Measuring Income Inequality using decomposition method - - PowerPoint PPT Presentation

rosalba.manna@uniparthenope.it

Rosalba Manna Measuring Income Inequality using decomposition method regression-based: Empirical analysis in a multidimensional perspective

Aim

 To measure the relative contributions of individual

as well as household factors to the explanation of the inequality in individual disposable incomes in Italy

 A regression-based decomposition technique was

implemented, following the Shapley approach

 The analysis exploited the potential of panel data,

with reference to the pooling of observations on a cross-section

• f

individuals

• ver

several time periods

2

Inequality Decomposition

Traditional Decomposition Income Source Population Subgroups Regression-based Decomposition Shapley-Value Approach Fields

3

Shapley-Value Approach

Regression-based decompositon Traditional decomposition Income Source Population Subgroups Fields Shapley-Value Approach

4

Income generating model

 First step:

– specification

and estimation

• f

an income generating function

– the choice of the functional form is dictated by the

standard Mincer model

– the log of income is regressed on some explanatory

variables accounting for individual (human capital) and household (physical capital) characteristics.

5

Model Selection

 Second step:

–

A model for longitudinal data with random effects is selected

–

This model allows to capture the heterogeneity between individuals in several time periods and between the same individual observed in several time periods

it i it it

u c x y + + = β log

N i T t ,..., 2 , 1 ,..., 2 , 1 = =

6

Model Assumptions

 The individual effects of this model are strictly

uncorrelated with the regressors

 This

hypotesis allows to include time-invariant regressors

 This is not possible when we estimate the fixed

effects model

7

Survey of Bank of Italy on Household Income and Wealth

 The data come from the Survey of Household

Income and Wealth (SHIW) from which we selected the information

• n

the income earners who have been successfully interviewed every two years from 2004 to 2014 (balanced panel)

 Such an information includes a large number

• f individuals N (N=1712) observed over a

short time period T (T=6 years covering on a whole time span of 10 years).

8

Variables in the model

Variables Variable name Description Log of Income Y Net individual disposable income (euro) Gender GENDER =1 for male =0 for female Education EDUCATION Years of completed study Age AGE age Age squared AGE2 Age squared Work status WORK =1 for employes =0 for self employed Geographical Area AREA =1 North and Center =0 South and Islands Household wealth WEALTH Household real and financial wealth (thousands of euro)

9

Model Estimation

10

Interpretation output

 The signs of the coefficients are in line with the

theory

 Significant income gaps are found by gender, level

• f education, work status and geographical area

 The concavity of income-age profile is confirmed  Larger income flows are associated with larger

stocks of wealth

11

Inequality decomposition through Shapley approach

 The Gini index I(Y) calculated on predicted

income values is expressed as the sum of the contributory factors

( ) ( )

I , X Y I

i k 1 i

Φ = ∑

=

12

Shapley Value Approach

 The contribution of each

is estimated through a sequence of regression models starting from the satured model and then proceeding by eliminating each regressor in succession. Since the contribution of any factor depends on the

• rder in which the factor appears in the elimination

sequence, the overall marginal contribution is given by the average calculated over all the possible elimination sequences

( )

I , Xi Φ

i

X

13

Gini Decomposition

Variables

Absolute Contribution Percentage Contribution

Gender

10.60 30.89

Education

11.80 32.29

Age

3.50 10.20

Work

3.19 9.30

Area

2.82 8.22

Wealth

0.53 1.54

Total Explained Inequality

31.72 92.45

Unexplained Inequality

2.59 7.55

Observed Inequality

34.31 100.00

14

Results

 Education plays a dominant role in the explanation of

• bserved inequality

 Gender has the second largest contribution  Less importance is accorded to household wealth  Age, work status and area of residence affect the

income differentials only in a residual way

15

Thanks for your attention

Rosalba Manna

rosalba_manna@yahoo.it