+ Growth and inequality: what are we talking about? A within-between distinction among inequality of opportunity and inequality of effort Geoffrey TEYSSIER (Supervisor: Charlotte Guénard; Co-supervisor: Sandra Poncet
Feb 22, 2016
+
Growth and inequality: what are we talking about? A within-between distinction among inequality of opportunity and inequality of effort
Geoffrey TEYSSIER
(Supervisor: Charlotte Guénard; Co-supervisor: Sandra Poncet
+PLAN
I. Quick overview of the thesis defended (empirical objectives and the benchmark specification)
II. Micro part(micro dataset construction and micro results)
III. Macro part(summary statistics and preliminary macro results)
IV. To be done…
+I. INEQUALITY IS LIKE CHOLESTEROL !!! Growth-inequality: a puzzling relationship
no agreement in the litterature are we in a dead-end? NO
Good (IE) vs bad inequality (IO)
income is function of 2 kinds of factors only: those factors under your control: study, work, … those factors outside your control: skin color, social
background (eg: parental education), …
income inequality due to what you can control is the inequality of effort (IE) and is morally fair
income inequality due to what you CANNOT control is the inequality of opportunity (IO) and is morally unfair
+I. Thesis defended: what is morally fair is also economically efficient(and vice-versa) By distinguishing between IO and IE, we can explain the
« growth – inequality paradox ». The benchmark specification is:
Growthi(t,t+9)= Ineqit + Control_variablesit + Ri + Tt +it+9
The empirical objective is to show that: If Ineq= IO
is negative and significant If Ineq= IE:
is positive and significant If Ineq=Itot
is not robust (negative if IO dominates, positive if IE dominates)
+II. MICRO PART : measures of inequality How to measure IO?
Total inequality= IO + IE
IO: inequality between groups defined by common « circumstances » (ie factors outside the individual’s control)
IE: inequality within groups
Need individual level data on income and circumstances (father’s education, father’s occupation, gender, skin color)
+II.MICRO DATASET
Sample restriction:
Positive income reported
Aged between 20 and 49 years old
With all circumstances observed
1980, 1991, 2000 Brazilian Censuses
taken from the IPUMS
Huge: million of observations for each year
Weights and survey design taken into account
+II. 2 adjustements for income
Adjustment for the time profile of the individual (« composition effect ») Because I do not want to take into account inequality due to
age
Adjustment for sample bias: because father’s education and occupation are only observed
for those people living in the same household as their father
+II. 2*2*4*4=64 groups defined
Race: White (or asian) Non white
Sex: Male Female
Father’s education No education Primary (1-4) Primary (5-8) Secondary or +
Father’s occupation 3 groups for active 1 group for not economicaly
active father (Direct question to the father) could not be entirely outside
the individual’s control One specification without
father’s occupation as a robustness check
So as to capture the most circumstances possible (otherwise, IO is underestimated), while having a reasonable number of observations within each group (otherwise, IO is not accurate)
II. Micro results: the inequality measures
80 observations: (26 regional states + 1 fedeferal state) -1 observation for the state of Toscantin which did not exist in 1980 « not adj »: inequality calculated on income distributions not adjusted for age and sample biasprior to the 2 adjustments: • IO was overestimated • total inequality was even more overestimated
+III. MACRO PART: growth inequality regression Growthi(t,t+9)= Ineqit + Control_variablesit + Ri + Tt
+it+9 unit of observration: Brazilian state i at time t+9 (or alternatively
t, depending on how we see things) Growthi(t,t+9): growth of GDP per capita data on GDP (at constant 2000 prices) and population,
Growthi(t,t+9) is computed as the difference of their growth rates multiplied by 100
Ri: regional dummy (central western region omitted) Tt: year dummies at time t (1980 and 1991, while 2000 omitted) Control variablesit:
State’s GDP at time t State’s public welfare expenditures (education and culture;
health and sanitation; social security and redistributive programs) at time t many more still to be included
+III. Evolution over time
IO and Itot have decreased since 1991but IO has decreased over the whole period, while Itot hase increased
Growth: economic stagnation in the 1990s
+III. Simple correlation: IO on growth (negative but not signigificant)
+III. Simple correlation: Itot on growth (negative but not signigificant)
+IV. To be done… ENDOGENEITY PB:
IV: Easterly but time invariant GMM Other leads to explore:
intergenerational mobility literature?
OTHER MEASURES OF IO Recent paper suggests an upper-
bound interesting because IO is otherwise
necessarily a lowe-bound Other measures based on another
definition of types and on a parametric method as a robustness
CONTROLS TO BE INCLUDED Other from Marrerro&Rodriguez
(benchmark paper) Proportion of people belonging
to each category of the circumstances variable (to be sure that IO does not capture the proportion of disadvantaged people)
Determinants of growth specific to Brazilian states
STRUCTURAL FORM EQUATIONS In order to investigate the
mechanism between IO and growth