Measuring

1

• Measuring
• Vulnerability to Poverty
• Cesar Calvo and Stefan Dercon
• Oxford University

2
Purpose?

• Develop a welfare concept for policy evaluation
  that focuses on downside risk and vulnerability
  (1)
• Present an axiomatic approach to measure
  individual vulnerability to poverty (2)
• Explore aggregation issues (3)
• Present elements of a Vulnerability Profile for
  Peru, using the approach developed (4)
• Inspire a little

3
Purpose?

• Part of a larger effort
• To focus on risk as a major cause of poverty and
  deprivation (beyond the dimension of a sense of
  insecurity).
• To argue for policies towards poverty alleviation
  that take risk in livelihoods more seriously.
• Here looking for a welfare concept that
  recognises risk

4
A paper in two parts

• In this session motivation, individual and
  aggregate measure of vulnerability plus
  discussion of empirical issues
• In session XV, discussion of the empirical
  application, with extension to multidimensional
  vulnerability (Cesar Calvo)

5
1. Vulnerability concept

• Term vulnerability is used as meaning
  risk-related vulnerability, whereby risk is
  constituting a threat to well-being
• A THREAT INDUCED BY RISK

6
1. Vulnerability concept

• PLEASE DONT BE CONFUSED
• Risk-related Vulnerability versus
• Other common meaning of vulnerability as
• state of being helpless or weak,
• as in vulnerable groups.

7
Developing a concept

• Policy research puts increased emphasis on
  multidimensional achieved outcomes in thinking
  about poverty and wellbeing
• Measuring incomes, health, education, social
  status, etc.
• Poverty is failure to secure acceptable living
  conditions
• BUT, these are EX-POST, devoid of the ex-ante
  uncertainty surrounding these outcomes.
• OR
• A MISSING DIMENSION THE UNCERTAINTY SURROUNDING
  FUTURE OUTCOMES

8
Uncertainty matters

• We may want to know about future poverty e.g.
  who will be poor? how poor will they be?
• Also people dislike uncertainty. The feeling of
  insecurity compounds the predicaments of the
  poor.
• Policy relevance!
• E.g. decisions about interventions
• Which one will be more effective in reducing
  poverty?
• It will require considering potential outcomes in
  different states of the world

9

Missing in literature

• Even in Sen (1981), with focus on drought and
  shocks leading to famine, the focus is on EX-POST
  outcomes , despite allusions
• For them a variation of the exchange
  relationships can spell ruin. () More modern,
  perhaps more vulnerable, certainly.

10
Developing a concept

• Lets try to measure an ex-ante concept, before
  the veil of uncertainty has been lifted.
• We define vulnerability as the threat of future
  poverty, and we mean it to relate
• (a) to the likelihood of suffering poverty in
  the future, and
• (b) to the severity of poverty in such a case.

11
Vulnerability measurement

• 1st issue?
• Can we design a measure with desirable and
  reasonable properties?
• a valuation ex-ante of possible welfare
  levels, taking into account
• -a poverty norm and
• -risk sensitivity

12
Vulnerability measurement?

• 2nd issue?
• Can we use it for policy-relevant analysis?
  empirical application
• Needs a forecast model of possible welfare
  outcomes
• Exploiting as effectively as possible increased
  panel data or repeated cross section
• Exciting and potentially important research area

13
2. Axiomatic approach to Vulnerability

• vj v (z, p, yj)
• What properties are desirable?

14
2. Axiomatic approach to Vulnerability

• Focus axiom we focus only on outcomes (in
  particular states) below some norm above the
  norm, they are censored at the norm.
• downside risk
• Good overall expectations do not imply low
  vulnerability.
• The threat of future poverty is not reduced by
  (ex-ante)
• possibilities of being well-off.
• A poor person with 1 probability of winning a
  100,000 lottery is as vulnerable as another with
  1 chance of winning 10 million.

15
2. Axiomatic approach to Vulnerability (ctd.)

• Symmetry over states of the world
• Continuity and Differentiability
• Scale invariance in norm and in outcomes
• Probability-dependent effect of outcomes
• changes in outcomes in states with higher
  probability have larger effects on vulnerability
• Probability Transfer
• vulnerability increases if bad states become
  relatively more prevalent
• Normalisation (between 0 and 1)

16
2. Axiomatic approach to Vulnerability (ctd.)

• Risk sensitivity if downside risk increases,
  vulnerability increases
• Put more structure on risk sensitivity
• constant relative risk sensitivity
• For instance, this could be replaced by constant
  absolute risk sensitivity (or any other
  conjecture about sensitivity towards risk)

17
2. Axiomatic approach to Vulnerability (ctd.)

• The only family of measures satisfying all our
  axioms is defined by
• With while lower a implies higher risk
  sensitivity

18
Existing vulnerability measures

• Low expected utility (Ligon and Schechter 2003,
  Elbers and Gunning 2003)
• VLS fails focus axiom
• Expected utility as a normative concept?
• Expected utility as a behavioural concept?
• note we assume LOSS AVERSION AT THE NORM
  consistent with evidence from shame experiments

19
Existing vulnerability measures

• Expected FGT poverty (Christiaensen and
  Subbarao 2004, Kamanou and Morduch 2004)
• a0 prob. of being poor
• a1 expected shortfall
• VEP fails Probability Transfer when a0, Risk
  Sensitivity when a1, and constant relative risk
  sensitivity when agt1
• (in fact, risk sensitivity increases in
  consumption for any agt1).

20
Discussion of our measure

• People who are certain to be poor are highly
  vulnerable.
• But some that are not poor at t can be highly
  vulnerable
• Some of the poor at t may have low vulnerability

21
3. Aggregating vulnerability

• Individualistic aggregation a society becomes
  (more) vulnerable when its individual members are
  (more) vulnerable
• Non-individualistic view society might not be
  indifferent between two scenarios, even if each
  individual feels equally vulnerable in either of
  them issue of covariance of outcomes in
  particular states

22
3. Aggregating vulnerability

• Individualistic aggregation
• r reflecting inequality aversion

23
3. Aggregating vulnerability

• Non- Individualistic aggregation
• Threat of widespread poverty versus Threat of
  contained poverty
• e.g. Sen, on China vs India on average as many
  hunger deaths in 1960s, but China mainly during
  59-61 and India spread across years. Which is
  better?

24
3. Aggregating vulnerability

• Non-Individualistic aggregation

25
4. Can we use these measures?

• One needs information about likely distribution
  of outcome (health, consumption)
• for t1, as seen at t
• Likely to be
• data intensive (statistical models, panel data)
  or
• data collection intensive (perceptions of risk)
• Use measures to derive profiles

26
CONCLUSION How useful?

• Application to Peru shows that currently poor
  is not the same as vulnerable
• But key contribution is also to highlight that
  current poverty measurement is too narrow
• Focus on ex-post measurement (realised states
  and backward-looking), so ignoring threat and
  insecurity
• The need to bring risk more to the fore
• Not least since risk-related vulnerability forces
  people to take actions that may reinforce poverty
  in the long run.