Dynamic Effects of Index Based Livestock Insurance on Household Intertemporal Behavior and Welfare
Dynamic Effects of Index Based Livestock
Insurance on Household Intertemporal
Behavior and Welfare
Munenobu Ikegami, Christopher B. Barrett, and Sommarat Chantarat
International Livestock Research Institute (ILRI), Cornell University, and The
Australian National University (ANU)
www.ilri.org/ibli
At “Mobile Pastoralism, Index Insurance, Computational Sustainability and Policy
Innovations for the Arid and Semi-arid Lands of East Africa”, ILRI, on June 10,
2015
0. Outline
1. Introduction
– Literature
– Central Question
2. Methodology
3. Model
a. Autarky
b. under IBLI
4. Results
a. Key Findings & policy implication based on model
prediction
b. Compare model prediction with data
• Next step
1. Introduction
• Index-based insurance
– Reduce vulnerability of agricultural households
– Increase their resilience and welfare
• The focus of previous/ealry studies
– Reduce Income fluctuation = direct effects
• The focus of this study
– Further positive effects:
• Reduce precautionary saving
• Encourage investment in productive asset
• How large are the indirect effects?
1. Introduction
• Previous studies on further positive effects
– Ex-ante evaluation
• Crop and technology adoption: De Nicola (2014)
• Demand and impact of IBLI under poverty trap:
– Chantarat, Mude, Barrett, Turvey (2014)
– Janzen, Carter, Ikegami (2015)
• Environmental feedback: Muller, Quaas, Frank, Baumgartner (2011)
– Reduced form ex-post evaluation of IBLI:
• Carter and Janzen (2014)
• Chebelyon, Lyons (2015)
• Jensen, Barrett, Mude (2015)
• Toth et al. (2014)
• This paper
– focus on joint decisions on investment and insurance purchase
– Structural ex-post evaluation
– Omit environmental feedback and poverty trap
1. Introduction
• Does Index Based Livestock Insurance (IBLI)
induce pastoralists to increase their herd size?
– Livestock as productive asset => increase their herd
– Livestock as precautionary saving => decrease their herd
– Does IBLI induce (further) over grazing and
environmental degradation?
• How much of their herd and under what conditions
do pastoralists insure?
– Is the current design of the insurance contract OK?
• How much would IBLI reduce pastoralists’
vulnerability and improve welfare in the long run?
– How large are “the further” effects?
2. Methodology
• Model
– household decisions on livestock investment and
insurance purchase
– Stochastic structure of livestock accumulation
• Data
– IBLI Marsabit Household Survey from 2009-2013
(Round 1-5)
– Vegetation condition: Normalized Differenced
Vegetation Index (NDVI) from 2001 to 2013 (MODIS
NDVI)
2. Methodology
• Fit the model to data
– 1st Step: Estimate milk production and stochastic
structure of livestock accumulation
– 2nd Step: Compute optimal decisions on livestock
investment and insurance purchase
• Counter-factual / ex ante policy simulation
• Compare data with model prediction (ex post impact
evaluation)
3.b. Model: under IBLI:
Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb
Period of continuing observation of NDVI
for constructing LRLD mortality index
LRLD season coverage SRSD season coverage
1 year contract coverage
Sale period
For SRSD
Predicted SRSD mortality is announced.
Indemnity payment is made if triggered
Period of NDVI observations
for constructing SRSD
mortality index
Prior observation of NDVI since
last rain for LRLD season
Sale period
For LRLD
Sale period
For SRSD
Predicted LRLD mortality is announced.
Indemnity payment is made if triggered
Prior observation of NDVI since last rain
for SRSD season
Short Rain Short Dry Long Rain Long Dry Short Rain Short Dry
Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb
Period of continuing observation of NDVI
for constructing LRLD mortality index
LRLD season coverage SRSD season coverage
1 year contract coverage
Sale period
For SRSD
Predicted SRSD mortality is announced.
Indemnity payment is made if triggered
Period of NDVI observations
for constructing SRSD
mortality index
Prior observation of NDVI since
last rain for LRLD season
Sale period
For LRLD
Sale period
For SRSD
Predicted LRLD mortality is announced.
Indemnity payment is made if triggered
Prior observation of NDVI since last rain
for SRSD season
Short Rain Short Dry Long Rain Long Dry Short Rain Short Dry
Temporal coverage of current IBLI (1 year with 2 seasons)
IBLI Contract Feature: Temporal Structure
4. Results
a. Key Finding & Policy Implication based on model
prediction
b. Compare model prediction with data
Key Findings & Policy Implications
• Does IBLI induce pastoralists to increase their herd
size?
– Livestock as productive asset => increase their herd
– Livestock as precautionary saving => decrease their herd
– Does IBLI induce (further) over grazing and
environmental degradation?
Net Livestock Investment in Upper Marsabit
-.2-.1
0
.1.2
i_t
0 10 20 30
k_t
autarky, CZNDVI_pos,t-1 = -15 autarky, CZNDVI_pos,t-1 = -5
autarky, CZNDVI_pos,t-1 = 5 IBLI, CZNDVI_pos,t-1 = -15
IBLI, CZNDVI_pos,t-1 = -5 IBLI, CZNDVI_pos,t-1 = 5
IBLI, CZNDVI_pos,t-1 = 15
Numbers of obsedrvations under autarky and IBLI are 2692 and 4596 respectively.
No observation with CZNDVI_pos_t-1 = 15 under autarky
investment by vegetation condition based on the model
Initial asset and last asset under autarky and IBLI by initial
asset level and non-livestock income in Upper Marsabit
Key Findings & Policy Implications
• Does IBLI induce pastoralists to increase their herd
size?
– Livestock as productive asset => increase their herd
– Livestock as precautionary saving => decrease their herd
– Does IBLI induce (further) over grazing and
environmental degradation?
– => Pastoralists would invest less in livestock in bad
seasons (in order to buy IBLI)
– => but insurance/safety net effects let pastoralists
accumulate more livestock
– => reduced risk exposure but might bring environmental
delegation and decreased productivity
IBLI purchase and non-livestock income
05
1015
0 10 20 30
k_t
data, ynl = 0 data, ynl = 10K
data, ynl = 20K model, ynl = 0
model, ynl = 10K model, ynl = 20k
number of observation is 3681
insured livestock in bad season by non-livestock income
Key Findings & Policy Implications
• How much would IBLI reduce pastoralists’
vulnerability and welfare in the long run?
– IBLI is more beneficial for vulnerable households with
less non-livestock income as an alternative insurance
mechanism
Key Findings & Policy Implications
• How much of their herd and under what conditions
do pastoralists insure?
– households may insure herd sizes larger than they own as
insurance against covariate income shocks more broadly
– households may divest livestock in order to buy
insurance
Results: IBLI purchase and forage condition
0
102030
0 10 20 30
k_t
data, CZNDVI_pos,t-1 = -15 data, CZNDVI_pos,t-1 = -5
data, CZNDVI_pos,t-1 = 5 data, CZNDVI_pos,t-1 = 15
model, CZNDVI_pos,t-1 = -15 model, CZNDVI_pos,t-1 = -5
model, CZNDVI_pos,t-1 = 5 model, CZNDVI_pos,t-1 = 15
The number of observations is 437. State variables are from observations with only k_tilde_t_t+1 > 0 in data.
insured livestock by vegetation condition
Key Findings & Policy Implications
• How much of their herd and under what conditions
do pastoralists insure?
– households seek to buy more insurance when current
vegetation conditions are bad and they expect poor range
conditions – and thus a higher livestock mortality rate –
in the following season
– Is the current design of the insurance contract OK?
• Do not adjust pricing as baseline range conditions change
• do not use an index that is conditional on range conditions as of
the contract sales date
– => Intertemporal opportunistic behavior problem
– => negative for insurance company
model predicts too large TLU insured in bad
seasons
0
102030
0 10 20 30
k_t
data, CZNDVI_pos,t-1 = -15 data, CZNDVI_pos,t-1 = -5
data, CZNDVI_pos,t-1 = 5 data, CZNDVI_pos,t-1 = 15
model, CZNDVI_pos,t-1 = -15 model, CZNDVI_pos,t-1 = -5
model, CZNDVI_pos,t-1 = 5 model, CZNDVI_pos,t-1 = 15
number of observation is 5519.
insured livestock by vegetation condition
model predicts too large offtake in bad season
under IBLI
-.2-.1
0
.1.2
i_t
0 10 20 30
k_t
data, CZNDVI_pos,t-1 = -15 data, CZNDVI_pos,t-1 = -5
data, CZNDVI_pos,t-1 = 5 data, CZNDVI_pos,t-1 = 15
model, CZNDVI_pos,t-1 = -15 model, CZNDVI_pos,t-1 = -5
model, CZNDVI_pos,t-1 = 5 model, CZNDVI_pos,t-1 = 15
Note: Seasons are those when IBLI are on sale.
Number of observations is 4305.
investment by vegetation condition under IBLI
model predicts too large offtake even under
autarky
-.1
-.05
0
.05
.1
i_t
0 10 20 30
k_t
model, CZNDVI_pos,t-1 = -15 model, CZNDVI_pos,t-1 = -5
model, CZNDVI_pos,t-1 = 5 data, CZNDVI_pos,t-1 = -15
data, CZNDVI_pos,t-1 = -5 data, CZNDVI_pos,t-1 = 5
Note: Seasons are the first 3 seasons before IBLI launch.
No observation with CZNDVI_pos_t-1 = 15
Number of observations is 2536.
investment by vegetation condition under autarky
Next step
• Model predicts large offtake and large TLU insured
in bad seasons but data do not show such
intertemporal opportunistic behaviour with such
magnitudes
• Model predicts too large offtake even under autarky
in good seasons under IBLI
• For too large offtake, we relax assumption of
constant TLU value over time
• For too large TLU insured, we introduce learning of
indemnity payout function (maybe in a separate
following paper)