Impacts of antidumping policies on Colombian domestic
protected industries
Arielle Knudsen
Department of Economics
Indiana University, Bloomington
November 16, 2016
Abstract
While the impacts of antidumping policies on trade have been studied on a wide
scale, relatively little has been done on the implications of antidumping on labor markets, even though job protection is often used as a motive when requesting protection.
This paper fills the gap by focusing on impacts to domestic firm employment and wages.
To identify the effect of antidumping measures, I use synthetic control groups constructed as a weighted average of unaffected industries to reflect similar pre-treatment
trends of industries receiving protection. I find that only two of the eight industries
receiving antidumping protection experience around 5-15% higher employment than
matched counterfactual control groups. The remaining six industries saw decreases
in employment relative to their control groups. In terms of average annual wages for
workers, three-quarters of the protected industries experienced wages around 5-15%
higher than their control groups.
1
1
Introduction
The impacts of antidumping policies have been studied on a wide scale that ranges from the
effects of the usage frequency on aggregate trade [Prusa, 2001] to the effect on the export
decisions of targeted firms [Lu et al., 2013]. One common thread between almost all of the
antidumping literature is the focus on import and export data. However, little has been
done to comment on the implications of antidumping or other temporary trade barriers on
domestic firms and industries, which are important because of the role firms play in initiating
the process. Antidumping petitions and measures are justified by firms as being necessary
for them to be able to compete with low price imports, but how much do they help?
In October 2016, Colombia’s steel associations joined with other regional associations
to press governments at the Ibero-American Summit to not grant China market economy
status due to competitive advantages enjoyed by Chinese steel producers. Market economy
status, among other things, would make it more difficult to enact temporary trade barriers.
The producers state “We are united by a common concern: our fears for the loss of jobs
and its impact on our communities, the closure of companies, the lack of incentives for
investment, financial losses and the destruction of our value chain.... In short, China exports
unemployment.” The Colombian manufacturing industry shares this sentiment. China, the
most frequent target of Colombian antidumping investigations, exports at such low prices
that even MFN maximum tariffs have little effect. In 2015, footwear from China, priced as
low as $0.50USD faced a maximum 15% tariff. Domestic producers, whose shoes were priced
over $5.50USD, were not competitive. Thus, the industry filed an antidumping petition for
additional protection from injury.
Officially, injury is determined in Colombian antidumping investigations through effects
of imports on several variables, including employment and wages.1 In 2016, the Ministry
1
Committee on Anti-Dumping Practices [1995, Chapter V Article 13.1.2] states “Effects of imports on
trends in the domestic industry with respect to factors such as, inter alia, prices, output, market share,
profits, utilization of installed capacity, inventories, sales, ability to raise capital or investment, employment
and wages.”
2
Table 1: Average Treatment Effect by Industry for Five Years After Petition, measured as the percent change
since 2000
Average Treatment Effect
Industry
ISIC codes
Number of workers
Average annual wage
Tableware
2520, 2691
-5.69
-1.10
Plastic balls
2520, 3693
-2.17
-1.96
Textiles
1711, 1721, 1730,
-48.86*
5.85
9.40
7.87
1810, 2899
Metal hardware
2893, 2899
*Textiles received a large persistant shock the year before it initiated a petition.
This negative impact is carried over from the shock.
of Commerce, Industry and Tourism, the entity responsible for antidumping investigations,
concluded dumping occurred in the aforementioned footwear case; however, dumping did not
cause injury to the domestic industry, so antidumping measures were not put in place.2 If
loss of employment and wages are in fact attributed to less-than-fairly priced imports, then
we would expect that antidumping remedies will increase or stabilize employment and wages
for the affected industries.
Table 1 summarizes the results of this paper, which show that antidumping protection
is not very effective at stabilizing employment, but can have an effect on average annual
wages. Only two of eight ISIC 4-digit industries saw definitive increases, around 5-15%
over employment levels of a counterfactual control group, after antidumping sanctions were
implemented. A third industry saw modest initial 5% increase over the control, which eventually became a large decrease in employment relative to the counterfactual. The remaining
industries experienced continued drops in employment relative to the control after sanctions
were in place. Six of the eight industries experienced average wage increases of 5-15% over
the counterfactual control. On the whole, antidumping policies have been more successful
in some industries than others, and the underlying reasons should be explored in the future.
2
The various tests for causality included the response of prices when Chinese import volumes fluctuated,
and what happened when domestically produced shoe prices fluctuated. As imports from China fell, the
price of certain types of shoes remained stable. Additionally, some domestic producers were able to raise
their own prices and increase their market share [El Director de Comercio Exterior, 2016].
3
Before considering changes to employment and wages, I first establish stylized facts about
firms that receive protection. At the productivity level, this paper demonstrates that firms directly involved with initiating antidumping cases tend to have relatively average productivity
in their industries before antidumping cases, as estimated by their total factor productivity
(TFP). These firms are usually rather large compared to other firms in their industries, with
each firm holding around 2.5% of assets in their industry, compared to the average firm
that holds less than 1% of industry assets. After successful antidumping cases, the relative
productivity of initiating firms falls below the median within 5 years of filing a petition. The
10th, 50th, and 90th percentiles of firm productivity in the protected industries fall, but not
by as much as the initiating firms. Previous literature suggests the lower productivity firms
should become more productive while the more productive firms are unaffected.
Existing literature currently addresses effects on domestic industries from two main angles: the impact on firm productivity; and the impact on industry employment. I describe
these investigations in the following section before proceeding to my analysis. After establishing differing trends on the domestic impact, the next logical step is to investigate
why differences exist between industries. Is there a signal that indicates when an industry
will benefit from antidumping protection? Are there trends in investment that should be
considered, or the types of employees hired while firms have a protection buffer in place?
Determining when duties will successfully protect an industry may yield insight on when
these distortions should be used, and this is where I will direct future research.
2
Literature Review
There are few existing studies on the domestic effects of antidumping protection. Of the
existing literature, I focus on three papers, Konings and Vandenbussche [2008], Pierce [2011],
and Chung et al. [2016], that address the effects on firm productivity and employment.
4
Konings and Vandenbussche [2008] answers the former question posed for European firms:
how does antidumping protection affect productivity, and how does the effect depend upon
firm heterogeneity? Using a difference in difference model, the average firm receiving protection sees a moderate improvement in productivity, measured by total factor productivity
as estimated through the Olley-Pakes [Olley and Pakes, 1996] methodology. However, the
productivity improvement is not enough to bring the firm to the same productivity as the
average firm in a similar unprotected industry. Furthermore, the “frontier firms” experience
productivity losses due to protection. These results are built on by Pierce [2011]. When productivity is based on revenue (value-added, as it is in Konings and Vandenbussche [2008]),
a difference in difference analysis of US data shows moderately increased productivity, similar to Konings and Vandenbussche [2008] at about 3 − 8% improvement, from protection.
If, instead, productivity is quantity-based, we see decreases in productivity from protection, suggesting that protected firms are increasing prices and markups to raise revenue, but
reducing output.
More surprising is the one paper addressing the impact of temporary trade barriers on
labor. As with the steel industry, advocates for protectionist policies frame it as protecting
domestic workers. Unfortunately, this seems to be something that has been neglected in the
research. The closest paper to this is Chung et al. [2016]: using a China-specific safeguard
case initiated by the US in the tire industry, the authors generated a synthetic control group
and used a difference in difference model to find that total employment and average wages
in the industry were unaffected by the safeguard. Intuitively, the higher tariff on Chinese
tires would reduce Chinese exports to the US in the tire industry and divert them elsewhere,
opening the door for other countries to export more tires to the US. So, the gap opened up
by fewer Chinese tire imports was filled by third-party countries, essentially leaving the US
domestic tire industry unchanged relative to a synthetic similar industry.
This paper will be similar to Chung et al. [2016] in that it focuses on the impact of
antidumping measures on Colombian labor. Following Chung et al. [2016], it would be
5
ideal to use global safeguard cases to determine if domestic labor is affected by a blanketed
trade barrier instead of a discriminatory one, blocking the possibility of trade diversion.
Unfortunately, Colombia has not been a major user of global safeguards- from 1999 through
2014, Colombia initiated 8 global safeguard cases, and only one of these resulted in final duties
in 2014, which is too recent to be able to adequately measure the impact on employment.
By using a relatively small economy, compared to the US and EU in previous studies, it
is possible that some diversion effects may still be mitigated, even without using a blanket
trade barrier. Since Colombia is not a major trade partner with larger economies like the
US, EU, and China, even though the US and China are major partners for Colombia, it
is possible that reducing trade flows from one country may not necessarily result in large
increases of goods from other countries.
Further, a benefit of using Colombia is the available data, collected and published online
by the Colombian government. Data is at the firm and industry level data3 . The level of
detail in surveys completed by firms allows for in-depth analysis of many different factors
that change throughout the years, providing a glimpse into what protected industries change
when they receive trade protection. Finally, a complexity that had to be dealt with in
Chung et al. [2016] that can be ignored using Colombia is the presence of labor unions: in
estimating the impact on employment and wages, Chung et al. [2016] had to consider the
effect of labor unions in the US tire industry. Labor unions negotiate wages, and then firms
make employment decisions based on output, cost of capital, cost of materials, negotiated
wages, and import penetration. In Colombia, labor unions during the period studied were
nearly non-existent.4 Hence, wage negotiations can be ignored.
3
The data collected is all at the firm level, but out of respect of the privacy of current firm operations,
recent data is only released to the public aggregated to the industry level
4
In 2006, less than 5% of Colombia’s labor force was in a union [Bureau of Democracy, Human Rights,
and Labor, 2006]. By 2013, the number rose to 6.5%, as estimated by the Greater Integrated Household
Survey administered by the Colombian government. Violence against union leaders by the government
and paramilitary groups contributes to low union membership. While it has since improved, Colombia
was considered one of the worst countries in the world for unionists not too long ago- more than half of
the worldwide murders of unionists (78 out of 144) in the year 2007 were committed in Colombia [ITU,
September 18, 2007]. Despite recent improvements in workers’ rights partially due to pressure from the
United States before signing a trade agreement and ceasefires between the government and rebel groups, this
6
3
Methodology
This paper uses two established methodologies: for computing TFP, I follow the Olley-Pakes
method [Olley and Pakes, 1996], and for the treatment effects, I use the synthetic control
method as developed by Abadie et al. [2010]. The following first describes the Olley-Pakes
method as adapted by Beveren [2010].
The Olley-Pakes method semi-parametrically estimates firm productivity while taking
into account the simultaneous decision of whether or not to produce and the quantities of
inputs to use. The model assumes productivity is the only unobservable state variable;
productivity is monotonically related to investment; and all firms face the same input and
output prices. Assuming each firm has a Cobb-Douglas production function, output is a
function of inputs and productivity.
yit = β0 + βk kit + βl lit + ωit + ηit ,
(1)
where yit is the log of output, measured by a firm’s (deflated) gross profit, kit is the log of
capital, and lit is log of labor expenditures. ωit is productivity: an ex-ante shock realized
before the firm makes any current-period decisions. Because ωit is known before the optimal
input selection, it is correlated with the input choices. The error, ηit , is an unknown expost shock. A firm’s capital investment for the next period depends on productivity and
current capital stock, iit = it (kit , ωit ). Inverting investment (thanks to the monotonicity
assumption), we have ωit = i−1
t (kit , iit ). Therefore, yit is a function of observables, nonlinear
in capital.
A two-step estimation procedure first finds a consistent estimator for βl , allowing capital
and investment to enter as a third-order polynomial. The second stage uses firm dynamics
paper studies the period of time when workers were generally not unionized. Even in instances where unions
could form, these typically were concentrated in the agriculture or mining industries, not the manufacturing
industry.
7
and the estimator for βl .
yit+1 − βl lit+1 = β0 + βk kit+1 + E[ωit+1 |ωit , χit+1 ] + ξit+1 + ηit+1
(2)
where ξit+1 , new information the firm has in t + 1, is uncorrelated with productivity and
capital in t+1, but is correlated with the variable input labor, which is why labor appears on
the left hand side of the equation. χit+1 is a survival indicator equal to one when the firm’s
productivity is sufficiently high. Otherwise, the firm shuts down. The expectation term is
approximated as a third order polynomial that depends on the probability of survival in t+1,
capital in t, and investment in t. Nonlinear least squares of Equation (2) yields a consistent
estimator of βk . Plugging βl and βk into Equation (1) determines productivity, ωit .
Part of my goal is to analyze the impact on employment and wages in Colombian manufacturing industries affected by successful (defined as resulting in a duty) antidumping petitions. To judge how these industries fare, I compare them to similar industries to identify
a treatment effect of antidumping duties. Identifying treatment effects requires comparison
between a treatment group (industries granted protection) and a control group (industries
not granted protection). The treatment and control industries should exhibit similar trends
before the treatment effect to ensure the effect is not due to some other industry-specific
variable. Typically, in trade, the control groups are composed of either similar products, or
other products that show similar characteristics to those that had antidumping investigations. The problem in the realm of antidumping is that in order for an antidumping petition
to be successful, there should, in fact, be some negative, harmful trend in the treated industry
that is not present in the control industry. Otherwise, one would expect the other industry
to initiate a similar petition. Thus, the assumption of similar pre-treatment trends is violated. One potential solution is to use other industries that have unsuccessfully petitioned
for protection, capturing unobserved effects of certain industries that make them more likely
to petition, and different trends. Unfortunately, Colombia does not have a sufficient number
8
of unsuccessful cases to generate enough of a control group. Another common approach is to
create a control group of other industries that are likely to initiate investigations. However,
there are not enough four digit ISIC industries to form a convincing control group.
To solve these problems, I use the synthetic control procedure from Abadie et al. [2010].
The synthetic control method identifies a weighted average of the potential groups (ISIC
codes, in this case) that could contribute to a control group. I exclude industries that
either have petitioned for antidumping protection in the past and resulted in no duties, or
will petition before 2015. The weights in the average are determined by minimizing the
distance between the preintervention characteristics of the treated group and those of the
untreated group. Once an appropriate distance-minimizing vector has been determined, the
treatment effect is identified as the difference between the outcome in the treated group
and the synthetic control group generated by the weights. This treatment effect can vary
between the years after the treatment has occurred, which is beneficial here because changing
antidumping duties and the temporary nature of the protection mean that the effects would
be expected to change over time.
The regressions used to fit the control group and determine the treatment effects are, as
in Chung et al. [2016]:
∆nit = β0 + βy ∆yit + βa ∆ait + βm ∆mit + βip ∆ipit
(3)
∆wit = β0 + βy ∆yit + βa ∆ait + βm ∆mit + βip ∆ipit
(4)
where all variables are measured as the log change since 2000, i is an ISIC four digit industry,
and t is the year. Equations (3) and (4) state that the changes in employement and wages
since 2000 depend on output, input, and import competition: the change in gross production
in the industry, yit , the change in the industry’s total assets (investment), ait , the change in
raw materials, mit , and the change in import penetration, ipit .
To determine the robustness of the synthetic control method, a placebo test is conducted.
9
A placebo test takes each manufacturing industry in Colombia and finds the “treatment
effect” as though each other industry was the subject of an antidumping duty. I am looking
for two outcomes in the placebo test: whether the root mean squared prediction error is
generally smaller than in the actual treated industries, and whether the magnitude of the
treatment effect on the industry that received protection is generally greater than that of
the placebo treatment effects. The former check is to ensure the pre-treatment fit is a good
match. The latter checks the significance of the treatment effects.
The synthetic control method has been applied to only a handful of international trade
papers before. In addition to Chung et al. [2016], Billmeier and Nannicini [2013] use the
method to assess the effect of economic liberalization on GDP trajectories, finding a generally
positive impact, although in more recent liberalizations in Africa, the impact was negligent.
Hosny [2012] applies the synthetic control method to estimate the impact of Algeria waiting
7 years before joining the Greater Arab Free Trade Area, and finds Algeria’s trade with other
Arab countries would have benefitted had it not delayed joining the FTA.
4
4.1
Data
Sources
The data comes from a variety of sources. Chad Bown’s Global Temporary Trade Database
collects information about antidumping cases, including the initiation of antidumping cases,
the outcomes (affirmative- duty, affirmative- no duty due to unestablished causality, negative,
and withdrawn), the products involved in cases, and the petitioning firms. The products,
given in HS codes, were matched to their corresponding ISIC Revision 3 four-digit industry. Notably, comparing the product industries to the industry where the petitioning firm
operates revealed that not all firms consider their primary industry the same as that of the
10
products identified.5
The remaining Colombia-specific data was collected by the Colombian government and
accessed either through the Departamento Administrativo Nacional de Estadı́stica (DANE)
or the Banco de la República. DANE conducts annual surveys of firms in the manufacturing
sector (the Encuesta Anual Manufacturera- EAM). The EAM covers firms with more than
10 employees or with production levels above a certain fluctuating threshold. In 2007, this
threshold was 120 million Colombian pesos (approximately $60, 000 in 2010 USD), while
in 2006, it was 121.7 million pesos. The data is collected at the firm level, but because of
privacy concerns of the firms, the data that is publicly released is at the industry level. The
data is available from 1995 through 2014, and is reported at the four-digit ISIC (Colombian
aggregation) level or the three digit ISIC level, with additional sub-categories, such as the region of the country or size of the firms6 . The ISIC has also undergone multiple revisions from
1995 to 2014. Most of the data I obtained is reported in ISIC Revision 2 industries through
1999. Then, it switches to ISIC Revision 3 through 2012, then finally to ISIC Revision 4
(the data skips Revision 3.1). To avoid concording industries through the years, I only use
data from 2000 to 2012 with the ISIC Revision 3 industry codes. Thus, the manufacturing
industry corresponds to ISIC Revision 3 two digit codes D15 through D37 (manufacturing).
The EAM survey provides detailed accounts of the industries: from the number of firms
within the 4-digit ISIC code, the number of employees (can be disaggregated into paid or
unpaid, permanent or temporary, type of contract, or even the number of employees that
are related to the owners), the compensation of employees (wages and benefits), gross production, intermediate goods used (including energy, land, materials, etc), value added, net
5
For example the products in Case 32, conventional tires (for cars, buses, or trucks) fall under the ISIC
industry “Manufacture of rubber tires and tubes; retreading and rebuilding of rubber tires” (code D2511).
However, the petitioning firm, Indústria Colombiana de Lllantas, identifies its principle operating industry
as “Sale of motor vehicle parts and accessories” (G5030). To account for this discrepancy, I consider Case
32 as affecting both industries and add variables across the two where applicable.
6
The Colombian aggregation of the ISIC industries is generally similar to the international ISIC version,
but some of the more important Colombian industries, such as the production of coffee, have their own fourdigit industry. Since I also use data reported in the ISIC international aggregation and multiple Colombian
industries may map into a single international industry, but not vice versa, I use the international ISIC codes.
Where needed, I concord the Colombian industries into the appropriate international industry classifications.
11
investment, and total assets. Monetary values are measured in thousands of current-priced
pesos. I deflate the values using a manuffacturing industry specific producer price index
compiled by DANE.
Total factor productivity of firms requires data on the labor, capital, and investment of
firms. I use balance sheet data provided by Superintendencia de Sociedades in the SIREM
database. These detailed financial accounts of “commercial companies, branches of foreign
corporations, sole proprietorships and other specified by law”7 report total assets, liabilities,
sales, and profits.
To consider the impact of antidumping investigations on domestic Colombian employment, it is important to control for the import competition. Colombian import data was
collected from the Comtrade database and product values concorded into ISIC Revision 3 industries in constant Colombian pesos to match the domestic data. Then, import penetration
is defined as the ratio of imports to domestic production plus imports in each industry.
4.2
Antidumping Cases
Antidumping cases will be used as the basis for forming groups in later sections of this paper.
Over the course of the years 2004-2008, 26 antidumping cases were initiated, half of which
resulted in price undertakings, where the firm exporting to Colombia pays the difference
between the price they charge and a base price. I restrict my attention to the 13 cases that
resulted in measures. In generating control groups to compare treatment groups (industries
with a case resulting in duties), I exclude from the potential donor pool any industry that
petitioned for protection, had the legal body determine imports were being sold at less-thanfair-value, but ultimately determined that the dumping prices were not harming the domestic
Colombian industry. I exclude these industries because of potential effects from the threat of
antidumping protection. As shown in [Prusa, 2001], much of the effects on trade of dumping
7
Function 2 of the Superintendencia de Sociedades. The companies included are companies not supervised by the Financial Superintendency in Colombia (banks and other financial firms). http://www.
supersociedades.gov.co/superintendencia/EstOrgTal/Fundeb/Paginas/default.aspx
12
policies occur during the initial phase. I want to focus on the longer term effects, so I only
consider industries where a final price undertaking was put in place. Table 2 summarizes the
included antidumping cases, and Table 3 lists the domestic Colombian firms that initiated
the cases.
13
Table 2: Colombian Antidumping Cases Initiated Between January 1, 2004 and December 31, 2008, Resulting in a Duty
Case
Country
Product
Product
Initiation
Final
Number
Named
Description
Industry
Date
Date
Duty
46
CHN
Ceramic Tableware
D2691, J6599, J6719
2/2004
11/2004
Diff btwn US$0.84/kg and price
47
CHN
Porcelain Tableware
D2520, D2691
2/2004
11/2004
Missing
49
CHN
Balls (except Golf, Table
D2520, D3693
5/2005
3/2006
Diff btwn US$0.46/kg and price
Tennis, or Inflatable)
14
50
CHN
Chain Links (polished or
galvanized)
D2899
6/2006
4/2007
Diff btwn US $1.82/kg and price
51
CHN
Socks
D1730, D1810
7/2006
5/2007
Diff btwn US$0.79/pair and price
52
CHN
Textiles- Blended Fabrics
D1711
7/2006
7/2007
Diff btwn US $0.75/sq meter and price
55
CHN
Textiles- Curtains
D1721
7/2006
7/2007
Diff btwn US$6.74/kg and price
56
CHN
Textiles- Bed Linen
D1721
7/2006
10/2007
Diff btwn US$6.64/unit and price
57
CHN
Textiles- Table Linen
D1711, D1721
7/2006
10/2007
Diff btwn US$5.29/unit and price
62
CHN
Textiles- Towels
D1711
7/2006
5/2007
Diff btwn US$5.13/kg and price
65
CHN
Bolts
D2893, D2899, G5143,
6/2008
2/2009
Diff btwn US$1.31/kg and price
G5234, J6599
67
CHN
Staples in Strips
D2899
6/2008
4/2009
Diff btwn US$1.66/kg and price
70
CHN
Hoes/ Digging Bars/ Picks
D2893
9/2008
6/2009
Diff btwn US$1.32/kg and price
Conveniently, in Colombia, it turns out that firms file petitions in bursts. Of the four
years where petitions were successful, 2004, 2005, 2006 and 2008, each year corresponds to
a different industry: ceramic and porcelain tableware, inflatable plastic balls, textiles, and
nuts, hoes, and picks (“other” metal tools industry), respectively. In the case of textiles
and metal tools, manufacturing associations called gremios (literally translates into guilds)
organized the petitions along with a few major firms, which is likely why, especially for
textiles, many petitions for different related products were all filed around the same time.
5
Results
5.1
Characteristics of Initiating Firms Within Industries
0
.01
.02
.03
.04
Firm’s Assets as Fraction of Industry’s Assets
−10
−5
0
TimeFromInitiation
90th Pctile
10th Pctile
5
10
50th Pctile
AD Firms
Figure 1: Percentiles of firm assets as fraction of total ISIC 4 digit industry assets in industries receiving
protection. Compares petitioning firms with the remainder of firms in their industries. The vertical line
indicates the year of initiation.
First, I establish general trends of the firms that initiate antidumping proceedings within
15
Table 3: Colombian Antidumping Cases Initiated Between January 1, 2004 and December 31, 2008, Resulting
in a Duty
Case ID Domestic Firms
46
Vajillas Corona S.A.
47
Loceria Colombiana S.A.
49
Kalusin Importing Company S.A.
50
Colcadenas Ltda.
51
Fábrica de Calcetines Crystal S.A.
Calceteria Nacional S.A.
Vestimundo S.A.
52
ASCOLTEX (Colombian Association of Textile Manufacturers)
55
ASCOLTEX (Colombian Association of Textile Manufacturers)
56
ASCOLTEX (Colombian Association of Textile Manufacturers)
57
ASCOLTEX (Colombian Association of Textile Manufacturers)
62
ASCOLTEX (Colombian Association of Textile Manufacturers)
65
Fábrica de Tornillos Gutemberto S.A.
Cato S.A.
Mundial de Tornillos S.A.
Tornillos y Partes Plaza S.A.
Redica Ltda.
Torhefe S.A.
Tornapar Ltda.
Ferreteria Barbosa & Cia S. en C.S.
Tornirap Ltda.
67
Sociedade General Metalica S.A.-GEMA S.A.
70
Herramientas Agricolas S.A.-HERRAGO S.A.
16
0
.02
.04
.06
Firm’s Labor Bill as Fraction of Industry’s Labor Bill
−10
−5
0
TimeFromInitiation
90th Pctile
10th Pctile
5
10
50th Pctile
AD Firms
Figure 2: Percentiles of firm short-term labor obligations as fraction of total ISIC 4 digit industry labor
obligations in industries receiving protection. Compares petitioning firms with the remainder of firms in
their industries. The vertical line indicates the year of initiation.
their industries. Note that this only covers cases when individual firms petitioned for protection, and therefore are specifically listed in the antidumping database, ignoring cases
where an association of manufacturers applied for protection.8 According to Figures 1 and
2, within an industry, the larger firms petition for protection, as measured by Total Assets
and short-term Labor Obligations.9 In the cases initiated by specific firms, the successful
petitioning firms are within the top 10% of asset holders and salary payers. Leading up to
filing a petition, they saw their control of assets falling: five years before initiating a case,
the average petitioning firm held about 3-4% of the industry’s assets. In the filing year,
8
A comprehensive list of firms involved in these associations is not available, so my next best option
would be to use the entire industry. For now, I will ignore these cases, and suffice it to say that typically, the
associations are made up of larger firms. That these would be the firms to initiate investigations is consistent
with my findings when specific firms are the petitioners rather than associations.
9
Short-term labor obligations refer to liabilities on the balance sheets for labor payments. Long-term labor
obligations includes items like severance packages and retirement benefits, while short-term labor obligations
accounts for current employees. To get a sense of the variable labor inputs, I restrict my attention to only
the short-term payments.
17
the average assets held by these firms was around 2.5%. Five years afterwards, their assets
increased again to 3-4% (followed by a sharp decline). In terms of profits, as measured by
Gross Profit Margin (GPM=Gross Profit/Sales Revenue), petitioning firms tend to perform
around the 50th to 90th percentiles, as seen in Figure 3. Profitability remains stable until
5 years after a case is initiated, then petitioning firms increased their profit margins. Similarly, petitioning firms, for the most part, had estimated total factor productivities, TFPs,
that were approximately average for their industry (Figure 4). TFP for all firms in protected industries fell within 5 years of the petition, but the decrease is more dramatic for
filing firms. Thus, petitioning firms are relatively large for their industries, but usually have
around average productivity and above average profit margins.
0
.2
.4
.6
.8
1
Industry Distribution of GPM
−10
−5
0
TimeFromInitiation
90th Pctile
10th Pctile
5
10
50th Pctile
AD Firms
Figure 3: Percentiles of gross profit margin of petitioning firms compared to gross profit margins of other
firms in same ISIC 4 digit industry that received protection. The vertical line indicates the year of initiation.
18
0
1
2
3
4
Industry Distribution of TFP
−10
−5
0
TimeFromInitiation
90th Pctile
10th Pctile
5
10
50th Pctile
AD Firms
Figure 4: Percentiles of estimated total factor productivity of petitioning firms compared to productivities
of other firms in same ISIC 4 digit industry that received protection. The vertical line indicates the year of
initiation.
5.2
Impacts on Industries with Successful Petitions
To determine the impact on employment and wages in industries with successful antidumping
petitions (ie, those that resulted in duties), I use the synthetic control method as described
previously. The employment variable used in the SCM is defined as
log
Employmentt
Employment2000
where Employmentt is the number of employees within the industry in the year t and
Employment2000 is the number of employees in the year 2000. Thus, I am measuring the
employment growth of these industries. Similarly, the wage per person variable is defined as
log
(T otalW agest )/(Employmentt )
(T otalW ages2000 )/(Employment2000 )
19
I also performed the described analysis on total wages paid and total benefits paid, but
omitted the results because most of the information gleaned from them are already reflected
in the total employment and average wages analyses. Since the analysis is now performed at
the ISIC four-digit industry level, I include cases where a manufacturing association initiated
the case.
Before the year of treatment, defined as when the antidumping case was initiated, we
are looking for similar values between the treated group and the synthetic group to ensure a
similar trend for comparability. The measure of this fit, the Root Mean Squared Prediction
Error (RMSPE) is given with each graph showing the comparison between the treated group,
synthetic control group, and manufacturing sector average in Figures 5 and 7. Post treatment, the difference between the treated average and the synthetic average is the treatment
effect. Using the method from Abadie et al. [2010] allows the treatment effect to change
through time. The vertical axes of these graphs show the magnitude of change- the percentage growth since 2000. The actual treatment effects (the difference between the average for
the treated group and the average for the control group) are plotted in Figures 6 and 8.
These graphs show a heterogeneous effect of antidumping policies across industries, especially in employment. Recall that in 2004, the impacted products were ceramic and
porcelain tableware. Relative to a control group with a similar pre-treatment trend, the
broader industries (ISIC codes 2520 manufacture of plastic products and 2691 manufacture
of non-structural non-refractory ceramic ware) experienced a decline in employment after
the imposition of protection measures. Similarly, for cases initiated in 2006 which focused
on textiles, the employment levels fell dramatically. However, in this instance, the decline in
employment is more likely attributed to what triggered the investigation in the first place.
Starting January 1, 2005, the WTO’s Agreement on Textile and Clothing mandated the
removal of quotas on textile and clothing in international trade. January 1, 2005 marked the
final round of quota elimination. The relative decrease in employment shown for the 2006
textile antidumping cases is actually picking up effects from the quota elimination, as firms
20
Percent Change in Employment from 2000
Percent change
0 20 40 60 80
Percent change
0 20 40 60 80100
Treated group and Control group
2000
2005
2010
2015
2002
2004
Year
2010
2012
RMSPE = 0.023
Percent change
−20
−100 102030
Percent change
−20 0 20 40 60
RMSPE = 0.026
2006
2008
Year
2002
2004
2006
2008
Year
2010
2012
RMSPE = 0.14
2004
2006
2008
Year
2010
2012
RMSPE = 0.14
Treated
Economy
Control
Figure 5: Comparison of the Treated Group and Control Group on Employment. Vertical lines indicate
treatment year. Vertical axis measures the percent change in employment from 2000 for the Treated group,
Control group, and the aggregate change in all of the manufacturing sector.
turn to cheaper Chinese textiles and others move their production to China, showing large
employment losses from the quota elimination. Finally, for the 2008 cases, covering bolts,
staples, hoes, digging bars, and picks, which fall into ISIC codes 2893 and 2899 (manufacture of cutlery, hand tools and general hardware, and manufacture of other fabricated metal
products not elsewhere specified), the result is more expected, where antidumping protection
allowed for more workers to be employed.
Figure 6 highlights the magnitude of the differences. Since the more interesting cases
are the ones initiated in 2005 and 2008, I focus on the two right-most graphs. The 2005
cases (balls, except golf, table tennis, or inflatable balls) saw, perhaps, a small initial spike in
21
Percent Change in Employment from 2000
Treatment Effect
0
5
10
−5
Treatment Effect
−30 −20 −10 0
10
Difference between treated group and control group
2000
2005
2010
2015
2002
2004
2006
2008
Year
2010
2012
0
−60
Treatment Effect
−40
−20
0
Treatment Effect
5
10
15
Year
2002
2004
2006
2008
Year
2010
2012
2004
2006
2008
Year
2010
2012
Figure 6: Treatment Effect on Employment. Vertical lines indicate treatment year. Vertical axis measures
the difference in the percent change in employment from 2000 between the treated group and the synthetic
control group.
relative employment of roughly 3-4% higher employment than the control group. However,
this spike quickly disappeared and led to nearly 5% lower employment levels. The duty was
in place until 2011, after which employment in the industry grew, but again, the magnitude
is small. In the assorted metal tools industries that initiated successful petitions in 2008,
the increase in employment ranged from 5-15% higher than the hypothetical control group.
In terms of average wages per person, there are again heterogeneous effects of antidumping
duties on different industries, where some industries see lower average wages (tableware and
inflatable balls) compared to their control groups, and others (textile and metal tools) see
higher average wages. Figure 7 shows the percent change in average wages compared to
22
Percent Change in Wages per Person from 2000
Percent change
−10 0 10 20 30
Percent change
−10 0 10 20 30
Treated group and Control group
2000
2005
2010
2015
2002
2004
Year
2010
2012
Percent change
−10 0 10 20 30 40
RMSPE = 0.0044
Percent change
−10 0 10 20 30
RMSPE = 0.0095
2006
2008
Year
2002
2004
2006
2008
Year
2010
2012
RMSPE = 0.016
2004
2006
2008
Year
2010
2012
RMSPE = 0.050
Treated
Economy
Control
Figure 7: Comparison of the Treated Group and Control Group on per person wages. Vertical lines indicate
treatment year. Vertical axis measures the percent change in average wages from 2000 for the Treated group,
Control group, and the aggregate change in all of the manufacturing sector.
2000 for the synthetic and control groups, as well as the manufacturing industry average.
Notably, in the upper two graphs that show wages in the tableware and ball industries,
there is essentially no effect for the first few years the duty is in place. This could possibly
indicate that the protection from the antidumping duty delayed the lowering of wages for
approximately 4 or 5 years until the duties approached their time for sunset reviews.
The magnitude of the treatment effects, found in Figure 8, is generally around a 10%
difference between the treated and control groups. That is, in the 2004 and 2005 cases, wages
in the treated group fell down to around 7% lower than the synthetic control groups, despite
increasing since 2000. Successful cases initiated in 2006 and 2008 saw higher average wages
23
Percent Change in Wages per Person from 2000
−8
Treatment Effect
−6 −4 −2 0
Treatment Effect
−8 −6 −4 −2 0 2
Difference between treated group and control group
2000
2005
2010
2015
2002
2004
2006
2008
Year
2010
2012
Treatment Effect
−10 −5 0 5 10 15
−5
Treatment Effect
0
5
10 15
Year
2002
2004
2006
2008
Year
2010
2012
2004
2006
2008
Year
2010
2012
Figure 8: Treatment Effect on Wages Per Person. Vertical lines indicate treatment year. Vertical axis
measures the difference in the percent change in average wages from 2000 between the treated group and
the synthetic control group.
in the protected industries relative to the control industry. However, part of the positive
effect in 2006 can be explained, again, by the effects from quota elimination. Since lower
cost Chinese producers could send more to the Colombian market, Colombian textile firms
became more specialized in higher quality textiles and more vertically integrated production
processes ([County and Regional Analysis Division, 2006]). With the move to higher-end
clothing, one would anticipate higher average wages.
Figures 9 through 11 give context on the effect of antidumping duties on import volumes.
For the most part, import volume measured in weight initially fell after duties were imposed,
but recovered shortly thereafter. The value in USD and value per kilogram shipped, however,
24
did not generally fall. Overall, this leads to higher import prices.
Value in USD
Balls
TableImports
14.5 15 15.5 16 16.5 17
BallImports
14 14.5 15 15.5 16
Tableware
2005
2010
2015
2000
2005
2010
Year
Year
Textiles
Metal Tools
2015
18
MetalImports
15.5 16 16.5 17 17.5 18
TextileImports
18.5
19
19.5
2000
2000
2005
2010
2015
2000
2005
Year
2010
2015
Year
Figure 9: Log of import value of protected industries in Colombia, in USD
Weight in Kg
Balls
BallImportsWt
12 12.5 13 13.5 14
TableImportsWt
15 15.5 16 16.5
Tableware
2005
2010
2015
2000
2005
2010
Year
Textiles
Metal Tools
2015
14
MetalImportsWt
15
16
17
Year
TextileImportsWt
17.2 17.4 17.6 17.8 18
2000
2000
2005
2010
2015
Year
2000
2005
2010
2015
Year
Figure 10: Log of import volume as measured by weight in kilograms of protected industries in Colombia
5.3
Robustness Checks
The placebo test is the standard robustness check used with the synthetic control method.
In the placebo test, the treatment effect is determined as though each industry in the pool
25
Value per Kg
−.5
BallUnitValue
1 1.2 1.4 1.6 1.8 2
Balls
TableUnitValue
0
.5
1
Tableware
2005
2010
2015
2000
2005
2010
Year
Year
Textiles
Metal Tools
2015
0
.8
TextileUnitValue
.5
1 1.5 2
MetalUnitValue
1 1.2 1.4 1.6 1.8
2000
2000
2005
2010
2015
Year
2000
2005
2010
2015
Year
Figure 11: Log of price per kilogram of goods in protected industries in Colombia
of potential control industries (ie, other manufacturing industries that did not receive antidumping protection) receives antidumping protection. The root mean squared prediction
errors are compared to determine the similarity to the synthetic control groups, and the
magnitudes of treatment effects to determine significance. As it turns out, the results do not
hold well against robustness checks.
Table 4 gives the percent of the placebo tests that fit the pre-treatment trends better than
the actual treated group. A lower RMSPE reflects a better fit, and a stronger result. Figures
12 and 13 graph out the treatment effects for the placebo tests and the true treated group.
As can be seen in these plots, the magnitude of the treatment effects are not largely outside
of the placebo industries, meaning the antidumping protection does not appear to have had
a huge impact on the protected industries. Each placebo test checks the “treatment” effects
in about 60 industries. To only compare similar industries, Figures 14 and 15 only graph
the effects in industries that have a median import elasticity within ± 0.5 of the protected
industry. Import elasticities are measured using elasticity estimates from Broda et al. [2006]
from Colombia. The graphs show the treatment effects are still not robust, meaning similar
changes in the number of workers and average annual wages were found in many other
26
Table 4: Root Mean Squared Prediction Error Compared with Placebo Tests
Dependent
Year of
Variable
Initiation
RMSPE
with a smaller RMSPE
2004
0.026
68%
2005
0.023
24%
2006
0.014
13%
2008
0.137
92%
2004
0.0095
38%
Average
2005
0.0044
10%
Wage
2006
0.0162
13%
2008
0.050
53%
Employment
Percent of placebo tests
industries, so the difference is not likely attributable to antidumping protection.
2
−1
−1
0
0
1
1
2
Treatment Effects Change in Total Employment from 2000
2000
2005
2010
2015
2002
2004
2006
2008
Year
2010
2012
−1
−1
0
0
1
1
2
2
3
3
Year
2002
2004
2006
2008
Year
2010
2012
2004
2006
2008
Year
2010
2012
Figure 12: Robustness Check: Compares the resulting treatment effect when each ISIC four digit industry
is considered individually as receiving the treatment. Each gray line is the treatment effect of a four digit
industry, and the black line is the treatment effect of the industries receiving antidumping protection. Vertical
lines indicate the treated year.
6
Conclusion
Overall, the impacts from antidumping duties vary across industries and measures. By
taking an approach that is typically applied in policy evaluation studies, but rarely applied
27
in international trade, I quantify the domestic impacts from antidumping protection. In
some instances, protected industries are better off than a synthetic control, and in others,
they are worse off, likely due to confounding events that led to an antidumping petition in the
first place. When antidumping duties help, they increase employment and wages per person
by up to 15% over the control group, artificially constructed to represent the pre-treatment
affected industry. Despite these magnitudes, there is not strong evidence that the results are
robust. Similar results can be obtained by conducting the same analysis on certain other
industries in the Colombian economy, even when restricting the sample to industries with
similar import elasticities.
Clearly, different firms have different responses to protection. In the future, I will explore
why some industries have more positive responses to protection than others. Are there ways
we can determine when duties will help the most? Do firms that invest more in research
and development fare better than those that do not? Employment is changing- are the
composition of workers changing with it? Using the rich surveys and balance sheet data
from Colombia can provide insight into these questions.
References
New ituc worldwide report reveals catalogue of murder, violence and intimidation
against trade unionists, September 18, 2007.
URL http://www.ituc-csi.org/
new-ituc-worldwide-report-reveals.
Alberto Abadie, Alexis Diamond, and Jens Hainmueller. Synthetic control methods for
comparative case studies: Estimating the effect of california’s tobacco control program.
Journal of the American Statistical Association, 105(490):493–505, 2010.
Ilke Van Beveren. Total factor productivity estimation: A practical review. Journal of
Economic Surveys, 26:98–128, 2010.
28
−1
−.5
−.6 −.4 −.2
0
0
.2
.5
.4
Treatment Effects Change in Average Wages from 2000
2000
2005
2010
2015
2002
2004
2006
2008
Year
2010
2012
−.5
−.5
0
0
.5
.5
1
Year
2002
2004
2006
2008
Year
2010
2012
2004
2006
2008
Year
2010
2012
Figure 13: Robustness Check: Compares the resulting treatment effect when each ISIC four digit industry
is considered individually as receiving the treatment. Each gray line is the treatment effect of a four digit
industry, and the black line is the treatment effect of the industries receiving antidumping protection. Vertical
lines indicate the treated year.
−1
−1
−.5
−.5
0
0
.5
.5
Treatment Effects Change in Total Employment from 2000
2000
2005
2010
2015
2002
2004
2010
2012
0
1
2
3
2006
2008
Year
−1
−1
0
1
2
3
Year
2002
2004
2006
2008
Year
2010
2012
2004
2006
2008
Year
2010
2012
Figure 14: Robustness Check: Placebo test including only industries with median import elasticities of ±0.5
of protected industries.
29
Andreas Billmeier and Tommaso Nannicini. Assessing economic liberalization episodes: A
synthetic control approach. Review of Economics and Statistics, 95(3):983–1001, 2013.
Christian Broda, Joshua Greenfield, and David Weinstein. From groundnuts to globalization:
A structural estimate of trade and growth. NBER Working Paper No. 12512, 2006.
Bureau of Democracy, Human Rights, and Labor. 2005 country reports on human rights
practices colombia. Technical report, 2006. URL http://www.state.gov/j/drl/rls/
hrrpt/2005/61721.htm.
Sunghoon Chung, Joonhyung Lee, and Thomas Osang. Did china tire safeguard save us
workers? European Economic Review, 85(C):22–38, 2016.
Committee on Anti-Dumping Practices. Notification of laws and regulations under articles
18.5 and 32.6 of the agreements: Colombia. Technical report, World Trade Organization,
1995.
County and Regional Analysis Division. The impact of the andean trade preference act.
Technical report, US International Trade Commission, 2006.
El Director de Comercio Exterior. Resolución número 097. Technical report, Ministerio de
Comercio, Industria Y Turismo, July 2016.
Amr Sadek Hosny. Algeria’s trade with gafta countries: A synthetic control approach.
Transition Studies Review, 19:35–42, 2012.
Jozef Konings and Hylke Vandenbussche. Heterogeneous response of firms to trade protection. Journal of International Economics, 76:371–383, 2008.
Yi Lu, Zhigang Tao, and Yan Zhang. How do exporters respond to antidumping investigations? Journal of International Economics, 91:290–300, 2013.
G. Steven Olley and Ariel Pakes. The dynamics of productivity in the telecommunications
equipment industry. Econometrica, 64:1263–1297, 1996.
30
Justin Pierce. Plant-level responses to antidumping duties: Evidence from us manufacturers.
Journal of International Economics, 85:222–233, 2011.
Thomas Prusa. On the spread and impact of anti-dumping. The Canadian Journal of
Economics, 34(3):591–611, 2001.
31
−.6
−.4
−.6 −.4 −.2
−.2
0
0
.2
.2
.4
Treatment Effects Change in Average Wages from 2000
2000
2005
2010
2015
2002
2004
2006
2008
Year
2010
2012
−.5
−.2
0
0
.2
.5
.4
1
.6
Year
2002
2004
2006
2008
Year
2010
2012
2004
2006
2008
Year
2010
2012
Figure 15: Robustness Check: Placebo test including only industries with median import elasticities of ±0.5
of protected industries.
32