Focused Targeting against Poverty Evidence from Tunisia

1
Focused Targeting against PovertyEvidence from
Tunisia

• Christophe Muller and Sami Bibi

2

• Anti-poverty transfer schemes (APTS)
• Major policy
• Building block
• Based on predictions of living standards
• Insufficient accuracy
• Poverty, Leakage, Undercoverage
• Riots

3

• Tunisian Food Subsidies Program
• Large part of the fiscal deficit
• APTS alternative
• Usually predictions based on OLS regression
  estimates
• We focus on the poor with
• Censoring the dependent variable
• Using quantile regressions
• Link with optimal transfer schemes (BF97)

4
Perfect Targeting

• Minti (1/N) ?i ((z-yi-ti)/z)a.1yitiltz
• ?i ti B
• ti 0, ?i
• ti Ta(yi, z, B), ?i
• Known solution and easy

5
Imperfect Targeting

• Mint(.) ??0z ((z-y-t)/z)a f(y x) dy
• ??0z t(y x) f(y x) dy B
• t(y x) 0
• Estimating f non-parametrically?
• Accuracy loss
• Estimating quantiles of f
• Quantile regressions, but there are many
• Eliminating many non-poor in regressions
• 1 estimation close to z may give good results

6
The chain of treatment

• 1 Estimation demand system
• 2 Calculus of equivalent-income and living
  standard indicators yi
• 3 Estimation of the predictions of living
  standards p_yi
• 4 Calculus of the optimal transfers ti(X)
• 5 Estimation of poverty and targeting efficiency
• 6 Tests of stochastic dominance

7

• Censored quantile regressions
• Minb (1/N) ?i ??(yi-Max(0,Xib))
• ??(u)?-1ult0u
• Algorithm and bootstrap

8
Estimation

• QAIDS for food
• Blundell-Robin estimator
• Equivalent-Income eliminating food subsidies and
  spatial price differences EI
• Poverty and efficiency based on EI
• Incorporation of Subsidies or Transfers
• OLS, Tobit, Quantile reg., Censored quantile reg.
• Various censorship and anchors

9
The Data

• 1990 Tunisian consumption survey
• 7734 households
• Budget of food subsidies allows
• Poverty line of 358 Tunisian Dinars (DT)
• Official poverty lines EI
• 4 sets of correlates of living standards

10
The Correlates

• Regions
• Demographic characteristics
• House characteristics
• Occupation
• Education

11
Regions

• Great Tunis
• Northeast
• Northwest
• Middle east
• Middle west
• Sfax
• Southeast
• Southwest

12
Demographic information

• Number of children in household old less than 2
  years old.
• Number of children aged between 3 and 6 years.
• Number of children aged between 7 and 11 years.
• Number of adults aged between 12 and 18 years.
• Number of adults old more than 19 years.
• Age of the household head (HH).
• Squared age of the HH.

13
House

• Number of rooms per capita
• 1 if household lives in a detached house, 0
  otherwise.
• 1 if household lives in a flat, 0 otherwise.
• 1 if household lives in an Arab house, 0
  otherwise.
• 1 if household lives in a hovel, 0 otherwise.

14
Occupation

• Dummy variable for HH living in Great Tunisia is
  unemployed.
• Dummy variable for HH living in the Northwest is
  unemployed.
• Dummy variable for HH living in the South (east
  or west) is unemployed.
• Dummy variable for HH living in another region is
  unemployed.
• Dummy variable for HH living in Southeast is
  agricultural laborer.
• Dummy variable for if HH living in Southwest is
  agricultural laborer.
• Dummy variable for if HH living in another region
  is agricultural laborer.
• Dummy variable for if HH is not agricultural
  laborer.
• Dummy variable for if HH is agricultural farmer.
• Dummy variable for if HH living in Northwest is
  agricultural farmer.
• Dummy variable for if HH is self-employed or
  manager.
• Dummy variable for if HH has another type of job.

15
Education

• Dummy variable for HH is illiterate.
• Dummy variable for HH has a primary schooling
  level.
• Dummy variable for HH has a junior secondary
  schooling level.
• Dummy variable for HH has a senior secondary
  schooling level.
• Dummy variable for HH has a higher educational
  level.

16
Results

• Normality and Homoscedasticity of perturbations
  are rejected
• Log real consumption per capita
• Expected signs in living standard equations
• Most coefficients are significant

17
Goodness-of-fit

• OLS have highest pseudo-R2 (0.5)
• Tobit have lower pseudo-R2 (0.33)
• (Censored) Quantile regressions much less
  efficient (0.12)
• But when focusing on 1st and 2nd quantiles of
  living standards
• Quantile regressions yields the highest
  prediction efficiency (for the poor)

18
Simulated Poverty Curves
19
Targeting Efficiency Statistics
20
Conclusion

• Using estimation methods focusing on the poor
  and near-poor can dramatically improve the
  performances of APT
• The small level of undercoverage may remove
  political obstacles to the implementation of APTS

21
Related projects

• QAIDS
• Price correction and anti-poverty targeting
• APTS and Calculus of variations
• Optimal estimator for APTS