Informing Policy

1
Informing Policy

• Prepared for WBI/DFID Workshop on Poverty Analysis

London, November 6-8, 2006
Jonathan Haughton Suffolk University,
Boston jonathan.haughton_at_suffolk.edu
2
Basics

• Poverty Reduction Strategy Papers are designed to
  force government to own the process (oxymoron?)
• Purpose here Examples of how poverty analysis is
  useful
• Benefit and tax incidence analysis
• Identifying determinants of poverty
• Targeting (especially poverty mapping)
• Trade and poverty

3
Tax and Benefit Incidence

• Figure out how much each person gains from
  government spending
• and provide a breakdown by the income or
  expenditure distribution (equity)
• Easier for some items (education, health,
  transfers) than others (roads, police, army,
  diplomatic service, public pensions)
• Applicable for ½ of government spending and for
  most taxes

4
Benefit Incidence Steps

• Step 1 Estimate unit subsidies
• Base on actual, not budget, spending
• May need survey (e.g. Ghana, health, 1995)
• Be skeptical (e.g. Uganda school subsidies)
• Factor in capital spending, admin. costs
• Focus on subsidy, so net out user charges
• Step 2 Identify users (coverage)
• Requires household survey data
• NB Survey data also needs information on
  spending and/or income, to allow for
  distributional analysis
• Poor recall may understate usage (Ghana
  hospitals, 12)
• Step 3 Aggregate users into groups
• Typically quintiles (5) or deciles (10)
• Sort by expenditure per capita or income per
  capita
• or by poor/non-poor
• Less common requires a poverty line.

5
Example Peru, education

• 15.5 of government spending
• Unit subsidies
• Pre-kindergarten S/. 583 per child per year
• Kindergarten and primary S/. 386 per pupil
  per year
• Secondary S/. 624 per pupil per year
• Tertiary S/. 2,506 per student per year.
• From central government budget (actuals) divided
  by official enrolment numbers.
• NB S/. 3.261 per USD, April 2005
• NB No adjustment for private schooling no
  disaggregation of costs by region no information
  on value of capital services
• Enrollment by level from ENNIV-2000
• Expenditure, income from ENNIV-2000

6
(No Transcript)
7
Income or Expenditure?

• Income overstates regressivity
• Some have low income, but transitory
• Expenditure permanent income
• But may understate regressivity
• At heart of debate in Latin America

8
(No Transcript)
9
(No Transcript)
10
Lorenz Concentration Curves
Gini 0.470 Quasi-Gini 0.102 Kakwani
-0.369 i.e. progressive Kakwani measure shows
progressivity of spending or tax
11
Reynolds-Smolensky
Gini 0.470 RS1 0.01282 Gini quasi-Gini of
net-of-subsidy disproportionality RS2
0.01234 Gini Gini of net-of-subsidy
redistributive capacity RS1 RS2 reflects
reordering a measure of horizontal inequity
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
Case Education in Africa (1)
16
Case Education in Africa (2)

• Who benefits from education subsidies?
• Progressive, but poorly targeted
• Demand depends on
• Income. So rich will send kids to poor more at
  all levels.
• Quality. Poorer quality for poor.
• Costs.
• Fees, etc. disproportionately deter poor.
• Distance (to secondary!) deters poor.
• Gender bias common

17
Tax case Lebanon 2004 (1)

• Govt. debt is 185 of GDP
• Debt trap looming
• Tax system is a muddle
• Goal of change
• Raise mobilization by 3 of GDP
• Make tax system more efficient
• Deadweight loss compliance costs administrative
  costs
• Maintain or enhance equity
• NB Need to treat changes as a package

18
Tax case Lebanon 2004 (2)

• Technique
• For each tax, determine incidence
• VAT in proportion to consumer spending, which
  needs survey data to measure
• Wage tax in proportion to wages
• Interest tax in proportion to assets
• Property tax in proportion to housing owned
• NB Needed survey data with both income and
  expenditure information
• Incorporate some behavior
• Elasticities for alcohol, tobacco excises
• VAT avoidance/evasion as function of rate
• Spreadsheet allowing easy simulation of packages
  of changes, invoking Stata see example on next
  page
• A very useful tool

19
(No Transcript)
20
Issue 1 Average vs. Marginal

• Results show average incidence
• Dont show incidence of additional (marginal)
  spending
• Dont show incidence of lower user fees with
  associated lower coverage
• Younger In Africa, early capture is by more
  affluent
• Implication marginal incidence is more
  progressive than average incidence
• Put another way, coverage and incidence are
  correlated
• To measure marginal effects
• Use spatial variation (Lanjouw Ravallion,
  India)
• Use time series variation
• Use demand estimates (Younger)

21
Issue 2 Incorporating Behavior

• Up to now, an accounting exercise
• beneficiary incidence (Demery)
• van de Walle
• VLSS 93 98 with N4,308 panel
• ?Cit ß ?Tit ? ?Xit ?dt ?eit
• Estimates ß0.45, t4.3.
• Evidence from elsewhere ß0.5.
• Incentive effects of welfare payments also strong
  in US, Ireland, etc.

22
Issue 3 Valuing benefits

• Cost of provision (unit costs) only OK if
• Government is efficient, honest
• AC MC
• Equals average of individual marginal valuations
• Contingent valuation
• How much is the service worth?
• With public goods, the least bad alternative
• Compensating variation
• Based on estimating demand curve
• 0/1 index
• 1 participate, 0 otherwise
• Equivalent to constant value for all beneficiaries

23
Issue 4 Pieces or Total?

• Demery
• Focus on one item at a time
• which expenditure items are most efficient at
  transferring income to the poor?
• Meerman, JH
• Tax-transfer system is a package makes sense to
  consider as a whole
• Jenkins Design taxes to collect revenue
  efficiently take care of distributional issues
  on the expenditure side

24
Issue 6 Deep causes

• Difficult to use findings for policy, unless one
  has a theory
• E.g. If girls get 30 of educational spending,
  what does it imply for policy? Perhaps parents
  are keeping girls home
• Why does government allocate its spending the way
  it does now?
• Political economy issues
• Recommendations must be politically feasible

25
Caveat on Incidence

• Demery cautions
• Public expenditures can be effective in reducing
  poverty only when the policy setting is right.
• E.g. Why spend on ag. Extension if overvalued
  exchange rate makes farming unprofitable.
• Analysis assumes public expenditure process is
  based on outcomes and impacts and not just line
  items.
• Public expenditure decisions must rest on a
  sound understanding of the needs and preferences
  of the population at large.
• Otherwise, benefits incidence analysis is
  unlikely to have much useful impact.
• JH Incidence analysis probably more helpful on
  the tax than on the spending side
• but needs survey data with income and
  expenditure

26
Targeting

• Universal provision is costly.
• Sri Lanka 1/5 of budget was for rice.
• How target?
• Means testing.
• Problem Getting reliable information, especially
  about chronically poor.
• Indirect indicator targeting
• Problem Cheaply collected information is poorly
  correlated with poverty (Ravallion on Jamaica)
• Self-targeting programs
• Food for Work, Maharashtra. Set wage low enough.

27
Targeting (2)

• Subsidize commodities consumed by poor
• Rough bread in Egypt (3 pence/round)
• Subsidize technology that helps the poor
• Geographic targeting

28
Modeling Determinants of Poverty

• Poverty profiles are descriptive
• The goal
• Find causes, not just correlations
• Find deep causes, not just proximate causes
• Easier said than done. Howard White the
  missing middle of clearly understanding of the
  fundamental causes of poverty

29
Proximate causes

• Typically divided into
• Regional level characteristics
• E.g. floods, climate, accessibility
• Community level characteristics
• Infrastructure, social capital (It takes a
  village)
• Household and Individual characteristics
• Demographic, incl. size, gender, age
• Economic, incl. assets, employment
• Social, incl. health, education, housing
• Description followed by regression

30
(No Transcript)
31
Determinants of malnutrition in Vietnam (Haughton
Haughton, EDCC, 1997)
32
Notes No gender bias in nutrition Weaning foods
problem High parity hurts Strong effect of
fathers education All
33
Growth and Poverty

• Dollar and Kraay 2000
• 236 episodes, 80 countries, 4 decades
• Regressed ln(income/capita) of poorest quintile
  on ln(income/capita) overall. Elasticity close
  to 1, high R2. Robust finding.
• More controversial other influences
• Rule of law, fiscal discipline, low inflation all
  associated with faster economic growth
• Democracy, higher public spending on health and
  education, globalization, have little or no
  measurable association with incomes of the poor.
• Debate now under rubric of pro-poor growth

34
A trade example

• Vietnam Effect of devaluation on incomes of
  poor?
• First, get effect of devaluation on relative
  prices domestically
• Second, link to 98 income sources, 269 spending
  headings, to find effect for each household
• Third, summarize distributional effects.
• By income, help poor and rich (see next slide)
• By expenditure, hurt poor
• By area, hurt urban
• By ethnicity, hurt minorities (remote)
• all these effects are small

35
(No Transcript)
36
In short

• Many ways to use the analysis to inform policy
• Also impact evaluation later today
• Sometimes relative simple tools are very useful
• E.g. Lebanon tax analysis
• But
• Causality is hard to determine
• Full theory of poverty is lacking
• In-country demand is essential
• And local analytical capacity helps