I{Publicly Traded} is a dummy if the firm is publicly traded, and I{Public} is a dummy if the firm is owned by the government. τs,t
the average AD tariff imposed on products produced by the sector of each firm. Standard errors are clustered at the firm level.
Table A.9: Effect of AD Tariffs on Downstream Firms Controlling for Shocks
down
τs,t
Control:
N
R2
# Firms
Mean Dep. Var
Mean Ind. Var
Firm FE
Year FE
(1)
log(# Workers)
-0.0263
(0.0220)
(2)
log(Wage Bill)
-0.0336
(0.0242)
(3)
log(# Workers)
-0.0553*
(0.0286)
(4)
log(Wage Bill)
-0.1000***
(0.0312)
Exchange Rate
182790
0.813
8686
2.412
.07
Yes
Yes
Exchange Rate
182790
0.834
8686
9.599
.07
Yes
Yes
Trade Liberalization
128745
0.808
6098
2.684
1.07
Yes
Yes
Trade Liberalization
128745
0.843
6098
10.062
1.07
Yes
Yes
This table presents the estimated parameters of model 3. The sample is composed of firms in sectors that produce the product under
an AD investigation. We limit the sample to the set of firms observed 5-years around the AD investigation. I{Campaign Contribution}
is a dummy taking one if the firm has made a campaign contribution in the past election, I{Gov. Demand} is a dummy taking one if the
firm has won a government procurement, I{Subsidize Loan} is a dummy if the firm has collected a subsidized loan from the government,
mid is
I{Publicly Traded} is a dummy if the firm is publicly traded, and I{Public} is a dummy if the firm is owned by the government. τs,t
the average AD tariff imposed on products produced by the sector of each firm. Standard errors are clustered at the firm level.
72
Table A.10: Effect of AD Tariffs on Upstream Firms Controlling for Shocks
up
τs,t
Control:
N
R2
# Firms
Mean Dep. Var
Mean Ind. Var
Firm FE
Year FE
(1)
log(# Workers)
0.0106
(0.00864)
(2)
log(Wage Bill)
0.00966
(0.00849)
(3)
log(# Workers)
0.000974
(0.0109)
(4)
log(Wage Bill)
-0.000780
(0.0120)
Exchange Rate
74735
0.821
3694
2.55
.29
Yes
Yes
Exchange Rate
74735
0.844
3694
9.8
.29
Yes
Yes
Trade Liberalization
67536
0.823
3352
2.55
.29
Yes
Yes
Trade Liberalization
67536
0.846
3352
9.8
.29
Yes
Yes
This table presents the estimated parameters of model 3. The sample is composed of firms in sectors that produce the product under
an AD investigation. We limit the sample to the set of firms observed 5-years around the AD investigation. I{Campaign Contribution}
is a dummy taking one if the firm has made a campaign contribution in the past election, I{Gov. Demand} is a dummy taking one if the
firm has won a government procurement, I{Subsidize Loan} is a dummy if the firm has collected a subsidized loan from the government,
mid is
I{Publicly Traded} is a dummy if the firm is publicly traded, and I{Public} is a dummy if the firm is owned by the government. τs,t
the average AD tariff imposed on products produced by the sector of each firm. Standard errors are clustered at the firm level.
A.5
A.5.1
Robustness
Controls
Table A.11: Robustness of the Effect of AD Tariffs on the National Producers
τs,t
N
R2
1 Digit Sector FE
2 Digit Sector FE
# Product Invest.
# Product AD
Tariffs
(1)
log(Wage Bill)
0.0111***
(0.00211)
(2)
log(Wage Bill)
0.0156***
(0.00358)
(3)
log(Wage Bill)
0.0121***
(0.00417)
(4)
log(Wage Bill)
0.0186***
(0.00390)
(5)
log(Wage Bill)
0.0130***
(0.00423)
(6)
log(Wage Bill)
0.0191***
(0.00397)
(7)
log(Wage Bill)
0.0156***
(0.00446)
119368
0.863
119368
0.863
X
119368
0.865
119368
0.863
X
119368
0.865
119368
0.863
X
119368
0.863
X
X
X
X
X
X
X
X
X
Description: This table presents the estimated parameters of model 3. The sample is composed of firms in sectors producing the product under AD investigation. we constrain the sample
to the set of firms observed after and before the AD investigation, that have more than 10 observations and more than one worker. These sample restrictions are maid to avoid compositional
change and special firms. log(Wage Bill) is the log of total labor expenditure of the firm. log(Number Workers) is the log of the total number of workers of the firm. log(Monthly Earnings) is
the average monthly earnings of workers at that firm. log(Establishments) is the log of number of establishments of the firm. I{exporter} is a dummy taking one if the exported any product
that year and I{importer} is a dummy for importing. AD tariff is the average AD tariff imposed against products produced by the sector of each firm. The sample goes from 1995 to 2016.
73
Table A.12: Effect of AD Tariffs on Downstream Firms
(1)
log(Wage Bill)
(2)
log(Wage Bill)
τs,t
-0.118***
(0.0239)
-0.0957***
(0.0247)
N
R2
182790
0.831
182790
0.832
τs,t
-0.0738***
(0.0181)
-0.0270
(0.0190)
969621
0.833
969619
0.834
X
N
R2
1 Digit Sector FE
2 Digit Sector FE
# Product Invest.
# Product AD
Tariffs
(3)
(4)
(5)
log(Wage Bill) log(Wage Bill) log(Wage Bill)
Sample: Main Downstream
-0.0124
-0.0857***
-0.0142
(0.0327)
(0.0244)
(0.0330)
182790
0.836
182790
182790
0.833
0.836
Sample: All Downstream
-0.0217
-0.0430**
-0.0249
(0.0220)
(0.0191)
(0.0221)
969611
0.835
969619
0.834
X
X
X
969611
0.835
X
X
(6)
log(Wage Bill)
(7)
log(Wage Bill)
-0.0820***
(0.0250)
-0.0998***
(0.0373)
182790
0.833
182790
0.833
-0.0372*
(0.0192)
-0.0362*
(0.0208)
969619
0.834
X
969619
0.834
X
X
X
X
X
Description: This table presents the estimated parameters of model 3. The sample is composed of firms in sectors that produce the product under an AD investigation. We limit the sample
to the set of firms observed 5-years around the AD investigation. log(Wage Bill) is the log of the firm’s total labor expenditure. Column 2 adds a 1 digit sector interacted with year as control,
column 3 has a 2 digit sector interacted with year as control, column 4 has a 1 digit sector-year FE with dummies for the number of product investigated, column 5 has a 2 digit sector-year
FE with dummies for the number of product investigated. column 6 has as control a 1 digit sector-year FE, number of product investigated, and number of products with AD. column 7 has as
control a 1 digit sector-year FE, number of product investigated, and tariff upstream and downstream. τs,t is the average AD tariff downstream. Standard errors are clustered at the firm level.
Table A.13: Effect of AD Tariffs on Upstream Firms
(1)
log(Wage Bill)
(2)
log(Wage Bill)
τs,t
0.00985
(0.00602)
-0.00219
(0.00707)
N
R2
74735
0.844
74735
0.844
τs,t
0.0174***
(0.00386)
0.00833**
(0.00395)
3238468
0.834
3238468
0.835
X
N
R2
1 Digit Sector FE
2 Digit Sector FE
# Product Invest.
# Product AD
Tariffs
(3)
(4)
(5)
log(Wage Bill) log(Wage Bill) log(Wage Bill)
Sample: Main Upstream
-0.0114
-0.000384
-0.0156
(0.0257)
(0.00809)
(0.0244)
74713
0.845
74735
74713
0.844
0.845
Sample: All Upstream
0.0115**
0.00637
0.0114**
(0.00569)
(0.00401)
(0.00576)
3238468
0.835
3238468
0.835
X
X
X
3238468
0.835
X
X
(6)
log(Wage Bill)
(7)
log(Wage Bill)
-0.00320
(0.00826)
-0.00364
(0.00902)
74735
0.844
74735
0.844
0.00486
(0.00403)
0.00283**
(0.00122)
3238468
0.835
X
3238468
0.835
X
X
X
X
X
Description: This table presents the estimated parameters of model 7. The sample is composed of firms in sectors that produce the product under an AD investigation. We limit the sample
to the set of firms observed 5-years around the AD investigation. log(Wage Bill) is the log of the firm’s total labor expenditure. Column 2 adds a 1 digit sector interacted with year as control,
column 3 has a 2 digit sector interacted with year as control, column 4 has a 1 digit sector-year FE with dummies for the number of product investigated, column 5 has a 2 digit sector-year
FE with dummies for the number of product investigated. column 6 has as control a 1 digit sector-year FE, number of product investigated, and number of products with AD. column 7 has as
control a 1 digit sector-year FE, number of product investigated, and tariff upstream and downstream. τs,t is the upstream tariff. Standard errors are clustered at the firm level.
74
A.5.2
All Connected Sectors
In this section, we identify both the effects of tariffs on firms and their propagation. Following
Acemoglu et al. (2014) and Bown et al. (2021), our specification is given by
up
down
0
yi,s,t = θτs,t + θup τ̃s(u),t
+ θdown τ̃s(d),t
+ Xi,s,t
κ + ηi + ηt + i,t ,
(A.8)
up
where τs,t is the average AD tariff against sector s, τ̃s(u),t
is the average exposure of firm i in
down
sector s to upstream tariffs, τ̃s(d),t
is the average exposure of firm i in sector s to downstream
0
tariffs. Xi,s,t
is a set of controls, which include a 1-digit sector fixed effect interacted with
year, and a dummy for the number of investigations. We run this regression on all firms–not
only the ones exposed to AD investigation as we studied before.
This specification has two drawbacks. The first one is that to identify the causal effect of
tariffs we have to assume that all sectors are in parallel trends. Given that we expect sectors
with AD investigations to be in a declining trend due to the institutions of AD investigations
discussed in Section 2, this is a strong assumption. A second drawback is that we cannot
test if sectors were in similar trends before the introduction of the tariffs.
Still, despite the drawbacks, Table A.14 confirms the result that AD tariffs increase
employment at midstream firms and do not significantly affects upstream firms.
A.5.3
Sectoral Regressions
In this section we study the effect of AD tariffs on sector-level aggregate variables. First, we
show that AD tariffs do not lead to the entry or exit of firms. Second, we show that, even on
the sector level (without exploiting firm-level variations as we did before), we find that AD
tariffs lead to an increase in employment midstream, and it decreases wages downstream.
AD tariffs do not lead to the entry or exit of firms or establishments, according to results
in Table A.15. Column 1 of Table A.15 shows the effect of tariffs on the number of firms in
the midstream, main downstream, and main upstream sectors. In none of these specifications
we find that tariffs lead to more or less firms in the sector. In column 2 of Table A.15 shows
the effect of tariffs on the total number of establishments. Once again, we find that AD
tariffs do not lead to more establishments midstream, downstream, or upstream.
75
Table A.14: Effect of AD Tariffs through the Input-Output Connection
mid
τs,t
up
τ̃d(s),t
down
τ̃d(s),t
N
R2
# Firms
Mean Mid. Tariff
Mean Up. Tariff
Mean Down. Tariff
(1)
log(# Workers)
0.00910***
(0.00163)
-0.00965*
(0.00510)
-0.0158
(0.0106)
(2)
log(Wage Bill)
0.00958***
(0.00184)
-0.00588
(0.00550)
-0.0551***
(0.0116)
(3)
IExporter
0.00325***
(0.000494)
-0.00445***
(0.00105)
0.0206***
(0.00234)
(4)
IImporter
0.00294***
(0.000503)
-0.00551***
(0.00106)
0.0250***
(0.00234)
3142280
0.814
180618
.1
.05
.05
3142280
0.840
180618
.1
.05
.05
3142280
0.586
180618
.1
.05
.05
3142280
0.600
180618
.1
.05
.05
Description: This table presents the estimated parameters of model 3. The sample is composed of firms in sectors
that produce the product under an AD investigation. We limit the sample to the set of firms observed 5-years around
the AD investigation. log(Wage Bill) is the log of the firm’s total labor expenditure. log(Number Workers) is the
log of the total number of workers in the firm. I{exporter} is a dummy that takes one if the protected firm exports
any product that year, I{importer} is a dummy taking one if the protected firm imports any product that year,
mid is the
log(Imports) is the log of expected imports of the firm, and log(Exports) is the log of expected exports. τs,t
up
down
average AD tariff imposed on products produced by the sector of each firm, τs,t
is the AD tariff upstream and τs,t
is the AD tariff downstream. Standard errors are clustered at the firm level.
The fact that AD tariffs do not cause entry or exit of firms is important for two reasons
– the identification of elasticities and our modeling assumptions. First, it guarantees that
our estimates of the effect of AD are not biased. If AD tariffs lead to the entry/exit of firms,
our estimates should be conditional on surviving. Second, in Section 5, based on the fact
that AD tariffs do not affect entry or exit, we build a model without this margin.
We also find that AD tariffs increase employment midstream and it decreases wages
downstream using sectoral aggregate data, as shown in Table A.15.
76
Table A.15: Effect of AD Tariffs on Firms using Sectoral Aggregates
(1)
log(N. Firms)
mid
τs,t
N
R2
0.0141
(0.00877)
(2)
(3)
log(N. Establishments) log(N. Workers)
Midstream
0.0139
0.0265***
(0.00918)
(0.00839)
1079
0.978
1079
0.974
(4)
log(Wage Bill)
0.0334**
(0.0128)
1079
0.927
1079
0.877
Downstream
down
τ̃d(s),t
N
R2
0.0304
(0.100)
0.0735
(0.125)
0.0717
(0.111)
0.00376
(0.0999)
1134
0.979
1134
0.977
1134
0.936
1134
0.905
Upstream
up
τ̃d(s),t
N
R2
-0.00812
(0.0121)
0.000793
(0.0168)
-0.0136
(0.0196)
0.0198
(0.0238)
944
0.988
944
0.981
944
0.970
944
0.964
Description: This table presents the estimated parameters of model 3 aggregated at sector level. The sample is
composed of sectors that produce the product under an AD investigation. log(Wage Bill) is the log of the sector’s
mid is the
total labor expenditure. log(Number Workers) is the log of the total number of workers in the sector. τs,t
up
down
average AD tariff imposed on products produced by the sector of each firm, τs,t
is the AD tariff upstream and τs,t
is the AD tariff downstream. Standard errors are clustered at the sector level.
A.5.4
Instrumental Variables
Exploiting the institutional setting discussed in Section A.4.1, we propose an instrument for
AD tariffs. We instrument AD tariffs in Brazil with the AD tariffs imposed other countries.
The rationale for that is the following. A supplier exporting with low prices to Brazil is
likely to export with low prices to other countries as well. Therefore, if a supplier meets the
requirements for an AD tariff in Brazil due to its low prices, it is also likely to meet these
requirements in other countries. Since the AD policy outside of Brazil is unlikely to directly
affect the Brazilian labor market, the instrument is exogenous to Brazilian employment.
77
We instrument τs,t , the average AD tariff on products of sector s, with a set of dummies
for the number of investigations and tariffs imposed against products of sector s in year t in
all other countries except Brazil. The first stage is
τs,t =
X
βoI It {o Countries Investigated Sector s} +
(A.9)
o
X
0
βoT It {o Countries Imposed AD Tariff on Sector s} + Xs,t
κ + s,t ,
o
where I {o Countries Investigated Sector s} equals one if countries except Brazil conduct
o AD investigations on sector s in year t. I {o Countries Imposed AD Tariff on Sector s}
equals one if countries except Brazil impose o AD tariffs on sector s in year t. We instrument
the exposure to tariffs downstream and upstream similarly.
Results in Table A.16 confirm the finding that AD tariffs increase employment midstream,
propagates downstream, but do not affect upstream firms. Column 1 and 2 of Table A.16
show the effect of AD tariffs on midstream firms using as instrument the AD policy of
countries outside Brazil. As in the baseline model, we limit the sample to the firms that
faced AD investigations. We find that imposing a 100% AD tariff causes a 3% increase
in employment. Column 3 and 4 shows the effect of tariffs downstream. It shows that a
100% AD tariff on all the inputs of a firm causes a 60% decrease in employment. This is
an order of magnitude larger than what we found in the main regressions. An instrument
variable regression identifies the effect of AD tariffs on compliers, i.e., on the set of sectors
that were targeted by both tariffs in Brazil and those outside of Brazil. These sectors are
not necessarily representative of the set of sectors targeted by tariffs in Brazil. Columns 5
and 6 show that there is no effect of tariffs upstream.
A.5.5
Regional Variation
In this section we study the effect of AD tariffs on local labor markets. We exploit heterogeneous sectoral composition across regions to create a measure of heterogeneous exposure of
regions to AD tariffs. We find that midstream tariffs significantly increase employment but
the propagation of tariffs through the input-output connection of firms is not significant.
78
Table A.16: Effect of AD Tariffs with Instruments
(1)
log(# Workers)
0.0319***
(0.00669)
τ
(2)
log(Wage Bill)
0.0316***
(0.00744)
τ down
(3)
log(# Workers)
(4)
log(Wage Bill)
-0.590***
(0.194)
-0.819***
(0.243)
τ up
Investigated Sectors
Sample
N
132816
R2
0.809
# Firms
6277
Mean Dep. Var
2.684
Mean Ind. Var
1.07
Investigated Sectors
132816
0.844
6277
10.062
1.07
Main Downstream
31748
0.831
1458
2.412
.07
Main Downstream
31748
0.843
1458
9.599
.07
(5)
log(# Workers)
(6)
log(Wage Bill)
0.0319
(0.0425)
0.0685
(0.0479)
Main Upstream
41424
0.818
2063
2.55
.29
Main Upstream
41424
0.838
2063
9.80
.29
Description: This table presents the estimated parameters of model 3. The sample is composed of firms in sectors that produce the product under an AD investigation. We limit the sample to
the set of firms observed 5-years around the AD investigation. log(Wage Bill) is the log of the firm’s total labor expenditure. I{exporter} is a dummy that takes one if the protected firm exports
any product that year, I{importer} is a dummy taking one if the protected firm imports any product that year, log(Imports) is the log of expected imports of the firm, and log(Exports) is the
mid is the average AD tariff imposed on products produced by the sector of each firm, τ up is the AD tariff upstream and τ down is the AD tariff downstream. Standard
log of expected exports. τs,t
s,t
s,t
errors are clustered at the firm level.
Denote τs,t the AD tariff impose against sector s in year t. The exposure of region r to
tariff τs,t equals:
reg
τr,t
P
s Employments,t−1 τs,t
= P
,
s Employments,t−1
reg,mid
where Employments,t−1 is employment and τr,t
is the exposure of region r to midstream
reg,down
tariffs. Similarly, we can calculate the exposure of region r to downstream tariffs, τr,t
,
reg,up
and to upstream tariffs, τr,t
.
The regional specification is the following:
reg,mid
reg,down
reg,up
0
yr,t = θmid τr,t
+ θdown τr,t
+ θup τr,t
+ Xr,t
κ + r,t
(A.10)
where yr,t is the log of a labor outcome in region r and year t and Xr,t is a set of controls
containing the weighted number of investigations, pre-period log employment interacted with
year, and pre-period log wage interacted with year.
Table A.17 shows the main results of this section. Column 1 and 2 show that midstream
tariffs have a large impact on employment and wage bill in local labor markets. We also
find that exposure to downstream tariffs has a large point estimate but it is not significant.
Columns 3 to 5 show that downstream tariffs decrease employment of workers with high
79
Table A.17: Effect of AD Tariffs on Regional Labor Markets
mid
τs,t
down
τ̃d(s),t
up
τ̃d(s),t
N
R2
# Regions
Mean Midstream Tariff
Mean Downstream Tariff
Mean Upstream Tariff
Mean Ind. Var
(1)
log(# Workers)
0.343***
(0.106)
-0.901
(0.859)
-0.101
(0.141)
(2)
log(Wage Bill)
0.312**
(0.126)
-1.428
(1.023)
-0.0998
(0.167)
(3)
log(# HS Dropout)
0.244**
(0.115)
0.259
(0.937)
0.155
(0.153)
(4)
log(# HS Complete)
0.312**
(0.126)
-3.476***
(1.024)
0.269
(0.168)
(5)
log(# More HS)
0.341**
(0.159)
-6.269***
(1.296)
-0.0884
(0.212)
14367
0.982
558
.01
0
.01
9.452
14367
0.981
558
.01
0
.01
16.806
14364
0.977
558
.01
0
.01
8.831
14358
0.979
558
.01
0
.01
8.177
14341
0.975
558
.01
0
.01
7.144
Description: This table presents the estimated parameters of model A.10. log(# Workers) is the log of total employment in the region, log(Wage Bill) is the log of total
wagebill in the region, log(# HS Dropout) is the log of high-school dropouts in the region, log(# HS Complete) is the log of employment of workers with high-school complete,
mid is the average AD tariff imposed on products produced by the sector of each firm,
and log(# More HS) is the log of employment of workers with more than high-school. τs,t
up
down is the AD tariff downstream. Standard errors are clustered at the firm level.
τs,t
is the AD tariff upstream and τs,t
school complete and more than high school but do not affect employment of workers with
less than high school. Finally, once again we find that tariffs do not propagate downstream.
B
B.1
Model Appendix
Model
Proof of Equation 13: Conditional on a sector, the household’s optimal sectoral consumption csr (ω) and labor supply ls (ω) are independent from their utility shocks z s (ω).
Dropping ω, the within-sector problem implies that the household’s sectoral consumption
equals the following:
P r csr =
1−θ
dr (PCr )1−θ
(1 − δ)ws ls
, s>0
1−θ
dr (PCr )1−θ
(1 − δ)b
, s = 0.
(P )
(P )
80
We denote the sectoral consumption shares with αr =
dr (P r )1−θ
,
(P C )1−θ
and the consumer price
index with P C :
S
X
PC =
1
! 1−θ
dr (P r )1−θ
.
r=1
Within a production sector s, we solve the household optimal labor supply which increases
in the sector’s real wage. The supply elasticity equals ψ s :
s
l =
ws
PC
ψs
.
(B.1)
Plugging Equation 9 and B.1 into Equation 10 gives the nonrandom component of welfare
associated with staying in sector s:
Us =
(1 − δ −
ψs
)
1+ψ s
s
ws 1+ψ
C
P
(1−δ)b
C
s>0
s = 0.
P
The elasticity of a household’s welfare in sector s > 0 with respect to the sector’s real
wage equals 1 + ψ s , and is greater than that of the outside sector’s welfare with respect to
the social insurance (which is 1). Real wage increase leads to higher labor supply for the
households that work in the sector and greater-than-unity increase in total real income and
thus welfare of staying in the sector.
With the familiar property of the Frechet distribution, we can solve for the probability
that a household chooses each sector, π s :
πs =
s ηs
ãs ( wC )
P
PS
ηs
µ
b
s ws
0
s=1 ã ( P C ) +ã ( P C )
, s>0
PS
, s = 0,
s
s=1 ã (
s
s
s
s
where η = µ(1+ψ ), ã = a (1 − δ −
µ
b
PC
s
ws η
+ã0
PC
ã0 (
)
)
µ
ψs
)
1+ψ s
81
( PbC )
µ
µ
, s > 0 and ã0 = (a0 (1 − δ)) are parameters.
The population in all sectors adds up to the total population:
S
X
π s ls = L.
(B.2)
s=0
This implies that the share of population in each sector, i.e. sectoral labor supply Ls ,
equals the following:
s λs
ãs ( wC )
P
PS
µL
s λs
s=1 ãs ( PwC ) +ã0 ( PbC )
π s ls
Ls = PS
L=
s s
s=0 π l
P
µ
b
PC
s
ws λ
+ã0
PC
ã0 (
S
s=1
ãs
(
)
)
µ
( PbC )
, s>0
L
, s = 0.
Using properties of the Frechet distribution, we also show that a household’s expected
welfare equals the following:
W =
S
X
! µ1
(ãs U s )µ
s=0
B.2
=
S
X
ãs
s=1
s
w
PC
λs
+ ã0
b
PC
µ ! µ1
.
(B.3)
Model in Changes
In order to compute counterfactuals, we rewrite the model in changes. By doing so we
eliminate the economic fundamentals that are often difficult to calibrate or estimate directly.
These fundamentals include productivity, foreign price, country and product preference,
among others. We use V 0 to denote the value of an ex-post (after a tariff shock) variable V ,
and V̂ =
V0
V
to denote the variable in changes.
First, the change in sectoral labor supply equals the following:
82
L̂s =
s λs
( P̂ŵC )
PS
µ
s λs
s=1 κs ( P̂ŵC ) +κ0 ( P̂1C )
, s>0
PS
, s = 0,
(B.4)
µ
( P̂1C )
λs
µ
1
0
s ŵs
s=1 κ ( P̂ C ) +κ ( P̂ C )
where κs =
Ls
L
denotes the population share in sector s in the baseline equilibrium. κ0
denotes the fraction of population that does not work.
The change in sectoral Brazilian output price is the following:
P̂0s =
ssL (ŵs )1−ρ +
S
X
1
! 1−ρ
0
0
s 1−ρ
sss
M (P̂ )
.
(B.5)
s0 =1
The change in input-output shares equals:
0
0
ŝss
M
=
(P̂ s )1−ρ
(P̂0s )1−ρ
.
0
0
0
ss ss
0
Therefore, the ex-post input-output shares equal: sss
M = ŝM sM .
The change in sector s expenditure shares on country i equals:
ŝsi
=
(P̂is )1−σ
s
(P̂ s )1−σs
,
where the change in sectoral input price equals:
s
(P̂ s )1−σ =
N
X
s
ssi (P̂is )1−σ .
(B.6)
i=0
The change in expenditure share on product line l in sector s import from country i is:
ŝsil
=
(t̂sil )1−ζ
s
(P̂is )1−ζ s
83
,
where the change in sector-origin level output price equals:
s
(P̂is )1−ζ =
X
s
ssil (t̂sil )1−ζ .
(B.7)
l∈Ωsi
Ex-post market clearing condition for sector s labor equates labor demand with labor
supply:
1 s0 s0 s0
s
s 1−σ s
s
s
X
+
E
(
P̂
)
= Ls0 = L̂s Ls .
F0
0
ws0 L 0
(B.8)
Similarly, ex-post market clearing condition for sector s input is the following:
s0
s0
s0
X =P C +
S
X
0
ssMs0
0
0
ss0 0 X s 0
0
+
0
0
s
EFs 0 (P̂0s )1−σ
,
(B.9)
s0 =1
where ex-post consumption:
P s0 C s0 = αs0 (1 − δ)
S
X
!
ws0 Ls0 + bL00 ,
s=1
in which αs0 = αs α̂s is the ex-post consumption expenditure share, and the expression for
α̂s is the following:
α̂s =
where (P C )1−θ =
PS
s=1
(P̂ s )1−θ
(P̂ C )1−θ
,
αs (P̂ s )1−θ .
The ex-post budget constraint for the government:
S
X
bL = δ(
ws0 Ls0 + bL00 ) + T D0 + T R0 ,
00
(B.10)
s=1
in which the ex-post trade deficit and ex-post tariff revenue equal:
T R0 =
S X
N X
X
s=1 i=1
l∈Ωsi
84
s0
X s0 ss0
i sil
τils0
,
ts0
il
(B.11)
0
TD =
S X
N X
X
s0 1
X s0 ss0
i sil s0
til
s=1 i=1 l∈Ωsi
−
S
X
s
(P̂0s )1−σ EFs 0 .
(B.12)
s=1
The change in a household’s expected welfare equals the following:
Ŵ =
s λs
µ ! µ1
S
s
0
X
L
ŵ
1
L
.
+
L P̂ C
L P̂ C
s=1
Equilibrium in changes Given government’s fiscal and tariff policy, {δ, b, {τils }i,l,s }, base0
s s
s
s s
line export, {EFs 0 }, market shares, {κs , αs , ssL , sss
M , si , sil }, elasticities {λ , µ, θ, ρ, σ , ζ }, the
equilibrium is defined as a set of changes in sectoral input prices, {P̂ s }s , and changes in
sectoral wages, {ŵs }s such that
1. Firms maximize profit (equation B.5);
2. The price index satisfied equations B.6 and B.7;
3. The goods market clear in the counterfactual equilibrium, satisfying equation B.9;
4. The labor market clears in the counterfactual equilibrium, satisfying equation B.8;
5. Government budget constraint in the counterfactual equilibrium (equation B.10) holds.
B.3
Calibration
We use the following algorithm to estimate the parameters. We guess a set of parameters,
{ρ, θ} and we provide sector-level annual tariffs, {τts }, to the model.67 For each year, we
solve the counterfactual equilibrium with the model in changes (Section B.2). Then we run
the same panel regression in the model as in the data:68
u
u
ys,t
= β u τ̃s,t
+ ηsu + ηtu + us,t , u ∈ {mid, down, up} .
67
(B.13)
We construct the sector-level tariffs with the country-sector-product level tariffs, {τils }i,l,s , as we discussed
in Section 6.
68
As AD tariffs are the only shock in this counterfactual exercise, parallel trends between treatment
and control group in the model simulated data is naturally guaranteed. Therefore, we do not control the
investigations in these regressions with model simulated data.
85
u
On the left-hand side, ys,t
denotes the sectoral variable of interest in the targeted and non-
targeted moments. They include employment, the wage bill, imports, and exports in the
midstream, main upstream and main downstream sectors (all in logs). On the right hand
u
side, τ̃s,t
denotes the exposures to midstream, downstream and upstream tariffs:
u
=
τ̃s,t
τs,t ,
u = mid
Input Demand of Sector d(s) from Sector s
× τs,t , u = down
Aggregate Input Demand of Sector d(s)
Sales to Sector s from Sector u(s) × τs,t ,
u = up.
Production of Sector u(s)
ηsu denotes the sector fixed effect and ηtu denotes the time fixed effect.
We also include in the non-targeted moments the elasticity of sectoral employment with
respect to midstream, average upstream, and average downstream tariffs when the three
tariffs enter the right-hand side of the regression at the same time. We apply the following
specification to model simulated data:
up
mid
down
ys,t = β1 τs,t
+ β2 τ̃s,t
+ β3 τ̃s,t
+ ηs + ηt + s,t ,
(B.14)
up
down
where τ̃s,t
denotes the downstream tariffs faced by upstream firms and τ̃s,t
denotes the
upstream tariffs faced by downstream firms. Similarly to how we construct them in the
empirical section, they equal the following:
up
τ̃s,t
=
down
τ̃s,t
=
X Sales of Sector s to Sector k
× τk,t ,
Aggregate
Sales
of
Sector
s
k
X Input Demand of Sector s from Sector k
× τk,t .
Aggregate
Input
Demand
of
Sector
s
k
With the model moments computed with model simulated data and these regressions, we
search for the parameters that minimize the sum of squared normalized distance between
these targeted moments in the model and in the data:
86
min(
mid,emp
mid,emp
βdata
− βmodel
mid,emp
βdata
ρ,θ
)2 + (
down,emp
down,emp
βdata
− βmodel
down,emp
βdata
)2
s.t. Equilibrium constraints in Section B.2.
To compute the standard errors of the estimated parameters, we bootstrap the AD tariffs
on the year level–that is, for each bootstrapped sample, we randomly draw years (with
replacement) from the original database and we impose all sectoral tariffs in that year.69 By
doing so, we ensure that every sector in the bootstrapped sample faces the factual midstream,
upstream and downstream tariffs in the year when the sample is drawn. With the standard
errors we can compute the 95% confidence interval of our estimates.
In Table B.1, we present how the model matches the targeted and non-targeted moments.
We also show how we calibrate alternative model specifications and their ability to match
these moments. We find that the baseline model performs better than alternative models in
matching most of the non-targeted moments.
Table B.1: Targeted and Non-targeted Moments, Data and Model
Moment names
Midstream employment
Main downstream employment
Main upstream employment
Midstream wage bill
Main downstream wage bill
Main upstream wage bill
Midstream firm exports
Midstream firm imports
1
2
3
4
5
6
7
8
9
10
11
Midstream tariffs
Upstream tariffs
Downstream tariffs
Targeted Moments
Estimated Parameter Values
(1)
Data
(2)
Baseline Model
0.0184
-0.0383
0.0032
0.0186
-0.0857
-0.0003
0.0133
0.0286
0.0184
-0.0383
0.0029
0.0218
-0.0769
0.0037
-0.0061
0.0167
0.009
-0.0158
-0.009
0.0117
-0.0256
-0.0093
1,2
ρ = 0.6694
θ = 4.4082
(4)
(5)
(3)
Sector-specific
Same Labor
Same Input and
Input Elasticity
Supply Elasticity
Final Elasticity
Elasticity with respect to midstream tariffs
0.0184
0.0184
0.0184
-0.0045
-0.0383
-0.0383
-0.0019
0.0029
-0.0004
0.0272
0.0218
0.0349
-0.0087
-0.0769
-0.0727
-0.0057
0.0037
-0.0008
-0.0103
-0.0061
-0.0099
-0.0585
0.0165
-0.0229
Employment elasticity with respect to average tariffs
0.0106
0.0117
0.0125
-0.0039
-0.0256
-0.0221
-0.0076
-0.0093
-0.0159
1
1,2,3
1,2
ρ=θ=
ρ = 0.6757 exp(−2.5e − 3 · U )
ρ = 1.1097
2.0127
θ = 4.4020
θ = 3.3340
(6)
Same Trade
Elasticity
(7)
No
Input-output
(8)
Cobb-Douglas Input
and Final Demand
0.0184
-0.0383
0.0074
0.0311
-0.0747
0.0147
-0.0123
0.001
0.0184
-0.0019
0.0006
0.0259
-0.0051
0.0013
0.0009
NA
0.0898
-0.0084
0.0167
0.1632
-0.0104
0.032
-0.0186
0.0085
0.012
-0.0268
0.0016
1,2
ρ = 0.4674
θ = 3.8915
0.0109
-0.0042
-0.0069
1
0.0513
-0.0123
0.0059
NA
θ = 2.1734
NA
Description: This table presents the targeted and non-targeted moments in the data and in the model. Moments 1-8 refer to the elasticity of midstream, main downstream and main upstream employment, wage bill, exports and imports
with respect to midstream tariffs. Moments 9-11 refer to the joint impact of midstream, average downstream and average upstream tariffs (see Section A.5.2). The data moments (Column 1) refer to the corresponding estimated coefficients
that are presented in the empirical section. The model moments (Column 2-8) refer to those estimated with model simulated data and Equations B.13 and B.14. Row “Targeted Moments” show the moments that the models target to estimate
the parameters, whose values are reported in Row “Estimated Parameter Values”. Specifically, Column 4 assumes sector-specific elasticity of substitution across inputs is log linear in sector upstreamness. The employment elasticity with
respect to average tariffs refers to the joint impact of own sector, average upstream, and average downstream tariffs that we document in Table A.14.
69
See Blume et al. (2008) a survey of indirect inference and bootstrap methods used in macroeconomics.
87
B.4
Changes in aggregate variables
The change in aggregate employment equals a weighted average of changes in sectoral employment. The weights are sector employment shares in total employment:
e
L̂ =
S
X
Ls
PS
s
s=1 L
s=1
L̂s .
AD tariffs that protect a sector draw additional labor from the pool of nonworking population
and from other sectors. As a result, the protected sectors observe an increase in employment.
With many sectors that buy from and sell to to each other, we need to solve the counterfactual
equilibrium to sign and quantify the aggregate effect.
The change in real GDP can also be written as a weighted average of the changes in
sectoral employment. However, different from the aggregate employment effect, the weights
are sector value-added shares in nominal GDP:
\ =
rGDP
S
X
s=1
ws Ls
L̂s .
PS
s Ls
w
s=1
(B.15)
Proof: A country’s nominal GDP equals the product of real GDP and a price index of
the real GDP. Alternatively, it can be written as the difference between the country’s gross
output and total intermediate input used.
GDP = P rGDP rGDP =
S
X
s=1
P0s Y s −
S
X
!
s0
P M
ss0
,
s0 =1
where P rGDP is the price index for real GDP. Consider the first-order approximation of
changes in real GDP while holding fixed the prices P rGDP , P0s and P s :
0
0
S
S
X
X
P0s Y s
P s M ss
0
s
d log(rGDP ) =
d log(Y ) −
d log(M ss ).
GDP
GDP
s=1
s0 =1
88
Note that the first-order approximation of the production function equals:
s
s
d log(Y ) = d log(A ) +
ssL d log(Ls )
+
S
X
0
0
ss
).
sss
M d log(M
s0 =1
As tariffs are the only exogenous shock to the model, we set d log(As ) = 0. Further note
0
0
0
s s
s s
s s s
that P s M ss = sss
M P0 Y and w L = sL P0 Y . These imply:
d log(rGDP ) =
S
X
w s Ls
d log(Ls ).
GDP
s=1
Taking the equation to discrete time leads to Equation B.15.
We measure real income (real GNI) with the ratio of nominal income (the sum of labor
income, foreign transfer and tariff revenue) to the consumer price index. Writing it in terms
of changes:
PS
\I =
rGN
s=1
s + TD + TR
ws L\
.
Pb
Extending Caliendo and Parro (2015) by considering varying aggregate labor supply, the
first order approximation of changes in real GNI equals the following:
Employment effect
Terms of trade effect
z
}|
{ z
}|
{
S
S
S
X
XXX
X ws Ls
bs + 1 (
\I =
EFs 0 ŶFs0 ) .
Tils tsil T̂ils −
rGN
L
GN I
GN I s=1 i∈Ξ l∈Ωs
s=1
s=1
F
(B.16)
i
On the right hand side, the employment effect summarizes changes in real income associated with sector employment changes. This term is identical to changes in real GDP. The
only difference is that the denominator for changes in real GNI is nominal GNI, whereas the
denominator for changes in real GDP is nominal GDP.
While sectoral employment changes are sufficient to summarize changes in real GDP (and
aggregate employment), the terms of trade effect indicates that tariffs contribute to real GNI
through not only the employment effect but also changes in foreign and domestic prices. In
this term Tils = psil yils denotes product-country level import value before tariffs, tsil = 1 + τils
89
where τils denotes tariffs, EFs 0 denotes the value of sectoral exports, and YFs0 denotes its
quantity. Lower import prices are associated with more import and higher export prices are
associated with less export. Both cases imply improvements in the terms of trade and an
increase in real GNI.
Proof: First order approximation of nominal GNI equals the following::
d log(rGN I) =
S
X
ws Ls
s=1
GN I
(d log(ws ) + d log(Ls )) +
dT D
dT R
+
− d log(P C ).
GN I GN I
(B.17)
The change in trade deficit equals:
dT D =
S X X
X
Tils d log(Tils )
−
s=1 i∈ΞF l∈Ωsi
S
X
EFs 0 (1 − σ s )d log(P0s ).
s=1
Changes in tariff revenue equal:
dT R =
S X X
X
τils Tils d log(Tils )
+
S X X
X
Tils tsil d log(tsil ).
s=1 i∈ΞF l∈Ωsi
s=1 i∈ΞF l∈Ωsi
Changes in consumer price equal:
C
d log(P ) =
S
X
αs d log(P s ),
s=1
in which changes in sectoral input prices equal:
d log(P s ) = ss0 d log(P0s ) +
XX
ssi ssli d log(tsil ).
i∈ΞF l∈Ωsi
Now we substitute final expenditure share in Equation B.18. Note that:
s
s
s
X =P C +
S
X
s0 =1
90
0
0
0
ssMs P0s Y s ,
(B.18)
where P0s Y0s denotes sector s output. Therefore,
αs =
S
X
1
P sC s
0
0
0
=
(X s −
ssMs P0s Y s ).
GN I
GN I
s0 =1
Plug this into Equation B.18:
d log(P C ) =
S
X
s=1
S
S
XX
X s
1 X X s0 s s0 s0
s0 d log(P0s ) +
ssi ssli d log(tsil ) −
sM P0 Y d log(P s ).
GN I
GN
I
s=1 s0 =1
i∈Ξ l∈Ωs
s
F
i
We can simplify the last term:
S X
S
X
s=1 s0 =1
S
X
0
0
0
ssMs P0s Y s d log(P s )
0
0
0
0
0
P0s Y s (d log(P0s ) − ssL d log(ws ))
s0 =1
0
0
0
P0s Y s d log(P0s ) −
=
=
S
X
s0 =1
S
X
0
0
0
ws Ls d log(ws ).
s0 =1
Plug these into Equation B.17:
d log(rGN I) =
S
X
ws Ls
s=1
+
GN I
s
d log(L ) +
S
X
w s Ls
s=1
S X X
X
Tils d log(Tils ) −
s=1 i∈ΞF l∈Ωsi
S
X
d log(ws )
EFs 0 (1 − σ s )d log(P0s )
s=1
S
1 XXX s s s
X si sli d log(tsil )
+
GN
I
s=1 i∈ΞF l∈Ωsi
s=1 i∈ΞF l∈Ωsi
S
s
X
X
X
X s
s0 d log(P0s ) +
ssi ssli d log(tsil )
−
GN
I
s=1
i∈Ξ l∈Ωs
1
+
GN I
S
X
GN I
XX
τils Tils d log(Tils )
F
1
+
GN I
S
X
P0s Y s d log(P0s )
s=1
i
S
1 X s s
−
w L d log(ws ).
GN I s=1
Collecting terms we get:
d log(rGN I) =
S
X
ws Ls
s=1
S
S
1 X s s
1 XXX s s
s
d log(L ) +
σ EF 0 d log(P0 ) +
t T d log(Tils ).
GN I
GN I s=1
GN I s=1 i∈Ξ l∈Ωs il il
s
F
91
i
Note that d log(YFs0 ) = −σ s d log(P0s ). Plug in and take the equation to discrete time we get
Equation B.16.
We can compute changes in household welfare based on Equation B.3. This is a known
variable after we solve for ŵs and P̂ C with the equilibrium conditions in changes:
Ŵ =
λs
µ ! µ1
S
X
1
Ls ŵs
L0
.
+
C
C
L
L
P̂
P̂
s=1
Now consider the consumption equivalent of these welfare changes. We let leisure/labor
decision should remain the same as the baseline equilibrium, and we compute the percentage
change in consumption that the resulting welfare change equals Ŵ . Use C s and C̃ s to denote
consumption of a sector s household before and after the change. The household budget
constraint and labor supply decision, Equations 9 and B.1, imply that:
ls =
Cs
1−δ
ψs
1+ψ
s
.
With this we can rewrite the baseline welfare in consumption terms:
W =
S
X
(1 −
s=1
1
ψs
)C s
s
1−δ1+ψ
µ ! µ1
.
The welfare when consumption becomes C̃ s but leisure remains the same as before, equals:
W̃ =
S
X
C̃ s −
s=1
1
ψs
Cs
1 − δ 1 + ψs
µ ! µ1
.
Taking the ratio and set it to Ŵ :
W̃
Ŵ =
=
W
S
X
s=1
ψs
ssL
1−δ
1+ψ s
˜
Ĉ s −
s
ψs
1 − δ − ψs +1
1 − δ − ψψs +1
!µ ! µ1
.
˜
In the end we solve Ĉ s , which is the consumption-equivalent welfare changes.
92
B.5
Sector upstreamness
We follow the procedure in Fally (2011), Antràs et al. (2012), and Antràs and Chor (2013)
to compute the sector upstreamness. Upstreamness measures the average number of sectors
that one dollar of a sector’s output passes through to reach final demand. If a sector’s output
is only used for final demand, the sector’s upstreamness equals 1. If a sector sells to other
sectors, its upstreamness will be greater than 1. The greater is the upstreamness measure,
the greater share of output the sector sells to other sectors and the more upstream is the
sector.
To compute the fraction of a sector’s output used in other sectors, we rely on the input0
output coefficients sss
M . Following the approach in the literature, we adjust the coefficients
0
0
ss
to take into account imports and exports with s̃ss
M = sM
P0s Y0s
s
s
s +X s (1−ss )
P0 Y0 −EF
0
0
, where P0s Y0s
denotes gross output, EFs 0 denotes total export in sector s and X s (1 − ss0 ) denotes sector s
total import. The denominator is thus total domestic absorption of sector s output. Finally,
the sector upstreamness equals:
~ = (I − Γ̃0 )−1 Y~ ./Y~ ,
U
0
where ./ denotes element-wise division, the s − s0 element of Γ̃ is s̃ss
M , and I is an identity
matrix.
In Figure B.1 we plot the correlations between the estimated elasticity of substitution
across products, trade elasticity, and labor supply elasticity, against sector upstreamness.
B.6
Optimal tariffs problem
A country’s policy maker maximizes changes in the following aggregate variables (defined in
Section B.4):
1. Total employment: L̂e , or
\ , or
2. GDP: rGDP
\I, or
3. Real income: rGN
93
Figure B.1: Correlation between product, trade and labor supply elasticities and
sector upstreamness
(b)
Product Elasticity
(c)
Trade Elasticity
7
Computer, Electrical and Machinery Equipment
s
Computer, Electrical and Machinery Equipment
Wood and Paper
Automobiles and Transportation Equipment
Petrochemicals
Minerals and Metals
Wood and Paper
Automobiles and Transportation Equipment
3
Trade elasticity σ
4
5
s
Product elasticity ζ
4
2
Labor Supply Elasticity
Computer, Electrical and Machinery Equipment
Minerals and Metals
s
6
Labor supply elasticity λ
.5
1
1.5
6
8
Agriculture, Mining, Food and Textile
2
(a)
Minerals and Metals
Agriculture, Mining, Food and Textile
Automobiles and Transportation Equipment
Petrochemicals
Wood and Paper
Service
1.5
2
2.5
Sector upstreamness
3
0
2
0
Agriculture, Mining, Food and
Textile
Petrochemicals
1.5
2
2.5
Sector upstreamness
3
1.5
2
2.5
Sector upstreamness
3
Description: This figure shows the correlation between the estimated elasticity of substitution across products, trade elasticity,
and labor supply elasticity, with sector upstreamness. To measure sector upstreamness on the broad sector level–the same level
on which the elasticities are estimated, we first compute the upstreamness measure on CNAE2.0 4-digit sector level with the
input-output table and sectoral imports and exports (see Section B.5 for details. Then we calculate the weighted average
upstreamness for each sector for which the weight equals a CNAE2.0 4-digit sector’s share in the broad sector.
4. Welfare: Ŵ
subject to the following equilibrium constraints: changes in prices summarized in Equations
B.5, B.6, B.7, market clearing conditions B.8 and B.9, as well as government budget constraint B.10. Furthermore, the government satisfy the additional fiscal constraint that the
government collects the same tariff revenue as from the benchmark tariffs:
T R0 = T R0,benchmark ,
where T R0 follows Equation B.11 and T R0,benchmark equals the value of T R0 under benchmark
tariffs.
B.7
B.7.1
Quantitative Results
Impact of Brazilian Annual AD tariffs
We calculate the AD tariffs imposed on each sector in each year by combining productcountry level AD tariffs and with Equations B.6 and B.7. We simulate the model with these
yearly tariffs and compute changes in the following aggregate variables: employment, real
GDP, real income (GNI), and welfare. We present the formulas for these variables in Section
B.4.
94
Figure B.2 shows that in all years and in all years except 2004-2006, AD tariffs cause
moderate aggregate employment gains and GDP gains. This indicates that the positive
midstream employment effect outweighs the decline in downstream employment. However,
AD tariffs cause larger annual real income and welfare losses. This indicates that the increase
in consumer price due to more expensive imports dominates the rise in nominal income. Table
B.2b shows that in an average year, Brazil gains from all AD tariffs 0.03% employment, 0.02%
GDP, but loses 0.49% real income, and 0.92% welfare.
Figure B.2: Aggregate Consequences of AD Tariffs
-.005
-.01
-.015
Changes
0
(a) Each Year
2000
2005
Employment
Year
GDP
2010
Real income
2015
Welfare
(b) Annual Average
Aggregate statistics
Annual average
95%
Confidence interval
Employment GDP Real income
0.03%
0.02%
-0.49%
0.02%
0.01%
-0.58%
0.03%
0.04%
-0.40%
Welfare
-0.92%
-1.06%
-0.78%
Description: Figure B.2a shows the impact of AD tariffs imposed in each year on aggregate employment, GDP, real income
and welfare. Table B.2b shows the annual average of these aggregate consequences and the 95% confidence intervals of the
means.
95
Alternative Model Specifications In Table B.2 we show that alternative models (except
the one with sector-specific input elasticity) substantially misunderstand the aggregate effects
of Brazilian AD policy.
Table B.2: Aggregate Consequences of Brazilian AD Policy in Different Model
Specifications
Aggregate
(1)
Consequence Baseline Model
Employment
GDP
Real income
Welfare
0.06%
0.05%
-1.32%
-2.43%
(2)
Same Input and
Final Elasticity
0.15% (126.97%)
0.11% (149.07%)
-1.35% (-2.48%)
-2.40% (1.20%)
(6)
(3)
(4)
(5)
Sector-specific
Same Labor
Same Trade
No
Input Elasticity Supply Elasticity
Elasticity
Input-output
0.06% (-2.35%)
-0.08% (-230.40%)
0.08% (22.02%) 0.15% (124.86%)
0.04% (-2.88%)
-0.13% (-377.84%) 0.12% (156.29%) 0.14% (199.65%)
-1.33% (-0.57%)
-1.54% (-16.45%)
-1.35% (-2.10%) -0.75% (42.95%)
-2.44% (-0.58%)
-2.65% (-8.90%)
-2.68% (-10.13%) -1.53% (37.10%)
(7)
Cobb-Douglas Input
and Final Demand
0.02% (-63.99%)
0.08% (71.23%)
-1.36% (-3.08%)
-2.36% (2.83%)
Description: This table shows the impact of Brazilian AD policy in different model specifications. The value outside the bracket refers to the level of the effect, and the value inside the bracket refers to the percentage difference
of the impact predicted by the alternative model relative to the absolute value of the impact predicted by the baseline model. The Brazilian AD policy refers to, for each sector, the maximum AD tariff of all years.
B.7.2
Impact of sectoral Tariffs
In Figure B.3a we plot the aggregate consequences of 200% sectoral tariffs imposed on every
CNAE 2.0 4-digit sectors. We plot them against how upstream the sectors are. While the
average impact of sectoral is small,70 imposing tariffs on downstream sectors, for example,
automobiles and transportation equipment, as well as computer, electrical and machinery
equipment, can significantly raise aggregate employment and GDP. On the other hand, tariffs
on upstream sectors, for example, petroleum and chemicals, significantly reduce aggregate
employment and GDP. Table B.3b shows that the associations between aggregate employment and GDP effects of sectoral tariffs with sector upstreamness are negative (-0.3513 and
-0.3193) and significant at 1% confidence interval. The negative correlations are robust to
sector characteristic controls. In Table B.3, Column 1 we show the simple regression of the
aggregate employment effects of sectoral tariffs on sector upstreamness. Column 2 and 3
control 2-digit sector fixed effects and broad sector fixed effects, respectively. Column 4 to
6 show that protecting the sectors that are smaller, import a larger share from abroad and
have larger elasticity of substitution between domestic and foreign output, can also lead to
larger aggregate employment gains. Across all specifications the negative correlation between
aggregate employment effect and sector upstreamness is negative and significant.
70
There are 297 CNAE 2.0 4-digit, non-service sectors. Therefore, the average share of each of these sectors
in the economy is small.
96
In contrast, the impact of sectoral tariffs on real income and welfare is negative for almost
all sectors. The associations between real income and welfare consequences of sectoral tariffs
with sector upstreamness are weakly positive. Taxing downstream sectors substitutes more
imports with domestic labor, increases domestic prices, and harms domestic welfare. On the
other hand, taxing upstream sectors decrease employment in more downstream sectors by
cutting their wages and lead to lower nominal income. Both forces contribute to lower real
income and welfare.
Figure B.3: Aggregate Consequences of 200% Sectoral Tariffs
.001
0
1
2
3
-.001
-.001
0
GDP
Employment
.001
(a) Aggregate Consequences
4
1
3
4
-.004-.002 0 .002.004
Welfare
Real income
-.004-.002 0
2
Sector upstreamness
.002 .004
Sector upstreamness
1
2
3
4
1
Sector upstreamness
2
3
4
Sector upstreamness
Agriculture, Mining, Food and Textile
Petroleum and Chemicals
Wood and Paper
Minerals and Metals
Computer, Electrical and Machinery Equip.
Automobiles and Transportation Equip.
(b) Correlation with Sector Upstreamness
Aggregate statistics
Correlation
Employment
GDP
-0.3513***
-0.3193***
Real income
0.0499
Welfare
0.0350
Description: This figure shows the aggregate consequences of 200% sectoral tariffs imposed on every CNAE 2-digit sector.
Panel (a) plots the employment, GDP, real income and welfare effects on the vertical axis, and sector upstreamness on the
horizontal axis. Each dot in the figure represents the average value in each 0.05 bin of sector upstreamness. Panel (b) shows the
correlation between the aggregate consequences of sectoral tariffs and the upstreamness of the sector. *, **, and *** represent
significance on the 0.1, 0.05, and 0.01 level.
97
To understand the sources of low correlation between the impact of sectoral tariffs on real
income with sector upstreamness, in Figure B.4 we show that both the impact of these tariffs
on nominal income and on consumer price are negatively correlated with sector upstreamness.
Protecting downstream sectors leads to greater increase in nominal income like the increase
in total employment and GDP. However, it also increases the consumer price more. The two
forces offset each other for real income, as it equals the ratio of nominal income to consumer
price. Figure B.5 shows that the relationship also holds when we take the average of CNAE
2.0 4-digit sectors for each broad sector.
.001
0
-.001
0
.002
.001
Price
Nominal income
.003
.002
.004
.003
.005
Figure B.4: Consequences of 200% Sectoral Tariffs on Nominal Income and Consumer Price
1
2
3
4
1
Sector upstreamness
2
3
4
Sector upstreamness
Agriculture, Mining, Food and Textile
Petroleum and Chemicals
Wood and Paper
Minerals and Metals
Computer, Electrical and Machinery Equip.
Automobiles and Transportation Equip.
Description: This figure shows the impact of 200% sectoral tariffs imposed on every CNAE 2.0 4-digit sector on nominal
income and consumer price. Changes in nominal income and consumer price due to the tariff changes are plotted on the vertical
axis, and sector upstreamness is plotted on the horizontal axis.
In Table B.2 we present the aggregate implications of Brazilian AD policy predicted by
98
Figure B.5: Consequences of 200% Sectoral tariffs on nominal income and consumer price, broad sector average
Consumer Price
.002
.0015
(b)
Nominal Income
Computer, Electrical and Machinery Equipment
Computer, Electrical and Machinery Equipment
.001
Price Change (%)
.001
.0005
Petrochemicals
.0005
Automobiles and Transportation Equipment
.0015
Automobiles and Transportation Equipment
Petrochemicals
Minerals and Metals
Agriculture, Mining, Food and Textile
Wood and Paper
0
Minerals and Metals
Agriculture, Mining, Food and Textile
Wood and Paper
0
Nominal Income Change (%)
(a)
1.5
2
2.5
Upstreamness
3
3.5
1.5
2
2.5
Upstreamness
3
3.5
Description: This figure shows the impact of 200% sectoral tariffs imposed on every CNAE 2.0 4-digit sector on nominal
income and consumer price. Changes in nominal income and consumer price due to the tariff changes are plotted on the vertical
axis, and sector upstreamness is plotted on the horizontal axis. Averages are taken on the broad sector level.
Table B.3: Correlation between Aggregate Employment Consequence of sectoral
Tariffs and Sector Characteristics
VARIABLES
Upstreamness
(1)
Employment
(2)
Employment
(3)
Employment
(4)
Employment
(5)
Employment
(6)
Employment
-6.55e-05***
(1.01e-05)
-5.61e-05***
(1.97e-05)
-2.75e-05**
(1.18e-05)
-6.59e-05***
(1.02e-05)
-0.000133
(0.000207)
-4.09e-05***
(1.04e-05)
4.38e-05
(3.85e-05)
3.97e-05***
(7.27e-06)
-4.11e-05***
(1.04e-05)
-3.84e-05
(0.000196)
4.35e-05
(3.86e-05)
3.96e-05***
(7.29e-06)
298
0.225
NA
298
0.225
NA
Employment share
Import share
Trade elasticity
Observations
298
R-squared
0.123
Fixed effect
NA
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
295
0.356
2-digit
297
0.260
Broad sector
298
0.125
NA
Description: This table shows the correlation between the aggregate employment consequence of sectoral tariffs and sector characteristics including
sector upstreamness, employment share in the economy, share of import, and trade elasticity.
alternative model specifications. We find that these alternative models (except the one with
heterogeneous elasticity of substitution across inputs) lead to incorrect conclusions about
the effects of Brazilian AD policy.
99
B.7.3
Optimal AD Tariff Policy
Figure B.6a shows CNAE 2.0 4-digit sectoral optimal tariffs that maximize employment and
GDP. They should be high for many downstream sectors in automobiles, transportation
equipment, as well as agriculture, mining, food and textile. Sometimes they even exceed
900%. Those on upstream sectors should be lower. For example, the employment-maximizing
tariffs on petroleum and chemical sectors should be negative, which means that to increase
employment Brazil should decrease their MFN tariffs for these sectors. In contrast, optimal
tariffs that maximize real income or welfare never exceed 100% and they should be set
negative for many sectors.
Table B.6b shows that employment- and GDP-maximizing tariffs strongly negatively correlate with sector upstreamness, whereas real income- and welfare-maximizing tariffs positively correlate with it. These findings are consistent with Sections B.7.2 and 7.2 which find
that compared to upstream sectors, imposing higher tariffs on downstream sectors increases
employment and GDP but decreases real income and welfare.
In Table B.4 we present the correlations of these optimal tariffs with one another, with
the benchmark tariffs and with sector upstreamness. Employment-maximizing tariffs are
strongly positively correlated with GDP-maximizing tariffs and negatively correlated with
real-income–maximizing tariffs. They weakly positively correlated with welfare-maximizing
tariffs.
Table B.6b also shows that the benchmark, factual Brazilian tariffs are negatively associated with sector upstreamness, which suggests that employment may be a strong motivation
that drives AD tariffs. However, the levels of all actual tariffs stay below 500% (see bottom
right panel of Figure B.6a). This suggests that either the Brazilian government is prevented
by WTO rules, bilateral/multilateral trade agreements, and domestic political institutions
from increasing tariffs further, or they are concerned that raising tariffs may impose additional harm on welfare.
100
Figure B.6: Optimal Tariffs
9
Optimal tariff for GDP x100%
-1
-1
1
1
3
3
5
5
7
7
9
Optimal tariff for employment x100%
(a) Sectoral Optimal Tariffs that Maximize Employment, GDP, Real Income
and Welfare
1
2
3
4
1
2
3
4
4
Baseline tariffs x100%
3
1
2
.5
1
0
1
2
3
4
0
-.5
-.5
0
.5
1
Optimal tariff for welfare x100%
1.5
Sector upstreamness
1.5
Optimal tariff for real income x100%
Sector upstreamness
1
Sector upstreamness
2
3
4
1
Sector upstreamness
2
3
4
Sector upstreamness
Agriculture, Mining, Food and Textile
Petroleum and Chemicals
Wood and Paper
Minerals and Metals
Computer, Electrical and Machinery Equip.
Automobiles and Transportation Equip.
(b) Correlations of Sectoral Optimal Tariffs with Sector Upstreamness
Optimal tariffs that maximize
Correlation
Employment
GDP
Real income Welfare Benchmark tariffs
Sector upstreamness -0.4979*** -0.4486***
0.4520***
0.1978***
-0.3381***
Description: This figure shows the sectoral optimal tariffs that maximize employment, GDP, real income and welfare. The
optimal tariffs solve a problem that maximize the respective aggregate variable, subject to the equilibrium constraints and the
additional constraint that the government collects the same tariff revenue as from the benchmark tariffs (see Section 7.3). The
benchmark tariffs refer to, for each sector, the sector’s maximum AD tariff of all years. Panel (a) plots these optimal tariffs
against sector upstreamness, and Panel (b) presents the correlations.
Table B.4: Correlations of Sectoral Optimal Tariffs
Optimal tariffs that maximize
(1)
Employment
(2)
GDP
(3)
Real income
(4)
Welfare
(5)
Benchmark tariffs
(6)
Sector upstreamness
Optimal tariffs that maximize
Employment
GDP
Real income Welfare
1.0000
0.6754***
1.0000
-0.3807***
-0.1054*
1.0000
0.0484
-0.2998***
-0.0936
1.0000
0.2575**
0.1250
-0.1912
0.2005
-0.4979*** -0.4486***
0.4520***
0.1978***
Benchmark tariffs
Sector upstreamness
1.0000
-0.3381***
1.0000
Description: This table shows the correlation of optimal tariffs that maximize employment, GDP, real income, welfare, as well as benchmark tariffs and sector upstreamness. The optimal
tariffs solve a problem that maximize the respective aggregate variable, subject to the equilibrium constraints and the additional constraint that the government collects the same tariff
revenue as from the benchmark tariffs (see Section 7.3) The benchmark tariffs refer to, for each sector, the the sector’s maximum AD tariff of all years.
101
@FedFRASER