Trade Reforms and Industry Wage

Premium: Evidence from Argentina

Guillermo Falcone y Luciana Galeano

Documento de Trabajo Nro. 212

Mayo, 2017
ISSN 1853-0168

www.cedlas.econo.unlp.edu.ar
 Trade Reforms and Industry Wage Premium:
Evidence from Argentina

Guillermo Falcone1 Luciana Galeano2

ABSTRACT

This paper studies the impact of Argentina trade liberalization during the nineties on
the industry wage premium structure. We find that accounting for unobserved time-
invariant industry characteristics is crucial. When we do not control for industry fixed
effects, we find that workers in protected sectors receive lower wages. However,
introducing industry fixed effects reverses the results; tariff protection creates sector
specific rents that are in part translated to workers in terms of greater wages. Since
Argentina’s tariff structure during this period protected relatively more sectors
employing higher proportions of skilled workers, nineties trade policy may have had
an adverse effect on Argentina’s income distribution.

RESUMEN

Este trabajo estudia el impacto de la liberalización comercial argentina en los noventa

sobre la estructura de las primas salariales sectoriales. Encontramos que resulta
crucial controlar por características sectoriales inobservables e invariantes en el
tiempo. Cuando no incluimos efectos fijos por industrias, encontramos que los
trabajadores en sectores protegidos reciben menores salarios. Al incluirlos se revierten
los resultados; la protección crea rentas sectoriales que se trasladan a los trabajadores
vía mayores salarios. Como la estructura tarifaria argentina durante este período
protegió relativamente más a sectores calificados, la política comercial podría haber
tenido un efecto adverso en la distribución del ingreso.

Key words: skill premium; industry premium; trade liberalization; Argentina

1 CEDLAS, Facultad de Ciencias Económicas, Universidad Nacional de La Plata, Calle 6 N777,

1900, La Plata. Email: guillermofalcone@gmail.com
2 CEDLAS, Facultad de Ciencias Económicas, Universidad Nacional de La Plata, Calle 6 N777,

1900, La Plata. Email: luchigaleano@gmail.com

1
1 Introduction

Trade economists have long investigated the relationship between trade

liberalization and wage premium paid to skilled workers. The natural question
to be answered is whether globalization and deepening of international trade
could be responsible for the growth in relative earnings of skilled workers
during the 1980s and 1990s in developed and developing countries respectively.
Maybe the most immediate way of thinking why trade could be associated with
movements in skill premium is to reason in a Heckscher-Ohlin framework.
Countries have different proportions of skilled and unskilled workers, and so
do the goods that each country produces and exports. If this is true, then
opening to trade is analogous to a skill transfer between economies. A country
with relative scarcity of skilled workers would export unskilled-labor-intensive
products and import skilled-labor-intensive products, and then opening the
economy would be equivalent to an increase in its relative supply of skilled
workers and a decline in its relative supply of unskilled workers. In this
framework, opening the economy to trade would have an overall impact over
the premium of skilled workers. We call this, an impact in the skill premium.
So if we rely in the Heckscher-Ohlin model for explaining the rise in the skill
premium in developing countries (and in Latin America in particular) in late
1980s and early 1990s, we find a priori a contradiction. As long as these
countries are unskilled-labor abundant, one would expect that their opening to
trade should lead to a decline in their skill premium and inequality during the
nineties. But, at least in Argentina, the effect we observe is the opposite.
Should we conclude that Heckscher-Ohlin is not working? Galiani and Porto
(2010) provide an example on how to study the relationship between trade and
wages in a Stolper-Samuelson framework with bargaining power of unions.
However, extra assumptions are not necessary to reconcile facts with the
classical model. Since tariff reductions in Latin American countries were
disproportionately larger in unskilled-labor intensive sectors (Attanasio,
2
Goldberg and Pavcnik, 2004; Hanson and Harrison, 1999; Pavcnik, Blom,
Goldberg and Schady, 2004), the classic model could still be working. Given
that tariff cuts were larger in unskilled-labor intensive sectors, the economy-
wide return to unskilled workers should have decreased. Nevertheless, the
evidence shown on those studies is not sufficient to conclude that the
mechanism through which trade affect wages is the one described by Stolper-
Samuelson years ago (Goldberg and Pavcnik, 2004).
More recent models of trade point out the relevant role that industry affiliation
plays in determining how trade affects wages (Goldberg and Pavcnik, 2005). In
these models, at least in the short run, labor is not perfectly mobile across
sectors as was in Stolper-Samuelson (Heckman and Pagés, 2000). This
assumption is important for the generation of sector specific rents that do not
evaporate in the short run due to labor stickiness. Workers with similar levels
of education or skills (even more, with similar levels of all observable
characteristics such as gender, age, formality, etc.), could be affected differently
by international trade just because of their sector or industry affiliation. When
two workers that are equal in all observable characteristics receive different
wages because they participate in different industries, we will say that there
exists an industry wage premium. In this type of models, since workers are not
completely mobile across sectors, tariff cuts translate into proportional
industry wage premiums reductions.
Also, classical models of trade suppose perfect competition in product and labor
markets. However, trade policy can affect industry wage premiums in
frameworks of imperfect competition. An example of that comes from
Levinsohn (1993), who use Turkish drastic manufacturing trade liberalization
in 1984 for testing the hypothesis of import as market discipline in an oligopoly
model. He finds that sectors with higher import penetration reduced markups
in a larger proportion. In this context, trade liberalization of the more
unskilled-labor-intensive sectors in Latin America should have also played a

3
 role in reducing industry wage premium in sectors with a low fraction of skilled
workers.
Our purpose in this paper is to estimate industry wage premiums in Argentina
during the 1990s and relate them to trade policy in each sector following a two-
step methodology estimation used by Goldberg and Pavcnik (2005). We find
that, once we control for unobserved time-invariant industry characteristics,
tariff reductions that took place in the 1990s reduced industry wage premiums
in the manufacturing sectors under analysis. Since tariff reductions were
larger for sectors who used a high proportion of unskilled workers, that trade
liberalization in Argentina may have contributed to raise inequality in the last
decade of 20th century. It is worth mentioning that our results do not refer to
changes in the skill premium per se, but to the premium that industries pay to
workers with the same level of skills. In this sense, if trade liberalization also
had a positive effect on the overall skill premium of the economy, low-skill
workers could have been hit twice.
The remainder of the paper is organized as follows. In Section 2, we motivate
the work by describing trade reforms made by Argentina in the nineties.
Section 3 is a brief description of the data used in the paper. In Section 4, we
present our regression model and explain the empirical methodology we follow.
In Section 5, we show our results, and Section 6 is left for conclusions.

2 Argentina's Background

Argentina’s economic performance during the 1980s was quite poor, and by the
end of the decade the economy was broken down. Growth of real output was
stagnated, financial markets were collapsed and prices were rising largely due
to currency depreciation. The hyperinflation of 1989 finally provided the
impetus for the economic reforms applied in the early nineties. Trade
liberalization policy was one of the basic instruments included in the

4
stabilization program which meant a new way of introducing the country in
international trade.
Trade liberalization started gradually as a unilateral policy in 1988 but made a
strong impact only after 1990. The program included both a reduction in
nominal protection and a significant reduction of tariff position that were
subject to quantitative restrictions. By the end of 1991, nominal tariffs had
been lowered to an average level of 12 percent and all import licenses had been
eliminated (Galiani and Sanguinetti, 2003).
Trade liberalization significantly impacted trade flows. As can be seen in
Figure I, between 1990 and 1998 the share of imports on GDP almost
triplicated, going from 4.63% to 12.93%.

Figure I. Evolution of imports (% of GDP)

14
12
10
8
6
4
2
0

Source: own elaboration based on WDI, World Bank.

Also, Table I reports import penetration defined as imports to value-added

disaggregated at industry level from 1993 to 1999. Import penetration
increased in all sectors during this period, with the exception of beverages and
footwear. On average, this index more than duplicated.

5
 Table I. Import Penetration by Manufacturing Sector
Manufacturing sector 1993 1994 1995 1996 1997 1998 1999
Food products 12.9 14.6 13.3 14.6 13.7 17.0 14.4
Beverages 4.7 4.1 3.8 3.3 4.4 4.7 4.1
Tobacco 0.1 0.1 0.1 0.2 - - -
Textiles 36.5 36.2 41.3 52.5 51.2 56.8 49.7
Apparel (except fur apparel) 28.4 30.0 26.8 26.7 - - -
Dressing and dyeing of fur; articles of
12.6 15.5 11.5 18.7 26.8 50.5 44.3
fur
Footwear 44.9 38.3 32.8 34.6 22.3 28.4 36.1
Wood production (non furniture) 28.6 40.4 55.4 63.1 61.5 67.0 62.4
Paper and paper products 79.7 86.5 98.4 119.2 82.2 80.2 83.4
Printing and related activities 7.5 9.3 11.6 11.2 - - -
Refine petroleum products 9.9 18.0 22.6 21.5 30.2 45.8 26.0
Basic chemicals 262.9 217.6 238.0 285.0 - - -
Other chemical products 32.5 38.7 39.2 44.5 - - -
Rubber products 64.8 66.5 70.5 88.4 109.4 107.6 153.7
Plastics products 23.2 29.5 30.9 29.5 40.3 37.9 31.9
Glass and glass products 26.4 31.6 30.9 32.9 39.9 51.7 47.2
Non-metallic mineral products n.e.c. 11.7 14.3 15.7 15.9 - - -
Basic iron and steel 51.1 48.8 45.7 44.6 59.6 63.6 74.7
Basic precious and non-ferrous
75.6 81.2 84.2 107.5 131.4 246.8 155.9
metals
Other metal products 33.2 44.9 53.5 52.9 67.6 69.5 67.1
General purpose machinery 184.9 264.8 299.2 291.2 351.8 356.2 403.2
Domestic appliances n.e.c. 218.2 268.7 296.2 307.5 369.0 411.8 359.0
Medical appliances and instruments
208.5 297.7 247.3 359.9 393.5 351.8 378.8
of precision
Motor vehicles 113.8 137.9 141.4 164.3 222.3 278.2 322.6
Furniture 18.2 17.5 13.3 19.9 47.2 46.3 52.7
Other manufacturing 171.6 188.5 217.5 234.0 448.0 422.6 415.1
Average 67.8 78.5 82.3 94.0 128.6 139.7 139.1
Source: own elaboration based on Nicita and Olarreaga database (2007).

Industry is well known to be a low-skill-labor sector relative to the average

skill level of the economy. This can be seen in Table II, which shows the share
of employment by skill group for the wide economy and for the manufacturing
sector. Thus, given that manufacturing as a whole is a low-skill-labor intensive
sector, if tariff reductions are larger for manufactures that employ high
proportions of unskilled workers, the impact of trade liberalization on income
distribution would be twice negative.
6
 Table II. Share of skilled workers by sector

Share of skilled workers (%)

1991 1995 1999
Manufacturing 9.99 12.01 14.96
Services 22.86 21.49 23.60
Total economy except manufacturing 22.87 21.56 21.12
Total economy 19.65 19.78 19.78
Source: own elaboration based on EPH (INDEC).

In Figure II we show the evolution of mean tariff for skilled and unskilled
industries. We define as unskilled workers those with [0-8] years of education,
and as skilled workers those with 9 or more. For each year, we classify a sector
as skilled if the share of skilled workers is greater than the share of unskilled
workers, and vice versa. In most cases we find that the sector classification is
stable along time. However, since we are working with household surveys
disaggregated at the 3 digit level, in some years we have sectors with few
observations, and therefore it could happen that an industry that seems to be
skilled abundant, in one year it has a larger share of unskilled workers than
usual. We decide to dismiss these cases and classify a sector in each group for
all years according to the category that predominates all over the panel. As a
result, a sector will be considered either skilled or unskilled for every year in
our data.
It can be seen that in the nineties tariffs reductions were relatively larger for
unskilled sectors, so Argentina’s trade policy during the nineties favored
skilled industries relatively more.

7
 Figure II. Evolution of tariffs by skill group
10
5
0
1989-1991 1992-1995 1996-1999
-5
-10
-15
-20
-25
Unskilled Skilled

Source: own elaboration based on Galiani and Porto (2010) and EPH (INDEC).

At the same time, inequality rose considerably, a common trend in Latin

American countries during this decade. Alvaredo and Gasparini (2015)
document that income inequality rose sharply in the region in the late 1980s
and in the 1990s, probably as the result of recurrent macroeconomic crisis and
some structural transformations. Figure III and IV show the wage gap between
high skilled (14 or more years of education) and low skilled workers (8 or less)
and the Gini coefficient on hourly wage in main job, respectively, for Argentina
between 1991 and 2000.

8
 Figure III. Wage gap
3.10

3.00

2.90

2.80

2.70

2.60

2.50
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000

Source: own elaboration based on EPH (INDEC).

Figure IV. Gini coefficient on hourly wages

0.440
0.430
0.420
0.410
0.400
0.390
0.380
0.370
0.360

Source: own elaboration based on EPH (INDEC).

9
 Given these two stylized facts, in the rest of this paper we investigate if trade
liberalization could have been one of the factors behind the sharp increase in
inequality.

3 Data

We use four different types of data: household-level data from the Permanent
National Household Survey (EPH) carried out by the Argentinian national
statistical office (INDEC) and standardized following the procedure used in
SEDLAC (CEDLAS-Universidad Nacional de La Plata and World Bank);
industry-level data from the trade and production database compiled by Nicita
and Olarreaga (2007); import tariff data used by Galiani and Porto (2010); and
multilateral real exchange rate series for Argentina published by the Central
Bank of Argentina.
Household level data from EPH (INDEC) contains information on earnings,
worker characteristics and industry affiliation. Information is homogenized to
be consistent across time. We restrict the sample to individuals in working age
(15 to 65 years old) who report positive labor earnings. We use surveys for the
period 1991-1999.
The variables that we use from the household surveys in our analysis are labor
earnings, age, gender, education, sector of employment, geographic location
and formality status. These variables are used in our Mincer regressions in
which we relate labor earnings with worker characteristics and include
dummies for sector affiliation. The coefficients accompanying these dummies
reflect industry wage premium, that is, differences in wages between workers
that are exactly the same, except for their industry affiliation. The definition of
sector of employment is expressed at the 3-digit level of the ISIC Revision 3
classification, so we match survey information to this classification using
United Nations Statistics Division concordance tables.

10
 Trade and production database from Nicita and Olarreaga (2007) collects sector
data at the 3-digit level of ISIC Revision 2 classification sector. We use
concordance tables again to translate them to Revision 3. This data contains
information about imports, exports, value-added, output, number of employees
and other variables, all at the sector level. With this data we construct two
variables of relevance: i) productivity, defined as the ratio of output (or value
added) to number of employees; and ii) trade openness, defined as the ratio of
imports to output (or value-added). Both productivity and trade openness are
interesting variables for explaining industry wage premiums. We have this
information for the years 1993-1999.
Import tariff data is the same used by Galiani and Porto (2010). It consists on
historical data collected by the authors from the Guía Práctica del Exportador e
Importador, a monthly publication that provides current tariffs at the most
disaggregated level of the National Import Tariff Classification (NADI). We
have this information for the period 1991-1999.
Finally, we use multilateral real exchange rate data series published by the
Central Bank of Argentina.

4 Empirical Strategy

We follow Goldberg and Pavcnik (2005) in investigating the effects of trade

policy on wage inequality, who suggest a two-stage estimation methodology. In
the first stage, the log of worker i’s wage is regressed on a vector of worker i’s
characteristics, including age, gender, level of education, dummies for formality
status and region of residence. Besides, we incorporate dummies for industry
affiliation. The coefficients on the industry dummies are interpreted as
industry wage premiums, and represent the part of the variation in wages that
is explained solely by workers’ industry affiliation, and not by workers’
individual characteristics. In other words, they represent the difference in
wages between workers that are exactly the same, except from the industry

11
where they work. In this stage eq. (1) is regressed separately for each year on
our sample:

ln(𝑤𝑖𝑗𝑡 ) = 𝐶𝑖𝑗𝑡 𝛽𝐶 + 𝐼𝑖𝑗𝑡 𝑝𝑗𝑡 + 𝜀𝑖𝑗𝑡 (1)

where ln(𝑤𝑖𝑗𝑡 ) is the log of worker i’s wage in sector j and time t; 𝐶𝑖𝑗𝑡 is the
vector of worker i’s characteristics in sector j and time t and 𝐼𝑖𝑗𝑡 are dummies
for industry affiliation of worker i in sector j and time t.
The use of individual characteristics allows us to control for compositional
differences across sectors. Average industry wages may vary across industries
because of the different types of workers that each sector usually employs, and
consequently sectors with a large share of skilled workers are likely to have
higher average wages. If these industries have low tariffs, one would wrongly
predict that protecting industries in the form of higher tariffs would lead to a
decrease in their worker’s wages. Note that this problem would not exist if
industry composition were constant over time, since industry fixed effects could
capture differences in composition between sectors.
Once we obtain the industry wage premium coefficients (𝑝𝑗𝑡 ), we compute a
normalized industry wage premium variable, expressing each 𝑝𝑗𝑡 coefficient as
deviation from the employment-weighted average industry wage premium (𝑤𝑡 ),
which is given by:

𝑤𝑡 = ∑ 𝑝𝑗𝑡 ∗ 𝑠𝑗𝑡
𝑗

where 𝑠𝑗𝑡 is the share of employment of sector j in year t. So, the normalized
industry wage premium (𝜑𝑗𝑡 ) is given by:

𝜑𝑗𝑡 = 𝑝𝑗𝑡 − 𝑤𝑡

12
This new variable is interpreted as the proportional difference in wages for a
worker in a given industry relative to an average worker in all industries with
the same observable characteristics.
Then we pool these normalized industry wage premiums over time and regress
them on trade industry characteristics (included in the vector 𝑇𝑗𝑡 ), namely
tariffs, value of imports, degree of openness, productivity, multilateral real
exchange rate and some interactions. Additionally, we include year and sector
fixed effects (𝐷𝑡 and 𝐷𝑗 , respectively).
Eq. (2) estimates the response of relative wages to trade policy changes,
controlling for other observables.

𝜑𝑗𝑡 = 𝑇𝑗𝑡 𝛽𝑇 + 𝐷𝑡 + 𝐷𝑗 + 𝑢𝑗𝑡 (2)

Although there probably are other variables affecting industry wage premiums,
data constraints prevent us from including them in the analysis. For instance,
if omitted variables are correlated with tariff changes, the OLS estimated
coefficients would be biased. However, if these variables affecting industry
wage premiums are not correlated with trade policy changes, or if they are time
invariant, our estimated trade policy effects would be correct.
Another problem arises if we think that policymakers take into account
industry characteristics when deciding the degree of protection they will apply.
This could be captured in the second stage of the estimation through industry
fixed effects, if political economy factors that are important do not change much
over time. However, if they do, for example responding to other variables that
also affect wages, our estimations would be biased. Consequently, we
instrument sector tariffs in Argentina with tariffs on the same sectors in Chile.
The identification assumption is that Chilean tariffs are not correlated with
any variable that affects industry wage premiums in Argentina other than
tariffs in Argentina. As is well known, we can check correlation between sector

13
tariffs in both countries looking at the estimator of the first stage of two stage
least squares.
Finally, as the dependent variable in eq. (2) is estimated, it is measured with
error. Although this does not affect consistency if this measurement error is
uncorrelated with the independent variables, additional noise is introduced in
eq. (2), which increases the variance of the estimators. To account for the fact
that usually the noise in the dependent variable differs across sectors and
depends on the variance of the estimated coefficients on industry dummies in
eq. (1), we use weighted least squares (WLS) following Krueger and Summers
(1988). This methodology consists on approximating the standard errors of the
normalized wage premium coefficients by the standard errors of the coefficients
of the original regression, and for the omitted base category by the standard
error of the constant term.

4.1 Instrumental variables

As is well known, the fixed effects estimation does not control for unobserved
time variant industry heterogeneity. Two problems may arise. First, if there
are omitted variables in equation (2) that vary over time and are correlated
simultaneously with industry wages and tariff protection, our OLS and WLS
estimations will be biased. For example, exogenous shocks to world prices.
Second, there could be time-varying selection into industries if trade
liberalization leads to industries’ compositional effects based on unobserved
workers’ characteristics. This bias could make coefficients be either under or
over estimated. If more able workers tend to leave sectors with larger tariff
cuts, the sample we get will be less able due to attrition. Therefore, our OLS
and WLS estimators will be biased upwards. On the contrary, if we think that
industries that experiment larger tariff cuts will dismiss less able workers, the

14
 sample will be composed by more able workers and the tariff coefficient will be
biased downwards.3
In order to solve these potential problems, we instrument sector tariffs in
Argentina with tariffs on the same sectors in Chile. The identification
assumption is that Chilean tariffs are not correlated with any variable that
affects industry wage premiums in Argentina other than Argentinean tariffs.
Correlation between sector tariffs in both countries can be examined by looking
at the estimator of the first stage of two stage least squares.

5 Results

The results we are interested in are those related to the second equation of the
empirical strategy4. We want to know if changes in trade policy affect industry
wage premiums through different hypothetical channels such as labor
stickiness or imperfect competition in labor or product markets.
Although we count on tariff information since 1991, output data is only
available since 1993, and that is why our regression sample consists on twenty-
six industries with available data for the period 1993-1999. Six industries lack
output information for the last three years but we decided to keep them
anyway, in order to avoid taking any arbitrary decision with possible
endogeneity bias behind.
Our most relevant results are shown in Table III. Columns (1) and (2) show
estimations of eq. (2) using OLS method, columns (3) and (4) using WLS and (5)
and (6) using IV.
We should remember that OLS has two potential problems. First, as the
dependent variable is estimated, it is measured with error. Although this does
not affect consistency if this measurement error is uncorrelated with the
independent variables, additional noise is introduced in eq. (2), which increases

3 See Goldberg and Pavcnik (2005).

4 Estimations of equation (1) are shown in Table A.I. in the Appendix.

15
the variance of the estimators. To account for the fact that usually the noise in
the dependent variable differs across sectors and depends on the variance of
the estimated coefficients on industry dummies in eq. (1), we use weighted
least squares (WLS). Second, there could be problems arising from omitted
variables or selection. If there are omitted variables in equation (2) that vary
over time and are correlated simultaneously with industry wages and tariff
protection, our OLS and WLS estimations will be biased. Additionally, there
could be time-varying selection into industries if trade liberalization leads to
industries’ compositional effects based on unobserved workers’ characteristics.
This bias could make coefficients be either under or over estimated. We address
these problems using IV.
Specifications (1), (3) and (5) are regressions of the normalized industry
premium in tariffs with three controls: openness (imports/value-added),
productivity (value-added/number of employees) and MRER, as well as year
fixed effects. Specifications (2), (4) and (6) are the same but include industry
fixed effects. The inclusion of productivity as a control variable results of
importance because otherwise we could be estimating spurious correlation. If
more productive industries pay higher wages, and they receive the lowest
protection because government thinks that they can survive international
competition without it, we could falsely conclude that lower tariffs raise
industry wage premiums when the factor behind the higher wages is just
higher productivity.
It is remarkable the importance of accounting for unobserved time-invariant
industry characteristics, since doing it reverses the sign of the tariff coefficient
independently of the method used, making it positive (and statistically
significant in WLS and IV) for explaining industry wage premiums. The
reversion of the sign of the tariff coefficient is in line with the results of
Goldberg and Pavcnik (2005), who also find that including industry fixed
effects turns the coefficient from negative to positive in their study for
Colombia’s liberalization.

16
 Table III. Estimation results

OLS WLS IV
(1) (2) (3) (4) (5) (6)

Tariffs -0.008** 0.008 -0.008*** 0.008** -0.030 0.067*

(0.003) (0.006) (0.002) (0.004) (0.038) (0.039)
Openness -0.001 0.037 0.018*** 0.039 -0.021 -0.017
(0.012) (0.035) (0.007) (0.025) (0.024) (0.051)
Productivity 0.001*** 0.000 0.001*** 0.001* 0.000 0.001
(0.000) (0.000) (0.000) (0.000) (0.001) (0.001)
MRER 0.009 0.004 0.013* 0.006 0.026 -0.051
(0.013) (0.014) (0.007) (0.009) (0.035) (0.038)

Year fixed effects Yes Yes Yes Yes Yes Yes

Industry fixed effects No Yes No Yes No Yes
Observations 158 158 158 158 132 132
Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Normalized Industry Wage Premium is the
dependent variable. Specifications (5) and (6) have less observations because data of Chilean tariffs is not
available for 1996.

As is expected, when we go from OLS to WLS, we gain efficiency, because WLS

uses the inverse of the standard deviation of the wage premium coefficients
estimated in the first stage as weights in the second regression. On the other
hand, going from WLS to IV results in a larger coefficient of tariffs. This is in
line with the hypothesis that industries that experimented larger tariff cuts
have dismissed less able workers, resulting in a sample composed by more able
workers, making the tariff coefficient biased downwards. Another possible
reason could be that there are omitted variables that are correlated both with
tariffs and industry wage premiums.
Our preferred specification is (6) as it solves the potential problems of
endogeneity and selection. The magnitude of the effects is economically
significant. For example, a reduction of 50-percentage points in tariffs leads to
a 3.35% decline in the wage premium in this industry. This positive correlation
between industry wage premiums and tariffs is consistent with the existence of

17
 sector rents that are reduced by trade liberalization, or, alternatively, with the
predictions of some models of trade in which there exists labor stickiness across
sectors.

6 Conclusions

Like many other Latin American countries, Argentina went through a broad
liberalization program during the nineties. At the same time, wage inequality
rose largely over this period.
Using data from household surveys and tariffs for Argentina, we find that
worker affiliation is an important factor for explaining labor earnings, and that
trade protection creates specific industry rents. We also find that accounting
for unobservable time-invariant industry characteristics is crucial for
understanding this relationship. Once we control for industry fixed effects in
our regressions, we find that tariff protection has a positive and statistically
significant effect on industry wage premiums.
Combining the above results with Argentina’s tariff structure during this
period, which protected relatively more sectors that employed higher
proportions of skilled workers, it could be possible that nineties trade policy
has contributed to the rise in inequality documented in the literature.

References

Alvaredo, F., and L. Gasparini (2015). “Recent trends in inequality and

poverty in developing countries”. Handbook of Income Distribution, 2A,
697-805.

18
Attanasio, O., P. Goldberg and N. Pavcnik (2004). “Trade reforms and
wage inequality in Colombia”. Journal of Development Economics 74,
331– 366.

Galiani, S. and G. Porto (2010). "Trends in Tariff Reforms and in the

Structure of Wages". The Review of Economics and Statistics, MIT Press,
vol. 92(3), pages 482-494, August.

Galiani, S. and P. Sanguinetti (2003). "The impact of trade liberalization

on wage inequality: evidence from Argentina". Journal of Development
Economics, Elsevier, vol. 72(2), pages 497-513, December.

Goldberg, P. and N. Pavcnik (2004). “Trade, Inequality, and Poverty:

What Do We Know? Evidence from Recent Trade Liberalization Episodes
in Developing Countries”. NBER Working Paper no. 10593. Cambridge,
MA: National Bureau of Economic Research, June.

Goldberg, P. and N. Pavcnik (2005). “Trade Protection and Wages:

Evidence from the Colombian Trade Reforms”. Journal of International
Economics, 66 (1), 75-105.

Harrison, A. and G. Hanson (1999). “Who gains from trade reform? Some
remaining puzzles”. Journal of Development Economics 50:125-154.

Heckman J. and C. Pages (2000). “The Cost of Job Security Regulation:

Evidence from the Latin American Labor Market.” NBER Working Paper
7773.

Krueger, A. B., and L. Summers (1988). "Efficiency Wages and the Inter-
Industry Wage Structure," Econometrica 56, 259-293.

Levinsohn, J. (1993). “Testing the Imports-as-Market-Discipline

Hypothesis”. Journal of International Economics, 35(1-2), 1-22.

19
Nicita, A. and M. Olarreaga (2007). "Trade, Production, and Protection
Database, 1976-2004". World Bank Economic Review 21 (1), 165-71.

Pavcnik, N., A. Blom, P.K. Goldberg, and N. Schady (2004). “Trade Policy
and Industry Wage Structure: Evidence from Brazil:” World Bank
Economic Review, 18(3), 319-344.

20
Appendix

Table A.I. first stage estimations

1993 1994 1995 1996 1997 1998 1999

Formality dummy (=1 0.113*** 0.114*** 0.168*** 0.051 0.222*** 0.192*** 0.314***
if formal) (0.034) (0.033) (0.031) (0.033) (0.031) (0.027) (0.030)
0.057*** 0.039*** 0.068*** 0.060*** 0.043*** 0.052*** 0.040***
Age
(0.007) (0.007) (0.007) (0.008) (0.007) (0.007) (0.007)
-0.001*** -0.000*** -0.001*** -0.001*** -0.000*** -0.000*** -0.000***
Age^2
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Gender dummy (=1 if 0.106*** 0.164*** 0.117*** 0.297*** 0.197*** 0.255*** 0.173***
men) (0.036) (0.034) (0.033) (0.035) (0.036) (0.031) (0.033)
0.297*** 0.312*** 0.097 -0.034 -0.055 0.171*** 0.066
Beverages
(0.070) (0.068) (0.068) (0.082) (0.077) (0.062) (0.067)
0.467 0.472** 0.596*** 0.333 0.325 0.485**
Tobacco
(0.662) (0.208) (0.160) (0.281) (0.225) (0.205)
-0.025 -0.036 -0.048 -0.240*** -0.100 0.147* -0.140
Textiles
(0.070) (0.073) (0.075) (0.089) (0.071) (0.087) (0.101)
Apparel (except fur 0.058 0.118** -0.220*** 0.055 -0.152*** 0.156*** 0.030
apparel) (0.060) (0.058) (0.053) (0.057) (0.054) (0.047) (0.052)
Dressing and dyeing 0.497 0.002 -0.048 -0.382 0.304
of fur; articles of fur (0.514) (0.304) (0.203) (0.274) (0.629)
0.145** 0.094 -0.033 -0.137** -0.095 0.066 -0.051
Footwear
(0.070) (0.065) (0.067) (0.064) (0.063) (0.063) (0.069)
Wood production (non 0.188* 0.202 0.071 -0.055 -0.099 0.069 0.099
furniture) (0.103) (0.130) (0.156) (0.144) (0.123) (0.109) (0.093)
Paper and paper 0.185** 0.210** 0.172** -0.060 0.121 0.105 -0.058
products (0.078) (0.082) (0.079) (0.094) (0.081) (0.073) (0.090)
Printing and related 0.192** 0.314*** 0.012 0.073 0.100 0.182*** 0.067
activities (0.085) (0.074) (0.067) (0.080) (0.070) (0.057) (0.066)
Refine petroleum 0.570*** 0.532*** 0.438*** 0.513*** 0.628*** 0.337*** 0.275**
products (0.159) (0.193) (0.122) (0.195) (0.173) (0.091) (0.125)
0.284*** -0.251 0.232* 0.036 -0.006 0.362*** 0.473***
Basic chemicals
(0.108) (0.189) (0.131) (0.143) (0.158) (0.095) (0.141)
Other chemical 0.203*** 0.257*** 0.336*** 0.140** 0.266*** 0.337*** 0.230***
products (0.062) (0.059) (0.053) (0.055) (0.056) (0.048) (0.054)
0.079 0.067 -0.040 0.215** 0.229** 0.044 0.156
Rubber products
(0.083) (0.107) (0.094) (0.096) (0.104) (0.098) (0.110)
0.011 0.119* -0.007 -0.037 -0.039 0.132** 0.078
Plastics products
(0.073) (0.064) (0.064) (0.063) (0.066) (0.055) (0.061)
Glass and glass 0.539*** 0.326*** 0.149 0.302*** 0.301** 0.323** 0.321***
products (0.118) (0.109) (0.106) (0.113) (0.133) (0.137) (0.116)
Non-metallic mineral 0.339*** 0.330*** -0.120 -0.155* -0.279*** 0.025 0.063

21
products n.e.c. (0.085) (0.089) (0.086) (0.086) (0.079) (0.065) (0.075)
0.097 0.104 -0.221* -0.085 0.093 0.155* 0.287***
Basic iron and steel
(0.114) (0.098) (0.120) (0.128) (0.104) (0.093) (0.099)
Basic precious and 0.114 0.206 0.267 0.164 0.173 0.350* 0.246
non-ferrous metals (0.408) (1.331) (0.209) (0.202) (0.592) (0.182) (0.156)
0.132** 0.152*** -0.005 0.014 0.086 0.165*** -0.014
Other metal products
(0.065) (0.053) (0.054) (0.068) (0.061) (0.056) (0.058)
General purpose -0.020 0.140* 0.148** 0.207** 0.309*** 0.262*** 0.177**
machinery (0.080) (0.075) (0.075) (0.091) (0.072) (0.067) (0.074)
Domestic appliances 0.156 0.010 0.203* 0.197 0.018 0.262*** -0.008
n.e.c. (0.125) (0.090) (0.115) (0.158) (0.097) (0.083) (0.107)
Medical appliances 0.127 -0.547** 0.057 0.256** -0.315*** -0.125 0.230
and instruments of
precision (0.154) (0.214) (0.115) (0.113) (0.111) (0.155) (0.149)
0.085 0.490*** 0.438*** 0.405*** 0.302*** 0.321*** 0.351***
Motor vehicles
(0.092) (0.080) (0.094) (0.093) (0.070) (0.064) (0.075)
0.161** 0.075 0.124* 0.063 0.009 0.157*** 0.077
Furniture
(0.071) (0.063) (0.067) (0.084) (0.076) (0.058) (0.069)
0.214* 0.137 -0.211** -0.157* -0.068 -0.115 0.186
Other manufacturing
(0.118) (0.093) (0.106) (0.090) (0.150) (0.092) (0.127)

Observations 1,265 1,397 1,395 1,321 1,344 1,885 1,669

R-squared 0.409 0.404 0.538 0.452 0.451 0.469 0.424
Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Regressions include region
and level of education dummies.

22