Does Insurance Improve
Resilience?
Research: Jennifer Denno Cissé
Presented by: Joanna Upton
Academic Workshop on Mobile Pastoralism, Index Insurance,
Computational Sustainability and Policy Innovations for the ASALs of East Africa
June 10, 2015

Outline
• Background on Development Resilience
• Data
• Insurance and Household Resilience
• Insurance and Aggregate Resilience
• Conclusion
Intro/Background Data Identification Aggregation Conclusion
Introduction

Barrett & Constas (2014)
• Proposes a theoretical framework for
“development resilience”
• Implications for measurement:
Recommends a moments-based,
probabilistic approach
Intro/Background Data Identification Aggregation Conclusion
Development Resilience

Cissé & Barrett (in production)
• Empirical implementation of B&C theory
of development resilience
• Builds on poverty measurement, poverty
dynamics, poverty traps, and
vulnerability literatures
• Resilient = a high probability of
maintaining acceptable level of well-
being
Intro/Background Data Identification Aggregation Conclusion
Development Resilience

Insurance & Resilience
• Use a development resilience approach
to evaluate impact of IBLI on well-being
• Explore household-level impacts of
insurance
• Construct aggregable FGT-type measure
to explore resilience by subgroup
Intro/Background Data Identification Aggregation Conclusion
Contribution

Data
Intro/Background Data Identification Aggregation Conclusion
IBLI
• 5 rounds of panel data (Marsabit)
• 800+ households
• HH demographics, livestock
accounting, income, consumption
• 2011 drought (before R3)

Data
Intro/Background Data Identification Aggregation Conclusion
Constructed Variable
• Shock (drought dummy) = 1 if
predicted livestock mortality > 15%
Instrumental Variables
• Insurance Coupon
• Coupon * Predicted Livestock

Data
Intro/Background Data Identification Aggregation Conclusion
Outcome Variables
• Mid-Upper Arm Circumference
(MUAC) in cm
• Tropical Livestock Units (TLU):
1 TLU = 1 cow, 0.7 camel, 10 sheep,
or 10 goats

Data
Intro/Background Data Identification Aggregation Conclusion
Table 1: Summary Statistics
Mean
Fully
Settled
Partially
Nomadic
Nomadic
MUAC 14.4 14.6 14.3 14.7
TLU 13.2 5.9 15.2 21.7
Drought 0.25 0.25 0.25 0.25
Female (Head) 0.38 0.35 0.39 0.10
Age (Head) 49.1 51.6 48.2 52.3
Educ (Head) 1.0 2.3 0.6 0
Dependency 1.02 1.04 1.01 1.17
N 3278 767 (23%) 2431 (74%) 80 (2%)
Rounds 4 4 4 4

Impact of Insurance
Data Identification Aggregation ConclusionIntro/Background
2SLS model for well-being
𝑊𝑖𝑡
= 𝛽 𝑀1𝑡 𝑊𝑖,𝑡−1 + 𝛽 𝑀2𝑡 𝑊𝑖,𝑡−1
2
+ 𝛽 𝑀3𝑡 𝑊𝑖,𝑡−1
3
+ 𝛽 𝑀4𝑡 𝑆𝑡 + 𝛽 𝑀5𝑡 𝐼𝑡 + 𝛽 𝑀6𝑡(𝑆𝑡∗ 𝐼𝑡) + 𝜷 𝑴𝟕𝒕 𝑯𝒊𝒕
+ 𝑢 𝑀𝑖𝑡
With well-being (𝑊𝑖𝑡) a function of lagged
𝑊 (here cubic polynomial), shocks (𝑆𝑡),
insurance holdings (𝐼𝑡), interaction term
(𝑆𝑡 ∗ 𝐼𝑡), HH characteristics (𝑯𝒊𝒕), & a

Impact of Insurance
Data Identification Aggregation ConclusionIntro/Background
Following Antle (1986)
𝑢 𝑀𝑖𝑡
2
= 𝛽 𝑉1𝑡 𝑊𝑖,𝑡−1 + 𝛽 𝑉2𝑡 𝑊𝑖,𝑡−1
2
+ 𝛽 𝑉3𝑡 𝑊𝑖,𝑡−1
3
+ 𝛽 𝑉4𝑡 𝑆𝑡 + 𝛽 𝑉5𝑡 𝐼𝑡 + 𝛽 𝑉6𝑡(𝑆𝑡∗ 𝐼𝑡) + 𝜷 𝑽𝟕𝒕 𝑯𝒊𝒕
+ 𝑢 𝑉𝑖𝑡
𝐸 𝑊𝑖𝑡 = 𝜇1𝑖𝑡 & 𝐸 𝑢 𝑀𝑖𝑡
2
= 𝜇2𝑖𝑡
 Parameters for our distribution

Data Identification Aggregation ConclusionIntro/Background
Impact of Insurance
Table 2: Pooled 2SLS Estimates of Well-being
(A) (B) (D) (E)
VARIABLES TLU V(TLU) MUAC V(MUAC)
W_lag 0.970*** 10.19*** -6.227** -0.936
W_lag2 -0.00383** -0.0728 0.444** 0.00559
W_lag3 7.04e-06 0.000786*** -0.00981** 0.00138
Shock -5.613*** -1.664e-30 -0.192* -0.477**
Insured TLU 0.424 1.026e-29 0.202* -0.184
Shock*Insrd 2.429 -5.250e-30 -0.0616 0.599**
Path dynamics of well-being.Negative Impact of Shocks.No impact of insurance or interaction on TLU
well-being. Slight impact on MUAC well-being.
Mixed impacts on variance. Appears to be
some heteroskedasticity.

Data Identification Aggregation ConclusionIntro/Background
Identification
Example from Cissé & Barrett
• Normal pdfs of TLU well-being for 2
HHs over time

Impact of Insurance
Data Identification Aggregation ConclusionIntro/Background
Normally distributed well-being
𝜌𝑖𝑡 ≡ 𝑃𝑟 𝑊𝑖𝑡 ≥ 𝑊 = 𝑊
∞
𝑓 𝑊𝑡
𝑤, 𝜇1𝑖𝑡, 𝜇2𝑖𝑡 =
𝑊
∞ 1
2𝜋𝜇2𝑖𝑡
𝑒
−
𝑤−𝜇1𝑖𝑡
2
2𝜇2𝑖𝑡
Resilience Model
𝜌𝑖𝑡
= 𝛽 𝑅1𝑡 𝑊𝑖,𝑡−1 + 𝛽 𝑅2𝑡 𝑊𝑖,𝑡−1
2
+ 𝛽 𝑅3𝑡 𝑊𝑖,𝑡−1
3
+ 𝛽 𝑅4𝑡 𝑆𝑡 + 𝛽 𝑅5𝑡 𝐼𝑡 + 𝛽 𝑅6𝑡(𝑆𝑡∗ 𝐼𝑡) + 𝛽 𝑅7𝑡 𝐻𝑖𝑡
+ 𝑢 𝑅𝑖𝑡

Data Identification Aggregation ConclusionIntro/Background
Identification
Table 2: Pooled 2SLS Estimates of Well-being
(C) (F)
VARIABLES TLU Resilience MUAC Resilience
W_lag 0.0285*** -2.070***
W_lag2 -0.00026*** 0.145***
W_lag3 5.06e-07*** -0.00323***
Shock -0.173*** -0.109***
Insured TLU 0.0212* 0.0795***
Shock*Insured 0.0545* -0.00455
Strong path dynamics with resilience.Negative impact of shock on resilience.Combining mean & variance allows us to see
positive impact of insurance on resilience,
even in non-shock years.
Having insurance during a drought years
further increase TLU resilience (not significant
for MUAC).

Aggregation
TLU Example (Pooled)
• 𝑊 = 6, 𝑃 = 0.8, 𝛼 = 0 (i.e., headcount)
• 𝑅0 𝝆 𝑻𝑳𝑼; 6, 0.8 ≡ 1 −
1
𝑛 𝑖=1
𝑞 𝑔 𝑖
0.8
0
=
0.297
Data Identification Aggregation ConclusionIntro/Background

Data Identification Aggregation ConclusionIntro/Background
Aggregation
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
2 3 4 5
ResilienceHeadcount
Round
TLU Resilience Headcount
(α=0, P=0.8)
Nomadic
Partially Nomadic
Fully Settled

Data Identification Aggregation ConclusionIntro/Background
Aggregation
0
0.1
0.2
0.3
0.4
0.5
0.6
2 3 4 5
ResilienceHeadcount
Round
TLU Resilience Headcount
(α=0, P=0.8)
Insured
Not
Insured

Conclusion
Current Work
• Resilience Measurement (Cisse & Barrett)
• Food Security Measurement using a
Development Resilience Approach
Future Directions
• Dynamic Optimization, Complex
Socioecological Systems
Data Identification Aggregation ConclusionIntro/Background

• Author: Jenn Cissé
jdc358@cornell.edu
• Presenter: Joanna Upton
jbu3@cornell.edu