WPS5711


Policy Research Working Paper                      5711




      Trade Liberalization, Firm Heterogneity,
                     and Wages
   New Evidence from Matched Employer-Employee Data

                              Pravin Krishna
                              Jennifer P. Poole
                             Mine Zeynep. Senses




The World Bank
Development Research Group
Trade and Integration Team
June 2011
Policy Research Working Paper 5711


  Abstract
  In this paper, the authors use a linked employer-                                 is random. This ignores the sorting of worker into firms
  employee database from Brazil to examine the impact                               and leads to a bias in estimates of the differential impact
  of trade reform on the wages of workers employed at                               of trade on workers at exporting firms relative to non-
  heterogeneous firms. The analysis of the data at the firm-                        exporting firms. Using detailed information on worker
  level confirms earlier findings of a differential positive                        and firm characteristics to control for compositional
  effect of trade liberalization on the average wages at                            effects and using firm-worker match specific effects to
  exporting firms relative to non-exporting firms. However,                         account for the endogenous mobility of workers, the
  this analysis of average firm-level wages is incomplete                           authors find the differential effect of trade openness on
  along several dimensions. First, it cannot fully account                          wages in exporting firms relative to domestic firms to be
  for the impact of a change in trade barriers on workforce                         insignificant. Consistent with the models of Helpman,
  composition especially in terms of unobservable (time-                            Itskhoki, and Redding (2010) and Davidson, Matusz and
  invariant) characteristics of workers (innate ability) and                        Schevchenko (2008), they also find that the workforce
  any additional productivity that obtains in the context of                        composition improves systematically in exporting firms
  employment in the specific firm (match specific ability).                         in terms of innate (time invariant) worker ability and in
  Furthermore, the firm-level analysis is undertaken under                          terms the quality of the worker-firm matches.
  the assumption that the assignment of workers to firms




  This paper is a product of the Trade and Integration Team, Development Research Group. It is part of a larger effort by
  the World Bank to provide open access to its research and make a contribution to development policy discussions around
  the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be
  contacted at Pravin_Krishna@jhu.edu.




         The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
         issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
         names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
         of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
         its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.


                                                       Produced by the Research Support Team
                                                                                                                           Trade	
  Liberalization,	
  Firm	
  Heterogeneity,	
  and	
  Wages:	
  	
  
                                                                                                                    New	
  Evidence	
  from	
  Matched	
  Employer–Employee	
  Data*	
  
                                                                                                                                                                                                                                       	
  
                                                                                                                                                                                                                          Pravin	
  Krishna	
  
                                                                                                                                                                                                              Johns	
  Hopkins	
  University	
  and	
  NBER	
  
                                                                                                                                                                                                                                       	
  
                                                                                                                                                                                                                         Jennifer	
  P.	
  Poole	
  
                                                                                                                                                                                                              University	
  of	
  California,	
  Santa	
  Cruz	
  
                                                                                                                                                                                                                                       	
  
                                                                                                                                                                                                                        Mine	
  Zeynep	
  Senses	
  
                                                                                                                                                                                                                   Johns	
  Hopkins	
  University	
  
                                                                                                                                                                                                                                       	
  
                                                                                                                                                                                                                                   June	
  2011	
  
                                                                                                                                                                                                                                   Abstract	
  	
  
In	
   this	
   paper,	
   we	
   use	
   a	
   linked	
   employer–employee	
   database	
   from	
   Brazil	
   to	
   examine	
   the	
  
impact	
   of	
   trade	
   reform	
   on	
   the	
   wages	
   of	
   workers	
   employed	
   at	
   heterogeneous	
   firms.	
   	
   Our	
  
analysis	
   of	
   the	
   data	
   at	
   the	
   firm-­‐level	
  confirms	
  earlier	
  findings	
   of	
  a	
  differential	
  positive	
  effect	
  
of	
   trade	
   liberalization	
   on	
   the	
   average	
   wages	
   at	
   exporting	
   firms	
   relative	
   to	
   non-­‐exporting	
  
firms.	
   However,	
   this	
   analysis	
   of	
   average	
   firm-­‐level	
   wages	
   is	
   incomplete	
   along	
   several	
  
dimensions.	
   First,	
   it	
   cannot	
   fully	
   account	
   for	
   the	
   impact	
   of	
   a	
   change	
   in	
   trade	
   barriers	
   on	
  
workforce	
  composition	
  especially	
  in	
  terms	
  of	
  unobservable	
  (time-­‐invariant)	
  characteristics	
  
of	
   workers	
   (innate	
   ability)	
   and	
   any	
   additional	
   productivity	
   that	
   obtains	
   in	
   the	
   context	
   of	
  
employment	
   in	
   the	
   specific	
   firm	
   (match	
   specific	
   ability).	
   	
   Furthermore,	
   the	
   firm-­‐level	
  
analysis	
   is	
   undertaken	
   under	
   the	
   assumption	
   that	
   the	
   assignment	
   of	
   workers	
   to	
   firms	
   is	
  
random.	
   This	
   ignores	
   the	
   sorting	
   of	
   worker	
   into	
   firms	
   and	
   leads	
   to	
   a	
   bias	
   in	
   estimates	
   of	
   the	
  
differential	
   impact	
   of	
   trade	
   on	
   workers	
   at	
   exporting	
   firms	
   relative	
   to	
   non-­‐exporting	
   firms.	
  
Using	
  detailed	
  information	
  on	
  worker	
  and	
  firm	
  characteristics	
  to	
  control	
  for	
  compositional	
  
effects	
   and	
   using	
   firm-­‐worker	
   match	
   specific	
   effects	
   to	
   account	
   for	
   the	
   endogenous	
   mobility	
  
of	
   workers,	
   we	
   find	
   the	
   differential	
   effect	
   of	
   trade	
   openness	
   on	
   wages	
   in	
   exporting	
   firms	
  
relative	
   to	
   domestic	
   firms	
   to	
   be	
   insignificant.	
   Consistent	
   with	
   the	
   models	
   of	
   Helpman,	
  
Itskhoki,	
  and	
  Redding	
  (2010)	
  and	
  Davidson,	
  Matusz	
  and	
  Schevchenko	
  (2008),	
  we	
  also	
  find	
  
that	
   the	
   workforce	
   composition	
   improves	
   systematically	
   in	
   exporting	
   firms	
   in	
   terms	
   of	
  
innate	
  (time	
  invariant)	
  worker	
  ability	
  and	
  in	
  terms	
  the	
  quality	
  of	
  the	
  worker-­‐firm	
  matches.	
  
	
  

Keywords:	
  Trade	
  liberalization,	
  firm	
  heterogeneity,	
  linked	
  employer-­‐employee	
  data.	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
JEL	
  Classification:	
  F16




	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
*	
   Our	
   special	
   thanks	
   to	
   Paulo	
   Furtado	
   de	
   Castro	
   for	
   help	
   with	
   the	
   RAIS	
   and	
   SECEX	
   data,	
   to	
   Marc-­‐

Andreas	
   Muendler	
   for	
   sharing	
   three-­‐firm	
   random	
   aggregates	
   from	
   the	
   PIA	
   data,	
   and	
   to	
   the	
  
Department	
   of	
   Economics	
   and	
   Academic	
   Computing	
   Services	
   at	
   the	
   University	
   of	
   California,	
   San	
  
Diego	
   for	
   assistance	
   with	
   data	
   access.	
   Pravin	
   Krishna	
   and	
   Jennifer	
   Poole	
   were	
   visiting	
   scholars	
   in	
   the	
  
International	
  Trade	
  Division	
  of	
  the	
  Development	
  Research	
  Group	
  of	
  the	
  World	
  Bank	
  while	
  work	
  on	
  
this	
  paper	
  was	
  conducted	
  and	
  acknowledge	
  the	
  funding	
  support	
  of	
  the	
  World	
  Bank	
  executed	
  Multi-­‐
Donor	
  Trust	
  Fund	
  on	
  Trade.	
  	
  



                                                                                                                                                                                                                                                                     1
	
  
1.	
                                      Introduction	
  
	
  
It	
   is	
   a	
   well-­‐established	
   empirical	
   regularity	
   in	
   the	
   international	
   economics	
   literature	
   that	
  
even	
  within	
  narrowly-­‐defined	
  industries,	
  globally-­‐engaged	
  firms	
  and	
  domestic	
  firms	
  differ	
  
in	
   terms	
   of	
   their	
   productivity,	
   size,	
   employment	
   composition,	
   and	
   wages	
   (Bernard	
   and	
  
Jensen	
   (1995,	
   1997)	
   and	
   Bernard,	
   Jensen,	
   and	
   Schott	
   (2009)).1	
   In	
   the	
   presence	
   of	
   such	
  
heterogeneity,	
   trade	
   liberalization	
   will	
   induce	
   inter-­‐firm	
   reallocations	
   within	
   an	
   industry	
   as	
  
the	
  more	
  productive,	
  exporting,	
  firms	
  expand	
  and	
  the	
  least	
  productive	
  firms	
  shrink	
  or	
  exit	
  
the	
   industry	
   (Melitz,	
   2003).	
   In	
   an	
   environment	
   with	
   labor	
   market	
   frictions,	
   this	
   within-­‐
industry	
  reallocation	
  can	
  have	
  distributional	
  consequences	
  for	
  workers	
  employed	
  in	
  firms	
  
with	
  differing	
  levels	
  of	
  global	
  engagement.	
  	
  
	
  
The	
  international	
  trade	
  literature	
  has	
  discussed	
  the	
  putative	
  impact	
  of	
  trade	
  liberalization	
  
on	
  wages	
  at	
  different	
  levels	
  of	
  aggregation,	
  examining	
  this	
  question	
  alternately	
  at	
  the	
  level	
  
of	
   the	
   firm	
   and	
   at	
   the	
   level	
   of	
   the	
   individual	
   worker.	
   	
   There	
   are	
   various	
   channels	
   through	
  
which	
  a	
  decline	
  in	
  trade	
  protection	
  could	
  result	
  in	
  differential	
  changes	
  in	
  average	
  wages	
  at	
  
exporting	
   firms	
   relative	
   to	
   firms	
   selling	
   only	
   to	
   the	
   domestic	
   market.	
   For	
   instance,	
   if	
  
liberalization	
   is	
   associated	
   with	
   a	
   change	
   in	
   relative	
   returns	
   to	
   worker	
   characteristics,2	
  
differences	
   in	
   work	
   force	
   composition	
   will	
   imply	
   that	
   average	
   firm-­‐level	
   wages	
   in	
   exporting	
  
firms	
   will	
   change	
   differentially	
   relative	
   to	
   domestic	
   firms	
   following	
   liberalization.	
   If	
   trade	
  
liberalization	
  does	
  not	
  affect	
  the	
  returns	
  to	
  worker	
  characteristics,	
  a	
  differential	
  wage	
  effect	
  
may,	
  nevertheless,	
  be	
  observed	
  if	
  changes	
  in	
  trade	
  policy	
  induce	
  compositional	
  changes	
  in	
  
the	
  workforce	
  of	
  exporting	
  firms	
  that	
  are	
  different	
  from	
  those	
  in	
  non-­‐exporting	
  firms.3	
  	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
1	
   Recent	
   work	
   extends	
   the	
   findings	
   of	
   Bernard	
   and	
   Jensen	
   (1995)	
   using	
   matched	
   employer-­‐employee	
  

data	
   for	
   Germany	
   (Schank,	
   Schnabel,	
   and	
   Wagner	
   (2007)	
   and	
   Klein,	
   Moser,	
   and	
   Urban	
   (2010)),	
  
Denmark	
   (Munch	
   and	
   Skaksen	
   2008),	
   and	
   Mexico	
   (Frías,	
   Kaplan,	
   and	
   Verhoogen	
   2009).	
   See	
   also	
  
Hummels,	
  Jorgensen,	
  Munch	
  and	
  Xiang	
  (2010)	
  for	
  analysis	
  of	
  the	
  impact	
  of	
  outsourcing	
  on	
  wages	
  and	
  
employment	
  using	
  matched	
  employer-­‐employee	
  data	
  from	
  Denmark.	
  
2	
   A	
   vast	
   empirical	
   literature	
   has	
   examined	
   the	
   effects	
   of	
   globalization	
   on	
   the	
   wage	
   outcomes	
   of	
  

workers	
   in	
   the	
   domestic	
   economy	
   with	
   a	
   particular	
   focus	
   on	
   the	
   important	
   question	
   of	
   how	
   trade	
  
affects	
   the	
   average	
   wages	
   of	
   workers	
   with	
   different	
   levels	
   of	
   skill.	
   Classic	
   papers	
   in	
   this	
   literature	
  
include	
  Lawrence	
  and	
  Slaughter	
  (1993),	
  Leamer	
  (1996),	
  and	
  Feenstra	
  and	
  Hanson	
  (1999).	
  Revenga	
  
(1997),	
  Currie	
  and	
  Harrison	
  (1997),	
  and	
  Trefler	
  (2004)	
  analyze	
  the	
  effect	
  of	
  trade	
  liberalization	
  on	
  
firm-­‐level	
   wages	
   in	
   Mexico,	
   Morocco,	
   and	
   Canada,	
   respectively,	
   with	
   mixed	
   results.	
   Feenstra	
   and	
  
Hanson	
  (2002)	
  and	
  Goldberg	
  and	
  Pavcnik	
  (2007)	
  provide	
  excellent	
  survey	
  treatments.	
  
3	
  An	
  expansion	
  in	
  exports	
  could	
  result	
  in	
  differential	
  labor	
  quality	
  upgrading	
  in	
  exporting	
  firms,	
  for	
  

example,	
  by	
  inducing	
  these	
  firms	
  to	
  adopt	
  technologies	
  favoring	
  highly	
  skilled	
  workers,	
  as	
  in	
  Yeaple	
  
(2005)	
  and	
  Bustos	
  (2011),	
  or	
  by	
  inducing	
  these	
  firms	
  to	
  upgrade	
  product	
  and	
  hence	
  labor	
  quality	
  as	
  
in	
  Verhoogen	
  (2008)	
  and	
  Kugler	
  and	
  Verhoogen	
  (2008).	
  



                                                                                                                                                                                                                                   2
	
  
	
  
Furthermore,	
   in	
   the	
   presence	
   of	
   labor	
   market	
   frictions,	
   changes	
   in	
   trade	
   protection	
   may	
  
result	
   in	
   differential	
   changes,	
   in	
   exporting	
   firms	
   relative	
   to	
   non-­‐exporting	
   firms,	
   in	
   the	
  
wages	
   of	
   ex-­‐ante	
   identical	
   workers.	
   For	
   instance,	
   if	
   firms	
   engage	
   in	
   some	
   form	
   of	
   rent	
  
sharing	
   with	
   their	
   workers,	
   the	
   wages	
   of	
   workers	
   employed	
   in	
   exporting	
   firms,	
   which	
  
experience	
   a	
   relative	
   improvement	
   in	
   their	
   profits	
   or	
   market	
   share	
   after	
   a	
   decline	
   in	
  
protection,	
  will	
  correspondingly	
  be	
  better	
  compared	
  to	
  workers	
  employed	
  in	
  firms	
  serving	
  
only	
   the	
   domestic	
   market	
   (Egger	
   and	
   Kreickemeier	
   (2009)	
   and	
   Amiti	
   and	
   Davis	
  
(forthcoming)).	
   	
   Equally,	
   if	
   the	
   labor	
   allocation	
   process	
   is	
   subject	
   to	
   search	
   and	
   matching	
  
frictions,	
  exporting	
  firms	
  may	
  screen	
  (ex-­‐ante	
  identical)	
  workers	
  more	
  intensively,	
  employ	
  
workers	
   of	
   higher	
   match-­‐specific-­‐ability,	
   and	
   pay	
   higher	
   wages	
   relative	
   to	
   non-­‐exporting	
  
firms	
   as	
   Helpman,	
   Itskhoki,	
   and	
   Redding	
   (2010)	
   have	
   shown	
   in	
   their	
   comprehensive	
  
theoretical	
   analysis	
   of	
   the	
   links	
   between	
   trade	
   and	
   labor	
   markets	
   in	
   this	
   context.	
   Here,	
  
opening	
   the	
   economy	
   to	
   trade	
   is	
   predicted	
   to	
   alter	
   this	
   wage	
   gap	
   between	
   exporting	
   and	
  
non-­‐exporting	
  firms	
  and	
  to	
  affect	
  wage	
  inequality.4	
  	
  
                                                      	
  
Does	
   trade	
   liberalization,	
   in	
   fact,	
   affect	
   differently	
   workers	
   employed	
   in	
   exporting	
   firms	
  
relative	
  to	
  non-­‐exporting	
  firms?	
  We	
  explore	
  this	
  issue	
  empirically	
  using	
  a	
  detailed	
  dataset	
  
with	
  matched	
  employer-­‐employee	
  data	
  on	
  workers	
  in	
  Brazil	
  for	
  the	
  years	
  1990-­‐1998.	
  The	
  
dataset	
   traces	
   individually	
   identifiable	
   workers	
   across	
   employers	
   over	
   time	
   and	
   contains	
  
detailed	
  information	
  on	
  worker	
  characteristics	
  such	
  as	
  age,	
  gender,	
  education,	
  occupation,	
  
and	
   tenure	
   at	
   the	
   firm	
   which	
   allows	
   us	
   to	
   suitably	
   account	
   for	
   the	
   role	
   of	
   unobservable	
  
worker	
   and	
   firm	
   characteristics	
   in	
   determining	
   wages.	
   We	
   complement	
   this	
   worker-­‐level	
  
information	
  with	
  firm-­‐level	
  data	
  on	
  exporter	
  status	
  from	
  the	
  Brazilian	
  Customs	
  Office,	
  and	
  
industry-­‐level	
   information	
   on	
   trade	
   protection	
   levels	
   to	
   capture	
   Brazil’s	
   main	
   trade	
   policy	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
	
  
4	
   More	
   generally,	
   the	
   various	
   channels	
   linking	
   trade	
   protection	
   to	
   wages	
   that	
   we	
   describe	
   are	
   each	
  

potentially	
   associated	
   with	
   changes	
   in	
   within-­‐group	
   inequality	
   (inequality	
   between	
   workers	
   with	
  
identical	
   observable	
   characteristics),	
   in	
   changes	
   in	
   between-­‐group	
   inequality	
   (inequality	
   between	
  
workers	
  with	
  different	
  characteristics,	
  such	
  as,	
  levels	
  of	
  education)	
  and	
  possibly	
  even	
  no	
  change	
  in	
  
inequality	
   at	
   all.	
   	
   For	
   instance,	
   under	
   the	
   rent-­‐sharing	
   mechanism,	
   within-­‐group	
   inequality	
   will	
  
change	
   following	
   trade	
   liberalization	
   if	
   otherwise	
   identical	
   workers	
   are	
   employed	
   at	
   firms	
   which	
  
experience	
   differential	
   profit	
   changes.	
   Alternatively,	
   if	
   trade	
   liberalization	
   changes	
   the	
   returns	
   to	
  
worker	
   characteristics,	
   this	
   will	
   result	
   in	
   a	
   change	
   in	
   between-­‐group	
   inequality.	
   Finally,	
   if	
   workers	
  
are	
  paid	
  competitive	
  wages	
  and	
  the	
  only	
  impact	
  of	
  trade	
  liberalization	
  is	
  to	
  reallocate	
  workers	
  across	
  
firms	
  without	
  any	
  changes	
  in	
  the	
  returns	
  to	
  worker	
  characteristics,	
  there	
  will	
  be	
  no	
  change	
  in	
  either	
  
within-­‐group	
  or	
  between-­‐group	
  wage	
  inequality.	
  



                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                       3
	
  
reforms.	
  	
  
	
  
We	
   begin	
   our	
   analysis	
   by	
   exploring	
   the	
   behavior	
   of	
   average	
   firm-­‐level	
   wages	
   and	
   find	
   that	
  a	
  
decline	
   in	
   trade	
   protection	
   is	
   associated	
   with	
   an	
   increase	
   in	
   average	
   wages	
   in	
   exporting	
  
firms	
   relative	
   to	
   domestic	
   firms,	
   consistent	
   with	
   the	
   findings	
   of	
   Amiti	
   and	
   Davis	
  
(forthcoming)	
   for	
   the	
   liberalization	
   period	
   in	
   Indonesia5.	
   	
   However,	
   we	
   argue	
   that	
   the	
  
analysis	
   of	
   average	
   firm-­‐level	
   wages,	
   although	
   informative,	
   is	
   incomplete	
   along	
   several	
  
dimensions.	
   First,	
   it	
   cannot	
   fully	
   account	
   for	
   the	
   impact	
   of	
   a	
   change	
   in	
   trade	
   barriers	
   on	
  
workforce	
   composition	
   in	
   terms	
   of	
   both	
   observable	
   worker	
   characteristics	
   (not	
   available	
   in	
  
firm-­‐level	
  datasets),	
  as	
  well	
  as	
  factors	
  that	
  are	
  observable	
  to	
  the	
  managers	
  of	
  the	
  firm	
  and	
  
hence	
   impact	
   wages	
   but	
   are	
   unobservable	
   in	
   the	
   data,	
   such	
   as	
   the	
   innate	
   ability	
   of	
   the	
  
worker	
  and	
  any	
  additional	
  productivity	
  (ability)	
  that	
  obtains	
  in	
  the	
  context	
  of	
  employment	
  
in	
   the	
   specific	
   firm	
   (match	
   specific	
   ability).	
   	
   Furthermore,	
   the	
   firm-­‐level	
   analysis	
   is	
  
undertaken	
   under	
   the	
   assumption	
   that	
   the	
   assignment	
   of	
   workers	
   to	
   firms	
   is	
   random	
   and	
  
ignores	
   the	
   sorting	
   of	
   worker	
   into	
   firms	
   and	
   the	
   resulting	
   change	
   in	
   the	
   distribution	
   of	
  
match-­‐specific	
   ability	
   in	
   different	
   firms.	
   Using	
   matched	
   employer-­‐employee	
   data,	
   we	
   test	
  
whether	
   wage	
   behavior	
   at	
   the	
   worker	
   level	
   confirms	
   the	
   maintained	
   assumption	
   of	
  
exogenous	
  worker	
  mobility,	
  as	
  in	
  Abowd,	
  McKinney	
  and	
  Schmutte	
  (2010).	
  Consistent	
  with	
  
theoretical	
   models	
   emphasizing	
   search	
   dynamics	
   or	
   other	
   frictions	
   as	
   important	
  
determinants	
  of	
  the	
  relationship	
  between	
  trade	
  liberalization	
  and	
  wages	
  such	
  as	
  Helpman,	
  
Itskhoki,	
  and	
  Redding	
  (2010)	
  and	
  Davidson,	
  Matusz	
  and	
  Schevchenko	
  (2008),	
  we	
  find	
  that	
  
the	
  data	
  decisively	
  reject	
  the	
  assumption	
  of	
  exogenous	
  job	
  mobility.	
  
                                                      	
  
Importantly,	
   using	
   detailed	
   information	
   on	
   worker	
   and	
   firm	
   characteristics	
   to	
   control	
   for	
  
compositional	
   effects	
   and	
   using	
   firm-­‐worker	
   match	
   specific	
   effects	
   to	
   account	
   for	
   the	
  
endogenous	
  mobility	
  of	
  workers,	
  we	
  find	
  the	
  differential	
  effect	
  of	
  trade	
  openness	
  on	
  wages	
  
	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
  To	
   our	
   knowledge	
   Amiti	
   and	
   Davis	
   (forthcoming)	
   is	
   the	
   first	
   paper	
   to	
   incorporate	
   firm-­‐level	
  
5	
  

heterogeneity	
  in	
  empirically	
  studying	
  the	
  impact	
  of	
  trade	
  liberalization	
  on	
  wages.	
  They	
  introduce	
  a	
  
general	
   equilibrium	
   model,	
   which	
   combines	
   firm	
   heterogeneity	
   in	
   exporting	
   patterns,	
   trade	
   in	
  
intermediate	
  inputs,	
  and	
  firm-­‐specific	
  wages.	
  The	
  latter	
  is	
  incorporated	
  into	
  the	
  model	
  by	
  assuming	
  a	
  
fair-­‐wage	
   specification	
   that	
   results	
   in	
   a	
   direct	
   link	
   between	
   firm	
   wages	
   and	
   firm	
   profitability.	
   The	
  
model	
   predicts	
   that	
   a	
   decline	
   in	
   final	
   goods	
   tariffs	
   reduces	
   the	
   wages	
   of	
   workers	
   at	
   firms	
   that	
   sell	
  
only	
  in	
  the	
  domestic	
  market,	
  but	
  raises	
  the	
  wages	
  of	
  workers	
  at	
  firms	
  that	
  export.	
  Consistent	
  with	
  the	
  
model,	
   they	
   find	
   differential	
   changes	
   in	
   average	
   firm-­‐level	
   wages	
   across	
   Indonesian	
   firms	
   with	
  
different	
   levels	
   of	
   global	
   engagement	
   following	
   liberalization,	
   which	
   they	
   attribute	
   wholly	
   to	
   firm	
  
performance.	
  
	
  


                                                                                                                                                                                                                                   4
	
  
in	
   exporting	
   firms	
   relative	
   to	
   domestic	
   firms	
   to	
   be	
   insignificant.	
   Thus,	
   our	
   findings	
   using	
  
matched	
   employer	
   employee	
   data	
   suggest	
   a	
   quite	
   different	
   picture	
   of	
   the	
   links	
   between	
  
trade	
   liberalization	
   and	
   wages	
   than	
   is	
   obtained	
   by	
   analyzing	
   the	
   data	
   at	
   a	
   more	
   aggregate	
  
(firm)	
   level.	
   Consistent	
   with	
   the	
   models	
   of	
   Helpman,	
   Itskhoki,	
   and	
   Redding	
   (2010)	
   and	
  
Davidson,	
   Matusz	
   and	
   Schevchenko	
   (2008),	
   we	
   find	
   that	
   the	
   workforce	
   composition	
  
improves	
   systematically	
   in	
   exporting	
   firms	
   in	
   terms	
   of	
   innate	
   (time	
   invariant)	
   ability	
   and	
   in	
  
terms	
   the	
   quality	
   of	
   the	
   worker-­‐firm	
   matches.	
   	
   This	
   finding	
   also	
   serves	
   to	
   explain	
   the	
  
difference	
   between	
   the	
   results	
   at	
   the	
   firm	
   level	
   and	
   those	
   at	
   the	
   worker	
   level.	
   If	
   average	
  
worker	
  ability	
  improves	
  systematically	
  in	
  exporting	
  firms	
  following	
  trade	
  liberalization,	
  and	
  
this	
   change	
   in	
   not	
   taken	
   into	
   account,	
   it	
   will	
   appear	
   that	
   trade	
   liberalization	
   leads	
   to	
   a	
  
differential	
   wage	
   improvement	
   for	
   workers	
   at	
   exporting	
   firms	
   even	
   when	
   this	
   is	
   not	
   the	
  
case.	
  	
  
                  	
  
The	
  remainder	
  of	
  this	
  paper	
  is	
  organized	
  as	
  follows.	
  In	
  Section	
  2,	
  we	
  present	
  a	
  background	
  
discussion	
  on	
  Brazil’s	
  trade	
  policy	
  reforms	
  and	
  describe	
  the	
  data.	
  We	
  present	
  the	
  empirical	
  
methodology	
  and	
  estimation	
  results	
  for	
  the	
  aggregate	
  (firm-­‐level)	
  analysis	
  in	
  Section	
  3	
  and	
  
for	
  the	
  analysis	
  at	
  the	
  worker	
  level	
  in	
  Section	
  4.	
  Section	
  5	
  concludes.	
  
	
  
	
  
2.	
            Data	
  and	
  Policy	
  Background	
  
	
  
Our	
   main	
   data	
   are	
   administrative	
   records	
   from	
   Brazil	
   for	
   formal-­‐sector	
   workers	
   linked	
   to	
  
their	
   employers.	
   We	
   combine	
   this	
   worker-­‐level	
   information	
   with	
   complementary	
   data	
  
sources	
   on	
   firm-­‐level	
   exporter	
   status	
   and	
   information	
   on	
   industry-­‐level	
   trade	
   protection	
  
during	
  Brazil’s	
  main	
  trade	
  policy	
  reform	
  period.	
  
	
  
2.1.	
   Brazil’s	
  Policy	
  Reforms	
  
	
  
The	
   1990s	
   were	
   a	
   period	
   of	
   dramatic	
   policy	
   reform	
   in	
   Brazil,	
   providing	
   a	
   particularly	
  
appropriate	
   setting	
   in	
   which	
   to	
   study	
   the	
   impact	
   of	
   trade	
   liberalization	
   on	
   wages.	
   As	
  
compared	
   to	
   the	
   gradual	
   process	
   of	
   globalization	
   in	
   many	
   developed	
   countries,	
   Brazil’s	
  
trade	
   reform	
   occurred	
   over	
   a	
   relatively	
   short	
   period	
   of	
   time,	
   and	
   with	
   substantial	
   cross-­‐
industry	
   variation.	
   Furthermore,	
   many	
   of	
   the	
   policy	
   reforms	
   were	
   arguably	
   unanticipated	
  




                                                                                                                                                             5
	
  
and	
  could	
  be	
  viewed	
  as	
  exogenous	
  to	
  changes	
  in	
  wages	
  at	
  the	
  firm	
  and	
  worker	
  level.6	
  
	
                                        	
  
The	
   second	
   half	
   of	
   the	
   20th	
   century	
   in	
   Brazil	
   was	
   characterized	
   by	
   tight	
   import	
   substitution	
  
industrialization	
   policies	
   designed	
   to	
   protect	
   the	
   domestic	
   manufacturing	
   sector	
   from	
  
foreign	
   competition.	
   Special	
   import	
   regimes	
   and	
   discretionary	
   import	
   controls	
   like	
   the	
   “law	
  
of	
   similars”,	
   under	
   which	
   goods	
   were	
   banned	
   if	
   they	
   too	
   closely	
   resembled	
   a	
   Brazilian	
  
product,	
  were	
  commonplace.	
  Coverage	
  of	
  these	
  quantitative	
  restrictions	
  remained	
  close	
  to	
  
100	
  percent	
  throughout	
  this	
  period,	
  leaving	
  Brazilian	
  manufacturers	
  highly	
  protected.	
  
	
                                        	
  
The	
  1990s,	
  however,	
  witnessed	
  sweeping	
  changes	
  in	
  Brazilian	
  trade	
  policy.	
  Beginning	
  with	
  
the	
   Collor	
   de	
   Melo	
   administration	
   and	
   continuing	
   with	
   President	
   Cardoso,	
   Brazil	
   began	
  
extensive	
  policies	
  of	
  trade	
  liberalization	
  in	
  1988,	
  which	
  paved	
  the	
  way	
  for	
  the	
  multilateral	
  
free	
   trade	
   area,	
   Mercosul,	
   with	
   its	
   Southern	
   Cone	
   neighbors	
   (Argentina,	
   Paraguay,	
   and	
  
Uruguay)	
   in	
   1991.	
   Average	
   ad	
   valorem	
   final	
   goods	
   tariff	
   rates	
   fell	
   from	
   41%	
   to	
   18%	
  
between	
   1988	
   and	
   1989.7	
   The	
   federal	
   government	
   abolished	
   all	
   remaining	
   non-­‐tariff	
  
barriers	
   inherited	
   from	
   the	
   import	
   substitution	
   era	
   and	
   brought	
   nominal	
   tariffs	
   further	
  
down	
   in	
   1990.	
   Effective	
   rates	
   of	
   protection	
   fell	
   by	
   over	
   70%	
   in	
   just	
   four	
   years—from	
  
approximately	
   42%,	
   on	
   average,	
   in	
   1988	
   to	
   12%,	
   on	
   average,	
   in	
   1994	
   (Kume,	
   Piani,	
   and	
  
Souza	
  2003).	
  	
  
	
                                        	
  
After	
   decades	
   of	
   high	
   inflation	
   and	
   several	
   unsuccessful	
   stabilization	
   attempts,	
   the	
   Brazilian	
  
government	
  succeeded	
  with	
  its	
  macroeconomic	
  stabilization	
  plan	
  (Plano	
  Real)	
  in	
  1994	
  and	
  
lastingly	
  ended	
  hyperinflation.	
  The	
  new	
  currency,	
  the	
  real,	
  began	
  officially	
  at	
  parity	
  with	
  the	
  
U.S.	
   dollar	
   on	
   July	
   1,	
   1994,	
   trading	
   freely	
   on	
   international	
   markets	
   and	
   appreciating	
   in	
   its	
  
first	
   months.	
   In	
   response,	
   the	
   government	
   partially	
   reversed	
   trade	
   reforms	
   in	
   1995	
   after	
  
manufacturing	
  industries	
  lost	
  competitiveness	
  due	
  to	
  the	
  real’s	
  appreciation;8	
  the	
  effective	
  
rate	
  of	
  protection	
  climbed	
  slightly	
  in	
  subsequent	
  years	
  from	
  an	
  average	
  of	
  12%	
  in	
  1994	
  to	
  
	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
6	
   The	
   same	
   argument	
   could	
   not	
   be	
   made	
   if	
   our	
   analysis	
   focused	
   on	
   industry-­‐level	
   wages,	
   given	
   the	
  

substantial	
   evidence	
   for	
   political	
   economy	
   factors	
   influencing	
   trade	
   policy	
   (see,	
   for	
   example,	
  
Grossman	
  and	
  Helpman	
  (1994)	
  and	
  Olarreaga	
  and	
  Soloaga	
  (1998)	
  for	
  evidence	
  from	
  Brazil’s	
  customs	
  
union,	
  Mercosul).	
  
7	
   These	
   reforms	
   had	
   little	
   impact	
   on	
   import	
   competition	
   however,	
   as	
   non-­‐tariff	
   barriers	
   remained	
  

highly	
  restrictive.	
  
8	
  Prior	
  to	
  1994	
  and	
  the	
  implementation	
  of	
  the	
  new	
  currency,	
  controls	
  on	
  Brazil’s	
  former	
  currency,	
  the	
  

cruzado,	
  had	
  served	
  as	
  yet	
  another	
  form	
  of	
  implicit	
  import	
  protection.	
  In	
  our	
  empirical	
  analysis,	
  we	
  
allow	
   for	
   any	
   differential	
   impacts	
   Brazil’s	
   exchange	
   rate	
   may	
   have	
   on	
   firms	
   with	
   differing	
   trade	
  
exposure.	
  	
  



                                                                                                                                                                                                                                   6
	
  
an	
  average	
  of	
  18%	
  in	
  1998.9	
  
	
  
2.2.	
   Worker	
  data	
  
	
  
The	
  linked	
  employer–employee	
  data	
  are	
  from	
  the	
  Brazilian	
  Labor	
  Ministry,	
  which	
  requires	
  
by	
   law	
   that	
   all	
   formally-­‐registered	
   firms	
   report	
   to	
   the	
   ministry	
   on	
   all	
   workers	
   in	
   every	
  
year.10	
   These	
   administrative	
   records	
   have	
   been	
   collected	
   in	
   the	
   Relação	
   Anual	
   de	
  
Informações	
   Sociais	
   (RAIS)	
   database	
   since	
   1986.	
   In	
   this	
   paper,	
   we	
   use	
   information	
   from	
  
RAIS	
   for	
   the	
   years	
   1990	
   through	
   1998,	
   when	
   we	
   also	
   have	
   complementary	
   data	
   on	
   the	
  
export	
  status	
  of	
  firms	
  and	
  industry-­‐level	
  tariff	
  rates.	
  
	
                                        	
  
The	
   main	
   benefit	
   of	
   the	
   RAIS	
   database	
   is	
   the	
   ability	
   to	
   trace	
   individually-­‐identifiable	
  
workers	
   over	
   time	
   and	
   across	
   firms.	
   The	
   data	
   include	
   a	
   unique	
   worker-­‐identification	
  
number	
  (which	
  remains	
  with	
  the	
  worker	
  throughout	
  his	
  work	
  history),	
  the	
  tax	
  number	
  of	
  
the	
  worker’s	
  firm,	
  the	
  industrial	
  classification	
  of	
  the	
  worker’s	
  firm11,	
  and	
  the	
  municipality	
  of	
  
the	
   worker’s	
   firm.	
   These	
   data	
   are	
   particularly	
   valuable	
   as	
   they	
   offer	
   variables	
   beyond	
   the	
  
available	
  information	
  in	
  many	
  other	
  firm-­‐level	
  databases,	
  often	
  used	
  in	
  studies	
  like	
  ours.	
  In	
  
particular,	
   the	
   data	
   contain	
   detailed	
   information	
   on	
   workers’	
   skill-­‐levels	
   (as	
   defined	
   by	
  
occupation12	
   and	
   education),	
   which	
   are	
   paramount	
   to	
   this	
   analysis.	
   Other	
   variables	
   of	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
9	
   Trade	
   policy	
   reforms	
   coincided	
   with	
   gradual	
   foreign	
   investment	
   liberalizations	
   and	
   the	
  

privatization	
   of	
   state-­‐owned	
   companies,	
   both	
   of	
   which	
   contributed	
   to	
   attracting	
   substantial	
   capital	
  
inflows	
   over	
   this	
   time	
   period.	
   Meanwhile,	
   the	
   government’s	
   regional	
   development	
   plans	
   also	
  
included	
  export	
  promotion	
  policies	
  as	
  explicit	
  elements,	
  helping	
  to	
  boost	
  exports	
  beginning	
  in	
  1995.	
  
In	
   each	
   specification,	
   we	
   include	
   region-­‐specific	
   year	
   dummies	
   to	
   capture	
   the	
   impact	
   of	
   these	
   and	
  
other	
  general	
  macroeconomic	
  trends	
  on	
  wages.	
  
10	
   The	
   process	
   for	
   firms	
   to	
   report	
   on	
   their	
   workers	
   is	
   extensive	
   and	
   costly.	
   However,	
   this	
   information	
  

is	
  used	
  to	
  administer	
  payments	
  of	
  the	
  annual	
  public	
  wage	
  supplements	
  to	
  every	
  formally-­‐employed	
  
worker,	
  thus	
  creating	
  a	
  strong	
  incentive	
  for	
  workers	
  to	
  urge	
  their	
  employers	
  to	
  report	
  accurately.	
  In	
  
practice,	
   however,	
   only	
   formally-­‐employed	
   workers	
   will	
   be	
   properly	
   recorded.	
   Goldberg	
   and	
   Pavcnik	
  
(2003)	
   estimate	
   informal	
   workers	
   represent	
   approximately	
   16%	
   of	
   Brazil’s	
   manufacturing	
   labor	
  
force	
   over	
   our	
   sample	
   period.	
   The	
   literature	
   on	
   the	
   impact	
   of	
   trade	
   reform	
   on	
   the	
   informal	
   sector	
  
offers	
  mixed	
  results	
  (see	
  Goldberg	
  and	
  Pavcnik	
  (2003),	
  Menezes-­‐Filho	
  and	
  Muendler	
  (2007),	
  and	
  Paz	
  
(2009)).	
  
11	
  The	
  sector	
  classification	
  used	
  in	
  this	
  paper	
  comes	
  from	
  Brazil’s	
  statistical	
  office	
  (Instituto	
  Brasileiro	
  

de	
   Geografia	
   e	
   Estatística,	
   IBGE)	
   and	
   is	
   roughly	
   comparable	
   to	
   a	
   3-­‐digit	
   NAICS	
   classification	
  
(Muendler	
  2002).	
  
12	
   Muendler,	
   Poole,	
   Ramey	
   and	
   Wajnberg	
   (2004)	
   map	
   the	
   Brazilian	
   classification	
   of	
   occupations	
  

(Classificação	
   Brasileira	
   de	
   Ocupações	
   (CBO))	
   to	
   the	
   International	
   Standard	
   Classification	
   of	
  
Occupations	
   (ISCO).	
   The	
   CBO	
   is	
   a	
   detailed,	
   task-­‐oriented	
   classification	
   system,	
   while	
   ISCO	
   reflects	
   a	
  
less-­‐detailed	
   and	
   more	
   skill-­‐oriented	
   classification	
   system.	
   The	
   skill	
   classification	
   is	
   intended	
   to	
  
incorporate	
   on-­‐the-­‐job	
   experience,	
   informal	
   training,	
   and	
   the	
   technological	
   skill	
   content	
   of	
   the	
  



                                                                                                                                                                                                                                   7
	
  
interest	
  are	
  the	
  worker’s	
  annual	
  real	
  wages	
  in	
  Brazilian	
  reais13,	
  tenure	
  at	
  the	
  firm	
  in	
  months,	
  
gender,	
  and	
  age.	
  
	
                                        	
  
We	
  restrict	
  observations	
  as	
  follows.	
  First,	
  we	
  draw	
  a	
  one	
  percent	
  random	
  sample	
  from	
  the	
  
complete	
   list	
   of	
   workers	
   ever	
   to	
   appear	
   in	
   the	
   national	
   records.	
   We	
   match	
   the	
   sampled	
  
workers	
   back	
   to	
   the	
   population	
   data	
   to	
   find	
   all	
   firms	
   in	
   which	
   these	
   workers	
   were	
   ever	
  
employed	
   over	
   time	
   to	
   create	
   a	
   complete	
   employment	
   history	
   of	
   this	
   one	
   percent	
   random	
  
sample	
   of	
   the	
   population	
   of	
   the	
   Brazilian	
   formal-­‐sector	
   labor	
   force.14	
   Next,	
   we	
   keep	
   only	
  
workers	
  with	
  valid	
  worker	
  identification	
  numbers	
  to	
  ensure	
  that	
  we	
  can	
  track	
  individuals	
  
over	
  time.	
  As	
  is	
  standard	
  in	
  the	
  literature,	
  we	
  include	
  only	
  prime-­‐age	
  workers	
  between	
  the	
  
ages	
   of	
   15	
   and	
   64	
   years,	
   workers	
   with	
   a	
   positive	
   average	
   monthly	
   wage,	
   and	
   workers	
   in	
  
private-­‐sector	
   jobs.	
   Finally,	
   for	
   workers	
   with	
   multiple	
   jobs	
   in	
   a	
   given	
   year,	
   only	
   the	
   most	
  
recent	
  job	
  is	
  included	
  in	
  the	
  sample.	
  If	
  a	
  worker	
  has	
  multiple	
  current	
  jobs,	
  only	
  the	
  highest	
  
paying	
  job	
  is	
  included.	
  
	
  
2.3.	
   Complementary	
  data	
  
	
  
Export	
  Status	
                                                                                                             Brazilian	
   firms’	
   tax	
   identification	
   numbers	
   are	
   common	
   across	
   many	
  
databases,	
   allowing	
   us	
   to	
   match	
   the	
   RAIS	
   data	
   to	
   complementary	
   firm-­‐level	
   data	
   sources.	
  
Information	
   on	
   firm-­‐level	
   export	
   transactions	
   is	
   available	
   from	
   the	
   Brazilian	
   Customs	
   Office	
  
(Secretaria	
  de	
  Comércio	
  Exterior,	
  SECEX).	
  SECEX	
  records	
  all	
  legally-­‐registered	
  firms	
  in	
  Brazil	
  
with	
  at	
  least	
  one	
  export	
  transaction	
  in	
  a	
  given	
  year.	
  We	
  match	
  the	
  SECEX	
  firm-­‐level	
  exporter	
  
status	
   data	
   to	
   our	
   RAIS	
   worker	
   data	
   by	
   the	
   firm’s	
   tax	
   identification	
   code	
   to	
   identify	
   workers	
  
at	
   exporting	
   and	
   non-­‐exporting	
   firms.	
   We	
   define	
   an	
   indicator	
   variable	
   equal	
   to	
   one	
   if	
   a	
  
worker	
  holds	
  a	
  job	
  at	
  a	
  firm	
  with	
  a	
  positive	
  dollar	
  value	
  of	
  free-­‐on-­‐board	
  exports	
  in	
  a	
  given	
  
year	
  and	
  zero	
  otherwise.	
  	
  
	
                                        	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
occupation	
   (Elias	
   and	
   Birch	
   1994).	
   ISCO	
   occupations	
   can	
   be	
   grouped	
   into	
   four	
   broad	
   occupational	
  
categories	
  following	
  Abowd,	
  Kramarz,	
  Margolis,	
  and	
  Troske	
  (2001)	
  to	
  reflect	
  the	
  skill-­‐intensity	
  of	
  the	
  
occupation.	
  
13	
   RAIS	
   reports	
   an	
   average	
   monthly	
   wage	
   (in	
   multiples	
   of	
   the	
   current	
   minimum	
   wage)	
   for	
   each	
   job	
   in	
  

which	
  a	
  worker	
  is	
  employed	
  in	
  each	
  year.	
  In	
  combination	
  with	
  information	
  on	
  the	
  number	
  of	
  months	
  
a	
  worker	
  was	
  employed	
  during	
  the	
  year	
  and	
  deflated	
  minimum	
  wage	
  information	
  in	
  reais	
  from	
  the	
  
Brazilian	
  Central	
  Bank,	
  we	
  calculate	
  an	
  annual	
  real	
  wage	
  for	
  each	
  worker.	
  
14	
   Note	
   that	
   any	
   firm-­‐level	
   aggregates	
   we	
   use	
   in	
   our	
   analysis,	
   however,	
   are	
   based	
   on	
   the	
   complete	
  

population	
  of	
  workers.	
  



                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                       8
	
  
    As	
   would	
   be	
   predicted	
   by	
   new	
   heterogeneous	
   firm	
   models	
   of	
   international	
   trade,	
   Brazil’s	
  
    trade	
  liberalization	
  in	
  the	
  early	
  1990s	
  was	
  associated	
  with	
  significant	
  firm-­‐level	
  entry	
  into	
  
    exporting.	
   The	
   share	
   of	
   exporting	
   firms	
   increased	
   over	
   50%—from	
   only	
   8.5%	
   in	
   1990	
   to	
  
    12.9%	
  in	
  1994	
  before	
  leveling	
  off.	
  
    	
  
    Tariffs	
                                                                            In	
  our	
  analysis	
  of	
  Brazil’s	
  trade	
  policy,	
  we	
  concentrate	
  on	
  two	
  trade	
  protection	
  
    measures:	
   the	
   final	
   goods	
   output	
   tariff	
   and	
   the	
   effective	
   rate	
   of	
   protection	
   (ERP).	
   The	
  
    effective	
  rate	
  of	
  protection	
  allows	
  us	
  to	
  incorporate	
  changes	
  in	
  tariffs	
  placed	
  on	
  inputs	
  into	
  a	
  
    firm’s	
   production	
   process.	
   Our	
   data	
   on	
   final	
   goods	
   tariffs	
   are	
   from	
   Muendler	
   (2003),	
   who	
  
    reports	
   monthly	
   nominal	
   final	
   goods	
   output	
   tariffs	
   at	
   the	
   Nível	
   80	
   Brazilian	
   industrial	
  
    classification	
   level.	
   We	
   access	
   monthly	
   ERP	
   at	
   the	
   Nível	
   80	
   industrial	
   classification	
   level	
  
    from	
  Kume,	
  Piani,	
  and	
  Souza	
  (2003).15	
  We	
  match	
  the	
  December	
  tariffs	
  from	
  1990	
  to	
  1998	
  
    with	
  our	
  annual	
  RAIS	
  worker	
  data	
  by	
  the	
  2-­‐digit	
  IBGE	
  subsector16	
  to	
  identify	
  workers	
  and	
  
    firms	
  in	
  industries	
  with	
  differential	
  rates	
  of	
  protection	
  and	
  liberalization	
  experiences.	
  	
  
    	
                                        	
  
    Figure	
   2.1	
   displays	
   both	
   the	
   mean	
   and	
   median	
   values	
   of	
   the	
   effective	
   rate	
   of	
   protection	
   in	
  
    the	
   manufacturing	
   sector	
   during	
   the	
   1990	
   to	
   1998	
   period.	
   The	
   early	
   1990s	
   experienced	
  
    sharp	
   declines	
   in	
   the	
   effective	
   rate	
   of	
   protection.	
   Mean	
   rates	
   fell	
   from	
   approximately	
   60	
  
    percent	
   to	
   20	
   percent,	
   while	
   median	
   rates	
   fell	
   from	
   40	
   percent	
   to	
   20	
   percent	
   in	
   a	
   half-­‐
    decade.	
  The	
  slight	
  aforementioned	
  protectionist	
  response	
  to	
  the	
  appreciation	
  of	
  the	
  real	
  is	
  
    also	
   evident.	
   Most	
   strongly	
   in	
   the	
   early	
   part	
   of	
   the	
   decade,	
   median	
   ERP	
   was	
   smaller	
   than	
  
    average	
   ERP,	
   suggesting	
   that	
   the	
   distribution	
   of	
   the	
   effective	
   rate	
   of	
   protection	
   is	
   skewed	
   to	
  
    the	
   right.	
   Over	
   time,	
   as	
   the	
   sectoral	
   variation	
   narrows,	
   the	
   mean	
   ERP	
   and	
   median	
   ERP	
  
    converge.	
  	
  
    	
                                        	
  
    This	
   substantial	
   cross-­‐industry	
   variation	
   in	
   both	
   levels	
   and	
   changes	
   in	
   the	
   ERP	
   is	
  

    	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
    15	
   Kume,	
   Piani,	
   and	
   Souza	
   (2003)	
   formally	
   measure	
   ERP	
   for	
   sector	
   k	
   as	
   the	
   increase	
   in	
   the	
   value-­‐

    added	
  due	
  to	
  the	
  structure	
  of	
  tariffs	
  relative	
  to	
  value-­‐added	
  at	
  free	
  trade	
  prices,	
  as	
  follows:	
  

           ERPk = τ k −
                                                                                                 ∑a τ                                mk k                                       d
                                                                                                                                                           ,	
   where	
   a = amk (1 + τ k ) 	
   is	
   the	
   input-­‐output	
   matrix	
   at	
   free-­‐trade	
  
                                                                                                1 − ∑a                                                  mk
                                                                                                                                                                            mk
                                                                                                                                                                                 1 + τm
    international	
  prices,	
   a 	
  is	
  the	
  input-­‐output	
  matrix	
  at	
  distorted	
  domestic	
  prices,	
  and	
   τ 	
  and	
   τ 	
  are	
  the	
  
                                                                                                                                                   d
                                                                                                                                                   mk                                                                                           k         m


    final	
  goods	
  and	
  intermediate	
  inputs	
  tariffs,	
  respectively.	
  
    16	
   We	
   follow	
   industry	
   concordances	
   available	
   at	
   http://econ.ucsd.edu/muendler/brazil	
   to	
   concord	
   the	
  
                               €            €                                                                                  €                  €
€   Nível	
   80	
   classification	
   to	
   the	
   2-­‐digit	
   IBGE	
   subsector	
   classification	
   used	
   in	
   RAIS.	
   This	
   constitutes	
   the	
  
    IBGE	
  subsectors	
  2	
  through	
  13.	
  



                                                                                                                                                                                                                                                                          9
    	
  
documented	
  in	
  more	
  detail	
  in	
  Figure	
  2.2	
  where	
  we	
  present	
  the	
  distribution	
  of	
  tariffs	
  across	
  
industries	
  in	
  1990	
  and	
  1998,	
  and	
  the	
  average	
  annual	
  change	
  in	
  ERP	
  during	
  this	
  period.	
  Note	
  
that	
  compared	
  to	
  1990,	
  the	
  distribution	
  of	
  the	
  effective	
  rate	
  of	
  protection	
  across	
  industries	
  
at	
   the	
   end	
   of	
   our	
   sample	
   is	
   much	
   more	
   compressed	
   around	
   a	
   lower	
   mean.	
   We	
   also	
   note	
  
substantial	
  variation	
  across	
  sectors	
  in	
  the	
  average	
  annual	
  changes	
  in	
  protection	
  rates.	
  	
  
	
  
2.4.	
   Descriptive	
  statistics	
  
	
  
The	
   complete	
   matched	
   data	
   include	
   3,932,297	
   worker-­‐firm-­‐year	
   observations,	
   with	
  
657,572	
  workers	
  in	
  490,884	
  firms.	
  Our	
  final	
  sample	
  of	
  the	
  manufacturing	
  sector,	
  however,	
  
has	
   504,660	
   worker-­‐firm-­‐year	
   observations,	
   with	
   114,042	
   workers	
   in	
   58,578	
   firms.	
   We	
  
report	
   detailed	
   descriptive	
   statistics	
   in	
   Table	
   2.1.	
   Roughly	
   three-­‐quarters	
   of	
   the	
   formal-­‐
sector	
   labor	
   force	
   have	
   at	
   most	
   a	
   primary	
   school	
   education.	
   An	
   additional	
   18%	
   of	
  
manufacturing	
   sector	
   workers	
   are	
   high	
   school	
   educated,	
   and	
   only	
   7%	
   of	
   workers	
   have	
   a	
  
college	
   degree.	
   The	
   majority	
   of	
   Brazil’s	
   labor	
   force	
   is	
   employed	
   in	
   skilled	
   blue-­‐collar	
  
occupations,	
   like	
   machine	
   operators	
   and	
   assemblers.	
   Almost	
   20%	
   of	
   the	
   manufacturing	
  
workforce	
  is	
  in	
  professional	
  or	
  managerial	
  positions,	
  with	
  11%	
  in	
  unskilled	
  blue-­‐collar	
  jobs.	
  
Other	
   white-­‐collar	
   workers	
   (for	
   example,	
   those	
   in	
   secretarial	
   and	
   office	
   assistant	
  
occupations)	
  represent	
  only	
  8%	
  of	
  the	
  formal	
  sector	
  manufacturing	
  labor	
  force.	
  The	
  average	
  
number	
  of	
  employees	
  in	
  manufacturing	
  firms	
  is	
  relatively	
  small	
  at	
  73.	
  
	
  
There	
   are	
   64,212	
   workers	
   working	
   in	
   11,143	
   exporting	
   firms,	
   and	
   80,895	
   workers	
   working	
  
in	
   53,537	
   domestic	
   firms	
   over	
   the	
   sample	
   period.	
   Exporters	
   pay	
   a	
   substantially	
   higher	
  
average	
   wage	
   than	
   do	
   non-­‐exporters.	
   Exporters	
   employ	
   a	
   higher	
   share	
   of	
   skilled	
   workers,	
  
on	
  average,	
  compared	
  to	
  non-­‐exporting	
  firms,	
  consistent	
  with	
  the	
  existing	
  literature.	
  Almost	
  
10%	
  of	
  workers	
  at	
  exporting	
  firms	
  are	
  college	
  educated	
  and	
  21%	
  are	
  high-­‐school	
  graduates.	
  
In	
   comparison,	
   in	
   non-­‐exporting	
   firms	
   the	
   share	
   of	
   college	
   educated	
   and	
   high-­‐school	
  
educated	
  workers	
  are	
  4%	
  and	
  17%,	
  respectively.17	
  Exporters	
  also	
  employ	
  a	
  higher	
  share	
  of	
  
workers	
  in	
  professional,	
  managerial,	
  and	
  technical	
  occupations.	
  Exporters	
  are	
  significantly	
  
larger	
   in	
   terms	
   of	
   their	
   average	
   employment	
   compared	
   to	
   non-­‐exporters.	
   The	
   average	
  
exporter	
   employs	
   346	
   employees,	
   while	
   the	
   average	
   non-­‐exporter	
   has	
   only	
   37	
   employees.	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
17	
   Note	
   that	
   in	
   a	
   country	
   like	
   Brazil,	
   the	
   share	
   of	
   workers	
   with	
   a	
   high-­‐school	
   education	
   is	
   a	
   more	
  

meaningful	
  representation	
  of	
  skill	
  (Gonzaga,	
  Menezes-­‐Filho,	
  and	
  Terra	
  2006).	
  



                                                                                                                                                                                                                                   10
	
  
    The	
  set	
  of	
  firm	
  level	
  variables	
  available	
  in	
  our	
  dataset	
  allow	
  us	
  to	
  appropriately	
  control	
  for	
  
    these	
   differences	
   between	
   exporters	
   and	
   domestic	
   firms	
   in	
   identifying	
   the	
   heterogeneous	
  
    impact	
  of	
  trade	
  policy	
  on	
  wages	
  of	
  these	
  firms.	
  	
  
    	
  
    3.	
  Firm-­‐Level	
  Analysis	
  
    	
  
    We	
  begin	
  our	
  analysis	
  at	
  the	
  firm	
  level	
  to	
  ensure	
  the	
  comparability	
  of	
  our	
  results	
  with	
  those	
  
    of	
   the	
   existing	
   literature	
   based	
   on	
   firm-­‐level	
   data	
   and	
   to	
   highlight	
   the	
   importance	
   of	
  
    introducing	
   worker	
   and	
   match	
   heterogeneity	
   into	
   the	
   analysis.	
   To	
   this	
   end,	
   we	
   aggregate	
  
    the	
   matched	
   employer-­‐employee	
   data	
   to	
   the	
   firm	
   level	
   and	
   estimate	
   the	
   following	
  
    specification:	
  
    	
  
       ln y jt = γ 1t kt + γ 2 t kt * Exp jt + γ 3 RERt * Exp jt + γ 4 Exp jt + Ψ j + δ tr + βZ jt + ε jt 	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  (1)	
  
    	
  

€   where	
  the	
  dependent	
  variable,	
   !"# ! "# $ ,	
  is	
  the	
  logarithm	
  of	
  average	
  wages	
  at	
  the	
  firm	
  level	
  

    for	
  firm	
  j	
  at	
  time	
  t.	
   ! "! 	
  denotes	
  the	
  level	
  of	
  protection	
  in	
  sector	
  k	
  in	
  which	
  the	
  firm	
  operates,	
  

    and	
   !"# $% 	
   is	
   an	
   indicator	
   variable	
   equal	
   to	
   one	
   if	
   firm	
   j	
   reports	
   a	
   positive	
   dollar	
   value	
   of	
  

    exports	
   at	
   time	
   t	
   and	
   zero	
   otherwise.	
   The	
   level	
   of	
   protection	
   at	
   the	
   sector-­‐level	
   is	
   measured	
  
    by	
  both	
  tariffs	
  and	
  the	
  effective	
  rate	
  of	
  protection	
  (ERP).	
  We	
  use	
  the	
  latter	
  measure	
  in	
  our	
  
    main	
  specifications,	
  since	
  in	
  an	
  environment	
  in	
  which	
  Brazilian	
  firms	
  face	
  declines	
  in	
  both	
  
    final	
   goods	
   and	
   intermediate	
   input	
   tariffs,	
   the	
   ERP	
   is	
   a	
   more	
   appropriate	
   measure	
   of	
  
    protection	
   faced	
   by	
   domestic	
   firms.	
   In	
   each	
   specification,	
   we	
   include	
   an	
   interaction	
   term	
  

    between	
   ! "! 	
   and	
   !"# $% 	
   to	
   allow	
   for	
   changes	
   in	
   protection	
   to	
   have	
   differential	
   effects	
   on	
  

    exporters	
  and	
  firms	
  serving	
  only	
  the	
  domestic	
  market.	
  
    	
  
    As	
   we	
   noted	
   earlier,	
   the	
   post-­‐liberalization	
   period	
   in	
   Brazil	
   coincided	
   with	
   a	
   period	
   of	
   an	
  
    appreciation	
   of	
   the	
   currency,	
   the	
   real,	
   making	
   Brazilian	
   goods	
   less	
   competitive	
   on	
  
    international	
   markets,	
   while	
   making	
   imported	
   goods	
   cheaper	
   in	
   real	
   terms.	
   Failing	
   to	
  
    incorporate	
   such	
   fluctuations	
   in	
   exchange	
   rates	
   into	
   our	
   analysis	
   could	
   bias	
   the	
   estimated	
  
    effect	
   of	
   liberalization	
   on	
   wages.	
   Henceforth,	
   in	
   each	
   specification,	
   we	
   also	
   include	
   an	
  




                                                                                                                                                                                                                                                    11
    	
  
interaction	
  of	
  Brazil’s	
  real	
  exchange	
  rate	
  (RER)18	
  and	
  the	
  firm’s	
  export	
  status.19	
  	
  
	
  

The	
   time-­‐varying,	
   firm-­‐level	
   controls,	
   ! "# ,	
   include	
   variables	
   available	
   in	
   standard	
   firm-­‐level	
  

data	
   sets	
   such	
   as	
   log	
   employment,	
   and	
   the	
   occupational	
   skill	
   composition20	
   of	
   the	
   firm	
   in	
  
addition	
   to	
   average	
   worker	
   tenure	
   at	
   the	
   firm,	
   and	
   controls	
   for	
   the	
   age21,	
   gender,	
   and	
  
educational	
   skill	
   composition22	
   of	
   the	
   firm.23	
   Each	
   specification	
   also	
   includes	
   firm	
   fixed	
  

effects,	
   ! ! ,	
   accounting	
   for	
   time-­‐invariant,	
   firm	
   characteristics	
   and	
   interactive	
   region-­‐year	
  

fixed	
   effects,	
   !!" ,	
   capturing	
   the	
   average	
   effect	
   of	
   policy	
   changes	
   that	
   may	
   differentially	
  

impact	
   wages	
   of	
   firms	
   in	
   different	
   regions	
   of	
   Brazil.24	
   Here,	
   ! !" 	
   is	
   an	
   error	
   term	
   that	
   is	
  

assumed	
   to	
   exhibit	
   no	
   serial	
   correlation,	
   and	
   to	
   be	
   orthogonal	
   to	
   all	
   regressors.	
   In	
   each	
  
specification,	
  the	
  standard	
  errors	
  are	
  clustered	
  at	
  the	
  industry-­‐year	
  level	
  to	
  account	
  for	
  the	
  
possibility	
  of	
  within-­‐industry,	
  across-­‐firm	
  correlation	
  in	
  errors	
  following	
  Moulton	
  (1990).	
  
	
  
In	
  interpreting	
  our	
  estimates	
  from	
  specification	
  (1)	
  we	
  focus	
  specifically	
  on	
  the	
  magnitude	
  
of	
   the	
   differential	
   change	
   in	
   average	
   firm-­‐level	
   wages	
   at	
   exporters	
   relative	
   to	
   non-­‐exporters	
  

( ! ! )	
   as	
   well	
   as	
   the	
   overall	
   wage	
   impact	
   of	
   a	
   decline	
   in	
   protection	
   for	
   exporters	
   ( ! ! + ! " )	
   and	
  

non-­‐exporters	
  ( !! ),	
  separately.	
  The	
  responsiveness	
  of	
  average	
  wages	
  in	
  firms	
  serving	
  only	
  



	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
18	
   The	
   real	
   exchange	
   rate	
   series	
   for	
   Brazil	
   is	
   constructed	
   in	
   Muendler	
   (2003)	
   and	
   is	
   available	
   at	
  

http://www.econ.ucsd.edu/muendler/html/brazil.html#brazdata.	
  
19	
   Since	
   the	
   overall	
   impact	
   of	
   the	
   time-­‐varying,	
   economy-­‐wide	
   RER	
   is	
   absorbed	
   by	
   the	
   region-­‐specific	
  

year	
   effects	
   ( δ tr ),	
   we	
   can	
   only	
   separately	
   identify	
   the	
   effect	
   of	
   the	
   RER	
   changes	
   on	
   exporting	
   firms	
  
relative	
  to	
  domestic	
  firms.	
  We	
  also	
  conduct	
  robustness	
  checks	
  using	
  industry-­‐specific	
  exchange	
  rates.	
  
20	
   We	
   define	
   the	
   firm’s	
   occupational	
   skill	
   composition	
   as	
   the	
   share	
   of	
   the	
   firm’s	
   workforce	
   in	
   four	
  

occupational	
  categories:	
  	
  unskilled	
  blue	
  collar,	
  skilled	
  blue	
  collar,	
  other	
  white	
  collar,	
  and	
  professional	
  
             €
and	
  managerial	
  workers.	
  Unskilled	
  blue-­‐collar	
  workers	
  are	
  the	
  omitted	
  category.	
  
21	
   We	
   define	
   the	
   firm’s	
   age	
   composition	
   as	
   the	
   share	
   of	
   the	
   firm’s	
   workforce	
   in	
   six	
   age	
   categories:	
  	
  

youth	
  (15-­‐17),	
  adolescent	
  (18-­‐24),	
  nascent	
  career	
  (25-­‐29),	
  early	
  career	
  (30-­‐39),	
  peak	
  career	
  (40-­‐49),	
  
and	
  late	
  career	
  (50-­‐64).	
  Youth	
  workers	
  are	
  the	
  omitted	
  category.	
  	
  	
  
22	
   We	
   define	
   the	
   firm’s	
   educational	
   skill	
   composition	
   as	
   the	
   share	
   of	
   the	
   firm	
   in	
   three	
   education	
  

categories:	
   less	
   than	
   high	
   school,	
   at	
   least	
   high	
   school,	
   and	
   more	
   than	
   high	
   school.	
   Less	
   than	
   high	
  
school	
  is	
  the	
  omitted	
  category.	
  	
  	
  
23	
  By	
  including	
  both	
  the	
  educational	
  and	
  the	
  occupational	
  skill	
  composition	
  of	
  the	
  firm,	
  we	
  are	
  able	
  to	
  

allow	
   for	
   the	
   possibility	
   that	
   firms	
   use	
   increasingly	
   higher-­‐skilled	
   individuals	
   (as	
   defined	
   by	
  
education)	
  in	
  lower-­‐skilled	
  occupations.	
  See	
  Muendler	
  (2008)	
  for	
  evidence	
  on	
  the	
  skill	
  upgrading	
  of	
  
occupations	
  in	
  response	
  to	
  trade	
  reform	
  in	
  Brazil.	
  
24	
  We	
  consider	
  Brazil’s	
  five	
  main	
  geographic	
  regions:	
  the	
  North,	
  Northeast,	
  Center-­‐West,	
  Southeast,	
  

and	
  South.	
  



                                                                                                                                                                                                                                   12
	
  
the	
   domestic	
   market	
   to	
   changes	
   in	
   protection	
   is	
   reflected	
   in	
   the	
   coefficient	
   !! .	
   A	
   positive	
   !! 	
  

would	
  suggest	
  that	
  a	
  decline	
  in	
  protection	
  is	
  associated	
  with	
  a	
  decrease	
  in	
  average	
  wages	
  in	
  
firms	
  serving	
  solely	
  the	
  domestic	
  market.	
  Note	
  that	
  when	
  ERP	
  is	
  the	
  measure	
  of	
  protection	
  

(instead	
  of	
  tariffs),	
   !! 	
   would	
  reflect	
  a	
  combined	
  effect	
  of	
  the	
  positive	
  impact	
  of	
  a	
  reduction	
  

in	
  input	
  tariffs	
  through	
  prices	
  and	
  access	
  to	
  enhanced	
  variety	
  and	
  quality	
  on	
  firm	
  profits25,	
  
as	
  well	
  as	
  any	
  negative	
  impact	
  of	
  increased	
  import	
  competition	
  due	
  to	
  a	
  decline	
  in	
  output	
  
tariffs.	
   Hence,	
   we	
   expect	
   the	
   coefficient	
   to	
   be	
   smaller	
   in	
   magnitude	
   when	
   the	
   measure	
   of	
  
protection	
   is	
   ERP	
   compared	
   to	
   the	
   estimated	
   coefficient	
   when	
   protection	
   is	
   measured	
   by	
  
(output)	
  tariffs.	
  	
  
	
  

The	
   coefficient	
   on	
   the	
   interaction	
   term,	
   ! ! ,	
   reflects	
   the	
   differential	
   effect	
   of	
   trade	
   policy	
  

changes	
   on	
   average	
   wages	
   in	
   exporting	
   firms	
   relative	
   to	
   firms	
   serving	
   only	
   the	
   domestic	
  
market.	
   If	
   a	
   decline	
   in	
   protection	
   results	
   in	
   a	
   differential	
   increase	
   in	
   firm-­‐level	
   average	
  

wages	
   in	
   exporting	
   firms,	
   we	
   expect	
   ! ! < " .	
   There	
   are	
   various	
   reasons	
   a	
   decline	
   in	
  

protection	
   could	
   result	
   in	
   a	
   differential	
   increase	
   in	
   firm-­‐level	
   average	
   wages	
   of	
   exporting	
  
firms.	
   For	
   example,	
   if	
   firms	
   engage	
   in	
   some	
   form	
   of	
   rent-­‐sharing	
   with	
   their	
   workers,	
  
workers	
  employed	
  in	
  exporting	
  firms,	
  whose	
  prospects	
  improve	
  as	
  a	
  result	
  of	
  a	
  decline	
  in	
  
protection,	
   could	
   experience	
   an	
   increase	
   in	
   their	
   wages.	
   Similarly,	
   if	
   liberalization	
   is	
  
associated	
  with	
  a	
  change	
  in	
  the	
  relative	
  returns	
  to	
  skill,	
  and	
  if	
  exporting	
  firms	
  differ	
  in	
  terms	
  
of	
   their	
   labor	
   force	
   composition,	
   average	
   firm-­‐level	
   wages	
   in	
   exporting	
   firms	
   will	
   increase	
  
relative	
   to	
   domestic	
   firms.	
   Also,	
   the	
   relative	
   increase	
   in	
   wages	
   in	
   exporting	
   firms	
   could	
   be	
   a	
  
reflection	
  of	
  a	
  change	
  on	
  workforce	
  composition	
  in	
  terms	
  of	
  factors	
  that	
  are	
  observable	
  to	
  
the	
  managers	
  of	
  the	
  firm	
  and	
  hence	
  impact	
  wages,	
  but	
  are	
  unobservable	
  in	
  the	
  data	
  (such	
  as	
  
innate	
  worker	
  ability	
  or	
  match-­‐specific	
  ability).	
  	
  	
  
	
  
3.1	
  Estimation	
  Results	
  
	
  




	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
25	
   See	
   Arkolakis,	
   Demidova,	
   Klenow,	
   and	
   Rodriguez-­‐Clare	
   (2008)	
   and	
   Goldberg,	
   Khandelwal,	
   Pavcnik,	
  

and	
   Topalova	
   (2008)	
   for	
   evidence	
   on	
   improved	
   input	
   variety,	
   and	
   Kugler	
   and	
   Verhoogen	
   (2008),	
  
Amiti	
   and	
   Konings	
   (2007),	
   Csillag	
   and	
   Koren	
   (2009),	
   and	
   Halpern,	
   Koren	
   and	
   Szeidl	
   (2009)	
   for	
  
evidence	
  on	
  improved	
  input	
  quality.	
  	
  



                                                                                                                                                                                                                                   13
	
  
Estimation	
   results	
   from	
   equation	
   (1)	
   with	
   tariffs26	
   as	
   the	
   measure	
   of	
   protection	
   are	
  
reported	
  in	
  Table	
  3.1.	
  The	
  results	
  suggest	
  that	
  a	
  decline	
  in	
  tariffs	
  is	
  associated	
  with	
  a	
  decline	
  
in	
  average	
  wages	
  at	
  non-­‐exporting	
  firms,	
  consistent	
  with	
  a	
  negative	
  impact	
  of	
  an	
  increase	
  in	
  
foreign	
  competition	
  on	
  these	
  firms.	
  We	
  find	
  that	
  a	
  ten	
  percentage	
  point	
  decrease	
  in	
  tariffs	
  
leads	
  to	
  a	
  decrease	
  in	
  average	
  firm-­‐level	
  wages	
  by	
  1.7%	
  for	
  these	
  firms.	
  The	
  negative	
  and	
  
significant	
   coefficient	
   on	
   the	
   interaction	
   term	
   between	
   tariffs	
   and	
   export	
   status	
   suggests	
  
that	
  the	
  wages	
  in	
  exporting	
  firms	
  increase	
  in	
  response	
  to	
  a	
  decline	
  in	
  tariffs	
  relative	
  to	
  firms	
  
serving	
  only	
  the	
  domestic	
  market.	
  We	
  find	
  that	
  a	
  change	
  in	
  tariffs	
  has	
  no	
  statistical	
  impact	
  
on	
  average	
  wages	
  at	
  the	
  firm-­‐level	
  for	
  exporting	
  firms.27	
  The	
  RER-­‐exporter	
  interaction	
  term	
  
suggests	
  that	
  a	
  depreciation	
  in	
  the	
  RER	
  (a	
  decrease	
  in	
  the	
  RER	
  as	
  is	
  it	
  defined	
  in	
  our	
  data)	
  
increases	
  the	
  wages	
  in	
  exporting	
  firms	
  relative	
  to	
  non-­‐exporting	
  firms,	
  as	
  expected.	
  All	
  the	
  
firm-­‐level	
   controls	
   are	
   statistically	
   significant	
   and	
   enter	
   with	
   the	
   expected	
   signs	
   and	
  
magnitudes.	
   Average	
   firm-­‐level	
   wages	
   are	
   increasing	
   with	
   the	
   average	
   tenure	
   and	
   age-­‐
profile	
  at	
  the	
  firm,	
  while	
  they	
  are	
  decreasing	
  in	
  the	
  share	
  of	
  female	
  workers	
  employed	
  at	
  the	
  
firm.	
  Wages	
  are	
  increasing	
  in	
  both	
  the	
  educational	
  and	
  occupational	
  skill	
  composition	
  of	
  the	
  
firm	
  and	
  the	
  size	
  if	
  the	
  firm.	
  
	
  
	
                                        Next,	
   we	
   test	
   whether	
   the	
   differential	
   impact	
   of	
   tariffs	
   we	
   document	
   on	
   exporting	
  
firms	
  holds	
  equally	
  for	
  firms	
  operating	
  in	
  all	
  industries.	
  More	
  specifically,	
  we	
  allow	
  for	
  the	
  
effect	
   of	
   a	
   change	
   in	
   tariffs	
   to	
   be	
   different	
   for	
   firms	
   operating	
   in	
   Brazil’s	
   comparative	
  
advantage	
   sectors.	
   We	
   expect	
   that	
   following	
   liberalization,	
   exporters	
   in	
   these	
   sectors	
   will	
  
experience	
   a	
   more	
   pronounced	
   increase	
   in	
   profitability	
   relative	
   to	
   exporters	
   in	
   low	
  
comparative	
   advantage	
   sectors	
   and	
   hence,	
   expect	
   a	
   stronger	
   impact	
   on	
   average	
   wages	
   at	
  
these	
  firms.	
  We	
  divide	
  our	
  sample	
  into	
  high	
  and	
  low	
  comparative	
  advantage	
  sectors	
  using	
  
data	
  on	
  each	
  industry’s	
  Balassa	
  (1965)	
  comparative	
  advantage.28	
  The	
  next	
  two	
  columns	
  in	
  
Table	
  3.1	
  report	
  our	
  results	
  by	
  the	
  comparative	
  advantage	
  of	
  the	
  firm’s	
  sector	
  where	
   high	
  
(low)	
  comparative	
  advantage	
  sectors	
  are	
  those	
  with	
  an	
  above	
  (below)	
  median	
  value	
  of	
  the	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
26	
  Our	
  estimation	
  results	
  are	
  essentially	
   the	
  same	
  when	
  we	
  use	
  tariffs	
  weighted	
  by	
  the	
  value	
  added	
  of	
  

the	
  industry	
  instead	
  of	
  un-­‐weighted	
  tariffs.	
  
27	
   The	
   absolute	
   effect	
   on	
   exporters,	
   the	
   F-­‐statistic	
   and	
   corresponding	
   p-­‐value	
   on	
   the	
   joint	
   significance	
  

of	
  the	
  coefficients	
  are	
  reported	
  at	
  the	
  bottom	
  of	
  Table	
  3.1.	
  
28	
  Industry	
  k’s	
  Balassa	
  (1965)	
  comparative	
  advantage	
  in	
  year	
  t	
  is	
  constructed	
  in	
  Muendler	
  (2007)	
  as	
  

follows:	
  	
  	
  

Comp_Advk,t	
  =	
  	
  
                            Brazil
                          X k,t /
                                       m
                                           Brazil
                                         X m,t
                                                     where	
  Xkt	
  are	
  exports.	
  	
  
                                                                                                                                                    ∑
                            World
                          X k,t /          World
                                         X m,t                                                                                                      ∑                 m




                                                                                                                                                                                                                                   14
	
                                    €
Balassa	
   (1965)	
   comparative	
   advantage	
   measure	
   across	
   all	
   merchandise	
   trade	
   sectors	
   in	
  
1986.29	
   30	
  Interestingly,	
  we	
  find	
  the	
  differential	
  impact	
  of	
  trade	
  liberalization	
  on	
  exporters	
  
to	
  be	
  prevalent	
  only	
  for	
  firms	
  operating	
  in	
  high	
  comparative	
  advantage	
  sectors.	
  Firms	
  in	
  low	
  
comparative	
   advantage	
   sectors	
   experience	
   a	
   decline	
   in	
   average	
   wages	
   regardless	
   of	
   their	
  
export	
  status.	
  
	
  
During	
   the	
   liberalization	
   period,	
   Brazilian	
   firms	
   faced	
   declines	
   in	
   both	
   final	
   goods	
   and	
   in	
  
intermediate	
   input	
   tariffs.	
   While	
   a	
   decrease	
   in	
   final	
   goods	
   tariffs	
   could	
   decrease	
   the	
  
profitability	
   of	
   domestic	
   firms	
   by	
   increasing	
   the	
   foreign	
   competition	
   that	
   they	
   face,	
   a	
  
decrease	
   in	
   input	
   tariffs	
   is	
   likely	
   to	
   have	
   the	
   opposite	
   effect	
   by	
   decreasing	
   the	
   price	
   of	
  
inputs,	
  and	
  improving	
  the	
  variety	
  and	
  possibly	
  quality	
  of	
  inputs	
  that	
  the	
  firm	
  has	
  access	
  to.	
  
If	
  the	
  industries	
  that	
  experienced	
  a	
  decline	
  in	
  final	
  goods	
  tariffs	
  also	
  experienced	
  a	
  decline	
  
in	
  input	
  tariffs,	
  the	
  product	
  tariff	
  is	
  likely	
  to	
  overestimate	
  the	
  actual	
  decrease	
  in	
  protection	
  
that	
   the	
   industry	
   has	
   experienced.	
   To	
   address	
   this	
   issue,	
   we	
   repeat	
   the	
   previous	
   analysis	
  
with	
   the	
   effective	
   rate	
   of	
   protection	
   (ERP)	
   instead	
   of	
   (output)	
   tariffs	
   as	
   the	
   measure	
   of	
  
protection.	
   The	
  estimation	
  results	
  suggest	
  that	
  a	
  decline	
  in	
  ERP	
  has	
  no	
  significant	
  impact	
  on	
  
average	
   wages	
   at	
   non-­‐exporting	
   firms.	
   Improved	
   access	
   to	
   imported	
   intermediates	
   from	
  
abroad	
  could	
  explain	
  the	
  difference	
  between	
  these	
  results	
  and	
  those	
  reported	
  in	
  first	
  half	
  of	
  
Table	
  3.1.	
  Note	
  that	
  this	
  coefficient	
  now	
  reflects	
  a	
  combined	
  effect	
  of	
  the	
  positive	
  impact	
  of	
  a	
  
reduction	
   in	
   input	
   tariffs	
   through	
   prices	
   and	
   access	
   to	
   variety	
   and	
   quality,	
   as	
   well	
   as	
   any	
  
negative	
   impact	
   of	
   increased	
   import	
   competition	
   due	
   to	
   a	
   decline	
   in	
   output	
   tariffs.	
   The	
  
negative	
   and	
   significant	
   coefficient	
   on	
   the	
   interaction	
   term	
   suggests	
   that	
   the	
   wages	
   in	
  
exporting	
   firms	
   increase	
   in	
   response	
   to	
   a	
   decline	
   in	
   the	
   ERP.	
   	
   A	
   ten	
   percentage	
   point	
  
decrease	
   in	
   the	
   ERP	
   increases	
   average	
   wages	
   by	
   1%	
   at	
   exporting	
   firms.	
   Similar	
   to	
   the	
  
results	
   for	
   output	
   tariffs,	
   in	
   the	
   high	
   comparative	
   advantage	
   sector,	
   we	
   find	
   that	
   a	
   decline	
   in	
  
ERP	
   is	
   associated	
   with	
   a	
   differential	
   increase	
   in	
   wages	
   in	
   exporting	
   firms	
   relative	
   to	
   their	
  
non-­‐exporting	
  counterparts.	
  The	
  increase	
  in	
  wages	
  associated	
  with	
  a	
  ten	
  percentage	
  point	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
29	
   We	
   choose	
   the	
   pre-­‐reform	
   year	
   of	
   1986	
   to	
   avoid	
   any	
   possible	
   endogeneity	
   between	
   the	
  

comparative	
  advantage	
  measure	
  based	
  on	
  exports	
  and	
  tariff	
  reforms	
  which	
  began	
  in	
  1988.	
  We	
  also	
  
experimented	
   with	
   generating	
   the	
   comparative	
   advantage	
   measure	
   across	
   only	
   manufacturing	
  
sectors	
   and	
   using	
   a	
   post-­‐reform	
   year	
   with	
   no	
   difference	
   in	
   the	
   results.	
   Brazil’s	
   Balassa	
   (1965)	
  
revealed	
  comparative	
  advantage	
  is	
  largely	
  time-­‐invariant.	
  
30	
   The	
   following	
   sectors	
   are	
   defined	
   as	
   high	
   comparative	
   advantage	
   sectors:	
   	
   Manufacture	
   of	
   non-­‐

metallic	
   mineral	
   products,	
   Manufacture	
   of	
   metallic	
   products,	
   Manufacture	
   of	
   transport	
   equipment,	
  
Manufacture	
   of	
   wood	
   products	
   and	
   furniture,	
   Manufacture	
   of	
   footwear,	
   and	
   the	
   Manufacture	
   of	
   food,	
  
beverages,	
  and	
  ethyl	
  alcohol.	
  



                                                                                                                                                                                                                                   15
	
  
decrease	
  in	
  ERP	
  is	
  2.9%	
  for	
  exporters	
  while	
  it	
  is	
  0.8%	
  (and	
  significant	
  only	
  at	
  10%	
  level)	
  for	
  
firms	
  serving	
  the	
  domestic	
  market.	
  The	
  impact	
  of	
  trade	
  liberalization	
  on	
  average	
  firm	
  wages	
  
is	
  not	
  statistically	
  significant	
  for	
  domestic	
  firms	
  or	
  exporters	
  in	
  low	
  comparative	
  advantage	
  
sectors.	
  Our	
  results	
  also	
  indicate	
  that	
  the	
  exporter	
  premium	
  is	
  slightly	
  higher	
  for	
  exporters	
  
in	
  the	
  high	
  comparative	
  advantage	
  sector.	
  
	
  
A	
  potential	
  concern	
  for	
  the	
  results	
  reported	
  in	
  Table	
  3.1	
  is	
  the	
  selection	
  bias	
  introduced	
  by	
  
the	
   heterogeneity	
   of	
   firms	
   exiting	
   the	
   sample	
   following	
   trade	
   liberalization.	
   If	
   a	
   decline	
   in	
  
ERP	
   results	
   in	
   the	
   exit	
   of	
   low	
   productivity	
   firms,	
   the	
   negative	
   effect	
   of	
   liberalization	
   on	
  
domestic	
   firms	
   will	
   be	
   underestimated,	
   as	
   the	
   remaining	
   firms	
   in	
   the	
   sample	
   would	
   have	
  
high	
   productivity	
   and	
   pay	
   high	
   wages.	
   To	
   evaluate	
   the	
   relevance	
   of	
   this	
   possibility	
   in	
  
explaining	
  the	
  differential	
  impact	
  we	
  find	
  on	
  exporting	
  firms,	
  we	
  repeat	
  our	
  analysis	
  for	
  a	
  
balanced	
   panel	
   of	
   firms,	
   which	
   were	
   under	
   operation	
   during	
   the	
   entire	
   sample	
   period.	
  
Results	
  reported	
  in	
  the	
  first	
  column	
  of	
  Table	
  3.2	
  suggest	
  that	
  restricting	
  our	
  sample	
  to	
  those	
  
firms,	
   which	
   do	
   not	
   enter	
   or	
   exit,	
   does	
   not	
   alter	
   the	
   magnitude	
   or	
   the	
   significance	
   of	
   our	
  
main	
  coefficient	
  of	
  interest,	
  suggesting	
  that	
  our	
  results	
  are	
  not	
  driven	
  by	
  sample-­‐selection.31	
  
	
                                        	
  
We	
  provide	
  further	
  robustness	
  checks	
  in	
  the	
  remaining	
  columns	
  of	
  Table	
  3.2.	
  As	
  we	
  address	
  
in	
   the	
   data	
   section,	
   our	
   main	
   firm-­‐level	
   regressions	
   draw	
   on	
   the	
   complete	
   employment	
  
history	
   of	
   a	
   1%	
   random	
   sample	
   of	
   the	
   population	
   of	
   Brazil’s	
   formal-­‐sector	
   labor	
   force.	
   To	
  
ensure	
   this	
   data	
   is	
   representative	
   of	
   Brazilian	
   firms	
   and	
   industries,	
   in	
   the	
   second	
   column	
   of	
  
Table	
   3.2,	
   we	
   repeat	
   the	
   analysis	
   drawing	
   on	
   the	
   complete	
   employment	
   history	
   of	
   a	
   5%	
  
random	
   sample	
   of	
   formal-­‐sector	
   males	
   living	
   in	
   metropolitan	
   areas.	
   In	
   the	
   third	
   column,	
   we	
  
restrict	
  the	
  data	
  to	
  Brazil’s	
  main	
  trade	
  reform	
  period	
  (1990-­‐1994)	
  when	
  average	
  protection	
  
levels	
   consistently	
   declined.	
   In	
   both	
   cases,	
   our	
   main	
   coefficient	
   of	
   interest	
   remains	
  
significant	
   with	
   little	
   change	
   in	
   its	
   magnitude.	
   The	
   results	
   reported	
   in	
   the	
   next	
   column	
  
suggest	
   that	
   our	
   findings	
   are	
   robust	
   to	
   replacing	
   the	
   economy-­‐wide	
   real	
   exchange	
   rate	
   with	
  
industry-­‐specific	
   real	
   exchange	
   rates	
   in	
   order	
   to	
   capture	
   differences	
   in	
   the	
   relative	
  


	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
31	
   Similarly,	
   if	
   the	
   least	
   productive	
   exporters	
   switch	
   out	
   of	
   exporting	
   following	
   liberalization,	
   the	
  

magnitude	
   of	
   the	
   differential	
   effect	
   of	
   a	
   decline	
   in	
   ERP	
   on	
   exporters	
   will	
   be	
   biased	
   upwards.	
   In	
  
unreported	
  regressions,	
  available	
  by	
  request,	
  we	
  include	
  log	
  total	
  factor	
  productivity	
  as	
  an	
  additional	
  
explanatory	
   variable	
   based	
   on	
   estimates	
   from	
   Muendler	
   (2004).	
   We	
   find	
   the	
   effect	
   of	
   log	
   TFP	
   on	
  
average	
  firm	
  wages	
  to	
  be	
  insignificant.	
  Importantly,	
  the	
  inclusion	
  of	
  log	
  TFP	
  does	
  not	
  alter	
  the	
  main	
  
coefficient	
  of	
  interest.	
  	
  



                                                                                                                                                                                                                                   16
	
  
importance	
  of	
  trading	
  partners	
  across	
  industries.32	
  	
  
	
  
We	
  also	
  test	
  whether	
  our	
  results	
  are	
  sensitive	
  to	
  the	
  exporting	
  thresholds	
  we	
  use	
  to	
  assign	
  
the	
  indicator	
  variable	
  denoting	
  a	
  firm’s	
  export	
  status.	
  In	
  the	
  main	
  firm-­‐level	
  specifications,	
  a	
  
firm	
   is	
   defined	
   as	
   an	
   exporter	
   if	
   it	
   exported	
   any	
   positive	
   dollar	
   amount	
   that	
   year.	
   Instead,	
   in	
  
columns	
  4	
  and	
  5,	
  we	
  only	
  consider	
  firms	
  with	
  an	
  export	
  value	
  more	
  than	
  the	
  5th	
  percentile	
  
(cutoff	
   5)	
   in	
   that	
   year,	
   and	
   more	
   than	
   the	
   10th	
   percentile	
   (cutoff	
   10)	
   as	
   exporting	
   firms,	
  
respectively.	
   Next,	
   we	
   replace	
   the	
   exporter	
   status	
   dummy	
   in	
   our	
   main	
   specification	
   with	
  
two	
  time-­‐invariant	
  measures	
  of	
  export	
  status.	
  Export90	
  takes	
  the	
  value	
  1	
  if	
  the	
  firm	
  was	
  an	
  
exporter	
  as	
  of	
  the	
  beginning	
  of	
  our	
  sample,	
  and	
  zero	
  otherwise.	
  Export_Once	
  takes	
  value	
  1	
  if	
  
the	
   firm	
   exported	
   for	
   at	
   least	
   one	
   year	
   during	
   the	
   sample	
   period.	
   Our	
   estimation	
   results	
  
continue	
   to	
   suggest	
   a	
   differential	
   positive	
   impact	
   of	
   liberalization	
   on	
   wages	
   at	
   exporting	
  
firms	
  relative	
  to	
   non-­‐exporters.	
   In	
   the	
   last	
   column,	
   we	
   report	
   results	
   from	
   a	
   specification	
   in	
  
which	
  we	
  include	
  the	
  logarithm	
  of	
  the	
  value	
  of	
  exports	
  for	
  firms	
  with	
  positive	
  exports.	
  Our	
  
results	
   indicate	
   that	
   conditional	
   on	
   exporting,	
   there	
   is	
   a	
   differential	
   increase	
   in	
   wages	
   at	
  
firms	
  with	
  higher	
  values	
  of	
  exports	
  following	
  liberalization.	
  	
  	
  
	
  
Next,	
   we	
   test	
   whether	
   the	
   differential	
   impact	
   on	
   exporters	
   we	
   document	
   could	
   simply	
   be	
  
attributed	
   to	
   compositional	
   differences	
   between	
   exporters	
   and	
   non-­‐exporters	
   in	
   an	
  
environment	
   in	
   which	
   the	
   returns	
   to	
   observable	
   worker	
   characteristics	
   are	
   changing	
   as	
   a	
  
result	
   of	
   liberalization.	
   Specifically,	
   we	
   allow	
   for	
   changes	
   in	
   the	
   premium	
   paid	
   by	
   firms	
   with	
  
different	
   workforce	
   skill	
   compositions,	
   by	
   interacting	
   ERP	
   with	
   measures	
   of	
   labor	
   force	
  
composition	
  at	
  the	
  firm	
  level	
  in	
  equation	
  (1)	
  and	
  find	
  no	
  impact	
  on	
  our	
  main	
  coefficient	
  of	
  
interest.	
  This	
  finding	
  suggests	
  that	
  the	
  differential	
  effect	
  that	
  we	
  find	
  cannot	
  be	
  explained	
  by	
  
changes	
  in	
  the	
  relative	
  returns	
  to	
  observable	
  characteristics	
  following	
  liberalization.33	
  
	
  
Finally,	
   we	
   consider	
   the	
   possibility	
   that	
   an	
   omission	
   of	
   the	
   importer	
   status	
   of	
   the	
   firm	
   could	
  
	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
32	
  We	
  construct	
  industry-­‐specific	
  real	
  exchange	
  rates	
  using	
  time-­‐varying	
  trade	
  weights,	
  as	
  in	
  

Goldberg	
  (2004).	
  More	
  specifically,	
  we	
  calculate	
  
                 ⎛ ⎛                                                                                                                                                                                                                             ⎞       ⎞
                 ⎜ ⎜     X tkc       M tkc                                  k
                                                                                                                                                                                                                                                   ⎟     c
                                                                                                                                                                                                                                                            ⎟
   RER _ ind = ∑ ⎜ ⎜ .5        + .5                                        t                                                                                                                                                                     ⎟ RERt ⎟ 	
  where	
   RER 	
  are	
  the	
  bilateral	
  exchange	
  
                                                                                                                                                                                                                                                                                             c



                          ∑ X tkc ∑ M tkc
                                                                                                                                                                                                                                                                                             t


               c ⎜ ⎜                                                                                                                                                                                                                             ⎟       ⎟
                 ⎝ ⎝ c              c                                                                                                                                                                                                            ⎠    €  ⎠
rates	
  for	
  trading	
  partner	
  c	
  of	
  Brazil,	
   X                                                                                                                                                                     t
                                                                                                                                                                                                                                    kc   	
  and	
   M   t
                                                                                                                                                                                                                                                          kc   	
  are	
  exports	
  and	
  imports	
  in	
  industry	
  k	
  to	
  or	
  from	
  
country	
  c	
  at	
  time	
  t.	
  Note	
  that	
   RER is	
  also	
  included	
  in	
  this	
  specification.	
                                                                                                             c
                                                                                                                                                                                                                              t


33	
  These	
  results	
  are	
  not	
  reported	
  to	
  conserve	
  space	
  and	
  are	
  available	
  upon	
  request.	
  
                                                                                                                          €




                                                                                                                                                                                                                                                                                                                                                     17
	
  
bias	
   our	
   results	
   if	
   improved	
   access	
   to	
   foreign	
   intermediate	
   inputs	
   increases	
   wages	
   at	
   the	
  
firm	
  level	
  and	
  if	
  the	
  firm’s	
  export	
  status	
  is	
  correlated	
  with	
  its	
  import	
  status.	
  Amiti	
  and	
  Davis	
  
(forthcoming)	
   find	
   this	
   effect	
   to	
   be	
   important	
   in	
   Indonesia:	
   following	
   a	
   decline	
   in	
   input	
  
tariffs,	
  average	
  wages	
  in	
  importing	
  firms	
  increase	
  relative	
  to	
  firms	
  that	
  do	
  not	
  import.	
  Our	
  
estimation	
   results	
   based	
   on	
   a	
   sub-­‐sample	
   of	
   firms	
   for	
   which	
   data	
   on	
   import	
   status	
   is	
  
available	
   suggest	
   no	
   such	
   differential	
   impact	
   on	
   importing	
   firms	
   in	
   Brazil.	
   Moreover,	
   the	
  
differential	
  impact	
  of	
  liberalization	
  on	
  average	
  wages	
  in	
  exporting	
  firms	
  is	
  robust	
  to	
  various	
  
specifications	
   controlling	
   for	
   the	
   import	
   status	
   of	
   the	
   firms.	
   Appendix	
   includes	
   a	
   more	
  
detailed	
  description	
  of	
  these	
  results.	
  
	
  
3.2	
  Discussion	
  of	
  Firm-­‐Level	
  Analysis	
  	
  
	
  
Our	
   firm-­‐level	
   analysis	
   confirms	
   findings	
   in	
   earlier	
   studies	
   regarding	
   the	
   differential	
   impact	
  
of	
   trade	
   reform	
   on	
   average	
   wages	
   at	
   firms	
   with	
   differing	
   degrees	
   of	
   trade	
   exposure,	
  
especially	
   in	
   high	
   comparative	
   advantage	
   sectors.	
   However,	
   this	
   analysis	
   of	
   average	
   firm-­‐
level	
   wages,	
   although	
   informative,	
   is	
   not	
   well	
   suited	
   to	
   examine	
   the	
   differential	
   impact	
   of	
  
liberalization	
  on	
  workers	
  in	
  heterogeneous	
  firms	
  for	
  a	
  number	
  of	
  inter-­‐related	
  reasons.	
  In	
  
general,	
  we	
  would	
  expect	
  that	
  in	
  addition	
  to	
  observable	
  worker	
  and	
  firm	
  characteristics,	
  the	
  
allocation	
   of	
   workers	
   to	
   firms	
   is	
   a	
   function	
   of	
   the	
   worker	
   characteristics	
   that	
   are	
  
unobservable	
   in	
   the	
   but	
   the	
   managers	
   can	
   observe	
   and	
   reward,	
   such	
   as	
   the	
   innate	
   (time-­‐
invariant)	
  ability	
  of	
  the	
  worker	
  and	
  any	
  additional	
  productivity	
  (ability)	
  that	
  obtains	
  in	
  the	
  
context	
   of	
   employment	
   in	
   the	
   specific	
   firm	
   (match-­‐specific	
   ability).	
   	
   Recent	
   theoretical	
  
models	
  (including	
  Mortensen	
  (2003),	
  Postel-­‐Vinay	
  and	
  Robin	
  (2002),	
  Lentz	
  (2010),	
  Shimer	
  
(2005))	
   and	
   Gibbons	
   et	
   al.	
   (2005))	
   variously	
   describe	
   the	
   central	
   role	
   played	
   by	
   ex-­‐ante	
  
unobservable	
   worker	
   characteristics	
   in	
   determining	
   the	
   equilibrium	
   assignment	
   of	
   workers	
  
to	
   firms.	
   In	
   the	
   context	
   of	
   international	
   trade	
   and	
   labor	
   markets,	
   Helpman,	
   Itskhoki,	
   and	
  
Redding	
   (2010),	
   who	
   model	
   the	
   labor	
   allocation	
   process	
   with	
   heterogeneous	
   firms	
   as	
   being	
  
subject	
  to	
  search	
  and	
  matching	
  frictions,	
  find	
  that	
  the	
  more	
  productive,	
  exporting,	
  firms	
  will	
  
screen	
   workers	
   more	
   intensively	
   and	
   employ	
   workers	
   of	
   higher	
   match-­‐specific	
   ability,	
   i.e.,	
  
the	
   equilibrium	
   assignment	
   of	
   workers	
   is	
   again	
   a	
   function	
   of	
   unobservable	
   match	
   specific	
  
worker	
  ability	
  and	
  is	
  thus	
  non-­‐random.	
  	
  
	
  
Importantly,	
  in	
  an	
  environment	
  in	
  which	
  firms	
  are	
  changing	
  the	
  composition	
  and	
  quality	
  of	
  




                                                                                                                                                       18
	
  
their	
  labor	
  force	
  in	
  response	
  to	
  liberalization,34	
  analysis	
  conducted	
  at	
  the	
  firm	
  level	
  faces	
  at	
  
least	
   two	
   problems.	
   If	
   exporting	
   firms	
   respond	
   to	
   liberalization	
   by	
   changing	
   the	
  
composition	
  of	
  their	
  workforce	
  systematically,	
  (for	
  example,	
  towards	
  workers	
  with	
  higher	
  
innate	
  ability	
  or	
  match-­‐specific	
  ability)	
  our	
  estimates	
  will	
  be	
  biased.	
  This	
  is	
  because	
  part	
  of	
  
the	
  differential	
  effect	
  we	
  find	
  for	
  exporters	
  at	
  the	
  firm-­‐level	
  could	
  be	
  due	
  to	
  compositional	
  
differences	
   between	
   firms	
   with	
   different	
   trade	
   orientation	
   and	
   not	
   because	
   otherwise	
  
identical	
  workers	
  are	
  being	
  paid	
  different	
  wages	
  in	
  different	
  firms.	
  Furthermore,	
  when	
  the	
  
job	
  mobility	
  of	
  workers	
  is	
  at	
  least	
  partly	
  determined	
  by	
  their	
  unobservable	
  characteristics	
  
(endogenous	
   mobility),	
   estimates	
   of	
   the	
   parameters	
   in	
   (1)	
   will	
   be	
   biased.	
   This	
   is	
   because	
  
non-­‐random	
   job	
   assignment	
   implies	
   a	
   correlation	
   between	
   the	
   error	
   term	
   ε jt (which	
  
                                                                                                                     	
  
subsumes	
   the	
   un-­‐observables	
   associated	
   with	
   workers	
   in	
   firm	
   j	
   at	
   time	
   t)	
   and	
   the	
   firm’s	
  
characteristics	
  represented	
  by	
  the	
  right	
  hand	
  side	
  variables, Ψ j 	
  and	
   Z jt ,	
  and	
  thus	
  a	
  failure	
  

of	
  the	
  maintained	
  assumption, E (ε jt | Ψ j , Z jt ) = 0 ,	
  underlying	
  the	
  estimation.	
  	
  	
  

	
  
In	
  the	
  following	
  section,	
  we	
  describe	
  the	
  worker-­‐level	
  analysis	
  we	
  undertake	
  using	
  matched	
  
employer-­‐employee	
   data	
   to	
   account	
   for	
   the	
   issues	
   of	
   unobservable	
   worker	
   ability	
  
differences	
   and	
   endogenous	
   job	
   mobility	
   that	
   we	
   have	
   just	
   described.	
   We	
   also	
   describe	
   in	
  
much	
  greater	
  detail	
  the	
  issue	
  of	
  endogenous	
  job	
  mobility	
  which	
  persists,	
  in	
  principle,	
  with	
  
worker-­‐level	
   analysis,	
   but	
   can	
   be	
   dealt	
   with	
   using	
   firm-­‐worker	
   match	
   effects	
   in	
   the	
   wage	
  
specification.	
  
	
  
	
  
	
  
4.	
  	
  	
  Worker-­‐Level	
  Analysis	
  
	
  
We	
   begin	
   our	
   worker-­‐level	
   analysis	
   by	
   considering	
   the	
   basic	
   specification	
   of	
   Abowd,	
  
Kramarz,	
  and	
  Margolis	
  (1999)	
  in	
  which	
  a	
  worker’s	
  wages	
  can	
  be	
  decomposed	
  as	
  follows:	
  
	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
34	
  In	
   unreported	
   results	
   available	
   upon	
   request,	
   we	
   provide	
   some	
   evidence	
   of	
   differential	
   skill	
  
upgrading	
  at	
  exporting	
  firms	
  relative	
  to	
  non-­‐exporting	
  firms	
  with	
  trade	
  liberalization.	
  Specifically,	
  we	
  
re-­‐estimate	
   equation	
   (1)	
   with	
   the	
   share	
   of	
   workers	
   with	
   different	
   levels	
   of	
   education	
   as	
   the	
  
dependent	
  variable.	
  While	
  a	
  change	
  in	
  the	
  ERP	
  has	
  no	
  significant	
  impact	
  on	
  the	
  skill	
  composition	
  of	
  
workers	
  in	
  non-­‐exporting	
  firms,	
  in	
  exporting	
  firms	
  a	
  decline	
  in	
  the	
  ERP	
  is	
  associated	
  with	
  an	
  increase	
  
in	
  the	
  share	
  of	
  college-­‐educated	
  workers	
  and	
  a	
  decrease	
  in	
  the	
  share	
  of	
  workers	
  with	
  less	
  than	
  high	
  
school	
  education.	
  



                                                                                                                                                                                                                                   19
	
  
  !" !"#$ = !" + " ##"$$ % + # % "$ + $ & #$ + %"#$ 	
   	
          	
        	
        	
        	
        	
  	
  	
  	
  	
  	
  	
  	
  	
  	
   	
     	
  	
  	
  	
  (2)	
  

	
  

where	
   i	
   indexes	
   the	
   individual,	
   j	
   indexes	
   the	
   firm,	
   t	
   indexes	
   time,	
   and	
   ln y ijt 	
   denotes	
  

individual-­‐level	
  log	
  wages.	
  The	
  panel	
  of	
  linked	
  worker-­‐firm	
  data	
  allows	
  us	
  to	
  control	
  for	
  a	
  
rich	
  array	
  of	
  fixed	
  factors	
  that	
  may	
  influence	
  a	
  worker’s	
  wages,	
  in	
  addition	
  to	
  time-­‐varying,	
  
                                                                                                                                      €
observable,	
  firm	
  characteristics	
  ( ! "# )	
  and	
  worker	
  characteristics	
  ( ! "# )	
  including	
  indicator	
  

variables	
  for	
  the	
  worker’s	
  occupation,	
  age,	
  and	
  education,	
  as	
  well	
  as	
  the	
  worker’s	
  tenure	
  at	
  
the	
   current	
   firm.	
   The	
   model	
   also	
   includes	
   individual	
   fixed	
   effects,	
   α ,	
   which	
   allow	
   us	
   to	
  
                                                                                                                                                    i




control	
   for	
   any	
   are	
   time-­‐invariant	
   unobservable	
   worker	
   characteristics,	
   and	
   firm	
   fixed	
  
                                                                                           €

effects,	
   ! j(i,t ) ,	
   for	
   firm	
   j	
   at	
   which	
   worker	
   i	
   is	
   employed	
   at	
   time	
   t,	
   representing	
   firm	
  

heterogeneity.	
   	
   Equation	
   (2)	
   could	
   then	
   be	
   augmented	
   to	
   include	
   sector	
   level	
   measures	
   of	
  
protection	
   interacted	
   with	
   the	
   firm’s	
   export	
   status	
   in	
   order	
   to	
   test	
   the	
   contribution	
   of	
  
unobservable	
  differences	
  across	
  workers	
  on	
  the	
  differential	
  response	
  to	
  trade	
  reform.	
  	
  
	
  
	
  
	
  
4.1	
  Endogenous	
  Worker	
  Mobility	
  
	
  
It	
   is	
   important	
   to	
   note	
   that	
   the	
   identifying	
   assumption	
   for	
   equation	
   (2)	
   is	
   that	
   the	
  
idiosyncratic	
  disturbance	
  term	
  in	
  each	
  period	
  is	
  independent	
  of	
  observable	
  worker	
  and	
  firm	
  
characteristics	
  as	
  well	
  as	
  firm	
  and	
  worker	
  fixed-­‐effects.	
  
	
  
   E(!it | "i , ! j(i,t ), Xit , Z jt ) = 0 	
  
	
  
Often	
   referred	
   to	
   in	
   the	
   literature	
   as	
   the	
   assumption	
   of	
   “conditional	
   exogenous	
   mobility”	
  
(see,	
  for	
  instance,	
  Abowd,	
  Kramarz,	
  and	
  Margolis	
  (1999),	
  Woodcock	
  (2008)	
  and	
  Soerensen	
  
and	
   Vejlin	
   (2009)),	
   the	
   assumption	
   implies	
   that	
   employment	
   mobility	
   could	
   depend	
   on	
  
time-­‐varying	
   observable	
   worker	
   or	
   firm	
   characteristics	
   and	
   firm	
   and	
   worker	
   fixed	
   effects,	
  
but	
  not	
   !it .	
  This	
  assumption	
  is	
  at	
  odds	
  with	
  many	
  well-­‐known	
  models	
  of	
  the	
  labor	
  market	
  

with	
   directed	
   search,	
   learning,	
   or	
   coordination	
   frictions.	
   For	
   example,	
   as	
   we	
   have	
   discussed	
  
before,	
   in	
   Helpman,	
   Itskhoki,	
   and	
   Redding	
   (2010),	
   workers	
   are	
   ex	
   ante	
   identical	
   and	
   job	
  




                                                                                                                                                                                       20
	
  
allocation	
  is	
  determined	
  on	
  the	
  basis	
  of	
  match	
  specific	
  ability	
  that	
  is	
  heterogenous	
  ex	
  post.	
  
Furthermore,	
   high	
   productivity	
   firms	
   (exporters)	
   screen	
   more	
   intensively	
   resulting	
   in	
  
higher	
   quality	
   firm-­‐worker	
   matches.	
   In	
   this	
   case	
   the	
   estimates	
   in	
   our	
   firm-­‐level	
   analyses	
  
will	
  be	
  biased,	
  due	
  to	
  an	
  omitted	
  worker-­‐firm	
  match	
  effect.	
  Similarly,	
  if	
  workers	
  with	
  certain	
  
observable	
  characteristics	
  are	
  more	
  successful	
  at	
  generating	
  good	
  matches	
  than	
  others	
  (for	
  
example,	
   because	
   the	
   return	
   from	
   search	
   is	
   higher	
   or	
   due	
   to	
   learning)	
   and	
   hence	
   higher	
  
wages,	
   omitted	
   match	
   effects	
   could	
   also	
   bias	
   the	
   estimated	
   returns	
   to	
   observable	
  
characteristics	
  in	
  equation	
  (2).	
  	
  
	
  
We	
   test	
   the	
   validity	
   of	
   the	
   exogenous	
   mobility	
   assumption	
   using	
   the	
   “match	
   effects	
   test”	
  
introduced	
   by	
   Abowd,	
   McKinney,	
   and	
   Schmutte	
   (2010).	
   The	
   test	
   statistic	
   is	
   based	
   on	
  
estimated	
   match	
   effects	
   that	
   are	
   computed	
   from	
   the	
   average	
   (over	
   time)	
   residual	
   for	
   a	
  
person	
   i	
   at	
   a	
   firm	
   j.	
   The	
   test	
   rests	
   on	
   the	
   logic	
   that	
   the	
   match	
   effect,	
   under	
   the	
   null	
   of	
  
exogenous	
   mobility,	
   should	
   not	
   predict	
   the	
   transitions	
   of	
   workers	
   between	
   firms.	
  

Specifically,	
   under	
   exogenous	
   mobility,	
   an	
   individual’s	
   average	
   residual,	
   !ijt!1 	
   (within	
  

quintiles	
   of	
   the	
   residual	
   distribution)	
   from	
   the	
   most	
   recently	
   completed	
   job	
   should	
   not	
  
predict	
   the	
   transition	
   across	
   firms	
   with	
   heterogeneous	
   values	
   of	
   ! j(i,t ) (say	
   from	
   a	
  

particular	
  quintile	
  of	
  the	
   ! j(i,t ) 	
  distribution	
  to	
  another).	
  The	
  test	
  is	
  implemented	
  as	
  follows.	
  

First,	
   equation	
   (2)	
   is	
   estimated	
   for	
   the	
   full	
   sample	
   of	
   workers.	
   Then,	
   for	
   workers	
   who	
  
switched	
   employers	
   between	
   time	
   t	
   and	
   t-­‐1,	
   the	
   average	
   residual	
   within	
   person	
   and	
   firm	
  	
  	
  	
  	
  	
  

( !ijt!1 )	
   is	
   calculated	
   across	
   all	
   years	
   throughout	
   the	
   duration	
   of	
   the	
   match	
   (i.e.,	
   until	
   t-­‐1);	
  

 !ijt!1 	
   represents	
   the	
   match	
   effect	
   for	
   the	
   firm	
   and	
   worker	
   pair.	
   Under	
   the	
   null	
   hypothesis,	
  

the	
  transition	
  rates	
  across	
  quintiles	
  of	
  the	
  firm	
  effects	
  distribution,	
  from	
  the	
   ! j(i,t"1) 	
  quintile	
  

to	
   the	
   ! j(i,t ) 	
   quintile,	
   are	
   independent	
   of	
   !ijt!1 .	
   Importantly,	
   if	
   the	
   null	
   is	
   rejected	
   for	
   our	
  

data,	
   this	
   would	
   suggest	
   that	
   the	
   estimation	
   results	
   from	
   equation	
   (2)	
   are	
   biased	
   and	
   that	
  
firm-­‐worker	
   match	
   effects	
   should	
   be	
   included	
   in	
   the	
   specification	
   to	
   account	
   for	
   the	
  
endogenous	
  mobility	
  of	
  workers.	
  
	
  
In	
  constructing	
  the	
  test	
  statistic,	
  we	
  restrict	
  the	
  sample	
  to	
  those	
  workers	
  with	
  at	
  least	
  two	
  
years	
  of	
  data	
  and	
  to	
  firms	
  in	
  the	
  manufacturing	
  sector.	
  Since	
  estimates	
  of	
  firm	
  fixed	
  effects	
  
for	
   firms	
   with	
   only	
   few	
   movers	
   are	
   likely	
   to	
   be	
   imprecise,	
   Abowd,	
   Kramarz	
   and	
   Margolis	
  



                                                                                                                                                                             21
	
  
(1999)	
   suggests	
   grouping	
   small	
   firms	
   together	
   and	
   estimating	
   one	
   ! j(i,t ) 	
   for	
   these	
   firms	
  

with	
   few	
   movers.	
   We	
   calculate	
   the	
   test	
   statistic	
   both	
   by	
   grouping	
   small	
   firms	
   with	
   less	
   than	
  
two	
  movers	
  into	
  one	
  firm	
  and	
  by	
  excluding	
  these	
  small	
  firms	
  from	
  the	
  sample.	
  In	
  both	
  cases,	
  
we	
  conduct	
  the	
  test	
  on	
  the	
  largest	
  mobility	
  group,	
  since	
  the	
  firm	
  fixed	
  effects	
  estimated	
  for	
  
unconnected	
   groups	
   are	
   not	
   directly	
   comparable	
   with	
   each	
   other.35	
   Consistent	
   with	
   the	
  
Abowd,	
   McKinney,	
   and	
   Schmutte	
   (2010)	
   finding	
   for	
   the	
   U.S.,	
   the	
   test	
   statistic	
   strongly	
  
rejects	
  the	
  null	
  hypothesis	
  of	
  exogenous	
  mobility	
  for	
  the	
  sample	
  of	
  job	
  switchers	
  in	
  our	
  data.	
  
                                                                                                                                                                                                                                       2
The	
  match	
  effects	
  test	
  statistic	
  has	
  a	
  value	
   ! = 25, 000 	
  and	
  is	
  distributed	
  chi-­‐sq	
  with	
  496	
  

of	
  degrees	
  of	
  freedom	
  (and	
  thus	
  a	
  p-­‐value	
  of	
  0.000).36	
  	
  This	
  finding	
  confirms	
  the	
  relevance	
  of	
  
models	
   of	
   labor	
   allocation	
   involving	
   search	
   dynamics	
   or	
   other	
   frictions	
   (such	
   as	
   Helpman,	
  
Itskhoki	
   and	
   Redding	
   (2010)	
   and	
   Davidson,	
   Matusz	
   and	
   Schevchenko	
   (2008))	
   and	
  
highlights	
   the	
   importance	
   of	
   allowing	
   for	
   the	
   possibility	
   of	
   firm-­‐worker	
   match	
  
heterogeneity	
  in	
  wage	
  determination	
  to	
  account	
  for	
  the	
  endogenous	
  mobility	
  of	
  workers.	
  
	
  
To	
   further	
   emphasize	
   this	
   point,	
   we	
   plot,	
   in	
   Figure	
   4.1,	
   the	
   conditional	
   distribution	
   of	
  
quintiles	
   of	
   the	
   firm	
   fixed	
   effects	
   for	
   the	
   previous	
   job,	
   ! j(i,t"1) ,	
   given	
   the	
   quintiles	
   of	
   the	
  

individual’s	
   average	
   residual	
   from	
   the	
   most	
   recently	
   completed	
   job,	
   for	
   the	
   sample	
   of	
   job	
  
changers.	
   Under	
   the	
   assumption	
   of	
   exogenous	
   mobility,	
   the	
   distribution	
   of	
   ! j(i,t"1) 	
   should	
  

not	
   show	
   any	
   variation	
   across	
   quintiles	
   of	
   the	
   average	
   residual.	
   That	
   is	
   to	
   say,	
   the	
  
estimation	
  strategy	
  requires	
  that	
  the	
  quality	
  of	
  the	
  firm-­‐worker	
  match	
  in	
  the	
  previous	
  job	
  
should	
   not	
   contain	
   any	
   information	
   about	
   the	
   estimated	
   firm-­‐fixed	
   effects	
   for	
   that	
   job.	
  
Figure	
  4.1	
  clearly	
  demonstrates	
  that	
  this	
  is	
  not	
  the	
  case	
  in	
  our	
  data.	
  For	
  example,	
  while	
  the	
  
job	
  changers	
  in	
  the	
  first	
  quintile	
  of	
  the	
  match	
  effect	
  (residual)	
  distribution	
  mostly	
  originate	
  
from	
  the	
  second	
  quintile	
  of	
  the	
   ! j(i,t"1) 	
  distribution,	
  most	
  switchers	
  in	
  the	
  second	
  quintile	
  

of	
  the	
  match	
  effect	
  originate	
  from	
  the	
  first	
  quintile	
  of	
  the	
   ! j(i,t"1) 	
  distribution.	
  	
  	
  
	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
35	
   A	
   connected	
   group	
   includes	
   all	
   the	
   workers	
   who	
   have	
   ever	
   worked	
   for	
   any	
   of	
   the	
   firms	
   in	
   that	
  

group	
   as	
   well	
   as	
   all	
   the	
   firms	
   at	
   which	
   any	
   of	
   these	
   workers	
   were	
   employed	
   at	
   during	
   the	
   sample	
  
period.	
   Since	
   within	
   each	
   group	
   the	
   mean-­‐deviated	
   firm	
   fixed	
   effects	
   sum	
   to	
   zero,	
   the	
   estimates	
   of	
  
! !!" "# # 	
   are	
   not	
   directly	
   comparable	
   across	
   unconnected	
   groups.	
   As	
   a	
   possible	
   solution,	
   one	
   can	
  
normalize	
  the	
  fixed	
  effects	
  so	
  they	
  have	
  the	
  same	
  mean	
  across	
  groups	
  (Abowd,	
  Creecy,	
  and	
  Kramarz	
  
(2002)).	
  Our	
  results	
  are	
  robust	
  to	
  using	
  the	
  full	
  sample	
  of	
  workers	
  and	
  correcting	
  the	
  fixed	
  effects	
  for	
  
the	
  unconnected	
  groups	
  in	
  this	
  fashion.	
  	
  
36	
  The	
  degrees	
  of	
  freedom	
  are	
  calculated	
  as:	
   !"!!!                                                                                                                                                            #$"!!" #!" %$!&# #$"!!" #!" %$ # #!&#$"!!#"$!& #!&# =(5*5*5-­‐
                                                                                                                                                                                                                                   "

1)*(5-­‐1)	
  where	
  #Q	
  denotes	
  the	
  number	
  of	
  quintiles.	
  



                                                                                                                                                                                                                                                                                                        22
	
  
	
  
In	
  Figure	
  4.2	
  we	
  plot	
  the	
  transition	
  rates	
  from	
  a	
  job	
  in	
  a	
  given	
   ! j(i,t"1) 	
  quintile	
  to	
  a	
  job	
  in	
  

   ! j(i,t ) 	
   quintile,	
   again	
   for	
   the	
   sample	
   of	
   job	
   changers.	
   Here,	
   we	
   find	
   strong	
   evidence	
   that	
   job	
  
transitions	
   are	
   not	
   random;	
   most	
   workers	
   move	
   between	
   jobs	
   within	
   the	
   same	
   employer-­‐
effect	
   quintile,	
   which	
   is	
   evident	
   from	
   the	
   rightward	
   movement	
   of	
   the	
   peak	
   of	
   the	
   ! j(i,t ) 	
  

distribution	
   with	
   higher	
   quintiles	
   of	
   the	
   original	
   job.	
   Moreover,	
   Figure	
   4.3	
   clearly	
   illustrates	
  
that	
   these	
   transition	
   probabilities	
   vary	
   considerably	
   by	
   different	
   quintiles	
   of	
   the	
   match	
  
effect	
  distribution.	
  Figures	
  4.3a,	
  4.3b,	
  and	
  4.3c	
  plot	
  the	
  transition	
  probabilities	
  for	
  the	
  first,	
  
third	
   and	
   fifth	
   quintiles	
   respectively,	
   of	
   the	
   residual	
   distribution.	
   For	
   example,	
   while	
   job	
  
switchers	
   are	
   much	
   more	
   likely	
   to	
   improve	
   their	
   employer	
   effect	
   in	
   the	
   middle	
   and	
   (to	
   a	
  
lesser	
  extent)	
  in	
  the	
  higher	
  end	
  of	
  the	
  residual	
  distribution,	
  we	
  find	
  this	
  not	
  to	
  be	
  the	
  case	
  in	
  
the	
   first	
   quintile	
   of	
   the	
   match	
   effect	
   distribution.	
   	
   This	
   illustrates	
   the	
   failure	
   of	
   the	
  
exogenous	
  mobility	
  assumption	
  as	
  the	
  estimated	
  match	
  effects	
  clearly	
  contain	
  information	
  
on	
  job-­‐to-­‐job	
  transitions	
  that	
  take	
  place	
  in	
  the	
  data.	
  	
  
               	
  
Our	
  findings	
  on	
  endogenous	
  worker	
  mobility	
  have	
  interesting	
  implications	
  for	
  the	
  analysis	
  
of	
   trade	
   and	
   labor	
   markets,	
   especially	
   since	
   the	
   workings	
   of	
   the	
   labor	
   market	
   have	
   been	
  
modeled	
  in	
  a	
  number	
  of	
  different	
  ways	
  in	
  the	
  recent	
  literature	
  on	
  international	
  trade	
  with	
  
heterogeneous	
   firms.	
   In	
   the	
   well-­‐known	
   paper	
   of	
   Melitz	
   (2003),	
   heterogeneous	
  
monopolistically	
   competitive	
   firms	
   pay	
   their	
   (homogeneous)	
   workers	
   an	
   identical	
   wage,	
  
with	
  the	
  assignment	
  of	
  particular	
  workers	
  to	
  firms	
  being	
  (effectively)	
  random.	
  Similarly,	
  in	
  
Egger	
   and	
   Kreickemeier	
   (2009)	
   and	
   Amiti	
   and	
   Davis	
   (forthcoming),	
   the	
   assignment	
   of	
  
workers	
   to	
   firms	
   is	
   random,	
   even	
   if	
   rent-­‐sharing	
   behavior	
   of	
   firms	
   implies	
   that	
   identical	
  
workers	
   may	
   be	
   paid	
   different	
   wages,	
   based	
   on	
   the	
   profits	
   of	
   the	
   firms	
   in	
   which	
   they	
   are	
  
employed.	
   The	
   decisive	
   rejection	
   of	
   the	
   assumption	
   of	
   exogenous	
   worker	
   mobility	
   in	
   our	
  
data	
   supports	
   a	
   picture	
   of	
   the	
   labor	
   market	
   that	
   is	
   closer	
   in	
   line	
   with	
   the	
   framework	
   of	
  
Helpman,	
  Itskhoki,	
  and	
  Redding	
  (2010),	
  where,	
  as	
  we	
  have	
  previously	
  discussed,	
  the	
  labor	
  
allocation	
   process	
   is	
   subject	
   to	
   search	
   and	
   matching	
   frictions	
   and	
   the	
   equilibrium	
  
assignment	
  of	
  workers	
  to	
  firms	
  is	
  clearly	
  non-­‐random.	
  	
  
	
  
4.2	
  Estimation	
  with	
  Match	
  Effects	
  
	
  




                                                                                                                                                                       23
	
  
If	
   match-­‐specific	
   effects	
   are	
   important	
   in	
   wage	
   determination,	
   the	
   Abowd,	
   Kramarz,	
   and	
  
Margolis	
   (1999)	
   specification	
   in	
   equation	
   (2)	
   including	
   only	
   worker	
   and	
   firm	
   fixed	
   effects	
  
will	
  result	
  in	
  both	
  biased	
  estimates	
  of	
  these	
  fixed	
  effects,	
  as	
  well	
  as	
  biased	
  estimates	
  of	
  the	
  
returns	
  to	
  observable	
  worker	
  and	
  firm	
  characteristics	
  (Woodcock	
  (2008)).	
  For	
  example,	
  if	
  
more	
  experienced	
  workers	
  are	
  likely	
  to	
  draw	
  better	
  matches,	
  omission	
  of	
  the	
  match	
  effect	
  
will	
   result	
   in	
   an	
   over	
   estimation	
   of	
   the	
   returns	
   to	
   experience.	
   In	
   the	
   context	
   of	
   the	
  
international	
   trade	
   literature,	
   if	
   the	
   labor	
   market	
   functions	
   in	
   the	
   manner	
   described	
   by	
  
Helpman,	
  Itskhoki	
  and	
  Redding	
  (2010),	
  the	
  screening	
  thresholds	
  for	
  match-­‐specific	
  ability	
  
will	
  be	
  different	
  in	
  the	
  resulting	
  equilibrium	
  after	
  liberalization,	
  shifting	
  the	
  distribution	
  of	
  
worker	
   abilities	
   (i.e.,	
   the	
   quality	
   of	
   matches)	
   within	
   each	
   firm.	
   Given	
   that	
   this	
   shift	
   varies	
  
systematically	
  with	
  the	
  export	
  status	
  of	
  the	
  firm,	
  not	
  controlling	
  for	
  match	
  quality	
  will	
  result	
  
in	
   biased	
   estimates	
   of	
   the	
   differential	
   effect	
   of	
   trade	
   liberalization	
   on	
   wages	
   in	
   exporting	
  
firms.	
  
	
  
To	
  account	
  for	
  the	
  fact	
  that	
  a	
  worker’s	
  employment	
  history	
  may	
  not	
  be	
  independent	
  of	
  the	
  
idiosyncratic	
   part	
   of	
   the	
   residual	
   in	
   equation	
   (2),	
   we	
   now	
   consider	
   a	
   more	
   elaborate	
  
specification	
  of	
  wages,	
  in	
  which	
  worker-­‐firm	
  match	
  fixed	
  effects	
  (or	
  spell	
  fixed	
  effects),	
  Mij,	
  
which	
   denote	
   a	
   given	
   worker	
   i’s	
   employment	
   at	
   a	
   given	
   firm	
   j,	
   are	
   included	
   on	
   the	
   right	
  
hand	
  side:	
  	
  
	
  

  !" !"#$ = !# $$ %$ + ! % $$ %$ & &'( #$ + ! ' $)&)$ & &'( #$ + ! ( $&'( #$ + *"# + "$+ + # , "$ + $ - #$ + %"#$ 	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  (3)	
  

	
  
Note	
  that	
  since	
  for	
  each	
  spell	
  of	
  a	
  worker	
  within	
  a	
  firm	
  neither	
  the	
  worker	
  nor	
  the	
  firm	
  fixed	
  
effect	
  varies,	
  inclusion	
  of	
  match	
  fixed	
  obviates	
  the	
  need	
  for	
  separate	
  inclusion	
  of	
  worker	
  and	
  
firm	
  fixed	
  effects.	
  	
  
	
  
4.3	
  Estimation	
  Results	
  	
  
	
  
Table	
   4.1	
   includes	
   estimation	
   results	
   from	
   equation	
   (3)	
   with	
   spell	
   fixed	
   effects	
   with	
   both	
  
tariffs	
  and	
  ERP	
  as	
  the	
  measure	
  of	
  protection.	
  Column	
  1	
  presents	
  results	
  with	
  tariffs	
  taken	
  as	
  
the	
  measure	
  of	
  trade	
  protection	
  in	
  the	
  sector.	
  Our	
  results	
  suggest	
  that	
  the	
  inclusion	
  of	
  spell	
  

fixed	
  effects	
  results	
  in	
  insignificant	
  estimates	
  of	
  both	
   ! ! " and	
   ! ! " .	
  The	
  insignificant	
  effect	
  of	
  a	
  

decline	
   in	
   protection	
   on	
   both	
   exporters	
   and	
   firms	
   serving	
   the	
   domestic	
   market	
   also	
   holds	
  



                                                                                                                                                                                                    24
	
  
separately	
   for	
   the	
   high	
   and	
   low	
   comparative	
   advantage	
   sectors,	
   although	
   the	
   interaction	
  
term	
   is	
   more	
   than	
   three	
   times	
   higher	
   in	
   magnitude	
   for	
   the	
   high	
   comparative	
   advantage	
  
sector.	
   These	
   results	
   continue	
   to	
   hold	
   when	
   we	
   use	
   effective	
   rate	
   of	
   protection	
   in	
   the	
   sector	
  
as	
  our	
  measure	
  of	
  trade	
  protection.	
  	
  
	
  
In	
  Table	
  4.2	
  we	
  present	
  a	
  wide	
  array	
  of	
  robustness	
  checks	
  we	
  have	
  conducted.	
  The	
  first	
  two	
  
columns	
  report	
  estimation	
  results	
  from	
  specification	
  (3)	
  for	
  various	
  alternative	
  samples.	
  In	
  
column	
   1,	
   we	
   use	
   a	
   5%	
   random	
   sample	
   of	
   males	
   in	
   metropolitan	
   areas	
   instead	
   of	
   a	
   1%	
  
random	
   sample	
   of	
   the	
   full	
   population.	
   In	
   column	
   2	
   we	
   restrict	
   our	
   analysis	
   to	
   only	
   the	
  
liberalization	
   period	
   (1990-­‐1994)	
   during	
   which	
   average	
   ERP	
   was	
   decreasing.	
   Our	
   results	
  
are	
   robust	
   to	
   these	
   alternative	
   samples	
   and	
   to	
   using	
   an	
   industry-­‐specific	
   real	
   exchange	
   rate	
  
measure	
   as	
   in	
   column	
   3,	
   instead	
   of	
   an	
   economy-­‐wide	
   measure.	
   Next,	
   we	
   test	
   whether	
   our	
  
result	
  of	
  insignificant	
  interaction	
  coefficient	
  is	
  sensitive	
  to	
  the	
  exporting	
  thresholds	
  we	
  use	
  
to	
  assign	
  the	
  indicator	
  variable	
  denoting	
  a	
  firm’s	
  export	
  status.	
  In	
  the	
  baseline	
  specifications	
  
a	
  firm	
  is	
  defined	
  as	
  an	
  exporter	
  if	
  it	
  exported	
  a	
  positive	
  dollar	
  amount	
  that	
  year.	
  Instead,	
  in	
  
columns	
  5	
  and	
  6,	
  we	
  only	
  consider	
  firms	
  with	
  an	
  export	
  value	
  more	
  than	
  the	
  5th	
  percentile	
  
(cutoff	
   5),	
   and	
   more	
   than	
   the	
   10th	
   percentile	
   (cutoff	
   10)	
   as	
   exporting	
   firms,	
   respectively.	
  
While	
   the	
   magnitude	
   of	
   the	
   interaction	
   term	
   increases	
   slightly	
   as	
   we	
   increase	
   the	
   cut	
   off	
  
threshold,	
   the	
   coefficient	
   remains	
   statistically	
   insignificantly	
   different	
   than	
   zero.	
   In	
   the	
   next	
  
two	
   columns	
   we	
   consider	
   two	
   time-­‐invariant	
   measures	
   of	
   export	
   status.	
   In	
   column	
   7,	
   a	
   firm	
  
is	
  defined	
  as	
  an	
  exporter	
  if	
  it	
  exported	
  a	
  positive	
  dollar	
  value	
  at	
  the	
  beginning	
  of	
  our	
  sample	
  
at	
   1990	
   (Export90).	
   The	
   export	
   status	
   variable	
   in	
   column	
   8	
   takes	
   the	
   value	
   1	
   if	
   the	
   firm	
  
reported	
   positive	
   exports	
   during	
   every	
   year	
   between	
   90	
   and	
   98.	
   Finally,	
   when	
   we	
   use	
   the	
  
magnitude	
   of	
   exports	
   for	
   the	
   sub-­‐sample	
   of	
   exporting	
   firms	
   to	
   represent	
   the	
   relevance	
   of	
  

exports	
   to	
   the	
   firm	
   instead,	
   we	
   obtain	
   yet	
   again	
   estimates	
   of	
   ! ! " and	
   ! ! " 	
   that	
   are	
  

insignificantly	
  different	
  from	
  zero.37	
  
	
  
Our	
  findings	
  using	
  worker	
  level	
  data	
  and	
  taking	
  endogenous	
  worker	
  mobility	
  into	
  account,	
  
suggests	
   an	
   insignificant	
   differential	
   effect	
   of	
   trade	
   policy	
   on	
   the	
   wages	
   of	
   workers	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
37	
   Other	
   robustness	
   checks	
   we	
   have	
   conducted	
   but	
   we	
   do	
   not	
   reported	
   to	
   conserve	
   space	
   include	
  

defining	
   a	
   firm	
   as	
   exporter	
   if	
   the	
   firm	
   exported	
   only	
   to	
   non-­‐Mercosur	
   countries	
   (and	
   omitting	
  
exporters	
  to	
  Mercosur	
  countries	
  from	
  the	
  sample)	
  and	
  using	
  a	
  more	
  detailed	
  industry	
  classification	
  
available	
   only	
   for	
   the	
   1994-­‐1998	
   period,	
   which	
   allows	
   us	
   to	
   identify	
   changes	
   in	
   industry	
   level	
   ERP	
   at	
  
a	
  more	
  disaggregated	
  level.	
  We	
  find	
  our	
  main	
  conclusions	
  to	
  be	
  robust.	
  	
  



                                                                                                                                                                                                                                   25
	
  
employed	
   in	
   exporting	
   firms.	
   This	
   finding	
   stands	
   in	
   sharp	
   contrast	
   to	
   results	
   obtained	
   using	
  
average	
   firm	
   level	
   wages	
   instead.	
   One	
   reason	
   for	
   this	
   difference	
   is	
   simply	
   that	
   the	
   use	
   of	
  
detailed	
  worker	
  level	
  data	
  allows	
  us	
  to	
  take	
  into	
  full	
  account	
  any	
  changes	
  in	
  the	
  composition	
  
of	
   the	
   workforce	
   (by	
   controlling	
   for	
   both	
   observable	
   and	
   time-­‐invariant	
   unobservable	
  
worker	
   characteristics)	
   following	
   trade	
   policy.	
   Furthermore,	
   by	
   taking	
   worker-­‐firm	
   match	
  
effects	
  into	
  account,	
  we	
  are	
  able	
  to	
  control	
  for	
  changes	
  in	
  the	
  composition	
  of	
  firms’	
  match	
  
quality	
   following	
   trade	
   policy	
   changes.	
   If,	
   following	
   trade	
   liberalization,	
   exporting	
   firms	
  
improve	
   their	
   average	
   match	
   quality	
   by	
   hiring	
   workers	
   with	
   better	
   match	
   quality,	
   the	
  

estimates	
   of	
   the	
   coefficient	
   ! ! " without	
   controlling	
   for	
   match	
   effects	
   would	
   mistakenly	
   be	
  

estimated	
  as	
  significant	
  even	
  in	
  the	
  absence	
  of	
  any	
  true	
  effect.	
  	
  
	
  
Table	
   4.3,	
   which	
   compares	
   the	
   changes	
   in	
   the	
   distribution	
   of	
   estimated	
   match	
   effects	
   in	
  
exporting	
  firms	
  relative	
  to	
  non-­‐exporting	
  firms,	
  confirms	
  this	
  point.38	
  Comparing	
  the	
  period	
  
1998	
  to	
  1990,	
  we	
  see	
  an	
  increase	
  in	
  both	
  the	
  mean	
  and	
  the	
  median	
  match	
  effect	
  in	
  exporting	
  
firms,	
   while	
   the	
   mean	
   and	
   median	
   match	
   effect	
   both	
   fall	
   in	
   non-­‐exporting	
   firms.	
   As	
   we	
   have	
  
noted	
  earlier,	
  match	
  fixed	
  effects	
  in	
  equation	
  (3)	
  absorb	
  both	
  worker	
  and	
  firm	
  fixed	
  effects	
  
in	
  addition	
  to	
  time-­‐invariant	
  match	
  quality	
  of	
  a	
  given	
  employment	
  spell.	
  Consequently,	
  our	
  
finding	
   of	
   an	
   improvement	
   in	
   average	
   match	
   effect	
   in	
   exporting	
   firms	
   between	
   1990	
   and	
  
1998	
  summarizes	
  the	
  combined	
  effect	
  of	
  a	
  change	
  in	
  workforce	
  composition	
  in	
  these	
  firms	
  
in	
  terms	
  of	
  improvement	
  in	
  worker	
  quality	
  in	
  time	
  invariant	
  worker	
  specific	
  characteristics,	
  
such	
  as	
  innate	
  ability	
  (captured	
  by	
  worker	
  fixed	
  effects)	
  and	
  improvement	
  in	
  the	
  quality	
  of	
  
the	
   worker-­‐firm	
   matches	
   (captured	
   by	
   match-­‐fixed	
   effects).39	
   The	
   improvement	
   in	
   the	
  
distribution	
   of	
   match-­‐specific	
   ability	
   in	
   exporting	
   firms	
   is	
   consistent	
   with	
   the	
   Helpman,	
  
Itskhoki,	
   Redding	
   (2010)	
   model	
   which	
   predicts	
   that	
   with	
   trade	
   liberalization,	
   	
   exporting	
  
firms	
   will	
   screen	
   more	
   intensively	
   and	
   set	
   a	
   higher	
   ability	
   threshold	
   for	
   employment.	
   It	
  is	
  
also	
   roughly	
   in	
   line	
   with	
   the	
   prediction	
   of	
   Davidson,	
   Matusz,	
   and	
   Shevchenko	
   (2008)	
   that	
  
greater	
  openness	
  leads	
  to	
  better	
  labor	
  market	
  sorting	
  of	
  higher	
  ability	
  workers	
  into	
  higher	
  
technology	
   firms.	
   This	
   finding	
   serves	
   to	
   explain	
   the	
   difference	
   between	
   the	
   results	
   at	
   the	
  
firm	
   level	
   and	
   those	
   at	
   the	
   worker	
   level:	
   If	
   average	
   quality	
   of	
   the	
   workforce	
   (in	
   terms	
   of	
  
match	
   specific	
   ability	
   or	
   innate	
   (time-­‐invariant)	
   ability)	
   improves	
   systematically	
   in	
  
	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
38	
  Table	
  4.3	
  is	
  for	
  a	
  group	
  of	
  firms,	
  which	
  were	
  included	
  in	
  our	
  sample	
  in	
  both	
  1990	
  and	
  1998	
  and	
  did	
  

not	
  switch	
  export	
  status	
  during	
  this	
  period.	
  	
  
39	
   Note	
   that	
   firm-­‐fixed	
   effect	
   for	
   a	
   given	
   firm	
   is	
   constant	
   across	
   time	
   and	
   cannot	
   account	
   for	
   the	
  

improvement	
  in	
  match	
  quality	
  between	
  1990	
  and	
  1998.	
  	
  



                                                                                                                                                                                                                                   26
	
  
exporting	
   firms	
   following	
   trade	
   liberalization,	
   not	
   controlling	
   for	
   match	
   effects	
   (as	
   is	
   the	
  
case	
   in	
   firm-­‐level	
   analysis),	
   will	
   incorrectly	
   suggest	
   that	
   trade	
   liberalization	
   leads	
   to	
   a	
  
differential	
   wage	
   improvement	
   for	
   workers	
   at	
   exporting	
   firms	
   even	
   when	
   this	
   is	
   not	
   the	
  
case.	
  
	
  
A	
  comparison	
  of	
  estimates	
  obtained	
  from	
  specifications	
  with	
  and	
  without	
  spell	
  fixed	
  effects	
  
suggests	
  that	
  match	
  effects	
  matter	
  both	
  qualitatively	
  and	
  quantitatively.	
  Table	
  4.4	
  compares	
  
estimates	
   obtained	
   from	
   alternate	
   specifications	
   with	
   only	
   firm	
   fixed	
   effects,	
   with	
   both	
  
worker	
   and	
   firm	
   fixed	
   effects	
   included,	
   and	
   finally	
   with	
   match	
   fixed	
   effects	
   included.	
   Note	
  
that	
   this	
   comparison	
   can	
   only	
   be	
   made	
   for	
   the	
   sample	
   of	
   workers	
   who	
   switch	
   jobs	
   during	
  
this	
   period,	
   as	
   worker	
   effects	
   cannot	
   be	
   separately	
   identified	
   from	
   match	
   effects	
   for	
   those	
  
workers	
   who	
   do	
   not.	
   As	
   expected,	
   the	
   inclusion	
   of	
   worker	
   fixed	
   effects	
   lowers	
   (absolute	
  

value	
  of)	
  the	
  magnitude	
  of	
   ! ! " 	
  from	
  0.242	
  to	
  0.151	
  for	
  tariffs,	
  and	
  from	
  0.086	
  to	
  0.050	
  for	
  

ERP	
   as	
   the	
   measure	
   of	
   protection.	
   The	
   inclusion	
   of	
   match	
   effects	
   lowers	
   the	
   coefficient	
  
further	
   from	
   0.151	
   to	
   0.104	
   and	
   0.050	
   to	
   0.018,	
   when	
   the	
   measure	
   of	
   protection	
   is	
   tariffs	
  
and	
  ERP,	
  respectively.	
  	
  
             	
  

Finally,	
   Table	
   4.5	
   reports	
   estimates	
   of	
   ! ! " and	
   ! ! " 	
   when	
   workers	
   with	
   different	
   levels	
   of	
  

human	
   capital	
   (grouped	
   into	
   two	
   educational	
   categories	
   –	
   high	
   education	
   and	
   low	
  
education)	
   are	
   examined	
   separately.	
   Interestingly,	
   there	
   are	
   heterogeneous	
   effects	
   across	
  
the	
  two	
  educational	
  groups.	
  We	
  find	
  that	
  the	
  relative	
  wages	
  of	
  workers	
  with	
  higher	
  levels	
  of	
  
education	
   employed	
   in	
   exporting	
   firms	
   improve	
   with	
   liberalization	
   compared	
   to	
   similar	
  
workers	
   employed	
   in	
   firms	
   serving	
   the	
   domestic	
   market.	
   One	
   explanation	
   for	
   this	
   finding	
   is	
  
that	
   the	
   rent	
   sharing	
   mechanism	
   we	
   have	
   discussed	
   earlier,	
   whereby	
   workers	
   share	
   a	
  
fraction	
   of	
   the	
   profits	
   made	
   by	
   the	
   firms,	
   may	
   be	
   relevant	
   in	
   the	
   case	
   of	
   high	
   education	
  
workers	
   even	
   it	
   remains	
   insignificant	
   in	
   workers	
   with	
   lower	
   levels	
   of	
   education.	
  
Alternatively,	
   the	
   improvement	
   in	
   the	
   quality	
   of	
   the	
   labor	
   force	
   in	
   exporting	
   firms,	
   could	
  
result	
   in	
   positive	
   productivity	
   spillovers	
   on	
   more	
   educated	
   workers	
   and	
   hence	
   higher	
  
wages.	
  	
  
	
  
	
  
	
  
	
  
	
  




                                                                                                                                                               27
	
  
5.	
  Conclusions	
  
                                                                                  	
  
In	
   this	
   paper,	
   we	
   use	
   a	
   linked	
   employer–employee	
   database	
   from	
   Brazil	
   to	
   examine	
   the	
  
impact	
   of	
   trade	
   reform	
   on	
   the	
   wages	
   of	
   workers	
   employed	
   at	
   heterogeneous	
   firms.	
   	
   Our	
  
analysis	
   of	
   the	
   data	
   at	
   the	
   firm-­‐level	
  confirms	
  earlier	
  findings	
  of	
  a	
  differential	
  positive	
  effect	
  
of	
   trade	
   liberalization	
   on	
   the	
   average	
   wages	
   at	
   exporting	
   firms	
   relative	
   to	
   non-­‐exporting	
  
firms.	
   However,	
   this	
   analysis	
   of	
   average	
   firm-­‐level	
   wages	
   is	
   incomplete	
   along	
   several	
  
dimensions.	
   First,	
   it	
   cannot	
   fully	
   account	
   for	
   the	
   impact	
   of	
   a	
   change	
   in	
   trade	
   barriers	
   on	
  
workforce	
  composition	
  especially	
  in	
  terms	
  of	
  unobservable	
  (time-­‐invariant)	
  characteristics	
  
of	
   workers	
   (innate	
   ability)	
   and	
   any	
   additional	
   productivity	
   that	
   obtains	
   in	
   the	
   context	
   of	
  
employment	
   in	
   the	
   specific	
   firm	
   (match	
   specific	
   ability).	
   Furthermore,	
   the	
   firm-­‐level	
  
analysis	
   is	
   undertaken	
   under	
   the	
   assumption	
   that	
   the	
   assignment	
   of	
   workers	
   to	
   firms	
   is	
  
random.	
   This	
   ignores	
   the	
   sorting	
   of	
   worker	
   into	
   firms	
   and	
   leads	
   to	
   a	
   bias	
   in	
   estimates	
   of	
   the	
  
differential	
   impact	
   of	
   trade	
   on	
   workers	
   at	
   exporting	
   firms	
   relative	
   to	
   non-­‐exporting	
   firms.	
  
Using	
  detailed	
  information	
  on	
  worker	
  and	
  firm	
  characteristics	
  to	
  control	
  for	
  compositional	
  
effects	
   and	
   using	
   firm-­‐worker	
   match	
   specific	
   effects	
   to	
   account	
   for	
   the	
   endogenous	
   mobility	
  
of	
   workers,	
   we	
   find	
   the	
   differential	
   effect	
   of	
   trade	
   openness	
   on	
   wages	
   in	
   exporting	
   firms	
  
relative	
   to	
   domestic	
   firms	
   to	
   be	
   insignificant.	
   Consistent	
   with	
   the	
   models	
   of	
   Helpman,	
  
Itskhoki,	
  and	
  Redding	
  (2010)	
  and	
  Davidson,	
  Matusz	
  and	
  Schevchenko	
  (2008),	
  we	
  also	
  find	
  
that	
   the	
   workforce	
   composition	
   improves	
   systematically	
   in	
   exporting	
   firms	
   in	
   terms	
   of	
  
innate	
  (time	
  invariant)	
  worker	
  ability	
  and	
  in	
  terms	
  the	
  quality	
  of	
  the	
  worker-­‐firm	
  matches.	
  
	
  




                                                                                                                                                                        28
	
  
References	
  	
  
	
  
Abowd,	
  John,	
  Kevin	
  McKinney,	
  and	
  Ian	
  M.	
  Schmutte,	
  2010.	
  “How	
  Important	
  is	
  Endogenous	
  
     Mobility	
   for	
   Measuring	
   Employer	
   and	
   Employee	
   Heterogeneity?,”	
   Mimeo,	
   November	
  
     2010.	
  
	
  
Abowd,	
   John,	
   Robert	
   Creecy,	
   and	
   Francis	
   Kramarz,	
   2002.	
   “Computing	
   Person	
   and	
   Firm	
  
     Effects	
   Using	
   Linked	
   Longitudinal	
   Employer–Employee	
   Data,”	
   U.S.	
   Census	
   Bureau	
  
     Technical	
  Paper	
  No.	
  2002-­‐06.	
  
	
  
Abowd,	
   John	
   M.,	
   Francis	
   Kramarz,	
   and	
   David	
   N.	
   Margolis,	
   1999.	
   “High	
   Wage	
   Workers	
   and	
  
     High	
  Wage	
  Firms,”	
  Econometrica	
  67	
  (2),	
  251–333.	
  
	
  
Abowd,	
   John.	
   M.,	
   Francis	
   Kramarz,	
   David	
   N.	
   Margolis,	
   and	
   Kenneth	
   R.	
   Troske,	
   2001.	
   “The	
  
     Relative	
   Importance	
   of	
   Employer	
   and	
   Employee	
   Effects	
   on	
   Compensation:	
   A	
  
     Comparison	
  of	
  France	
  and	
  the	
  United	
  States,”	
  Journal	
  of	
  the	
  Japanese	
  and	
  International	
  
     Economies	
  15,	
  419–436.	
  
	
  
Amiti,	
  Mary	
  and	
  Donald	
  R.	
  Davis,	
  “Trade,	
  Firms,	
  and	
  Wages:	
  Theory	
  and	
  Evidence,”	
  Review	
  
     of	
  Economic	
  Studies,	
  forthcoming.	
  
	
  
Amiti,	
   Mary	
   and	
   Jozef	
   Konings,	
   2007.	
   "Trade	
   Liberalization,	
   Intermediate	
   Inputs,	
   and	
  
     Productivity:	
  Evidence	
  from	
  Indonesia,"	
  American	
  Economic	
  Review	
  97	
  (5),	
  1611-­‐1638.	
   	
  
	
  
Arkolakis,	
   Costas,	
   Svetlana	
   Demidova,	
   Peter	
   J.	
   Klenow	
   and	
   Andres	
   Rodriguez-­‐Clare,	
   2008.	
  
     “Endogenous	
  Variety	
  and	
  the	
  Gains	
  from	
  Trade,”	
   American	
  Economic	
  Review	
   Papers	
  and	
  
     Proceedings,	
  98	
  (4),	
  444-­‐450.	
  	
  
	
  
Balassa,	
   Bela,	
   1965.	
   "Trade	
   Liberalization	
   and	
   'Revealed'	
   Comparative	
   Advantage,"	
  
     Manchester	
  School	
  of	
  Economic	
  and	
  Social	
  Studies,	
  33,	
  99-­‐123.	
  	
  
	
  
Bernard,	
   Andrew	
   B.	
   and	
   J.	
   Bradford	
   Jensen,	
   1995.	
   “Exporters,	
   Jobs	
   and	
   Wages	
   in	
   U.S.	
  
     Manufacturing,	
   1976-­‐87,”	
   Brookings	
   Papers	
   on	
   Economic	
   Activity:	
   Microeconomics,	
   67–
     112.	
  
	
  
Bernard,	
  Andrew	
  B.	
  and	
  J.	
  Bradford	
  Jensen,	
  1997.	
  "Exporters,	
  Skill	
  Upgrading,	
  and	
  the	
  Wage	
  
     Gap,"	
  Journal	
  of	
  International	
  Economics,	
  42	
  (1),	
  3-­‐31.	
   	
  
	
   	
  
Bernard,	
   Andrew	
   B.,	
   J.	
   Bradford	
   Jensen,	
   and	
   Peter	
   K.	
   Schott,	
   2009.	
   “Importers,	
   Exporters,	
  
     and	
   Multinationals:	
   	
   A	
   Portrait	
   of	
   Firms	
   in	
   the	
   U.S.	
   that	
   Trade	
   Goods,”	
   in	
   T	
   Dunne,	
   JB	
  
     Jensen,	
  MJ	
  Roberts	
  eds.,	
  Producer	
  Dynamics:	
  	
  New	
  Evidence	
  from	
  Micro	
  Data,	
  University	
  
     of	
  Chicago	
  Press.	
  
	
        	
  
Bustos,	
   Paula,	
   2011.	
   “Trade	
   Liberalization,	
   Exports,	
   and	
   Technology	
   Upgrading:	
   	
   Evidence	
  
     on	
   the	
   Impact	
   of	
   MERCOSUR	
   on	
   Argentinean	
   Firms,”	
   American	
   Economic	
   Review,	
   101	
  
     (1),	
  304-­‐340.	
   	
  
	
  
Csillag,	
  Márton	
  and	
  Miklós	
  Koren,	
  2009.	
  “Machines	
  and	
  Machinists:	
  The	
  Effect	
  of	
  Imported	
  
     Capital	
  on	
  Wage	
  Inequality,”	
  Central	
  European	
  University,	
  unpublished	
  manuscript.	
  	
  




                                                                                                                                                                  29
	
  
	
   	
  
Currie,	
   Janet	
   and	
   Ann	
   Harrison,	
   1997.	
   “Sharing	
   the	
   Costs:	
   The	
   Impact	
   of	
   Trade	
   Reform	
   on	
  
     Capital	
  and	
  Labor	
  in	
  Morocco,”	
  Journal	
  of	
  Labor	
  Economics,	
  15	
  (3),	
  44–71.	
  
	
  
Davidson,	
   Carl,	
   Steven	
   Matusz	
   and	
   Schevchenko,	
   2008.	
   “Globalization	
   and	
   Firm	
   Level	
  
     Adjustment	
  with	
  Imperfect	
  Labor	
  Markets,”	
  Journal	
  of	
  International	
  Economics,	
  July,	
  295	
  
     –	
  309.	
  
	
  
Davidson,	
   Carl,	
   Fredrik	
   Heyman,	
   Steven	
   Matusz,	
   Fredrik	
   Sjöholm	
   and	
   Susan	
   Zhu,	
   2010.	
  
     “Globalization	
  and	
  Imperfect	
  Labor	
  Market	
  Sorting,”	
  IFN	
  Working	
  Paper	
  No.	
  856.	
  
	
  
Egger,	
   Hartmut	
   and	
   Udo	
   Kreickemeier,	
   2009.	
   “Firm	
   Heterogeneity	
   and	
   the	
   Labor	
   Market	
  
     Effects	
  of	
  Trade	
  Liberalization,”	
  International	
  Economic	
  Review,	
  50,	
  187–216.	
  	
  
	
  
Elias,	
   Peter	
   and	
   Margaret	
   Birch,	
   1994.	
   “Establishment	
   of	
   Community-­‐Wide	
   Occupational	
  
     Statistics,”	
  Technical	
  Report,	
  Institute	
  for	
  Employment	
  Research,	
  University	
  of	
  Warwick.	
  
	
   	
   	
  
Feenstra,	
   Robert	
   and	
   Gordon	
   Hanson,	
   1999.	
   “The	
   Impact	
   of	
   Outsourcing	
   and	
   High-­‐
     Technology	
   Capital	
   on	
   Wages:	
   Estimates	
   for	
   the	
   U.S.,	
   1979-­‐1990,”	
   Quarterly	
   Journal	
   of	
  
     Economics,	
  114	
  (3),	
  907-­‐940.	
  
	
  
Feenstra,	
  Robert	
  and	
  Gordon	
  Hanson,	
  2002.	
  “Global	
  Production	
  and	
  Inequality:	
  A	
  Survey	
  of	
  
     Trade	
  and	
  Wages,”	
  in	
  James	
  Harrigan	
  and	
  E.	
  Kwan	
  Choi,	
  eds.,	
   Handbook	
  of	
  International	
  
     Trade,	
  Basil	
  Blackwell.	
  
	
  
Frías,	
   Judith.	
   A.,	
   David	
   Kaplan,	
   and	
   Eric	
   Verhoogen,	
   2009.	
   “Exports	
   and	
   Wage	
   Premia:	
  
     Evidence	
   from	
   Mexican	
   Employer-­‐Employee	
   Data,”	
   Discussion	
   paper,	
   Columbia	
  
     University.	
  
	
  
Gibbons,	
   Robert,	
   Katz,	
   Lawrence.	
   F.,	
   Lemieux,	
   Thomas	
   and	
   Parent,	
   Daniel,	
   2005.	
  
     “Comparative	
  Advantage,	
  Learning	
  and	
  Sectoral	
  Wage	
  Determination,”	
  Journal	
  of	
  Labor	
  
     Economics	
  23:	
  681-­‐723.	
  
	
  
Goldberg,	
   Linda	
   S.,	
   2004.	
   “Industry-­‐Specific	
   Exchange	
   Rates	
   for	
   the	
   United	
   States,”	
   Federal	
  
     Reserve	
  Bank	
  of	
  New	
  York	
  Economic	
  Policy	
  Review.	
  
	
  
Goldberg,	
   Pinelopi,	
   Amit	
   Khandelwal,	
   Nina	
   Pavcnik,	
   and	
   Petia	
   Topalova,	
   2008.	
   "Imported	
  
     Intermediate	
   Inputs	
   and	
   Domestic	
   Product	
   Growth:	
   Evidence	
   from	
   India,"	
   NBER	
  
     Working	
  Paper	
  No.	
  14416.	
  
	
  
Goldberg,	
   Pinelopi	
   and	
   Nina	
   Pavcnik,	
   2003.	
   “The	
   Response	
   of	
   the	
   Informal	
   Sector	
   to	
  
     Liberalization,”	
  Journal	
  of	
  Development	
  Economics,	
  72,	
  463-­‐496.	
  
	
  
Goldberg,	
   Pinelopi	
   and	
   Nina	
   Pavcnik,	
   2007.	
   “Distributional	
   Effects	
   of	
   Globalization	
   in	
  
     Developing	
  Countries,”	
  Journal	
  of	
  Economic	
  Literature,	
  45	
  (1),	
  39–82.	
  
	
  
Gonzaga,	
   Gustavo,	
   Naércio	
   Aquino	
   Menezes-­‐Filho,	
   and	
   Cristina	
   Terra,	
   2006.	
   “Trade	
  
     Liberalization	
   and	
   the	
   Evolution	
   of	
   Skill	
   Earnings	
   Differentials	
   in	
   Brazil,”	
   Journal	
   of	
  
     International	
  Economics,	
  68	
  (2),	
  345–367.	
  
	
  



                                                                                                                                                      30
	
  
Grossman,	
   Gene	
   and	
   Elhanan	
   Helpman,	
   1994.	
   “Protection	
   for	
   Sale,”	
   American	
   Economic	
  
     Review,	
  84	
  (4),	
  833–850.	
  
	
  
Helpman,	
   Elhanan,	
   Oleg	
   Itskhoki,	
   and	
   Stephen	
   Redding,	
   2010.	
   “Inequality	
   and	
  
     Unemployment	
  in	
  a	
  Global	
  Economy,”	
  Econometrica,	
  78	
  (4),	
  1239–1283.	
  
	
  
Halpern	
  László,	
  Miklós	
  Koren,	
  and	
  Adam	
  Szeidl,	
  2009.	
  "Imported	
  Inputs	
  and	
  Productivity,"	
  
     Central	
  European	
  University,	
  unpublished	
  manuscript.	
  
	
  
Hummels,	
   David,	
   Rasmus	
   Jorgensen,	
   Jakob	
   Munch	
   and	
   Chong	
   Xiang,	
   2010.	
   “The	
   Wage	
   and	
  
     Employment	
  Effects	
  of	
  Offshoring:	
  Evidence	
  from	
  Danish	
  Matched	
  Worker-­‐Firm	
  Data,”	
  
     mimeo.	
  
	
  
Klein,	
   Michael	
   W.,	
   Christoph	
   Moser,	
   and	
   Dieter	
   M.	
   Urban,	
  2010.	
   “The	
   Skill	
   Structure	
   of	
   the	
  
     Export	
   Wage	
   Premium:	
   	
   Evidence	
   from	
   German	
   Manufacturing,”	
   unpublished	
  
     manuscript.	
  
	
  
Kugler,	
   Maurice	
   and	
   Eric	
   A.	
   Verhoogen,	
   2008.	
   “The	
   Quality-­‐Complementarity	
   Hypothesis:	
  
     Theory	
  and	
  Evidence	
  from	
  Colombia,”	
  NBER	
  Working	
  Paper	
  No.	
  14418.	
  
	
  
Kume,	
  Honório,	
  Guida	
  Piani,	
  and	
  Carlos	
  Frederico	
  Bráz	
  de	
  Souza,	
  2003.	
  “A	
  Política	
  Brasileira	
  
     de	
   Importacão	
   no	
   Período	
   1987-­‐98:	
   Descricão	
   e	
   Avaliacão,”	
   in	
   Carlos	
   Henrique	
   Corseuil	
  
     and	
  Honório	
  Kume,	
  eds.,	
  A	
  abertura	
  comercial	
  brasileira	
  nos	
  anos	
  1990:	
  Impactos	
  sobre	
  
     emprego	
  e	
  salários,	
  Rio	
  de	
  Janeiro:	
  MTE	
  and	
  IPEA,	
  chapter	
  1,	
  9–37.	
   	
  
	
  
Lawrence,	
  Robert	
  and	
  Matthew	
  Slaughter,	
  1993.	
  “International	
  Trade	
  and	
  American	
  Wages	
  
     in	
   the	
   1980s:	
   Giant	
   Sucking	
   Sound	
   or	
   Small	
   Hiccup?,”	
   in	
   Martin	
   Neil	
   Baily	
   and	
   Clifford	
  
     Winston,	
  eds.,	
  Brookings	
  Papers	
  on	
  Economic	
  Activity:	
  Microeconomics	
  2,	
  161–211.	
  
	
   	
  
Leamer,	
   Edward,	
   1996.	
   “In	
   Search	
   of	
   Stolper-­‐Samuelson	
   Effects	
   on	
   US	
   Wage,”	
   NBER	
  
     Working	
  Paper	
  No.	
  5427.	
  
	
  
Lentz,	
   Rasmus,	
   2010.	
   “Sorting	
   by	
   search	
   intensity,”	
   Journal	
   of	
   Economic	
   Theory	
   145(4):	
  
     1436-­‐1452.	
  
	
  
Melitz,	
   Marc	
   J.,	
   2003.	
   “The	
   Impact	
   of	
   Trade	
   on	
   Intra-­‐Industry	
   Reallocations	
   and	
   Aggregate	
  
     Industry	
  Productivity,”	
  Econometrica,	
  71	
  (6),	
  1695–1725.	
  
	
  
Menezes-­‐Filho	
   Naércio	
   Aquino	
   and	
   Marc-­‐Andreas	
   Muendler,	
   2007.	
   "Labor	
   Reallocation	
   in	
  
     Response	
   to	
   Trade	
   Reform,"	
   University	
   of	
   California,	
   San	
   Diego,	
   unpublished	
  
     manuscript.	
  
	
  
Mortensen,	
   Dale,	
   2003.	
   “Wage	
   Dispersion:	
   Why	
   Are	
   Similar	
   Workers	
   Paid	
   Differently?,”	
  
     Zeuthen	
  lecture	
  book	
  series,	
  MIT	
  Press,	
  Cambridge,	
  Mass.	
  
	
  
Moulton,	
   Brent	
   R.,	
   1990.	
   “An	
   Illustration	
   of	
   a	
   Pitfall	
   in	
   Estimating	
   the	
   Effects	
   of	
   Aggregate	
  
     Variables	
  on	
  Micro	
  Units,”	
  The	
  Review	
  of	
  Economics	
  and	
  Statistics,	
  72,	
  334–338.	
  
	
  
Muendler,	
   Marc-­‐Andreas,	
   2003.	
   “Nominal	
   and	
   Real	
   Exchange	
   Rate	
   Series	
   for	
   Brazil,	
   1986-­‐
     2001,”	
  University	
  of	
  California,	
  San	
  Diego,	
  unpublished	
  manuscript.	
  



                                                                                                                                                            31
	
  
	
  
Muendler,	
   Marc-­‐Andreas,	
   2003.	
   “The	
   Database	
   Pesquisa	
   Industrial	
   Anual	
   1986-­‐2001:	
   A	
  
     Detective’s	
  Report,”	
  University	
  of	
  California,	
  San	
  Diego,	
  unpublished	
  manuscript.	
  	
  
	
  
Muendler,	
   Marc-­‐Andreas,	
   Jennifer	
   P.	
   Poole,	
   Garey	
   Ramey,	
   and	
   Tamara	
   Wajnberg,	
   2004.	
   “Job	
  
     Concordances	
  for	
  Brazil:	
  Mapping	
  the	
  Classificação	
  Brasileira	
  de	
  Ocupações	
  (CBO)	
  to	
  the	
  
     International	
  Standard	
  Classification	
  of	
  Occupations	
  (ISCO-­‐88),”	
  University	
  of	
  California,	
  
     San	
  Diego,	
  unpublished	
  manuscript.	
  
	
  
Muendler,	
  Marc-­‐Andreas,	
  2004.	
  “Trade,	
  Technology,	
  and	
  Productivity:	
  	
  A	
  Study	
  of	
  Brazilian	
  
     Manufacturers,	
   1986-­‐1998,”	
   University	
   of	
   California,	
   San	
   Diego,	
   unpublished	
  
     manuscript.	
  
	
  
Muendler,	
   Marc-­‐Andreas,	
   2008.	
   “Trade	
   and	
   Workforce	
   Changeover	
   in	
   Brazil,”	
   in	
   Stefan	
  
     Bender	
  and	
  Julia	
  Lane	
  eds.,	
  Analysis	
  of	
  firms	
  and	
  employees:	
  Quantitative	
  and	
  qualitative	
  
     approaches,	
  NBER	
  conference	
  report	
  series,	
  Chicago	
  and	
  London,	
  ch.	
  9,	
  267-­‐308.	
  
	
  
Munch,	
  Jakob	
  Roland	
  and	
  Jan	
  Rose	
  Skaksen,	
  2008.	
  “Human	
  capital	
  and	
  wages	
  in	
  exporting	
  
     firms,”	
  Journal	
  of	
  International	
  Economics,	
  75,	
  363-­‐372.	
  
	
  
Olarreaga,	
   Marcello	
   and	
   Isidro	
   Soloaga,	
   1998.	
   “Endogenous	
   Tariff	
   Formation:	
   The	
   Case	
   of	
  
     MERCOSUR,”	
  World	
  Bank	
  Economic	
  Review,	
  12	
  (2),	
  297–320.	
  
	
  
Paz,	
   Lourenço,	
   2009.	
   “The	
   Impacts	
   of	
   Trade	
   Liberalization	
   on	
   Informal	
   Labor	
   Markets,”	
  
     Syracuse	
  University,	
  unpublished	
  manuscript.	
  
	
  
Postel-­‐Vinay,	
   F.	
   and	
   J.-­‐M.	
   Robin,	
   2002,	
   “Equilibrium	
   Wage	
   Dispersion	
   with	
   Worker	
   and	
  
     Employer	
  Heterogeneity”,	
  Econometrica,	
  70(6),	
  2295-­‐350.	
  
	
  
Revenga,	
   Ana,	
   1997.	
   “Employment	
   and	
   Wage	
   Effects	
   of	
   Trade	
   Liberalization:	
   The	
   Case	
   of	
  
     Mexican	
  Manufacturing,”	
  Journal	
  of	
  Labor	
  Economics,	
  15	
  (3).	
  
	
  
Schank,	
   Thorsten,	
   Claus	
   Schnabel,	
   and	
   Joachim	
   Wagner,	
   2007.	
   “Do	
   exporters	
   really	
   pay	
  
     higher	
  wages?	
  	
  First	
  evidence	
  from	
  German	
  linked	
  employer-­‐employee	
  data,”	
  Journal	
  of	
  
     International	
  Economics,	
  72,	
  52-­‐74.	
  
	
  
Shapiro,	
   Carl	
   and	
   Joseph	
   E.	
   Stiglitz,	
   1984.	
   “Equilibrium	
   Unemployment	
   as	
   a	
   Worker	
  
     Discipline	
  Device,”	
  American	
  Economic	
  Review,	
  74	
  (3),	
  433-­‐444.	
  
	
  
Shimer,	
   Robert,	
   2005.	
   “The	
   Assignment	
   of	
   Workers	
   to	
   Jobs	
   in	
   an	
   Economy	
   with	
  
     Coordination	
  Frictions,”	
  Journal	
  of	
  Political	
  Economy,	
  113(5):	
  996-­‐1025.	
  	
  
	
  
Kenneth	
   L.	
   Sørensen	
   &	
   Rune	
   M.	
   Vejlin,	
   2010.	
   "Worker	
   and	
   Firm	
   Heterogeneity	
   in	
   Wage	
  
     Growth:	
  An	
  AKM	
  approach,"	
  Economics	
  Working	
  Papers	
  2010-­‐13,	
  School	
  of	
  Economics	
  
     and	
  Management,	
  University	
  of	
  Aarhus.	
  
	
  
Trefler,	
   Daniel,	
   2004.	
   “The	
   Long	
   and	
   Short	
   of	
   the	
   Canada-­‐U.S.	
   Free	
   Trade	
   Agreement,”	
  
     American	
  Economic	
  Review,	
  94	
  (4),	
  870–895.	
  
	
  
Verhoogen,	
   Eric	
   A.,	
   2008.	
   “Trade,	
   Quality	
   Upgrading	
   and	
   Wage	
   Inequality	
   in	
   the	
   Mexican	
  



                                                                                                                                             32
	
  
     Manufacturing	
  Sector,”	
  Quarterly	
  Journal	
  of	
  Economics,	
  123	
  (2),	
  489–530.	
  
	
  
Woodcock,	
  Simon	
  D.,	
  2008.	
  “Match	
  Effects,”	
  mimeo.	
  	
  
	
  
Yeaple,	
   Stephen	
   Ross,	
   2005.	
   “A	
   Simple	
   Model	
   of	
   Firm	
   Heterogeneity,	
   International	
   Trade,	
  
     and	
  Wages,”	
  Journal	
  of	
  International	
  Economics,	
  65,	
  1–20.	
  
	
  
	
                                           	
  




                                                                                                                                          33
	
  
Appendix	
  
	
  
For	
  a	
  sub-­‐sample	
  of	
  firms,	
  we	
  can	
  combine	
  RAIS	
  with	
  data	
  from	
  the	
  Brazilian	
  manufacturing	
  
survey,	
  Pesquisa	
  Industrial	
  Anual	
  (PIA).40	
  We	
  construct	
  an	
  indicator	
  variable	
  that	
  takes	
  the	
  
value	
   one	
   if	
   a	
   firm	
   reports	
   positive	
   acquisitions	
   of	
   foreign	
   intermediate	
   goods	
   or	
   foreign	
  
machinery.	
  This	
  measure	
  captures	
  acquisitions	
  by	
  firms	
  which	
  use	
  imported	
  intermediates	
  
even	
  if	
  they	
  don’t	
  import	
  them	
  directly,	
  and	
  purchase	
  them	
  from	
  local	
  distributors	
  instead.	
  
In	
   order	
   to	
   establish	
   a	
   comparison	
   with	
   our	
   results	
   for	
   the	
   full	
   sample,	
   we	
   start	
   by	
  
replicating	
  our	
  main	
  specification	
  for	
  the	
  smaller	
  sample	
  of	
  firm-­‐groups	
  for	
  which	
  we	
  also	
  
have	
  data	
  on	
  the	
  importer	
  status	
  of	
  the	
  firm.	
  We	
  find	
  the	
  differential	
  effect	
  on	
  exporters	
  to	
  
be	
  prevalent	
  for	
  the	
  restricted	
  sample	
  of	
  PIA	
  firm-­‐groups	
  as	
  well.	
  Our	
  estimates	
  suggest	
  that	
  
a	
  one	
  percentage	
  point	
  decrease	
  in	
  ERP	
  is	
  associated	
  with	
  an	
  increase	
  in	
  firm	
  average	
  wages	
  
by	
   0.18%.	
   In	
   column	
   (2)	
   we	
   include	
   the	
   firm’s	
   importer	
   status	
   and	
   the	
   interaction	
   terms	
  
between	
   importer	
   status	
   with	
   ERP	
   and	
   RER	
   in	
   our	
   main	
   specification.	
   Our	
   main	
   findings	
   on	
  
the	
  differential	
  effect	
  of	
  liberalization	
  on	
  exporters	
  continue	
  to	
  hold	
  in	
  this	
  specification.	
  We	
  
find	
   that	
   while	
   importing	
   firms	
   pay	
   higher	
   wages	
   on	
   average,	
   the	
   decline	
   in	
   ERP	
   does	
   not	
  
have	
  a	
  differential	
  effect	
  on	
  importing	
  firms.	
  
	
                                        In	
  the	
  PIA	
  sample	
  there	
  is	
  significant	
  overlap	
  in	
  firms	
  that	
  export	
  and	
  import.	
  In	
  fact,	
  
conditional	
   on	
   importing,	
   most	
   firms	
   also	
   export.	
   50	
   percent	
   of	
   firms	
   that	
   export	
   also	
  
import,	
   while	
   only	
   11%	
   of	
   non-­‐exporters	
   import.	
   In	
   the	
   specification	
   reported	
   in	
   column	
  
(3),	
   we	
   test	
   the	
   importance	
   of	
   this	
   overlap	
   on	
   our	
   results.	
   More	
   specifically,	
   we	
   define	
   an	
  
exporter	
   dummy	
   which	
   takes	
   the	
   value	
   one	
   for	
   firms	
   that	
   export	
   but	
   do	
   not	
   import;	
   an	
  
importer	
  dummy	
  which	
  takes	
  the	
  value	
  one	
  for	
  firms	
  that	
  import	
  but	
  do	
  not	
  export	
  and	
  an	
  
exporter-­‐importer	
   dummy	
   which	
   takes	
   the	
   value	
   one	
   for	
   firms	
   that	
   both	
   export	
   and	
  
import.41	
  The	
  differential	
  effect	
  of	
  a	
  change	
  in	
  ERP	
  on	
  exporters	
  is	
  significant	
  and	
  negative	
  
for	
   firms	
   that	
   only	
   export	
   (and	
   do	
   not	
   import)	
   as	
   well	
   as	
   for	
   firms	
   that	
   both	
   export	
   and	
  

	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
40	
  PIA	
   is	
   a	
   confidential	
   data	
   set,	
   based	
   on	
   a	
   survey	
   of	
   Brazilian	
   manufacturing	
   firms	
   conducted	
  
annually	
   by	
   the	
   Brazilian	
   Census	
   Bureau	
   between	
   1986	
   to	
   the	
   present	
   (excluding	
   1991).	
   Since	
   we	
   do	
  
not	
   have	
   access	
   to	
   this	
   confidential	
   dataset,	
   instead,	
   we	
   use	
   random	
   three-­‐to-­‐five	
   firm	
   cells	
   with	
  
similar	
   characteristics,	
   constructed	
   by	
   Muendler	
   (2003)	
   to	
   meet	
   confidentiality	
   requirements.	
   The	
  
data	
   we	
   use	
   are	
   based	
   on	
   these	
   random	
   aggregates	
   of	
   PIA	
   (firm-­‐groups)	
   as	
   described	
   in	
   Muendler	
  
(2003)	
  and	
  Muendler	
  (2004).	
  In	
  our	
  discussion	
  of	
  the	
  PIA	
  data,	
  we	
  use	
  the	
  terms	
  “firms”	
  and	
  “firm-­‐
groups”	
  interchangeably.	
  
41	
  Although	
  we	
  report	
  the	
  estimated	
  coefficients	
  for	
  column	
  (3)	
  in	
  the	
  same	
  row	
  as	
  columns	
  (1)	
  and	
  

(2),	
  we	
  note	
  that	
  in	
  the	
  first	
  two	
  columns,	
  the	
  export	
  variable	
  takes	
  the	
  value	
  one	
  for	
  firms	
  that	
  report	
  
positive	
  exports	
  and	
  the	
  import	
  variable	
  takes	
  the	
  value	
  one	
  for	
  firms	
  that	
  report	
  positive	
  imports.	
  In	
  
column	
   (3),	
   the	
   export	
   and	
   import	
   variables	
   take	
   the	
   value	
   one	
   for	
   firms	
   that	
   export	
   but	
   do	
   not	
  
import,	
  and	
  import	
  but	
  do	
  not	
  export,	
  respectively.	
  



                                                                                                                                                                                                                                   34
	
  
import.	
  We	
  find	
  no	
  differential	
  impact	
  of	
  a	
  decline	
  in	
  ERP	
  on	
  firms	
  that	
  only	
  import.	
  We	
  find	
  
the	
   exporter-­‐premium	
   on	
   wages	
   to	
   be	
   higher	
   for	
   exporters	
   that	
   import,	
   compared	
   to	
  
exporters	
   that	
   do	
   not.	
   The	
   higher	
   average	
   wages	
   for	
   importers	
   relative	
   to	
   domestic	
   firms	
  
reported	
   in	
   column	
   (2)	
   seems	
   mainly	
   to	
   be	
   due	
   to	
   the	
   fact	
   that	
   most	
   of	
   these	
   firms	
   are	
  
exporters	
  as	
  well.42	
  
	
  
	
  
	
  
	
  




	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
     We	
   note,	
   however,	
   that	
   there	
   is	
   heterogeneity	
   in	
   this	
   coefficient	
   based	
   on	
   the	
   comparative	
  
42	
  

advantage	
   of	
   the	
   firm’s	
   sector.	
   Unreported	
   results	
   suggest	
   that,	
   in	
   high	
   comparative	
   advantage	
  
sectors,	
  a	
  decline	
  in	
  ERP	
  increases	
  wages	
  of	
  domestic	
  firms	
  using	
  imported	
  intermediates	
  relative	
  to	
  
domestic	
   firms	
   which	
   do	
   not	
   use	
   imported	
   intermediates	
   (at	
   the	
   10%	
   level	
   of	
   significance);	
   a	
   one	
  
percentage	
  point	
  decrease	
  in	
  ERP	
  increases	
  wages	
  at	
  high	
  comparative	
  advantage	
  importers	
  which	
  
do	
   not	
   export	
   by	
   0.21%.	
   By	
   contrast,	
   average	
   wages	
   at	
   importers	
   which	
   do	
   not	
   export,	
   are	
   not	
  
statistically	
   different	
   than	
   purely	
   domestic	
   firms	
   post-­‐liberalization	
   in	
   the	
   low	
   comparative	
  
advantage	
   sector.	
   Moreover,	
   average	
   wages	
   at	
   exporters	
   (which	
   do	
   not	
   import)	
   and	
   exporter-­‐
importers	
  in	
  the	
  low	
  comparative	
  advantage	
  sector	
  increase	
  with	
  a	
  decline	
  in	
  ERP	
  relative	
  to	
  purely	
  
domestic	
  firms.	
  These	
  results	
  are	
  available	
  upon	
  request.	
  



                                                                                                                                                                                                                                   35
	
  
Figure	
  3.1	
  Time	
  Variation	
  in	
  Effective	
  Rates	
  of	
  Protection,	
  1990-­‐1998	
  
	
  
	
  
       0.60	
  

       0.50	
  

       0.40	
  

       0.30	
  

       0.20	
  

       0.10	
  

       0.00	
  
                  1988	
   1989	
   1990	
   1991	
   1992	
   1993	
   1994	
   1995	
   1996	
   1997	
   1998	
  

                                               Average	
  ERP	
      Median	
  ERP	
  
                                                                                                                       	
  




                                                                                                                              36
Figure	
  3.2:	
   Cross-­Industry	
  Variation	
  in	
  Effective	
  Rates	
  of	
  Protection	
  
	
  




                                                                                      	
  	
  	
  	
  	
                                                                                                	
  
	
       Effective	
  Rates	
  of	
  Protection	
  in	
  1990	
                	
                            	
  	
  	
  	
  	
  	
  	
  	
  Effective	
  Rates	
  of	
  Protection	
  in	
  1998	
  




	
       	
        	
        	
                                                                                                                           	
  
	
       	
        	
        	
       	
  	
  	
  	
  	
  	
  	
  	
  	
  Annual	
  Change	
  in	
  ERP,	
  1990-­‐1998	
  
Source:	
  Kume,	
  Piani,	
  and	
  Souza	
  (2003).	
  
	
  
	
  




                                                                                                                                                                                                               37
    Figure	
  4.1	
  Distribution	
  of	
  Psi	
  ( Ψj(i,t ))	
  Quintile	
  by	
  Residual	
  ( ε ijt −1)	
  Quintile	
  




                                            €                                        €
           35	
  

           30	
                                                                                                                                 30-­‐35	
  

           25	
                                                                                                                                 25-­‐30	
  

            20	
                                                                                                                                20-­‐25	
  
                                                                                                                                                15-­‐20	
  
            15	
  
                                                                                                                                                10-­‐15	
  
             10	
                                                                                                          Residual_5	
  
                                                                                                                        Residual_4	
            5-­‐10	
  
                5	
  
                                                                                                                    Residual_3	
                0-­‐5	
  
               0	
  
                                                                                                               Residual_2	
  
              psi_lag=1	
  
                                      psi_lag=2	
  
                                                         psi_lag=3	
                                      Residual_1	
  
                                                                            psi_lag=4	
  
                                                                                                   psi_lag=5	
  

                                                                                                                                                              	
  
    Figure	
  4.2	
  Probability	
  of	
  Transition	
  to	
  Each	
  Quintile	
  of	
  the	
  Psi	
  ( Ψj(i,t ))	
  Distribution,	
  by	
  
    Psi	
  ( Ψj(i,t −1))	
  Quintile	
  of	
  Origin,	
  Full	
  Sample	
  of	
  Job	
  Changers	
  


                                                                                            €
€              0.5	
  
             0.45	
  
                0.4	
  
              0.35	
                                                                                                                        psi_lag=1	
  
                0.3	
                                                                                                                       psi_lag=2	
  
              0.25	
                                                                                                                        psi_lag=3	
  
                    0.2	
  
                                                                                                                                            psi_lag=4	
  
               0.15	
  
                                                                                                                                            psi_lag=5	
  
                    0.1	
                                                                                           psi_lag=5	
  
                                                                                                                 psi_lag=4	
  
               0.05	
  
                                                                                                               psi_lag=3	
  
                        0	
  
                                                                                                           psi_lag=2	
  
                                psi=1	
  
                                             psi=2	
                                                    psi_lag=1	
  
                                                           psi=3	
  
                                                                         psi=4	
  
                                                                                       psi=5	
  
                                                                                                                                                              	
  
    	
  
    	
  


                                                                                                                                               38
Figure	
  4.3	
  Probability	
  of	
  Transition	
  to	
  Each	
  Quintile	
  of	
   Ψj(i,t ) 	
  ,	
  	
  by	
   Ψj(i,t −1)	
  Quintile	
  
Figure	
  4.3a	
  Job	
  Changers	
  in	
  Q( ε ijt −1)=1	
  


                                                                              €                 €
    50	
                           €
    40	
                                                                                            40-­‐50	
  

    30	
                                                                                            30-­‐40	
  

     20	
                                                                                           20-­‐30	
  
                                                                           psi_lag=5	
              10-­‐20	
  
     10	
  
                                                                        psi_lag=3	
                 0-­‐10	
  
        0	
  
        psi=1	
   psi=2	
  
                            psi=3	
                            psi_lag=1	
  
                                             psi=4	
  
                                                          psi=5	
  
                                                                                                                  	
  
Figure	
  4.3b	
  Job	
  Changers	
  in	
  Q( ε ijt −1)=3	
  




    50	
                           €
    40	
                                                                                            40-­‐50	
  

    30	
                                                                                            30-­‐40	
  

     20	
                                                                                           20-­‐30	
  
                                                                           psi_lag=5	
              10-­‐20	
  
     10	
  
                                                                        psi_lag=3	
                 0-­‐10	
  
        0	
  
        psi=1	
   psi=2	
  
                            psi=3	
                            psi_lag=1	
  
                                             psi=4	
  
                                                          psi=5	
  
                                                                                                                  	
  
Figure	
  4.3c	
  Job	
  Changers	
  in	
  Q( ε ijt −1)=5	
  




    50	
                           €
    40	
                                                                                            40-­‐50	
  

    30	
                                                                                            30-­‐40	
  

     20	
                                                                                           20-­‐30	
  
                                                                           psi_lag=5	
              10-­‐20	
  
     10	
  
                                                                        psi_lag=3	
                 0-­‐10	
  
        0	
  
        psi=1	
   psi=2	
  
                            psi=3	
                            psi_lag=1	
  
                                             psi=4	
  
                                                          psi=5	
  
                                                                                                                  	
  


                                                                                                                                               39
Table	
  2.1	
  Descriptive	
  Statistics

                                            Sample    Exporters   Domestic
Worker	
  Characteristics
Average	
  Wage	
  in	
  reais               3,917     5,512       2,456
Average	
  Log	
  (Wages)                    7.630     8.019       7.273
Share	
  of	
  Workers
Less	
  than	
  High	
  School               0.745     0.696       0.790
High	
  School	
  Graduates                  0.184     0.206       0.164
College	
  Graduates                         0.066     0.094       0.039

Unskilled	
  Blue	
  Collar                  0.109     0.099       0.119
Skilled	
  Blue	
  Collar	
                  0.609     0.599       0.617
Other	
  White	
  Collar	
                   0.078     0.079       0.077
Professional	
  and	
  Managerial            0.179     0.206       0.154
Firm	
  Characteristics
Average	
  Employment                          73       346          37
Average	
  Log	
  (Employment)                2.82      4.75        2.57
Number	
  of	
  Workers                     114,042    64,212      80,895
Number	
  of	
  Firms                        58,578    11,143      53,537




                                                                             40
Table	
  3.1	
  Trade	
  Protection	
  and	
  Firm-­‐Level	
  Average	
  Wages

                                                                  Tariffs                                                    ERP
                                                                    	
  High	
           Low	
                           	
  High	
        Low	
  
                                          Full	
  Sample     Comparative	
         Comparative	
     Full	
  Sample   Comparative	
     Comparative	
  
                                                              Advantage             Advantage                          Advantage         Advantage
                                              0.170**             -­‐0.058           0.279***
Tariff
                                              (0.079)            (0.182)             (0.095)
                                           -­‐0.248***        -­‐0.562***             -­‐0.117
Export*Tariff
                                              (0.074)            (0.109)             (0.094)
                                                                                                          -­‐0.011        -­‐0.082*           0.067
ERP
                                                                                                         (0.033)          (0.045)           (0.046)
                                                                                                      -­‐0.103***      -­‐0.206***           -­‐0.008
Export*ERP
                                                                                                         (0.037)          (0.054)           (0.064)
                                           -­‐0.258***         -­‐0.293***          -­‐0.230***       -­‐0.233***      -­‐0.247***        -­‐0.199**
Export*RER
                                              (0.058)             (0.076)              (0.081)           (0.055)          (0.074)           (0.081)
                                            0.328***            0.414***             0.275***          0.279***         0.318***           0.219**
Export
                                              (0.074)             (0.094)              (0.104)           (0.067)          (0.088)           (0.100)
N                                            505,369             258,374              246,995           505,369          258,374          246,995
Firm-­‐Fixed	
  Effects                         YES                 YES                  YES                 YES              YES               YES
Region-­‐Specific	
  Year	
  Dummies            YES                 YES                  YES                 YES              YES               YES
Detailed	
  Firm-­‐Level	
  Controls            YES                 YES                  YES                 YES              YES               YES
Impact	
  on	
  Exporters
Tariff                                       -­‐0.078          -­‐0.620***             0.162          -­‐0.114**       -­‐0.289***          0.058
F-­‐Statistic                                   0.50                9.37                1.40              4.86             14.58             0.59
p-­‐value                                       0.48                0.00                0.24              0.03              0.00             0.45




                                                                                                                                                          41
Table	
  3.2	
  ERP	
  and	
  Firm-­‐Level	
  Wages:	
  Robustness

                                            Alternative	
  Samples                                                     Export	
  Status
                                                                                    Industry	
                                                                   Size	
  of	
  
                                 Balanced	
   5%	
  Metro	
   Lib.	
  period	
  
                                                                                 Specific	
  RER    Cutoff5       Cutoff10       Export90       Export_All      Exports
                                      Panel          Sample           90-­‐94
                                     0.004            -­‐0.021       -­‐0.000          -­‐0.031        -­‐0.011       -­‐0.011       -­‐0.013        -­‐0.001    0.284***
ERP
                                    (0.038)          (0.033)        (0.029)          (0.031)          (0.033)        (0.033)        (0.034)         (0.033)        (0.091)
                                  -­‐0.095**       -­‐0.083**    -­‐0.125***         -­‐0.055*     -­‐0.108***    -­‐0.110***    -­‐0.115***      -­‐0.086**    -­‐0.025***
Export*ERP
                                    (0.042)          (0.033)        (0.034)          (0.031)          (0.038)        (0.039)        (0.042)         (0.037)        (0.007)
                                -­‐0.009***      -­‐0.210***     -­‐0.098***      -­‐0.005***      -­‐0.233***    -­‐0.238***    -­‐0.261***    -­‐0.190***     -­‐0.022***
Export*RER
                                    (0.002)          (0.050)        (0.032)          (0.001)          (0.056)        (0.056)        (0.057)         (0.046)        (0.008)
                                  0.284***        0.251***         0.145***        0.466***         0.279***       0.283***                                       0.034***
Export
                                    (0.067)          (0.060)        (0.039)          (0.060)          (0.067)        (0.068)                                       (0.010)
N                                 204,437          354,564         270,400          505,369          505,369        505,369       505,369        505,369           58,418
Firm-­‐Fixed	
  Effects                YES               YES            YES               YES             YES            YES        YES            YES               YES
Region-­‐Specific	
  Year	
  
                                   YES              YES              YES             YES              YES            YES            YES            YES             YES
Dummies
Detailed	
  Firm-­‐Level	
  
                                   YES              YES              YES             YES              YES            YES            YES            YES             YES
Controls




                                                                                                                                                                         42
Table	
  4.1	
  Trade	
  Protection	
  and	
  Worker-­‐Level	
  Wages

                                                               Tariffs                                                ERP
                                                                 	
  High	
        Low	
                          	
  High	
      Low	
  
                                          Full	
  Sample    Comparative	
   Comparative	
     Full	
  Sample   Comparative	
   Comparative	
  
                                                             Advantage        Advantage                         Advantage       Advantage
Tariff                                         0.133           -­‐0.018          0.273
                                             (0.177)          (0.300)          (0.256)
Export*Tariff                                 -­‐0.110         -­‐0.180         -­‐0.055
                                             (0.090)          (0.129)          (0.127)
ERP                                                                                               -­‐0.020        -­‐0.083          0.198
                                                                                                (0.073)          (0.074)          (0.130)
Export*ERP                                                                                        -­‐0.045        -­‐0.042          0.004
                                                                                                (0.045)          (0.051)          (0.074)
Export                                        0.133*             0.067           0.184            0.113*           0.024            0.154
                                            (0.072)            (0.074)         (0.125)          (0.061)          (0.064)          (0.115)
Export*RER                                  -­‐0.094*           -­‐0.025        -­‐0.147        -­‐0.085*         -­‐0.004         -­‐0.132
                                            (0.056)            (0.059)         (0.095)          (0.051)          (0.055)          (0.091)
N                                           504,424            266,463         237,961          504,424          266,463          237,961
Detailed	
  Worker	
  Level	
  Controls         YES                YES             YES               YES             YES              YES
Detailed	
  Firm	
  Level	
  Controls           YES                YES             YES               YES             YES              YES
Region-­‐Specific	
  Year	
  Dummies            YES                YES             YES               YES             YES              YES
Spell-­‐Fixed	
  Effects                        YES                YES             YES               YES             YES              YES




                                                                                                                                                 43
Table	
  4.2	
  ERP	
  and	
  Worker-­‐Level	
  Wages:	
  Robustness

                                                          Alternative	
  Samples                                               Export	
  Status
                                                                        Liberalizatio Industry	
                                                                    Size	
  of	
  
                                                        5%	
  Metro	
  
                                                                        n	
  period	
  90-­‐ Specific	
  RER   cutoff5     cutoff10      Export90     Export_All   Exports
                                                       Male	
  Sample
                                                                                  94
ERP                                                       -­‐0.044            -­‐0.040            -­‐0.025      -­‐0.015     -­‐0.012      -­‐0.026     -­‐0.030    -­‐0.170
                                                         (0.059)             (0.092)             (0.074)       (0.073)     (0.073)        (0.072)      (0.074)     (0.196)
Export*ERP                                                -­‐0.031            -­‐0.046            -­‐0.035      -­‐0.054     -­‐0.058      -­‐0.037     -­‐0.021     0.007
                                                         (0.050)             (0.045)             (0.044)       (0.046)     (0.044)        (0.047)      (0.040)     (0.015)
Export                                                     0.054               0.076          -­‐0.003***       0.115*     0.122**                                   0.006
                                                         (0.055)             (0.058)             (0.001)       (0.062)     (0.062)                                 (0.012)
Export*RER                                                -­‐0.019            -­‐0.053        -­‐0.003***       -­‐0.084   -­‐0.089*     -­‐0.099**     -­‐0.024     0.001
                                                         (0.047)             (0.051)             (0.001)       (0.051)     (0.051)         (0.043)     (0.026)     (0.009)
N                                                        447,957             269,951            504,424        504,424     504,424       504,424       504,424     241,147
Detailed	
  Worker	
  Level	
  Controls                     YES               YES                YES             YES         YES            YES          YES          YES
Detailed	
  Firm	
  Level	
  Controls                       YES               YES                YES             YES         YES            YES          YES          YES
Region-­‐Specific	
  	
  	
  	
  	
  Year	
  Dummies        YES               YES                YES             YES         YES            YES          YES          YES
Spell-­‐Fixed	
  Effects                                    YES               YES                YES             YES         YES            YES          YES          YES




                                                                                                                                                                                44
Table	
  4.3	
  Change	
  in	
  Estimated	
  Match	
  Effects	
  Over	
  Time

                                                       Match	
  Effect
                                       1990                                     1998
                           Median                  Mean              Median              Mean
Exporters                  0.145                  0.135               0.169             0.160
Non-­‐exporters            0.018                 -­‐0.019            -­‐0.056          -­‐0.068
All	
  Firms               0.065                  0.031               0.059             0.034




                                                                                                  45
Table	
  4.4	
  Alternative	
  Specifications	
  for	
  Switchers

                                                       Tariffs                                                 ERP
                                                   Both	
  Firm	
                                        Both	
  Firm	
  
                                 Only	
  Firm	
                     Match	
  Fixed	
   Only	
  Firm	
                     Match	
  Fixed	
  
                                                  and	
  Worker	
                                       and	
  Worker	
  
                                Fixed	
  Effects                      Effects         Fixed	
  Effects                      Effects
                                                  Fixed	
  Effects                                      Fixed	
  Effects
Tariff                            0.552***             0.158*          0.056
                                   (0.180)            (0.149)        (0.232)
Export*Tariff                      -­‐0.242*        -­‐0.151**        -­‐0.104
                                   (0.124)            (0.092)        (0.101)
ERP                                                                                        0.087             0.020          -­‐0.011
                                                                                         (0.084)           (0.063)         (0.095)
Export*ERP                                                                                -­‐0.086          -­‐0.050        -­‐0.018
                                                                                         (0.070)           (0.050)         (0.058)
Export                            0.255***             0.182           0.133             0.198**             0.144           0.098
                                   (0.094)            (0.074)        (0.083)             (0.086)           (0.066)         (0.076)
Export*RER                         -­‐0.140*        -­‐0.117**        -­‐0.089            -­‐0.109          -­‐0.097        -­‐0.071
                                   (0.072)            (0.057)        (0.066)             (0.070)           (0.054)         (0.063)
N                                  226,193            226,193            226,193         226,193            226,193           226,193
Firm	
  Controls                      YES                YES              YES               YES               YES               YES
Worker	
  Controls                    YES                YES              YES               YES               YES               YES
Firm	
  Fixed	
  Effects              YES                YES              NO                YES               YES               NO
Worker	
  Fixed	
  Effects            NO                 YES              NO                NO                YES               NO
Spell-­‐Fixed	
  Effects              NO	
               NO               YES               NO	
              NO                YES




Table	
  4.5	
  ERP	
  and	
  Worker	
  Level	
  Wages,	
  by	
  Skill

                                 Less	
  than	
   More	
  than	
  
                                High-­‐School High-­‐School
ERP                                -­‐0.005          -­‐0.055
Export*ERP                         -­‐0.035        -­‐0.094*
Export                            0.116*              0.127
Export*RER                         -­‐0.088          -­‐0.088
Firm	
  Controls                      YES               YES
Worker	
  Controls                    YES               YES
Spell-­‐Fixed	
  Effects              NO	
               NO




                                                                                                                                    46
Table	
  A.1	
  ERP	
  and	
  Firm-­‐Level	
  Wages:	
  Importers	
  and	
  Exporters

                                         Only	
  Export	
      Export	
  and	
   Detailed	
  Trade	
  
                                              Status          Import	
  Status          Status
                                              0.090                  0.084               0.089
ERP
                                             (0.073)                (0.074)           (0.072)
                                          -­‐0.178***           -­‐0.175***        -­‐0.185***
Export*ERP
                                             (0.036)                (0.038)           (0.045)
                                          -­‐0.154***             -­‐0.117**          -­‐0.088*
Export*RER
                                             (0.051)                (0.051)           (0.051)
                                           0.226***              0.184***             0.163**
Export
                                             (0.064)                (0.065)           (0.065)
                                                                 0.128***               -­‐0.034
Import
                                                                    (0.042)           (0.135)
                                                                     0.001              -­‐0.034
Import*ERP
                                                                    (0.042)           (0.135)
                                                                -­‐0.112***              0.065
Import*RER
                                                                    (0.034)           (0.101)
                                                                                   -­‐0.175***
Export_Import	
  *ERP
                                                                                      (0.045)
                                                                                   -­‐0.219***
Export_Import	
  *RER
                                                                                      (0.050)
                                                                                    0.306***
Export_Import
                                                                                      (0.064)
N                                           11,372                   11,372            11,372
Firm-­‐Fixed	
  Effects                      YES                      YES                  YES
Year	
  Dummies                              YES                      YES                  YES
Detailed	
  Firm-­‐Level	
  Controls         YES                      YES                  YES




                                                                                                         47