Index Insurance and Cash Transfers: A Comparative Analysis

Download Report

Transcript Index Insurance and Cash Transfers: A Comparative Analysis

Index Insurance and Cash Transfers:
A Comparative Analysis from Northern Kenya
Nathaniel D. Jensen, Andrew G. Mude and Christopher B. Barrett
Presented by Nathan Jensen
Minneapolis, MN
July, 2014
Motivation
Both cash transfers and index insurance are often endorsed as effective tools
for reducing poverty and providing social protection.
Cash transfers have been extensively studied while the rapid proliferation of
index insurance programs in developing countries has progressed without a
parallel growth in knowledge of the quantity or impacts of such programs.
In one of the most extensive synthesis written about the impacts if index
insurance, Cole et al. (2012) conclude the following:
“The field is in urgent need of evaluations analysing take-up and
impact of marketed products…
…at this stage, research on the impact of index-based insurance
should be the key priority. It cannot be emphasised enough that
very few empirical evaluations of marketed index-based microinsurance programmes exist” (Cole et al. 2012, p. 46-47).
NATHANIEL JENSEN
JULY 2014 | CORNELL UNIVERSITY
2
What impact does index insurance coverage have on the production strategies
and welfare of pastoralists? How do those outcomes compare to that of an
unconditional cash transfer program?
Behavioral changes to investment & production strategies in response to changes
in risk profile and base income
•
•
Cash Transfers: Bianchi & Bobba 2013; Covarrubias et al. 2012; Gertler et al. 2012;
Stoeffler & Mills 2014
Index Insurance: Cai et al. 2010; Karlan et al. 2014; Mobarak & Rosenzweig 2012
Welfare impacts of behavioral changes and direct financial transactions
• Cash Transfers: An abundance of encouraging although not necessarily consistent
studies. See Arnold (2011) and Fiszbein & Schady (2009) for surveys of the literature
• Index Insurance: Karlan et al. 2014; Janzen & Carter 2013
NATHANIEL JENSEN
JULY 2014 | CORNELL UNIVERSITY
3
Setting: Pastoralists in Marsabit, Kenya
• Pastoralists generate a large portion of their income from livestock and livestock
byproducts. (43% of our observations are 100% livestock dependent)
• Drought is the largest killer of livestock.
• Droughts periodically decimate herds.
Causes of Livestock
Mortality Marsabit, Kenya
Source: Author’s calculation (2009-2012)
NATHANIEL JENSEN
JULY 2014 | CORNELL UNIVERSITY
4
Index Based Livestock Insurance (IBLI)
• Introduced in northern Kenya, January 2010
• Objective: To insure households against livestock mortality associated
with droughts
• Signal: Remotely sensed normalized differenced vegetation index
(NDVI) as an indicator of forage availability
• Index: Predicts division average seasonal livestock mortality rate
• Privately provided with public support (DFID, GoK, ILRI, USAID, WB)
Source: Esri
• See http://livestockinsurance.wordpress.com/, Chantarat et al. (2013)
for details
1 year contract coverage
LRLD season coverage
Jan
Feb
Sale period
For LRLD
Mar
Apr
May
Jun
Jul
Aug
SRSD season coverage
Sep
Oct
Nov
Dec
Jan
Feb
Period of NDVI observations for
constructing LRLD mortality index
Sale period
For SRSD
Period of NDVI observations
For constructing SRSD
mortality index
Predicted LRLD mortality is announced.
Indemnity payment is made if IBLI is triggered
Predicted SRSD mortality is announced.
Indemnity payment is made if IBLI is triggered
NATHANIEL JENSEN
JUNLY 2014 | CORNELL UNIVERSITY
For more information on the IBLI project, visit
http://livestockinsurance.wordpress.com/
5
Hunger Safety Net Program (HSNP)
• Part of the larger GoK National Safety Net Program
• Phase I: 2009-2013 (funded by DFID)
• Unconditional bi-monthly cash transfers (~$28/transfer1)
• 3 targeting strategies randomized at the community level:
1.
Social pension: All members over the age of 54
2.
Depends ratio: Ratio of members that are dependent > 57%
3.
Community based targeting: 50% of the community, selected by the community
• No retargeting or graduation
• See http://www.hsnp.or.ke/, Hurrell & Sabates-Wheeler (2013) for details.
1
$ indicates USD, which is calculated using exchanges rates from 1/1/2010.
As a reference point, the average monthly income in our data is $51.57.
NATHANIEL JENSEN
JUNLY 2014 | CORNELL UNIVERSITY
6
Research Design & Data
Legend
MarsabitIBLI
Survey data
HSNP, IBLI Game_HSNP, No IBLI Game_No H
HSNP, IBLI Game
• Annual longitudinal survey of 924 households for 4
rounds
• 4/5 IBLI index divisions
• Seasonal data collected for the most relevant
variables
HSNP, No IBLI Game
No HSNP, IBLI Game
SABARET
ILLERET
EL-HADI
DARADE
NORTH HORR
HURRI HILLS
MOITE
EL GADE
GALAS
KALACHA
GAS
MAIKONA
LOIYANGALANI
TURBI
ARAPAL
LARACHI
KURUGUM
KURUNGU
BUBISA
MAJENGO(MARSABIT)
KARGI
JIRIMEQILTA
HULAHULA
SAGANTE
OGUCHODIRIB GOMBO
KITURUNI
SONGA
KARARE JALDESA
SOUTH HORR(MARSA)HAFARE
• Overlap with a cash transfer program (HSNP) in 8
of 16 communities
• Randomized distribution of coupons providing from
10-60% discount on IBLI policies to 60% of sample
each sales window
NATHANIEL JENSEN
FUROLE
BALESA
OLTUROT
MT. KULAL
Research Design
No HSNP, No IBLI Game
DUKANA
JUNLY 2014 | CORNELL UNIVERSITY
KAMBOYE
KORR
ILLAUT(MARSABIT)
LOGOLOGOGUDAS/SORIADI
LONYORIPICHAU
NGURUNIT
SHURA
LAISAMIS
LONTOLIO
KOYA
IRIRMERILLE
7
Econometric Strategy
′
𝑦𝑖𝑡 = 𝛽0 + 𝛽1 𝑉𝑂𝐼𝑖𝑡 + 𝑥𝑖𝑡
𝛽2 + 𝑐𝑖 + 𝜀𝑖𝑡
𝑉𝑂𝐼𝑖𝑡 = {(𝐻𝑆𝑁𝑃𝐶𝑖𝑡 , 𝐻𝑆𝑁𝑃𝑖𝑡 ), (𝐼𝐵𝐿𝐼𝐶𝑖𝑡 , 𝐼𝐵𝐿𝐼𝑖𝑡 )}
(HSNPC𝑖𝑡 , HSNP𝑖𝑡 ) = (Cumulative prior seasons as an HSNP participant, Current HSNP participant)
(IBLIC𝑖𝑡 , IBLI𝑖𝑡 ) =(Cumulative prior seasons with IBLI coverage, Current IBLI coverage in TLUs)
Unobserved 𝑐𝑖  Use household fixed effect model
Assume Ǝ unobserved x𝑖𝑡 that are correlated with 𝑉𝑂𝐼𝑖𝑡 & 𝑦𝑖𝑡  Instrument for 𝑉𝑂𝐼𝑖𝑡
a. IBLI: Randomly distributed discount coupons
b. HSNP: Exogenous eligibility thresholds
NATHANIEL JENSEN
JUNLY 2014 | CORNELL UNIVERSITY
8
HSNP (FE-IV)
Previous
Current
Participation
Participant
Production strategies:
Herd Size
Veterinary Expenditures (KSH)
Ratio of Herd Held at Home
Household is Partially/Fully MobileA
Production outcomes:
Milk income per TLU (KSH)
Livestock Mortality Rate
IBLI (FE-IV)
Previous
Current Coverage
Coverage
(TLU)
-0.167
(0.453)
11.06
(59.95)
0.0126
(0.0185)
0.0322
(0.0206)
-3.216
(4.121)
371.3
(316.4)
0.106
(0.0965)
0.185**
(0.0855)
-5.912**
(2.776)
955.3**
(462.2)
0.169
(0.165)
-0.0871
(0.147)
0.139
(0.662)
5.039
(167.5)
-0.118*
(0.0711)
0.0934
(0.0658)
74.08**
(30.47)
-0.0260**
(0.0115)
-103.1
(127.8)
0.0556
(0.0348)
760.8***
(211.1)
-0.00838
(0.0599)
118.8**
(57.25)
-0.00280
(0.0161)
A complete list of covariates, coefficient estimates, and model statistics can be found in Jensen, Mude & Barrett (2014). A A
linear probability model is used to estimate the likelihood that a household is partially or fully mobile. Clustered and robust
standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
NATHANIEL JENSEN
JUNLY 2014 | CORNELL UNIVERSITY
9
The impact of covariate shocks and program participation on livestock
sales (TLUs)
Shock
Participation/Coverage (P/C)
P/C*Shock
H0: P/C + P/C *Shock=0 (tstatistic)
Observations
Model F-statistic
Current HSNP
Participant
(FE-IV)
0.261***
(0.087)
0.215
(0.166)
0.0163
(0.203)
Current IBLI Coverage
(FE-IV)
1.118
6,564
4.993
-0.200
6,570
5.054
0.362***
(0.082)
0.188**
(0.083)
-0.215
(0.158)
The shock is an indicator that the division average livestock mortality rates in the current season are equal to
or above 15%. A complete list of covariates, coefficient estimates, and model statistics can be found in
Jensen, Mude & Barrett (2014). Clustered robust standard errors in parentheses. *** p<0.01, ** p<0.05, *
p<0.1.
NATHANIEL JENSEN
JUNLY 2014 | CORNELL UNIVERSITY
10
HSNP (FE-IV)
Previous
Current
Participation
Participant
Indicators of Welfare:
Consumption per Adult
Equivalent (AE)
IBLI (FE-IV)
Previous
Current
Coverage
Coverage (TLU)
-60.36*
(34.01)
-32.17
(372.9)
[15.67]
355.7
(150.7)
-180.7
(150.7)
[13.03]
-8.715***
(3.031)
16.98*
(8.765)
[18.26]
-16.13
(23.26)
-6.077
(4.697)
[21.04]
Income per AE
0.595
(41.35)
374.5
(293.5)
[17.99]
132.9
(304.1)
420.7***
(150.2)
[19.13]
School Enrollment
0.0242
(0.0244)
-0.0392
(0.0941)
[4.077]
0.0931
(0.115)
0.0629
(0.0501)
[3.785]
MUAC
0.0729
(0.109)
0.539
(0.651)
[7.086]
-0.417
(0.846)
-0.0185
(0.226)
[6.504]
Asset Index
A complete list of covariates, coefficient estimates, and model statistics can be found in Jensen, Mude & Barrett (2014).
Clustered robust standard errors in parentheses. Model F-statistic in brackets. *** p<0.01, ** p<0.05, * p<0.1.
NATHANIEL JENSEN
JUNLY 2014 | CORNELL UNIVERSITY
11
Impacts Normalized by Cost Among Clients
Cost structure
Total Program Cost/Participant:
Marginal Cost of an Additional
Participant:
HSNP
IBLI
HSNP
IBLI
Income from Milk
Impact
Impact/Cost
1,585
0.0333
2,536
0.0587
1,585
2,536
0.0469
1.1660
Income per AE
Impact
Impact/Cost
336
0.0071
361
0.0084
336
361
0.0099
0.1662
All values in real 2009 Kenya Shillings. Impacts are estimated using the average client value and costs provided
below, and parameter estimates in the previous two slides.
Average values in the final survey round (clients)
VOI
Mean VOI in Final Period
HSNP
0.89
HSNPC
3.89
IBLI
0.48
IBLIC
1.20
Values are calculated for the subset of clients in each
program.
NATHANIEL JENSEN
Average cumulative cost per client by the final round (KSH)
Total Program
Cost/Participant
Marginal Cost of an
Additional Participant
JUNLY 2014 | CORNELL UNIVERSITY
HSNP
IBLI
47,600
43,200
(2.7BN/57,811HH) (99MM/3,297contracts*1.
44 contract/HH)
33,800
2,175
(14.6 transfers)
(4.41TLUs)
12
Impacts Normalized by Cost Among Clients
Cost structure
Total Program Cost/Participant:
Marginal Cost of an Additional
Participant:
HSNP
IBLI
HSNP
IBLI
Income from Milk
Impact
Impact/Cost
1,585
0.0333
2,536
0.0587
1,585
2,536
0.0469
1.1660
Income per AE
Impact
Impact/Cost
336
0.0071
361
0.0084
336
361
0.0099
0.1662
All values in real 2009 Kenya Shillings. Impacts are estimated using the average client value and costs provided
below, and parameter estimates in the previous two slides.
Average values in the final survey round (clients)
VOI
Mean VOI in Final Period
HSNP
0.89
HSNPC
3.89
IBLI
0.48
IBLIC
1.20
Values are calculated for the subset of clients in each
program.
NATHANIEL JENSEN
Average cumulative cost per client by the final round (KSH)
Total Program
Cost/Participant
Marginal Cost of an
Additional Participant
JUNLY 2014 | CORNELL UNIVERSITY
HSNP
IBLI
47,600
43,200
(2.7BN/57,811HH) (99MM/3,297contracts*1.
44 contract/HH)
33,800
2,175
(14.6 transfers)
(4.41TLUs)
13
Conclusions
• Households with IBLI coverage reduce herd size (precautionary
savings), are more active in livestock markets during non-shock
seasons, increase investments in livestock health services and
realize greater productivity.
• HSNP participants are more mobile, experience reduced
livestock mortality and increased productivity.
• Both programs are likely to improve income/AE.
• HSNP & IBLI produce similar improvements/TPC/participant.
• IBLI generates greater improvements/MC/participant.
NATHANIEL JENSEN
JUNLY 2014 | CORNELL UNIVERSITY
14
IBLI IVs
• A test of balance between coupon recipients and non-recipients
finds few significant differences (<10% of the characteristics
observed).
• Receiving a coupon has a positive and significant impact on
Dummy
Level
demand.
(=1 if purchased)
(TLUs insured)
Coupon Dummy
Observations
F(2,1008)
R2
0.206***
(0.028)
0.541***
(0.076)
7,042
33.5
0.225
7,042
32.7
0.128
Regression includes the following covariates: adult equivalence, age of head, age of
head squared, maximum education in household, a dummy indicating the head of
household is a widow, the current season’s predicted livestock mortality rate, the
current season’s predicted livestock mortality rate squared, division-period dummies
and the three HSNP targeting characteristics to the first, second, and third power.
Clustered and robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
NATHANIEL JENSEN
JUNLY 2014 | CORNELL UNIVERSITY
15
HSNP IVs
• Tests for distortions in the responses around the thresholds does
not reveal misreporting to meet eligibility requirements.
• The intent to treat indicator has a positive and significant impact
on the likelihood of participating in HSNP.
ITT
Observations
Pseudo R2
HSNP Participant
0.614***
(0.050)
7,036
0.524
Regression includes the following covariates: adult equivalence, age of head,
age of head squared, maximum education in household, a dummy indicating
the head of household is a widow, the current season’s predicted livestock
mortality rate, the current season’s predicted livestock mortality rate squared,
division-period dummies and the three HSNP targeting characteristics to the
first, second, and third power. Clustered and robust standard errors in
parentheses. *** p<0.01, ** p<0.05, * p<0.1
NATHANIEL JENSEN
JUNLY 2014 | CORNELL UNIVERSITY
16