Presentation is loading. Please wait.

Presentation is loading. Please wait.

Informal Insurance in the Presence of Poverty Traps: Evidence from Southern Ethiopia Paulo Santos and Christopher B. Barrett Cornell University September.

Similar presentations


Presentation on theme: "Informal Insurance in the Presence of Poverty Traps: Evidence from Southern Ethiopia Paulo Santos and Christopher B. Barrett Cornell University September."— Presentation transcript:

1 Informal Insurance in the Presence of Poverty Traps: Evidence from Southern Ethiopia Paulo Santos and Christopher B. Barrett Cornell University September 14, 2006 seminar Michigan State University

2 Core Question Models of consumption smoothing and informal insurance typically rely on the assumption of stationary income processes. Our question: what happens when that assumption does not hold?

3 Outline 1: What do we know 2: Asset shocks and insurance 3: Data 4: Who gives to whom 5: Who knows whom 6: Conclusions

4 1: What do we know Lybbert et al (2004 EJ)  Evidence of multiple equilibria  Asset risk is largely idiosyncratic  But asset transfers are quite small

5 What do we know Santos and Barrett (2006)  Asset shocks associated with adverse rainfall events are the source of non-linear asset dynamics (multiple equilibria)  Boran pastoralists perceive this.  Ability matters !

6 2: Asset shocks and insurance Poverty trap models emphasize assets and thresholds. So we focus on asset dynamics, risk and transfers around thresholds. Basic intertemporal decision model: Max {ct, ijt} E{  t=0…T  t U(c t (k t ))|  } subject to:k t = g( k t-1 +  t + ji t - ij t ) c T (k T ) = k T k 0 given,  ~[-k t,0], t ={0, } Transfers ( ) and asset shocks (  ) affect asset (k) dynamics, underlying income generation and consumption (c).

7 Asset shocks and insurance Growth dynamics are key to understanding the nature of the resulting informal insurance arrangements. k c t = g c l (k t-1 +  t + ji t - ij t ) if i  c, k t-1 <  = g c h (k t-1 +  t + ji t - ij t ) if i  c, k t-1   for clubs c=1,…,C The most general specification allows for: 1) different clubs w/o thresholds (C>1,  =0), 2) unique club w/ threshold (C=1,  >0), 3) canonical convergence model (C=1,  =0, g(.) concave) that implicitly underpins the standard consumption smoothing and informal insurance literatures

8 Asset shocks and insurance Convergence: every match is in insurance pool (standard literature) Precautionary savings: only capacity to reciprocate (but not actual losses) matters (McPeak JDE 2006) Poverty traps due to multiple equilibria: 1) exclude the poorer and those with lower ability (i.e, those at lower level equilibria) because it is harder to punish them if they don’t reciprocate. 2) privilege those at the threshold (because maximizes gains from transfer). Losses YesNo Herd size YesPoverty trapsPrecautionary savings NoConvergence?

9 3: Data Pastoral Risk Management (PARIMA) project (USAID GL CRSP) 119 households, 2000-2003 Data on insurance networks  5 Random matches [X] within sample : Question 1: Do you know [X]? Question 2: Would you give to [X] if s/he asked?  Advantage/(potential) disadvantages: no bias because lack of knowledge of one side of the relation data on links, not transfers: but transfers are small potential, not real, links: but inference based on this information is reliable (Santos and Barrett, 2006)

10 Data 1) Gifts  Loans 2) Not everyone knows everyone else 3) Doesn’t know  Doesn’t give 4) Know (not  ) Give Know Give YesNo Yes653 No370123 Gift Loan YesNo Yes4253 No10123

11 4: Who gives to whom l ij * = α i +  1 f(h j )+  L j + Σ t=1…4 β t E tj + δ X ij + λZ i + ε ij Key variables: h j (recipient herd size), L j (recipient herd loss), E j (recipient equilibrium regime) X ij = (possibly asymmetric) differences between i and j Z i = characteristics of the respondent Assumptions on ε ij : ε ij ~ log(0,  2 /3) E (ε ij,ε ih ) ≠ 0 if j ≠ h E (ε ih,ε jh ) = 0 if i ≠ j  Logit model, observations clustered on the respondent

12 Who gives to whom Alternative assumption: E (ε ih, ε jh ) ≠ 0 if i ≠ j Ways to check/correct for this possibility: - Udry & Conley (2005), Fafchamps and Gubert (JDE forthcoming) use Conley’s estimator to correct for correlated error structures - Quadratic Assignment Procedure (QAP): nonparametric permutation test that gives correct p-values Ultimately, these more complex error structures matter little

13 Who gives to whom (1)(2)(3)(4) h j =00.3570.2750.1401.207 hjhj -0.014-0.021-0.024-0.020 E2-0.0920.275-0.387 E30.2030.6550.005 E4-0.611-0.019-0.734 LjLj 0.919 L j * E10.465 L j * E21.711 L j * (h j =0)-1.188 Bold indicates statistical significance at 5% level or lower. Result: Transfers respond to losses – i.e., they are state-contingent insurance claims – but also depend on ex post herd size. We thus reject the precautionary transfers and insurance under convergence hypotheses in favor of the insurance in the presence of poverty traps.

14 Who gives to whom Conclusion: Asset transfers are best understood as insurance of permanent income, preventing recipients from falling into persistent poverty and excluding those who are not expected to be able to reciprocate.

15 Who gives to whom Does “ability club” membership matters? A priori expectation: those with low ability should not receive gifts, if match’s ability is observed by respondents. Approach followed:  Get estimates of efficiency (high, medium, low)  Re-estimate previous model  Bootstrap results to get correct SE

16 Who gives to whom (1)(2)(3) Low1.1370.3761.334 Medium2.5420.4352.616 E2*low1.372-0.248 E2*medium-1.1451.588 E2*high1.6072.720 L j * E2* lowDropped L j * E2* medium2.856 L j * E2* high2.500 Result: As predicted: transfers related to losses and ex post herd size for those facing multiple equilibria.

17 Who gives to whom Does the threshold play a role in targeting?  No if transfers are given to those with maximal capacity to reciprocate  Yes if transfers are intended to maximize expected gains from transfer The predictions of the two models diverge for those herders who suffered losses but are above the threshold  Helped in the 1 st model  Not helped in the 2 nd model  Problem: no data in the region where the predictions differ (above the threshold)  Solution: use simulation results on expected gains from transfers

18 Who gives to whom Simulated expected herd growth (and long-term herd size)

19 Who gives to whom Result: Transfers seem ex post insurance that takes into account recipient’s expected gains but not his/her expected wealth … a non-monotonic relation between recipient’s wealth and transfers. (1)(2)(3)(4)(5) E (wealth)-0.487-0.723-0.023 E (gains)0.2770.2100.418 E (wealth) * Loss 20.724-15.608 E (gains) * Loss 1.5242.144

20 Who gives to whom Conclusions: 1) Transfers are influenced: By the existence of thresholds By the existence of ability clubs 2) Asset transfers seem to be best understood as insurance of the permanent component of income and driven largely by expected recipient gains

21 5: Who knows whom: Social exclusion and poverty traps “[t]o be poor is one thing, but to be destitute is quite another, since it means the person so judged is outside the normal network of social relations and is consequently without the possibility of successful membership in ongoing groups, the members of which can help him if he requires it. The Kanuri [in the West African savannah] say that such a person is not to be trusted”. (Iliffe, 1987, The African Poor) Coef. No cattle since 2000-1.106 E1 since 2000-0.145 E2 since 2000-0.127 E3 since 2000-0.581 E4 since 2000-1.297 Lost cattle 2000-20030.203 More cattle-0.014 Less cattle0.040 Use same logit estimation approach, with “know” as dependent variable now.

22 6: Conclusions Implications for public transfers - is crowding out really a concern for the poorest? No Our results:  The poorest are (rationally) not recipients of informal transfers: no risk of crowding out at very low levels of assets  Possibility of crowding in (by moving people nearer the threshold, where private transfers can be triggered … see Chantarat and Barrett, 2006)  Targeting may be especially difficult: public transfers must consider [needs * dynamics * ability]  Social invisibility of the poorest makes community based targeting a challenge

23 Thank you for your attention … I welcome your comments and questions.


Download ppt "Informal Insurance in the Presence of Poverty Traps: Evidence from Southern Ethiopia Paulo Santos and Christopher B. Barrett Cornell University September."

Similar presentations


Ads by Google