Chapter 2 Behavioral micro-simulation models: the case of labor - - PowerPoint PPT Presentation

1

Chapter 2 Behavioral micro-simulation models: the case

• f labor

supply

François Bourguignon Paris School

• f Economics

M2-PPD, M2-APE, 2009-10

2

Outline

1. The case of labor supply

a) The problem

• f the non-linear

budget constraint b) Simple linear approaches c) Continuous non-linear rational models d) Discrete non-linear models e) Example: the WFTC in UK

2. The case of child labor and the demand for schooling in developing countries

Example: the Bolsa Familia program in Brazil

3. Other types of behavior

3

d) Using discrete labor choice models

• Present

models use a discrete choice formulation (Van Soest, 1995; Hoynes, 1996; Blundell and MaCurdy, 1999)

• More flexibility

can be

• btained

when individuals are supposed to choose from a finite set of working times (including zero) Dj .

• Conventional

multi-logit model: where ε distributed as independent double exponential and u( ) an unrestricted quadratic function

• f its

(z, C) arguments.

j k ε ; β ; C u(z U ε ) ; β ; C u(z U D L

k i k k i i k i j i j j i i j i j i

≠ + = ≥ + = = all for ) if

) ; ; , , ( γ

i j j i

• i

j i

• i

j i

z D D w y TB D w y C − + =

4

Using discrete labor choice models (2)

• Estimates
• f coefficients of quadratic

form

• btained

from multinomial logit

• model. Likelihood
• f an observation:

where is the (observed) duration

• f work

by agent i

• "Pseudo-residuals" obtained

from drawing randomly in the double exponential law under conditions

{ }

[ ]

) ( ; ] ; ; ( [ ) ; ; ( Pr

* * *

TB D w y C C z u Exp C z u Exp D L

k i i k i k k k i i j j i i j i

i i i

− + = = =

∑

β β

i j k i k k i i j i j j i i

L D with j k ε β ; C ; u(z ε ) β ; C ; u(z = ≠ + ≥ + all for ) ) ) ) )

j

β )

j i

ε )

* i

j

5

Using discrete labor choice models (3)

• Simulation of modifying

the tax-benefit system to TB*(…, γ* ): obtained by solving numerically the utility maximizing problem: Disposable income is given by: *) ; ; , , ( * γ

i j j i

• i

j i

• i

js i

z D D w y TB D w y C − + =

j k all for ) ε , β ; C ; u(z ) ε , β ; C ; u(z such that D L

k i k ks i i j i j js i i j s i

≠ ≥ = ˆ ˆ ˆ ˆ

*) ; ; , , ( * γ

i j j i

• i

j i

• i

s i

z D D w y TB D w y C − + =

6

Example: Effect

• f the Working

Family Tax Credit in UK (Blundell et al. 2000)

7

10 20 30 50 100 200 300 400 L C

Budget constraint example for lone parent with childcare costs

w

8

9

10

11

Final remarks

• n labor

supply models

• The issue of participation and non-observed

wages (Heckman, 1974, 1979)

• The role
• f labor

demand

– Are unemployed "unemployable" or voluntarily inactive? Identifying through specification

• f the distribution of heterogeneity

(Laroque and Salanié, 1999) ?

• Households
• vs. Individuals

– Options = duration

• f work

for both members

• f a couple, easily

modeled with discrete choice model (Aaberge et al., 1998) – Intensive (hours) vs. extensive (participation) labor supply elasticity (Saez, 2002)

• Identifying behavioral response to reform with cross-sectional

income-wage variations

12

• 2. The case of child

labor and the demand for schooling in developing countries

• Bolsa

Escola = Conditional cash transfer (CCT) program to alleviate poverty and incentivize schooling:

– Eligible: poor households with school age kids – Cash transfer for each kid at school age attending school

• Can we

think

• f a simple econometric

model that permits simulating ex-ante this program?

• Ex-ante model by Bourguignon, Ferreira and Leite

(2004)

• Note: Bolsa

Escola transformed into Bolsa Familia

13

The Bolsa Escola Programme

• Means-test : income

per capita less than R$90 (50 per cent of the minimum wage)

• Conditionality

: 6-15 year-olds must attend school.

• Transfer : R$15 per child

in school

• Limit

: R$45 per household

• Monitoring at the local and federal

levels (1 US $ ~ 2 reais)

14

A minimal schooling and child labor supply model

• Child’s occupational choice

Choose the alternative with maximum utility among:

(0) Not going to school (paid

• r unpaid

work); (1) Going to school and paid work; (2) Going to school and no paid work Ui (0) = Zi .γ0 + α0 .(Y-i + yi0 ) + vi0 Ui (1) = Zi .γ1 + α1 .(Y-i + yi1 ) + vi1 Ui (2) = Zi .γ2 + α2 .(Y-i + yi2 ) + vi2 Zi = characteristics

• f household

i, Y-i = income without child' work, yij = income

• f the child

in alternative j

15

… model (ctd.)

• Child i’s

Contribution to Household Income in alternatives j = 0, 1 or 2:

yi0 = wi ; yi1 = Mwi ; yi2 = Dwi (not observed)

with wi = ("full time") market earnings: Log yi0 = Xi .δ + ui Log yi1 = Xi .δ + m*Ind(Si =1) +ui and M = Exp(m) = "part-time" wage reduction coefficient

16

full model:

(yi0 = wi ; yi1 = Mwi ; yi2 = Dwi )

Household (kid ?) i chooses the alternative j that yields the highest utility Ui (j):

Ui (0) = Zi .γ0 + α0 .Y-i + β0 .wi + vi0 Ui (1) = Zi .γ1 + α1 .Y-i + β1 .wi + vi1 Ui (2) = Zi .γ2 + α2 .Y-i + β2 .wi + vi2

with β0 = α0 ; β1 = α1 M; β2 = α2 D

17

Estimating the model

Vij = double exponential leading to multi-logit:

Likelihood With It is possible to identify all parameters (rather than only the differences between two alternatives). Then necessary to draw random values for v's consistent with observed choice (see above)

{ }

[ ]

[ ]

∑

= − −

+ + + + = ≠ ≥

2 , 1 , * *

. . . . . . ), ( ) ( Pr

* * *

k k i k i k i j i j i j i i i i i

w Y Z Exp w Y Z Exp j k k U j U

i i i

β α γ β α γ D M

2 2 1 1

α β α β α β = = =

18

The simulation framework

Introducing the Cash Conditional Transfer: Easily evaluated when αj , βj , γj , vij are known. ° > + + + = ° ≤ + + + + = ° > + + + + = ° ≤ + + + + + = + + + =

− − − − − − − − −

Y Y if v w D Y Z U Y Y if v w D T Y Z U Y w M Y if v w M Y Z U Y w M Y if v w M T Y Z U v w Y Z U

i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i i 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1

ˆ ˆ ˆ ˆ ˆ . ) 2 ( ˆ ˆ ˆ ) ( ˆ ˆ . ) 2 ( ˆ ˆ ˆ ˆ ˆ ˆ . ) 1 ( ˆ ˆ ˆ ˆ ) ( ˆ ˆ . ) 1 ( ˆ ˆ ˆ ˆ . ) ( α α γ α α γ α α γ α α γ β α γ

19

Descriptive Statistics and Estimation Results

20

21

22

23

Simulation Results.

MODEL BY AGE Not going to school Going to school and working Going to school and not working Total Not going to school 65,7% 10,2% 24,1% 5,8% Going to school and working 0,0% 99,8% 0,2% 16,9% Going to school and not working 0,0% 0,0% 100,0% 77,3% Total 3,8% 17,4% 78,8% 100,0% Not going to school Going to school and working Going to school and not working Total Not going to school 52,2% 14,2% 33,6% 9,1% Going to school and working 0,0% 99,6% 0,4% 23,7% Going to school and not working 0,0%

• 100,0%

67,2% Total 4,7% 24,9% 70,3% 100,0% Poor Households

AFTER REFORM Before reform

24

Simulation of Alternative Programme Designs

1. Transfer doubled from R$15 to R$30. 2. Age Progressive Transfer (R$10 to R$35). 3. Means-Test raised to R$120. 4. Transfer doubled & Means-Test raised. 5. Age Progressive Transfer & Means-Test raised. 6. No conditionality.

25

26

Table 7. Simulated distributional effects of alternative specifications of the conditional cash transfer program

Original Bolsa escola's program Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Scenario 6 Mean Income per capita 253.9 255.0 256.1 255.8 255.2 256.5 256.3 255.1 Inequality measures Gini coefficient 0.594 0.589 0.584 0.585 0.588 0.583 0.584 0.589 Mean logarithmic deviation 0.704 0.670 0.647 0.652 0.669 0.644 0.649 0.668 Theil index 0.710 0.700 0.690 0.692 0.699 0.687 0.689 0.699 Generalized Entropy (2) 1.605 1.589 1.572 1.575 1.585 1.565 1.569 1.587 Poverty measures Poverty headcount 30.5% 29.5% 28.2% 28.5% 29.5% 28.2% 28.5% 29.4% Poverty gap 13.5% 12.4% 11.2% 11.5% 12.4% 11.2% 11.5% 12.3% Total square deviation from poverty line 8.1% 7.1% 6.2% 6.4% 7.1% 6.2% 6.4% 7.0% Annual cost of the program (million Reais) 1,668 3,228 3,000 1,944 3,984 3,720 1,668

Source: PNAD/IBGE 1999 and author's calculation

27

Conclusions

1. Bolsa Escola’s conditionality is mildly effective: likely to send to school one third of 10-15 year-

• lds currently not attending (and up to one-half

among the poor). 2. Likely to INCREASE number of children studying and working. (Time at school?) 3. The program’s impact

• n overall poverty and

inequality is modest, due largely to low transfer amounts.

28

Conclusions (ctd.)

4.

Impact on poverty appears to be elastic w.r.t. transfer amount, but not to level of means-test. 5. Simple micro-econometric estimation and simulation procedures can be usefully deployed to investigate likely impact of alternative program designs.

29

References

• Aaberge, R., U. Colombino

and S. Strom (1998), Evaluating Alternative Tax Reforms in Italy with a Model of Joint Labor Supply of Married Couples, Structural Change and Economics Dynamics, 9(4): 415-33

• Blundell R.W, MaCurdy T., (1999), “Labour Supply: a Review of Alternative

Approaches” in “Handbook of Labour Economics” vol 3a, Ashenfelter and Card eds, North Holland.

• Blundell, R. , A. Duncan, McCrae J., C. Meghir

(2000), “The Labour Market Impact of The Working Families’ Tax Credit”, Fiscal Studies nº 21(1).

• Bourguignon, F., F. Ferreira and P. Leite (2004), Conditional Cash Transfers,

Schooling, andChild Labor: Micro-Simulating Brazil’s Bolsa Escola Program, World Bank Economic Review, 17 (2)

• Creedy J., Duncan A., (2002), “Behavioural Microsimulation

with Labour Supply Responses”, Journal of Economic Surveys n16, pp. 1-38.

• Hausman

J., (1980), “The effect on Wages, taxes and Fixed Costs on Women’s Labour Force Participation”, Journal of Public Economics, n.14, pp. 161-194.

• Hausman, J., (1985), “The Econometrics of Nonlinear Budget Set”,

Econometrica, vol 53, pp.1255-82

30

References

• Heckman, J. (1974), Shadow prices, market wages and labor supply,

Econometrica, 679-94

• Heckman, J. (1979), Sample Selection Bias as a Specification Error,

Econometrica, Vol. 47, pp. 153-161

• Hoynes H., (1996), “Welfare Transfers in two Parent Families: Labour Supply

and Welfare Participation Under AFDC-UP”, Econometrica, vol 64, pp. 295- 332

• Laroque, G. and B. Salanié, (1999), Breaking Down Married Female Non-

Employment in France, CEPR Discussion Papers N° 2239

• MaCurdy, T., D. Green and H. Paarsch

(1990), Assessing Empirical Approaches for Analyzing Taxes and Labor Supply, Journal of Human Resources, 25(3): 415-90

• Saez, E. (2002), Optimal income transfer programs: intensive vs. Extensive

labor supply responses, Quarterly Journal of Economics, 117(3), 1039-73

• Van Soest A., (1995) “

A Structural Model of Family Labour Supply: a Discrete Choice Approach”, Journal of Human Resources n 30 pp. 63-88.

31

Chapter 3 Ex-post program evaluation: experimental and quasi-experimental methods

François Bourguignon Paris School

• f Economics

M2-PPD, M2-APE, 2009-10

32

Introduction

• Ex-post

means that 'results' from policy, or more realistically program being evaluated can be 'observed'

• Experimental

methods based

• n randomly

chosen 'treatment' and 'control' groups

• With

experimental methods, effect

• f a policy

given by simple difference between treatment and control group: essentially a 'reduced form' approach to policy.

33

Introduction

• Quasi-experimental

methods: treatment and control groups not chosen randomly, issue of the selection bias.

• Correction of selection

bias can be

• btained

with various approaches, based

• ne way
• r another
• n specific

assumptions.

• This chapter

gives

• nly

a cursory view

• n these

methods (interested students will have seen

• r will

see it in the econometrics course)

34

Outline

1. Experimental methods: full randomization and simple differences 2. Quasi-experimental methods: correcting for the selection bias

• Differences

in differences

• Instrumental variables
• Propensity

Score Matching

3. Examples

• f application

4. Generalization to the case of externalities 5. Limitations of pure experimental approach

35

• 1. Experimental

methods: full randomization and simple differences

• Basic model:

Where yi is the outcome

• f interest, Xi

a set of individual characteristics, Pi a dummmy variable indicating participation to the program being evaluated, and ui the effect

• f all unobserved

variables

• The principle
• f randomization: Pi

=0 and Pi =1

• bserved
• n fully

random samples.

• This

guarantees that ui and Pi are independent so that OLSQ on (1) yields an unbiased estimates

• f α

(In effect first term in RHS of (1) not needed)

i i i i

u P X y + + = α β

(1)

36

Simple differences in means as estimated effect

• f intervention

Averaging:

Randomization implies: With Caution: observations may not be independent – e.g. geographical rather than individual

• randomization. In that

case, (3) is under-estimating variance.

) ( ) 1 ( ˆ = − = =

i i

P y P y α

(2)

) / ( 1 ) 1 / ( 1 ) ˆ (

1

= + = =

i i

P y V N P y V N V α

(3)

) ( ) ( ) ( ) 1 ( ) 1 ( ) 1 ( = + = = = = + + = = =

i i i i i i

P u P Z P y P u P Z P y α

37

Applications

Numerous applications:

Biology Economics:

Negative income tax (example of RSA) Self-sufficiency program Active Labor Market Programs (Bloom et al., 1997) Deworming (Miguel and Kemer, 2004) Teacher incentives Vouchers (Angrist et al., 2002) …

For social programs, however, randomization is

• ften difficult (ethical/political resistance; selective

non-compliance).

38

Y1 Impact = Y1- Y1

*

Y1

*

Y0 t=0 t=1 time

Program starts

Simple differences: graphical illustration

39

Example : Progresa (Oportunidades) in Mexico

• Progresa

= Conditional Cash Transfer Program (Education + health care and nutrition)

• Implemented

in rural Mexico starting in 1998

• Sequencing
• f program starts

done randomly: randomization design

• Intensively

and extensively analyzed program

40

The program: Eligibility criterion and cash transfer schedule

Eligibility: "score" based

• n multiple

criteria below some threshold (defined at locality level)

41

Size of the program

42

Evaluation surveys

43

The various situations of households in experimental localities

44

Usual estimators

• f treatment

effects

Simple difference: Cross section Simple difference Before-after From what was seen before, however, BADIF likely to be biased and inferior estimator in comparison with CSDIF

45

Education effect (Skoufias, 2005)

CSDIF BADIF

46

Health (Skoufias, 2005)

CSDIF BADIF