Online appendixes for “Regional Effects of Trade Reform: What is theCorrect Measure of Liberalization?” by Brian K. Kovak

Specific Factors Model

A1. Factor prices

This section closely follows Jones (1975), but deviates from that paper’s resultby allowing the amount of labor available to the regional economy to vary. Con-sider a particular region, r, suppressing that subscript on all terms. Industriesare indexed by i = 1...N . L is the total amount of labor and Ti is the amount ofindustry i-specific factor available in the region. aLi and aT i are the respectivequantities of labor and specific factor used in producing one unit of industry ioutput. Letting Yi be the output in each industry, the factor market clearingconditions are

(A1) aT iYi = Ti ∀i,

(A2)∑i

aLiYi = L.

Under perfect competition, the output price equals the factor payments, where wis the wage and Ri is the specific factor price.

(A3) aLiw + aT iRi = Pi ∀i

Let hats represent proportional changes, and consider the effect of price changesPi. θi is the cost share of the specific factor in industry i.

(A4) (1− θi)w + θiRi = Pi ∀i,

which follows from the envelope theorem result that unit cost minimization implies

(A5) (1− θi)aLi + θiaT i = 0 ∀i.

Differentiate (A1), keeping in mind that Ti is fixed in all industries.

(A6) Yi = −aT i ∀i

Similarly, differentiate (A2), let λi = LiL be the fraction of regional labor utilized

in industry i, and substitute in (A6) to yield

(A7)∑i

λi(aLi − aT i) = L.

21

By the definition of the elasticity of substitution between Ti and Li in production,

(A8) aT i − aLi = σi(w − Ri) ∀i.

Substituting this into (A7) yields

(A9)∑i

λiσi(Ri − w) = L.

Equations (A4) and (A9) can be written in matrix form as follows.

(A10)

θ1 0 . . . 0 1− θ1

0 θ2 . . . 0 1− θ2...

. . ....

...0 0 . . . θN 1− θN

λ1σ1 λ2σ2 . . . λNσN −∑

i λiσi

R1

R2...

RNw

=

P1

P2...

PNL

Rewrite this expression as follows for convenience of notation.

(A11)

[Θ θLλ′ −

∑i λiσi

] [Rw

]=

[P

L

]Solve for w using Cramer’s rule and the rule for the determinant of partitionedmatrices.

(A12) w =L− λ′Θ−1P

−∑

i λiσi − λ′Θ−1θL

Note that the inverse of the diagonal matrix Θ is a diagonal matrix of 1θi

’s. Thisyields the effect of goods price changes and changes in regional labor on regionalwages:

(A13) w =−L∑i′ λi′

σi′θi′

+∑i

βiPi

(A14) where βi =λiσiθi∑

i′ λi′σi′θi′

This expression with L = 0 yields (1). Changes in specific factor prices can becalculated from wage changes by rearranging (A4).

(A15) Ri =Pi − (1− θi)w

θi22

Plugging in (A13) and collecting terms yields the effect of goods price changesand changes in regional labor on specific factor price changes.

(A16) Ri =(1− θi)θi

L∑i′ λi′

σi′θi′

+

(βi +

1

θi(1− βi)

)Pi −

(1− θi)θi

∑k 6=i

βkPk

Setting L = 0 in (A13) and (A16) yields the equivalent expressions in Jones(1975).

A2. Graphical representation of the model

The equilibrium adjustment mechanisms at work in the model are demon-strated graphically in Figure A1, which represents a two-region (r = 1, 2) andtwo-industry (i = A,B) version of the model.29 Region 1 is relatively well en-dowed with the industry A specific factor. In each panel, the x-axis representsthe total amount of labor in the country to be allocated across the two industriesin the two regions, and the y-axis measures the wage in each region. Focusingon the left portion of panel (a), the curve labeled PAF

AL is the marginal value

product of labor in industry A, and the curve labeled PBFBL is the marginal value

product of labor in industry B, measuring the amount of labor in industry Bfrom right to left. Given labor mobility across sectors, the intersection of the twomarginal value product curves determines the equilibrium wage, and the alloca-tion of labor in region 1 between industries A and B, as indicated on the x-axis.The right portion of panel (a) is interpreted similarly for region 2. For visualclarity, the figures were generated assuming equal wages across regions beforeany price changes. This assumption is not necessary for any of the theoreticalresults presented in the paper.

Panel (a) of Figure A1 shows an equilibrium in which wages are equalized acrossregions. Since region 1 is relatively well endowed with industry A specific factor,it allocates a greater share of its labor to industry A. Panel (b) shows the effectof a 50% decrease in the price of good A, so the good A marginal value productcurve in both regions moves down halfway toward the x-axis. Consistent with(1), the impact of this price decline is greater in region 1, which allocated a largerfraction of labor to industry A than did region 2. Thus, region 1’s wage falls morethan region 2’s wage. Now workers in region 1 have an incentive to migrate toregion 2. For each worker that migrates, the central vertical axis moves one unitto the left, indicating that there are fewer laborers in region 1 and more in region2. As the central axis shifts left, so do the two marginal value product curvesthat are measured with respect to that axis. This shift raises the wage in region1 and lowers the wage in region 2. Panel (c) shows the resulting equilibrium

29Figure A1 was generated under the following conditions. Production is Cobb-Douglas with specific-factor cost share equal to 0.5 in both industries. L = 10, T1A = 1, T1B = 0.4, T2A = 0.4, and T2B = 1.Initially, PA = PB = 1, and after the price change, PA = 0.5.

23

assuming costless migration and equalized wage across regions. Again, none ofthe theoretical results in the paper require this assumption.

A3. Nontraded goods prices

As above, consider a particular region, omitting the r subscript on all terms.Industries are indexed by i = 1...N . The final industry, indexed N , is nontraded,while other industries (i 6= N) are traded. The addition of a nontraded indus-try does not alter the results of the previous section, but makes it necessary todescribe regional consumers’ preferences to fix the nontraded good’s equilibriumprice.

Assume a representative consumer with Cobb-Douglas preferences over goodsfrom each industry. This implies the following relationship between quantitydemanded by consumers, Y c

i , total consumer income, m, and price, Pi.

(A17) Y ci = m− Pi

Consumers own all factors and receive all revenue generated in the economy;m =

∑i PiY

pi where Y p

i is the amount of good i produced in equilibrium. Apply-ing the envelope theorem to the revenue function for the regional economy, onecan show30

(A18) m = ηLL+∑i

ϕiPi.

where ηL is the share of total factor payments accounted for by wages and ϕiis the share of regional production value accounted for by industry i. Plugging(A18) into (A17) for the nontraded industry and rearranging terms,

(A19) PN −∑i 6=N

ϕi∑i′ 6=N ϕi′

Pi −ηL

1− ϕNL =

−1

1− ϕNY cN .

Holding labor fixed, this expression shows that consumption shifts away from thenontraded good if the nontraded price increases relative to a weighted averageof traded goods prices, with weights based on each good’s share of traded sectoroutput value.

A similar expression can be derived for the production side of the model. Wageequals the value of marginal product, so w = PN+FNL . The specific factor is fixed,

so Y pN = (1 − θN )LN . Combining these two observations with Euler’s theorem

30The revenue maximization problem is specified as follows.

r(P1...PN , L) = maxLi

∑i

PiFi(Li, Ti) s.t.

∑i

Li ≤ L

24

and the definition of the elasticity of substitution as in footnote 6 yields

(A20) w = PN −θN

σN (1− θN )Y pN

Substitute in the expression for w in (A13) and rearrange.

(A21) PN −∑i 6=N

βi∑i′ 6=N βi′

Pi +1

(1− βN )∑

i′ λi′σi′θi′

L =θN

(1− βN )σN (1− θN )Y pN

Holding labor fixed, this expression shows that production shifts toward the non-traded good if the nontraded price increases relative to a weighted average oftraded goods prices, with weights based on the industry’s size and labor demandelasticity captured in βi.

Equations (A19) and (A21) relate very closely to the intuition described inSection I.B for the case of only one traded good. Consumption shifts away fromthe nontraded good if its price rises relative to average traded goods prices whileproduction shifts toward it. Since regional consumption and production of thenontraded good must be equal, with a single traded good the nontraded pricemust exactly track the traded good price. Combining (A19) and (A21) showsthat in the many-traded-good case, the nontraded price change is a weightedaverage of traded goods price changes.

(A22) PN =

ηL − σNθN

(1−θN )∑i λi

σiθi∑

i′ 6=N

[σNθN

(1− θN )βi′ + ϕi′] L+

∑i 6=N

ξiPi

(A23) where ξi =

σNθN

(1− θN )βi + ϕi∑i′ 6=N

[σNθN

(1− θN )βi′ + ϕi′]

A4. Restriction to Drop the Nontraded Sector from Weighted Averages

Under Cobb-Douglas production with equal factor shares across industries, θi =θ, and σi = 1 ∀i. This implies that βi = λi, the fraction of labor in each industry.Similarly,

(A24) ϕi ≡PiY

pi∑

i′ Pi′Ypi′

=λiηL1− θ

= λi.

since ηL = 1− θ when θi = θ. When βi = ϕi, the weights ξi in (A22) are

(A25) ξi =βi∑

i′ 6=N βi′

25

Plug these weights into (3) and then plugging that into (1) yields a result equiv-alent to dropping the nontraded sector from the weighted average in (1) and(2).

(A26) w =

∑i 6=N βiPi∑i′ 6=N β

′i

A5. Wage Impact of Changes in Regional Labor

This section shows that an increase in regional labor decreases the regionalwage. Substituting (A22) into (A13), holding traded goods prices fixed (Pi =0 ∀i 6= N), and rearranging yields

(A27) w =

−σNθN

(1− θN )− (1− ϕN ) + λNσNθNηL

(∑

i λiσiθi

)[σNθN

(1− θN )(1− βN ) + (1− ϕN )] L.

Since the denominator is strictly positive, the sign of the relationship between Land w is determined by the numerator. An increase in regional labor will decreasethe wage if an only if

(A28) (1− ϕN ) >σNθN

(λNηL − (1− θN )).

Using the fact that ηL = (∑

i λi(1−θi)−1)−1 the previous expression is equivalentto

(A29) (1− ϕN ) >σNθN

(1− θN )

(λN (1− θN )−1∑i λi(1− θi)−1

− 1

).

The left hand side is positive while the right hand side is negative. Thus, theinequality always holds, and an increase in regional labor will always lower theregional wage.

26

Figure A1. Graphical Representation of Specific Factors Model of Regional Economies

(a) Initial Equilibrium

(b) Response to a Decrease in PA – Prohibiting Migration

(c) Response to a Decrease in PA – Allowing Migration

 

27

Data Appendix

B1. Industry Crosswalk

National accounts data from IBGE and trade policy data from (Kume, Pianiand de Souza 2003) are available by industry using the the Nıvel 50 and Nıvel80 classifications. The 1991 Census reports individuals’ industry of employmentusing the atividade classification system, while the 2000 Census reports industriesbased on the newer Classificacao Nacional de Atividades Economicas - Domiciliar(CNAE-Dom). Table B1 shows the industry definition used in this paper and itsconcordance with the various other industry definitions in the underlying datasources. The concordance for the 1991 Census is based on a crosswalk betweenthe national accounts and atividade industrial codes published by IBGE. Theconcordance for the 2000 Census is based on a crosswalk produced by IBGE’sComissao Nacional de Classificacao, available at http://www.ibge.gov.br/concla/.

B2. Trade Policy Data

Nıvel 50 trade policy data come from Kume, Piani and de Souza (2003). De-pending on the time period, they aggregated tariffs on 8,750 - 13,767 individualgoods first using unweighted averages to aggregate individual goods up to theNıvel 80 level, and then using value added weights to aggregate from Nıvel 80 toNıvel 50.31 In order to maintain this weighting scheme, I weight by value addedwhen aggregating from Nıvel 50 to the final classification listed in Table B1. Forreference, Figure B1 and Figure B2 show the evolution of nominal tariffs and effec-tive rates of protection in the ten largest sectors by value added. Along with thegeneral reduction in the level of protection, the dispersion in protection was alsogreatly reduced, consistent with the goal of aligning domestic production incen-tives with world prices. It is clear that the move from a high-level, high-dispersiontariff structure to a low-level low-dispersion tariff distribution generated substan-tial variation in protection changes across industries; industries with initially highlevels of protection experienced the largest cuts, while those with initially lowerlevels experienced smaller cuts.

Recall from Section III that the 1987-1990 period exhibited substantial tariff re-dundancy due to the presence of selective import bans and special import regimesthat exempted many imports from the full tariff charge. The tariff changes duringthis period did not generate a change in the protection faced by producers, butrather reflected a process called called tarifacao, or “tariffization,” in which theimport bans and special import regimes were replaced by tariffs providing equiv-alent levels of protection (Kume, Piani and de Souza 2003). The nominal tariffincreases in 1990 shown in Figure B1 reflect this tariffization process in whichtariffs were raised to compensate for the protective effect of the import bans thatwere abolished in March 1990 (Carvalho 1992).

31Email correspondence with Honorio Kume, March 12, 2008.

28

B3. Cross-Sectional Wage Regressions

In order to calculate the regional wage change for each microregion, I estimatestandard wage regressions separately in 1991 and 2000. Wages are calculated asan individual’s monthly earnings / 4.33 divided by weekly hours in their mainjob (results using all jobs are nearly identical). I regress the log wage on age,age-squared, a female indicator, an inner-city indicator, four race indicators, amarital status indicator, fixed effects for each year of education from 0 to 17+,21 industry fixed effects, and 494 microregion fixed effects. By running these re-gressions separately by survey year, I control for changes in regional demographiccharacteristics and for changes in the national returns to those characteristics.The results of these regressions are reported in Table B2. All terms are highlystatistically significant and of the expected sign. The regional fixed effects arenormalized relative to the national average log wage, and their standard errorsare calculated using the process described in Haisken-DeNew and Schmidt (1997).The regional wage change is then calculated as the difference in these normalizedregional fixed effects between 1991 and 2000:

(B1) d lnwr = (lnw2r000− lnw1

r991)− (lnw2000r − lnw1991

r )

The regional wage changes are shown in Figure 2. As mentioned in the maintext, some sparsely populated regions have very large measured regional wagechanges. To demonstrate that these results are driven by the data and not someartifact of the wage regressions, Figure B3 shows a similar map plotting regionalwage changes that were generated without any demographic or industry controls.They represent the change in the mean log wage in each region. The amount ofvariation across regions is similar, and these unconditional regional wage changesare highly correlated with the conditional versions used in the analysis, with acorrelation coefficient of 0.93.

B4. Region-Level Tariff Change Elements

The fixed factor share of input costs is measured as one minus the labor shareof value added in national accounts data. The resulting estimates of θi are plottedin Figure B4. The distribution of laborers across industries in each region (λri)are calculated in the 1991 Census using the sample of employed individuals, notenrolled in school, aged 18-55. The tariff changes in 5 are calculated from thenominal tariff data in Kume, Piani and de Souza (2003) and are shown in FigureB5.

29

FigureB1.NominalTariffTim

eline

Text

iles

Auto

, Tra

nspo

rt, V

ehic

les

Food

Pro

cess

ing

Non

met

allic

Min

eral

Man

f

Elec

tric

, Ele

ctro

nic

Equi

p.

Mac

hine

ry, E

quip

men

t M

etal

s Ag

ricul

ture

Ch

emic

als

Petr

oleu

m R

efin

ing

0 10

20

30

40

50

60

70

80

90

1984

19

85

1986

19

87

1988

19

89

1990

19

91

1992

19

93

1994

19

95

1996

19

97

1998

nominal tariff (percent)

30

FigureB2.EffectiveRateofProtectionTim

eline

Agric

ultu

re

Food

Pro

cess

ing

Met

als

Petr

oleu

m R

efin

ing

Mac

hine

ry, E

quip

men

t

Elec

tric

, Ele

ctro

nic

Equi

p.

Chem

ical

s

Auto

, Tra

nspo

rt, V

ehic

les

Text

iles

Non

met

allic

Min

eral

Man

f

0 20

40

60

80

100

120

140

160

1984

19

85

1986

19

87

1988

19

89

1990

19

91

1992

19

93

1994

19

95

1996

19

97

1998

effective rate of protection (percent)

31

Figure B3. Unconditional Regional Wage Changes

!!

!

!

!

!

!

!

!

!

!

BelémBelém

RecifeRecife

ManausManaus

CuritibaCuritiba

BrasíliaBrasília

SalvadorSalvador

FortalezaFortaleza

São PauloSão Paulo

Porto AlegrePorto Alegre

Belo HorizonteBelo Horizonte-15% to -5%-5% to 5%5% to 15%15% to 49%

-45% to -15%

Proportional wage change by microregion - normalized change in microregionfixed effects without demographic controls

32

Figure B4. Fixed Factor Share of Input Costs

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Non

trad

ed

Agric

ultu

re

Food

Pro

cess

ing

Met

als

Petr

oleu

m R

efin

ing

Mac

hine

ry, E

quip

men

t

Elec

tric

, Ele

ctro

nic

Equi

p.

Chem

ical

s

Auto

, Tra

nspo

rt, V

ehic

les

Text

iles

Non

met

allic

Min

eral

Man

uf

Pape

r, Pu

blish

ing,

Prin

ting

Petr

oleu

m, G

as, C

oal

Appa

rel

Woo

d, F

urni

ture

, Pea

t

Plas

tics

Phar

ma.

, Per

fum

es, D

eter

gent

s

Oth

er M

anuf

.

Min

eral

Min

ing

Foot

wea

r, Le

athe

r

Rubb

er

Fixe

d Fa

ctor

Sha

re o

f Inp

ut C

osts

om

ittin

g th

e co

st o

f int

erm

edia

te in

puts

Source: IBGE National Accounts data 1990 Sorted by industry value added in 1990 (largest to smallest)

33

Figure B5. Tariff Changes

-0.3

-0.25

-0.2

-0.15

-0.1

-0.05

0

0.05

Agric

ultu

re

Food

Pro

cess

ing

Met

als

Petr

oleu

m R

efin

ing

Mac

hine

ry, E

quip

men

t

Elec

tric

, Ele

ctro

nic

Equi

p.

Chem

ical

s

Auto

, Tra

nspo

rt, V

ehic

les

Text

iles

Non

met

allic

Min

eral

Man

uf

Pape

r, Pu

blish

ing,

Prin

ting

Petr

oleu

m, G

as, C

oal

Appa

rel

Woo

d, F

urni

ture

, Pea

t

Plas

tics

Phar

ma.

, Per

fum

es, D

eter

gent

s

Oth

er M

anuf

.

Min

eral

Min

ing

Foot

wea

r, Le

athe

r

Rubb

er

Chan

ge in

log

(1 +

tarif

f rat

e)

Source: Author's calculations based on Kume et al. (2003) Sorted by industry value added in 1990 (largest to smallest)

34

TableB1—

Indust

ryAggregationand

Concordance

Fina

l Ind

ustr

ySe

ctor

Nam

eN

ível

50

Nív

el 8

019

91 C

ensu

s (at

ivid

ade)

2000

Cen

sus (

CNAE

-Dom

)1

Agr

icul

ture

110

1-19

901

1-03

7, 0

41, 0

42, 5

8111

01-1

118,

120

1-12

09, 1

300,

140

1, 1

402,

200

1, 2

002,

50

01, 5

002

2M

iner

al M

inin

g (e

xcep

t com

bust

ible

s)2

201-

202

050,

053

-059

1200

0, 1

3001

, 130

02, 1

4001

-140

043

Petro

leum

and

Gas

Ext

ract

ion

and

Coa

l Min

ing

330

1-30

205

1-05

210

000,

110

004

Non

met

allic

Min

eral

Goo

ds M

anuf

actu

ring

440

110

026

010,

260

91, 2

6092

5Ir

on a

nd S

teel

, Non

ferr

ous,

and

Oth

er M

etal

Pro

duct

ion

and

Proc

essi

ng5-

750

1-70

111

027

001-

2700

3, 2

8001

, 280

028

Mac

hine

ry, E

quip

men

t, C

omm

erci

al In

stal

latio

n M

anuf

actu

ring,

and

Tra

ctor

Man

ufac

turin

g8

801-

802

120

2900

110

Elec

trica

l, El

ectro

nic,

and

Com

mun

icat

ion

Equi

pmen

t and

Com

pone

nts M

anuf

actu

ring

10-1

110

01-1

101

130

2900

2, 3

0000

, 310

01, 3

1002

, 320

00, 3

3003

12A

utom

obile

, Tra

nspo

rtatio

n, a

nd V

ehic

le P

arts

Man

ufac

turin

g12

-13

1201

-130

114

034

001-

3400

3, 3

5010

, 350

20, 3

5030

, 350

9014

Woo

d Pr

oduc

ts, F

urni

ture

Man

ufac

turin

g, a

nd P

eat P

rodu

ctio

n14

1401

150,

151

, 160

2000

0, 3

6010

15Pa

per M

anuf

actu

ring,

Pub

lishi

ng, a

nd P

rintin

g15

1501

170,

290

2100

1, 2

1002

, 220

0016

Rub

ber P

rodu

ct M

anuf

actu

ring

1616

0118

025

010

17C

hem

ical

Pro

duct

Man

ufac

turin

g17

,19

1701

-170

2, 1

901-

1903

20

023

010,

230

30, 2

3400

, 240

10, 2

4090

18Pe

trole

um R

efin

ing

and

Petro

chem

ical

Man

ufac

turin

g18

1801

-180

620

1, 2

02, 3

52, 4

7723

020

20Ph

arm

aceu

tical

Pro

duct

s, Pe

rfum

es a

nd D

eter

gent

s Man

ufac

turin

g20

2001

210,

220

2402

0, 2

4030

21Pl

astic

s Pro

duct

s Man

ufac

turin

g21

2101

230

2502

022

Text

iles M

anuf

actu

ring

2222

01-2

205

240,

241

1700

1, 1

7002

23A

ppar

el a

nd A

ppar

el A

cces

sorie

s Man

ufac

turin

g23

2301

250,

532

1800

1, 1

8002

24Fo

otw

ear a

nd L

eath

er a

nd H

ide

Prod

ucts

Man

ufac

turin

g24

2401

190,

251

1901

1, 1

9012

, 190

2025

Food

Pro

cess

ing

(Cof

fee,

Pla

nt P

rodu

cts,

Mea

t, D

airy

, Sug

ar, O

ils, B

ever

ages

, and

Oth

er)

25-3

125

01-3

102

260,

261

, 270

, 280

1501

0, 1

5021

, 150

22, 1

5030

, 150

41-1

5043

, 150

50,

1600

032

Mis

cella

neou

s Oth

er P

rodu

cts M

anuf

actu

ring

3232

0130

033

001,

330

02, 3

3004

, 330

05, 3

6090

, 370

0099

Non

trade

d G

oods

and

Ser

vice

s33

-43

3301

-430

134

0, 3

51-3

54, 4

10-4

24, 4

51-4

53, 4

61, 4

62, 4

64,

471-

476,

481

, 482

, 511

, 512

, 521

-525

, 531

, 533

, 54

1-54

5, 5

51, 5

52, 5

71-5

78, 5

82-5

89, 6

10-6

19,

621-

624,

631

, 632

, 711

-717

, 721

-727

, 901

, 902

1500

, 400

10, 4

0020

, 410

00, 4

5001

-450

05, 5

0010

, 50

020,

500

30, 5

0040

, 500

50, 5

3010

, 530

20, 5

3030

, 53

041,

530

42, 5

3050

, 530

61-5

3068

, 530

70, 5

3080

, 53

090,

531

01, 5

3102

, 531

11-5

3113

, 550

10, 5

5020

, 55

030,

600

10, 6

0020

, 600

31, 6

0032

, 600

40, 6

0091

,60

092,

610

00, 6

2000

, 630

10, 6

3021

, 630

22, 6

3030

, 64

010,

640

20, 6

5000

, 660

00, 6

7010

, 670

20, 7

0001

, 70

002,

710

10, 7

1020

, 710

30, 7

2010

, 720

20, 7

3000

, 74

011,

740

12, 7

4021

, 740

22, 7

4030

, 740

40, 7

4050

, 74

060,

740

90, 7

5011

-750

17, 7

5020

, 800

11, 8

0012

, 80

090,

850

11-8

5013

, 850

20, 8

5030

, 900

00, 9

1010

, 91

020,

910

91, 9

1092

, 920

11-9

2015

, 920

20, 9

2030

, 92

040,

930

10, 9

3020

, 930

30, 9

3091

, 930

92, 9

5000

, 99

000

35

Table B2—Cross-Sectional Wage Regressions - 1991 and 2000 Census

Year 1991 2000

Age 0.057 0.061(0.000)** (0.000)**

Age2 / 1000 -0.575 -0.601(0.004)** (0.004)**

Female -0.392 -0.335(0.001)** (0.001)**

Inner City 0.116 0.101(0.001)** (0.001)**

RaceBrown (parda) -0.136 -0.131

(0.001)** (0.001)**Black -0.200 -0.173

(0.002)** (0.001)**Asian 0.154 0.122

(0.006)** (0.006)**Indigenous -0.176 -0.112

(0.010)** (0.006)**

Married 0.186 0.163(0.001)** (0.001)**

Fixed EffectsYears of Education (18) X XIndustry (21) X XMicroregion (494) X X

Observations 4,721,996 5,135,618R-squared 0.518 0.500

Robust standard errors in parentheses+ significant at 10%; * significant at 5%; ** significant at 1%Omitted category: unmarried white male with zero years of education, outside inner city,

working in agriculture

dependent variable: log wage = ln((monthly earnings / 4.33) / weekly hours) at main job

0

0.5

1

1.5

2

2.5

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

Educ

atio

n fix

ed e

ffect

est

imat

e(r

elat

ive

to n

o ed

ucat

ion)

Years of education

1991

2000

36