Climate Change, Climate Variability And Poverty Traps:

The Role (and Limits) of Index Insurance for East African Pastoralists

Christopher B. BarrettCornell University

Presentation at the Brown International Advanced Research Institute on Climate Change and Its Impacts

Providence, RIJune 13, 2011

Arid and semi-arid lands (ASAL) cover ~ 2/3 of Africa, home to ~20mn pastoralists – who rely on extensive livestock grazing.

Pastoralist systems adapted to variable climate, but very vulnerable to severe drought events. Big herd losses cause humanitarian and environmental crisis.

Motivation

Poverty traps in the southern Ethiopian rangelands

Standard policy response to climate shocks in the ASAL: food aid (slow, insufficient, inefficient, even insulting).

Pay attention to the risk and dynamics that cause destitution!

Motivation

Large economic/human costs of uninsured risk, esp. in presence of poverty traps.

Sustainable insurance can:• Prevent downward slide of vulnerable

populations• Stabilize expectations & crowd-in investment

and accumulation by poor populations• Induce financial deepening by crowding-in

credit supply and demand • Reinforce extant social insurance mechanisms

But conventional (individual) insurance rarely works in remote rural areas like the ASAL:

• High transactions costs• Moral hazard/adverse selection

Insurance to manage risk

Index insurance can avoid problems that make individual insurance infeasible in ASAL:

• No transactions costs of measuring individual losses

• Preserves effort incentives (no moral hazard) as no single individual can influence index.

• Adverse selection does not matter as payouts do not depend on the riskiness of those who buy the insurance

• Available on near real-time basis: faster response than conventional humanitarian relief

In principle, index insurance can help create an effective safety net to alter poverty dynamics and help address climate shocks.

Index insurance

‘Big 5’ Challenges of Sustainable Index Insurance1. High quality data (reliable, timely, non-

manipulable, long-term) to calculate premium and to determine payouts

2. Minimize uncovered basis risk through product design

3. Innovation incentives for insurance companies to design and market a new product

4. Establish informed effective demand, especially among a clientele with little experience with any insurance, much less a complex index insurance product

5. Low cost delivery mechanism for making insurance available for numerous small and medium scale producers

Index insurance

Solutions to the ‘Big 5’ Challenges

1. High quality data: – Satellite data (remotely sensed vegetation

index: NDVI) 2. Minimize uncovered basis risk:

– Analysis of household data on herd loss3. Innovation incentives for insurers:

– Researchers do product design, develop awareness materials and assist with capacity building

4. Establish informed effective demand– VIPs; Simulation games; comic books; radio

shows5. Low cost mechanism

– Delivery through partners

Index insurance

IBLI

New commercial Index-Based Livestock Insurance (IBLI) product launched commercially in January 2010 in Marsabit District in northern Kenya

Based on technical design developed at Cornell, refined and led in the field by the International Livestock Research Institute (ILRI) in collaboration with university and private sector partners.

Now being adapted and extended to Ethiopia and expanded to other ASAL districts in Kenya.

NASA NDVI Image Produced By: USGS-EROS Data Center. Source: FEWS-NET

ZNDVI: Deviation of NDVI from long-term average

Laisamis Cluster

-3-2-1012345

19

81

19

82

19

83

19

84

19

85

19

86

19

87

19

87

19

88

19

89

19

90

19

91

19

92

19

93

19

94

19

95

19

96

19

97

19

98

19

98

19

99

20

00

20

01

20

02

20

03

20

04

20

05

20

06

20

07

20

08

Karare

Logologo

Ngurunit

Korr

Laisamis Cluster, zndvi (1982-2008)

Historical droughts

NDVI (Feb 2009, Dekad 3)

IBLI insures against area average herd loss predicted based on NDVI data fitted to past livestock mortality data.

IBLI

NDVI-based Livestock Mortality Index

The IBLI contract is based on area average livestock mortality predicted by remotely-sensed (satellite) information on vegetative cover (NDVI):

IBLI

Spatial Coverage– Two separate area-specific “response

functions” map NDVI into predicted livestock mortality.

– Five separate index coverage regions (2 in one area, 3 in the other).

Upper Marsabitcluster

Lower Marsabitcluster

KARGI

SHURA

MAIKONA

BUBISA

TURBI

ILLERET

GALAS

SABARET

KOYA

DARADE

NORTH HORR

DUKANA

EL GADE

KORR

KURUGUM

BALESA

LAISAMIS

EL-HADI

FUROLE

KALACHA

HAFARE

GAS

HURRI HILLS

LOIYANGALANI

KURUNGU

LONTOLIO

ARAPAL

LOGOLOGO

QILTA

MT. KULAL

MOITE

GUDAS/SORIADI

KARARE

IRIR

NGURUNIT

LARACHI

KAMBOYESOUTH HORR(MARSA)

LONYORIPICHAU

SONGA

MERILLE

ILLAUT(MARSABIT)

HULAHULA

MAJENGO(MARSABIT)

OGUCHO

OLTUROT

JALDESA

KITURUNIDIRIB GOMBO

JIRIME

SAGANTE

IBLI

Temporal Coverage– Year-long contract, with two prospective

indemnity payment dates, following each dry season.

– Two marketing campaigns, just prior to rainy season.

– NDVI observed and index updated continuously.

IBLI

Risk Coverage and PricingPayoffs for predicted losses above 15% (“strike point”). Trade off: Higher Strike Lower Risk Coverage Lower Cost

Contract Cluster Consumer Price

Upper Marsabit 5.5%

Lower Marsabit 3.25%

IBLI

Testing the Index PerformancePerformance of predicted herd mortality rate in predicting area-average livestock mortality observed in longitudinal data

– Out-of-sample prediction errors within 10% (especially in bad years)– Predicts historical droughts well

0%

10%

20%

30%

40%

50%

19

82

19

83

19

84

19

85

19

86

19

87

19

88

19

89

19

90

19

91

19

92

19

93

19

94

19

95

19

96

19

97

19

98

19

99

20

00

20

01

20

02

20

03

20

04

20

05

20

06

20

07

20

08

20

09

20

10

Actual Vs. Predicted Seasonal Mortality Rate - Laisamis Cluster

PredictedActual

0%

10%

20%

30%

40%

50%

19

82

19

83

19

84

19

85

19

86

19

87

19

88

19

89

19

90

19

91

19

92

19

93

19

94

19

95

19

96

19

97

19

98

19

99

20

00

20

01

20

02

20

03

20

04

20

05

20

06

20

07

20

08

20

09

20

10

Actual Vs. Predicted Seasonal Mortality Rate - Chalbi Cluster

PredictedActual

Out of sample

IBLI

IBLI ImplementationCommercially launched in January 2010Two sales periods of varying experience:

• Jan/Feb 2010: Sold ~2000 contracts: Premiums collected ~ $46,000: Value of livestock covered ~$1,200,000

• Jan/Feb 2011: Sold ~750 contracts: Premiums collected ~ $9,500

Key ongoing considerations/challenges:• Delivery Channel• Extension/Education• Information Dissemination and Trust Building• Regulation

IBLI

Impact Evaluation Under Way

Confounding factor: ongoing implementation of cash transfer (HSNP)

Encouragement design• Insurance education game: played among 50% sample in game

site• Discount coupon on the first 15 TLU insured: (no subsidy for

40% of sample, 10%-60% subsidies for the rest)

Sample selection: 924 households • Sample/site proportional to relative pop. sizes• For each site, random sampling stratified by livestock wealth

class

KARGI

SHURA

MAIKONA

BUBISA

TURBI

ILLERET

GALAS

SABARET

KOYA

DARADE

NORTH HORR

DUKANA

EL GADE

KORR

KURUGUM

BALESA

LAISAMIS

EL-HADI

FUROLE

KALACHA

HAFARE

GAS

HURRI HILLS

LOIYANGALANI

KURUNGU

LONTOLIO

ARAPAL

LOGOLOGO

QILTA

MT. KULAL

MOITE

GUDAS/SORIADI

KARARE

IRIR

NGURUNIT

LARACHI

KAMBOYESOUTH HORR(MARSA)

LONYORIPICHAU

SONGA

MERILLE

ILLAUT(MARSABIT)

HULAHULA

MAJENGO(MARSABIT)

OGUCHO

OLTUROT

JALDESA

KITURUNIDIRIB GOMBO

JIRIME

SAGANTE

Legend

MarsabitIBLI

HSNP, IBLI Game_HSNP, No IBLI Game_No HSNP, IBLI Game_No HSNP, N

HSNP, IBLI Game

HSNP, No IBLI Game

No HSNP, IBLI Game

No HSNP, No IBLI Game

IBLI Game No IBLI Game

HSNP 4 sites 4 sites

No HSNP

5 sites 3 control sites

IBLI

Core impact evaluation questions

1) For whom is IBLI most attractive and effective?- simulation-based answer: IBLI most valuable among the vulnerable non-poor

- simulation-based and WTP survey based answer: Highly price elastic demand for IBLI

2) Does IBLI induce increased asset accumulation and escapes from poverty? Does it reduce asset loss and falls into poverty? How does it perform relative to cash transfers? Are there spillover effects on the stockless poor?- simulation-based answers: Yes on first two points. Don’t know on latter two questions.

Use survey data to test these hypotheses in quasi-experimental setting with real insurance in a survey designed to test IBLI vs./with cash transfers under Kenya’s new Hunger Safety Nets Program.

IBLI

IBLI is a promising option for putting climate risk-based poverty traps behind

us

Thank you for your time, interest and comments!Let’s take a short break.

Break

Much attention to climate change impacts in Africa.

But focus falls mainly on the likely effects of changes in average rainfall and temperature on crop output.

Little study of the likely consequences of increased climate variability, nor to the likely effects on livestock systems.

Threat of Climate Change

What happens to east African pastoralists if the frequency of extreme rainfall events changes?

Barrett & Santos (2011) explore the likely consequences of more frequent drought in the African ASAL on pastoralists’ livestock herd dynamics. - Use original primary data on rainfall-conditional herd growth dynamics collected among Boran pastoralists in S.Ethiopia - Demonstrate state-dependence of herd growth- Reproduce unconditional herd dynamics previously observed - Simulate herd dynamics under changed climate distributions

The results demonstrate how vulnerable pastoralists systems are to relatively modest increases in the frequency of drought.

Core Question

Previous results

Past herd dynamics studies from the region find nonlinear, bifurcated wealth dynamics. For example, among the southern Ethiopia Boran pastoralists we study, Lybbert et al. (2004 EJ) find:

Data

Data

Collected subjective herd growth expectations data, conditional on anticipated rainfall regime, from 116 households in four villages from same Boran region.

Each household asked subjective dist’n of 1 year ahead herd size based on 4 randomly assigned initial herd sizes. Thus multiple observations per hh.

Methods

Methods

1) Nonparametrically explore differences in rainfall-conditional herd dynamics.

2) Fit parametric herd growth functions.

3) Use estimation results from 2) and historical rainfall data to simulate decadal herd dynamics. Compare against previous results.

4) Use estimation results from 2) to simulate herd dynamics under different climate distributions.

Key findings 1

Key findings 1) Not surprisingly, herd dynamics differ

markedly between good and poor rainfall states.

Figure 1. Expected one year ahead herd dynamics with (A) poor rainfall or (B) good rainfall. Points reflect herder-specific observations based on randomly assigned initial herd sizes. The solid line reflects stable herd size. The dashed line depicts the nonparametric kernel regression.

Parametric herd growth estimates match NP results

Table 1. Estimates of expected one year ahead herd size conditional on rainfall regime (columns) and randomly assigned initial herd size (h0). P-values in parentheses; estimates statistically significant at the five percent level in bold.

Variable

Rainfall Regime

Very good Good/Normal Bad Very Bad

h0 1.293 (0.00) 1.477 (0.019) 0.538 (0.224) 0.246 (0.246)

h02 0.026 (0.010) 0.009 (0.010)

h03 -0.00039 (0.0001) -0.00017 (0.0001)

Constant 0.897 (0.448) 0.179 (0.416) 0.513 (1.185) -0.575 (1.083)

N 61 96 192 61

R2 0.986 0.994 0.792 0.589

Key findings 1

0

10

20

30

40

50

60

0 10 20 30 40 50 60

Exp

ecte

d h

erd

siz

e 10

yea

rs a

hea

d

Initial herd size

Key findings 2

Key findings 2) Simulated herd dynamics using parametric

herd growth function estimates and historical (N(490, 152)) rainfall distribution generates unconditional herd dynamics very similar to observed patterns.

So pastoralists seem to grasp clearly the underlying herd dynamics of he current system.

Key findings 3

Key findings 3) Herd dynamics change with drought (rainfall

<250 mm/year) risk. Halving the current risk would enhance resilience and eliminate apparent poverty trap. By contrast, doubling drought risk would eliminate high-level equilibiurm and lead to system collapse in expectation.

Simulated using the parametric herd growth function estimates and mean-preserving changes of rainfall variance, defined by π= prob(rainfall<250 mm/yr)

0

10

20

30

40

50

60

0 10 20 30 40 50 60

Exp

ecte

d h

erd

siz

e 10

yea

rs a

hea

d

Initial herd size

Prob. = 0.03

Prob. = 0.12

Prob. = 0.06

The main store of wealth of Africa’s pastoralists is at risk if climate change brings increased drought, as expected.

Climate variability adaptation is crucial ASAL pastoral systems highly vulnerable to potential system change due to quite plausible changes in rainfall variability. Need more than just food aid in response to disasters. Need to alter herd dynamics to cope with increasing drought risk.

Must begin addressing: - range and water management- resource tenure (e.g., dry season reserve

access) and reconcile with biodiversity conservation

goals- livestock insurance. IBLI one possible tool.

Policy implications

Thank you for your time, interest and comments!