Explaining the decline in earnings

inequality in Brazil, 1995-2012

Francisco H.G. Ferreira*

Sergio Firpo #

Julian Messina*

* The World Bank and IZA # Sao Paulo School of Economics at FGV and IZA

Plan of the talk

1. The question

2. The suspects

3. Data

4. Methodology

5. Results

2

The question:

What accounts for the decline in the Gini

coefficient for labor incomes between

1995 (0.50) and 2012 (0.40) in Brazil?

3

Context: A broader decline in inequality (in

household income per capita, 2001-2011)

��� ����

����

1����

�

�∈�� 1�����

�

�∈�

-8-18-20-45-13 4

Sha

ple

y V

alu

e E

stim

ate

s o

f th

e C

on

trib

uti

on

s to

th

e d

ecl

ine

in

the

Gin

i Co

eff

icie

nt

Others non labor income by adult Income from pension by adult Income from transfer by adult

Labor income by occupied Share of adults Income from capital by adult

Share of occupied by adults

Change in

Gini

coefficient:

6.5 pts

Source: figure based on analysis in Azevedo, Inchauste and Sanfelice (2013)

Three key factors drove the fall in inequality: demographics, labor incomes, and transfers.

Labor income levels:1995-2012

5Note: median and average earnings calculated over estimating sample (formal, informal and self employed of ages 18-65)

781.5

445.0

486.4

671.9

431.3

467.1

893.6

637.1

670.0

400

500

600

700

800

900R

eale

s (2

005)

1995 2000 2005 2010Year

Average laborincome

Median laborincome

Average householdper capita income

Labor income inequality: 1995-2012

6Note: labor income Gini coefficient calculated over estimating sample (formal, informal and self employed aged 18-65). Household per

capita income Gini calculated over the entire population. The confidence intervals were computed using jack-knife standard errors.

0.59

0.50

0.58

0.47

0.52

0.40.4

.45

.5

.55

.6

Gin

i In

dex

1995 2000 2005 2010year

Labor income 95% CIHousehold percapita income

95% CI

7

0.94

0.89

0.92

0.80

0.66

0.63

.5

.6

.7

.8

.9

119

95=

1

1995 2000 2005 2010Year

Gini (1995=0.50) Theil (1995=0.45) P90/P10 (1995=10.00)

Labor income inequality: 1995-2012

The suspects:

• Human Capital

- Increases in the supply of education

- Declines in the returns to education

- Aging population

• Labor market institutions

- A rising real minimum wage

- Changes in the formal / informal composition of the labor force

• Demographics

- Female participation

- Racial gaps

• Spatial

- Urban/Rural gaps

- Regional disparities 8

Human Capital

9

10

Increases in the supply of education

0

.2

.4

.6

.8

1fr

acti

on

in t

he

po

pu

lati

on

age

d 1

8-65

0 5 10 15 20years of education

1996 2003 2012

cdf years of education

The evolution of (potential) experience

11

52.7 50.745.8

89.8 89.886.6

25.0 25.220.4

0

20

40

60

80

100

% p

opula

tion b

etw

een 1

8-6

5 ye

ars

1995 2000 2005 2010Year

56-65 46-55 36-45

26-35 18-25

Labor market institutions

12

A rising real minimum wage

13Note: median and average earnings calculated over estimating sample (formal, informal and self employed of ages 18-65)

0.86

0.97

1.201.14

1.43

2.03

.5

1

1.5

2

1995

=1

1995 2000 2005 2010Year

Average Earnings Median Earnigs Minimum Wage

11.916.8 16.2

0

20

40

60

80

100

% o

f th

e o

ccup

ied

pop

ulat

ion

bet

wee

n 1

8-65

year

s

1995 2000 2005 2010Year

At or above min W Below

14

Density of the distribution of monthly labor

earnings (with relevant minimum wages)

95-96min w

02-03min w

11-12min w

0

.2

.4

.6

.8

1kd

ensi

ty ln

w

4 6 8 10log of wage in 2005 Reales

1995-96 2002-03 2011-12

The bandwith is .07 for all the periods

Formalization

15

27.4

51.8

25.3

52.5

21.3

42.3

0

20

40

60

80

100%

of

occ

upie

d p

op

ulat

ion

bet

wee

n 1

8-65

year

s

1995 2000 2005 2010Year

Formal employees Informal employees Self employment

Demographics

16

Gender and racial composition of the labor force

17

43.5 45.5

53.3

5.2 6.08.7

0

20

40

60

80

100

% o

f p

op

ulat

ion

bet

wee

n 1

8-65

yea

rs

1995 2000 2005 2010Year

Withe Mestizo Indigena&other African american

37.9 39.942.6

0

20

40

60

80

100

% o

f th

e o

ccup

ied

pop

ula

tio

nbet

wee

n 1

8-65

year

s

1995 2000 2005 2010Year

Male Female

Geography

18

Urbanization and regional composition

19

77.2 77.3 77.7

93.1 92.8 92.3

31.1 31.9 34.5

26.8 26.5 26.5

0

20

40

60

80

100

% o

f p

op

ula

tion

bet

wee

n 1

8-65

yea

rs

1995 2000 2005 2010Year

Center west South Southest

North other Northest

18.914.4 14.0

0

20

40

60

80

100

% p

op

ula

tio

n b

etw

een

18-

65 y

ears

1995 2000 2005 2010Year

Urban Rural

Market Structure (OLS estimators)

20

age 18-25

age 26-35

age 36-45

age 46-55

primary or less

Secontary incomplete

Secondary

Terciary Incomplete

Below minimum wage

Self employment

Informal

Mestizo

Indigenous& other

African american

Female

Rural

Northest

North

Southest

South

-1.5 -1 -.5 0 .5

1995-96 2011-12 95% CI

Geography

Demography

Labor market institutions

Human Capital

21

Returns to observable worker characteristics

1995-96 2002-03 2004-05 2011-12

coef/se coef/se coef/se coef/se

18-25 -0.189*** -0.283*** -0.285*** -0.285***

(0.007) (0.006) (0.005) (0.005)

26-35 0.028*** -0.060*** -0.077*** -0.110***

(0.007) (0.005) (0.005) (0.005)

36-45 0.149*** 0.046*** 0.030*** -0.026***

(0.007) (0.005) (0.005) (0.005)

46-55 0.121*** 0.074*** 0.070*** 0.018***

(0.007) (0.006) (0.006) (0.005)

Primary or less -1.228*** -1.150*** -1.082*** -0.907***

(0.007) (0.006) (0.005) (0.005)

Secondary incomplete -0.959*** -0.976*** -0.922*** -0.790***

(0.008) (0.006) (0.006) (0.005)

Secondary -0.696*** -0.777*** -0.738*** -0.654***

(0.008) (0.006) (0.006) (0.004)

Tertiary Incomplete -0.466*** -0.479*** -0.453*** -0.417***

(0.010) (0.008) (0.007) (0.006)

Below minimum wage -1.082*** -1.073*** -1.070*** -0.974***

(0.004) (0.003) (0.003) (0.003)

Self employment 0.049*** -0.031*** -0.038*** 0.067***

(0.004) (0.003) (0.003) (0.003)

Informal -0.193*** -0.131*** -0.103*** -0.045***

(0.004) (0.003) (0.003) (0.003)

‘Mestiço’ -0.137*** -0.102*** -0.097*** -0.081***

(0.003) (0.003) (0.002) (0.002)

Indigenous & other 0.197*** 0.061*** 0.073*** 0.034**

(0.025) (0.018) (0.018) (0.013)

African-American -0.183*** -0.119*** -0.109*** -0.092***

(0.006) (0.004) (0.004) (0.003)

Female -0.413*** -0.328*** -0.323*** -0.287***

(0.003) (0.002) (0.002) (0.002)

Rural -0.235*** -0.120*** -0.103*** -0.125***

(0.004) (0.003) (0.003) (0.003)

Northeast -0.214*** -0.218*** -0.229*** -0.216***

(0.005) (0.004) (0.003) (0.003)

North -0.055*** -0.094*** -0.075*** -0.106***

(0.007) (0.004) (0.004) (0.004)

Southeast 0.085*** 0.024*** -0.005 -0.018***

(0.005) (0.004) (0.003) (0.003)

South -0.001 -0.025*** -0.019*** -0.008**

(0.005) (0.004) (0.004) (0.004)

Data

• Pesquisa Nacional por Amostra de Domicílios (PNAD). Annual household survey carried out by the

Instituto Brasileiro de Geografía e Estatística (IBGE).

• Periods compared are “paired years”: 1995-1996, 2002-2003,2004-2005 and 2011-2012

• Wage measure: total (gross) individual monthly labor earnings.

Sample for analysis:

• Working age population: 18-65

• Men and women, in rural and urban areas.

• Employees and self employed.

• Further, we distinguish between formal and informal employees

– An employee is informal if (s)he does not have Carteira de Trabalho.

• Trimming: top and bottom percentiles of the distribution omitted.

• We exclude the rural North of the country, which was included in PNAD only after 2004

• Monetary values are in constant 2005 R$.

22

MethodologyA generalized Oaxaca-Blinder decomposition

• Consider two time periods, A and B. We can express the overall change in the

distributional statistic � of wages Y over time as

∆��� � ���|���� − � ���|���� ,

where F is the cumulative distribution of wages �� � and ! � 1 is an

indicator of group B membership.

– Think of ���|���� as the integral of the density function

– Define a counterfactual distribution ���|���� analogously, as the integral of:

23

( ) ( ) ( )dXXXygyf BBB φ∫∫∫=

( ) ( ) ( )dXXXygyf BAs φ∫∫∫=

MethodologyA generalized Oaxaca-Blinder decomposition

• Adding and subtracting the counterfactual distribution statistic � ���|���� :

– ∆��= � ���|���� − � ���|���� + � ���|���� − � ���|����

– Structure effect: � ���|���� − � ���|����

– Composition effect: � ���|���� − � ���|����

• If interested in mean wages the OB decomposition is straightforward.

Applying the law of iterated expectations we can use the regression

coefficients and sample means to perform this decomposition:

– ∆�"

= #$! %&! − %&� + #$! − #$� %&�

24

Methodology:Firpo, Fortin and Lemieux (2009)

• Other distributional statistics are more complex. Non-linearities preclude us

from applying the law of iterated expectations.

• FFL suggest using the recentered influence function (RIF)

• The influence function of a distributional statistic represents the influence of

an individual observation to that distributional statistic: '�(�; �)

• The RIF adds back the distributional statistic: +'� = � + '�(�; �)

• A convenient feature of this RIF function is that E(+'�) = �

• Hence, we can apply OLS to obtain regression coefficients from RIF

transformed variables and go back to our traditional Oaxaca-Blinder

decompositions. This allows us to build counterfactuals for any distributional

statistic with a known influence function.

• In our case, we build counterfactuals for the mean and the Gini coefficient.25

Methodology: Detailed Specification

With indicator variables representing a categorical variable (dummies), the

Oaxaca decomposition is not invariant to the choice of the excluded category

(Oaxaca and Ransom (1999) and Yun (2005)).

26

-./ 01 2345 = 6 + 789:31 ;3<=>30 + ?@5:A4B3<C=DE +

FGHIJKLMN� + O�PQ�QRSR TLJH + OU'�VIKRLWQX� + Y

Omitted Categories: Best performers (white males , tertiary education completed in the age

bracket 56-65, urban center-west, being at or above the minimum wage and being a formal

employee)

Changes in Average Earnings in Brazil.

Detailed Decompositions

27

-.2

0

.2

.4

Total Endowments Structure Total Endowments Structure Total Endowments Structure

1995-2012 1995-2003 2004-2012

Human Capital Gender&Race Urban/Rural&Regions Min wage Informality Constant

Ch

ange

in

lo

g(W

)

Changes in Gini Earnings Inequality in

Brazil. Detailed Decompositions

28

-.15

-.1

-.05

0

.05

Total Endowments Structure Total Endowments Structure Total Endowments Structure

1995-2012 1995-2003 2004-2012

Human Capital Gender&Race Urban/Rural&Regions Min wage Informality Constant

Ch

ange

in

Gin

i

Summary of main findings

• Decline in earnings inequality between 1995 and 2012 was driven

primarily by changes in the structure of remuneration in the Brazilian

labour market.

• In fact, changes in the distribution of worker characteristics were

inequality enhancing, in particular human capital (‘paradox of growth’?).

• The negative pay structure effect occurred because of declines in various

different wage premia:

• Falling schooling premia;

• Reductions in the gender wage gap;

• Reductions in the racial wage gap;

• Reductions in the urban-rural wage gap;

• Reductions in the pay gap between formal and informal workers.

29

The role of Minimum Wage

• Real minimum wage, which more than doubled over the period.

• That increase generated a formidable spike in the density function of

earnings by 2012

• As suspected, this rise in the minimum wage contributed to falling

inequality in the 2004-2012 sub-period.

• However, for the period between 1995 and 2003, increases in MW rose

inequality through composition effect.

• There was a steady increase in the proportion of employed workers

earning strictly less than MW.

• Thus, because labour market was ‘softer’ (higher unemployment rates) in

the first sub-period, it meant that the overall impact of minimum wages in

the whole period was inequality-increasing.

30

Conclusions

• In contrast to earlier documented periods – the story of these seventeen years was a happy one in Brazilian labour markets:

– Unemployment fell and earnings rose.

– Not only did average earnings rise, but they rose by most for those groups of workers who used to earn the least.

– There was a compression in the schooling wage premia, which used to be unusually large in Brazil.

– Even more impressive were the reductions in wage gaps among workers that are observationally equivalent in terms of their human capital, but differ along such dimensions as race, gender, location and type of job.

31

Implications for Africa

• Schooling: (a) increases in productivity and (b) if focus is on primary and secondary levels, it leads to greater prosperity and greater equity;

• Discrimination: gender, ethnic/racial, or other forms – tend to be both inefficient and inequitable. – Encouraging female education, reduction in fertility rates, and greater labour

force participation has contributed to growth in average earnings, and to a less unequal distribution in Brazil.

• Regional integration: integration of rural areas, and the workers who live there:– Greater connectivity and less labour market segmentation between cities and

the countryside are an ongoing part of Brazil’s recently successful fight against poverty and inequality.

• Fiscal redistribution is still important (not to be the only tool for policy):– Well-designed transfer programs are perfectly consistent with vibrant labour

markets, with rising average wages and declining dispersion.

32

Thank you!

• Questions or comments:

sergio.firpo@fgv.br

• More on interaction between Africa and Brazil coming soon:

– The CLEAR Regional Center for Brazil and Lusophone Africa -to

be launched soon- hosted by Sao Paulo School of Economics at

FGV.

33