160 likes | 308 Views
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
E N D
Growthand inequality: what are wetalkingabout? A within-between distinction amonginequality of opportunity and inequality of effort Geoffrey TEYSSIER (Supervisor: Charlotte Guénard; Co-supervisor: Sandra Poncet
PLAN • I. Quick overview of the thesisdefended (empirical objectives and the benchmark specification) • II. Micro part (micro dataset construction and micro results) • III. Macro part (summarystatistics and preliminary macro results) • IV. To bedone…
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) • incomeisfunction of 2 kinds of factorsonly: • thosefactorsunderyour control: study, work, … • thosefactorsoutsideyour control: skin color, social background (eg: parental education), … • incomeinequality due to whatyoucan control is the inequality of effort (IE) and ismorallyfair • incomeinequality due to whatyouCANNOT control is the inequality of opportunity (IO) and ismorallyunfair
I. Thesisdefended: whatismorallyfairisalsoeconomically efficient(and vice-versa) • By distinguishingbetween IO and IE, wecanexplain the « growth – inequalityparadox ». The benchmark specificationis: Growthi(t,t+9)= Ineqit + Control_variablesit + Ri + Tt +it+9 • Theempirical objectiveis 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: inequalitybetweengroups defined by common « circumstances » (iefactorsoutside the individual’s control) • IE: inequalitywithingroups Needindividuallevel data on income and circumstances(father’seducation, father’s occupation, gender, skin color)
II.MICRO DATASET • Sample restriction: • Positive incomereported • Agedbetween 20 and 49 yearsold • With all circumstancesobserved • 1980, 1991, 2000 BrazilianCensuses • takenfrom the IPUMS • Huge: million of observations for eachyear • Weights and survey design takenintoaccount
II. 2 adjustements for income • Adjustment for the time profile of the individual (« composition effect ») • Because I do not want to takeintoaccountinequality due to age • Adjustment for samplebias: • becausefather’seducation and occupation are onlyobserved for those people living in the samehousehold as theirfather
II. 2*2*4*4=64 groups defined So as to capture the mostcircumstances possible (otherwise, IO isunderestimated), whilehaving a reasonablenumber of observations withineach group (otherwise, IO is not accurate) • Race: • White (or asian) • Non white • Sex: • Male • Female • Father’seducation • 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 beentirelyoutside the individual’s control • One specificationwithoutfather’s occupation as a robustness check
II. Micro results: the inequalitymeasures • 80 observations: (26 regional states + 1 fedeferal state) -1 observation for the state of Toscantinwhichdid not exist in 1980 • « not adj »: inequalitycalculated on income distributions not adjusted for age and samplebias • prior to the 2 adjustments: • IO wasoverestimated • total inequalitywaseven 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 weseethings) • Growthi(t,t+9): growth of GDP per capita • data on GDP (at constant 2000 prices) and population, Growthi(t,t+9) iscomputed as the difference of theirgrowth rates multiplied by 100 • Ri: regionaldummy (central western regionomitted) • Tt: yeardummiesat time t (1980 and 1991, while 2000 omitted) • Control variablesit: • State’s GDP at time t • State’s public welfareexpenditures (education and culture; health and sanitation; social security and redistributive programs) at time t many more still to beincluded
III. Evolution over time • IO and Itot have decreasedsince 1991 • but IO has decreased over the wholeperiod,whileItot 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 bedone… • ENDOGENEITY PB: • IV: Easterly but time invariant • GMM • Otherleads to explore: intergenerationalmobilityliterature? • OTHER MEASURES OF IO • Recentpapersuggestsan upper-bound • interestingbecause IO isotherwisenecessarily a lowe-bound • Othermeasuresbased on anotherdefinition of types and on a parametricmethod as a robustness • CONTROLS TO BE INCLUDED • OtherfromMarrerro&Rodriguez (benchmark paper) • Proportion of people belonging to eachcategory of the circumstances variable (to be sure that IO does not capture the proportion of disadvantaged people) • Determinants of growthspecific to Brazilian states • STRUCTURAL FORM EQUATIONS In order to investigate the mechanismbetween IO and growth