Productive Spillovers of the Takeup of Index

advertisement
Introduction | Setting | Design | Data | Results | Conclusion
Main Question
What is the impact of index-based livestock
insurance (IBLI) on herd stocking and movement
choices of East African pastoralists (livestock
herders)?


IBLI could enhance welfare by helping pastoralists
smooth shocks to herd stocks due to weather.
Concern that scale-up of IBLI could lead to
unsustainable environmental degradation.
Introduction | Setting | Design | Data | Results | Conclusion
Theoretical Possibilities
• Herd sizes follow boom-and-bust cycles between
normal weather and shocks due to drought.
• What happens if we introduce weather-indexed
insurance that pays out during droughts?
Livestock
Drought
Drought
Time
Introduction | Setting | Design | Data | Results | Conclusion
Theoretical Possibilities
• Herd sizes follow boom-and-bust cycles between
normal weather and shocks due to drought.
• What happens if we introduce weather-indexed
insurance that pays out during droughts?
Livestock
Drought
Drought
Time
Introduction | Setting | Design | Data | Results | Conclusion
Theoretical Possibilities
• Predictions for behavioural response depend on
what motivates herd accumulation choices:
Precautionary savings? Lack of investment
alternatives?
o If precautionary savings motives dominate,
then IBLI  herd size ↓
o If risk-adjusted investment motives dominate,
then IBLI could  herd size ↑
• Will they feel less need to move (increasing
grazing intensity), due to substituting IBLI for selfinsurance through costly movement, or perceiving
less risk? Or trickle-through effects of herd size –
e.g., more vigilant asset protection?
Introduction | Setting | Design | Data | Results | Conclusion
Theoretical Possibilities
• Predictions for behavioural response depend on
what motivates herd accumulation choices:
Precautionary savings? Lack of investment
alternatives?
o If precautionary savings motives dominate,
then IBLI  herd size ↓
o If risk-adjusted investment motives dominate,
then IBLI could  herd size ↑
• Will they feel less need to move (increasing
grazing intensity), due to substituting IBLI for selfinsurance through costly movement, or perceiving
less risk? Or trickle-through effects of herd size –
e.g., more vigilant asset protection?
Introduction | Setting | Design | Data | Results | Conclusion
Existing Literature is Mixed
• Janzen and Carter (2013): Insured HH with larger
asset stocks show less asset decumulation after a
drought (in Kenya).
• Jensen et al (2014): IBLI leads to decumulation of
herd assets (in Kenya).
• Cole et el (2013): weather index insurance leads
Indian farmers to more risky production choices.
Introduction | Setting | Design | Data | Results | Conclusion
Mobile Pastoralism
• Arid and semi-arid rangelands of southern
Ethiopia (Borena plateau).
• Over 8 million migrant pastoralists in Ethiopia,
accounting for significant proportion of
agricultural GDP and grazing land.
• Implications for food
security and use of
otherwise
unproductive land.
Introduction | Setting | Design | Data | Results | Conclusion
Mobile Pastoralism
• Bi-annual dry seasons in which local forage and
water resources are not enough to sustain most
herds.
• Hence mobile pastoralists temporarily migrate to
remote water points and pastures.
o
Occupational ladder: sedentarism  mobile
pastoralism  diversification
Introduction | Setting | Design | Data | Results | Conclusion
Mobile Pastoralism
Introduction | Setting | Design | Data | Results | Conclusion
Mobile Pastoralism
Introduction | Setting | Design | Data | Results | Conclusion
Mobile Pastoralism
• Every 3-5 years there have been more severe
droughts.
<wwalert.wordpress.com>
<earthtimes.org>
<uk.oneworld.net>
Introduction | Setting | Design | Data | Results | Conclusion
Index Insurance (IBLI)
• In response to this, colleagues at Cornell, UC Davis
and the International Livestock Research Institute
(Nairobi, Kenya) began developing an index-based
livestock insurance (IBLI) product around 2005,
and rolled it out in 2009.
• Index-based: uses weather data to detect high
probability of herd loss.
• Livestock insurance: households can directly
insure cattle.
Introduction | Setting | Design | Data | Results | Conclusion
What We Do
• Randomly allocate free IBLI contracts covering 15
cattle, to 10 households (10 control).
o
Sample stratified to cover 4 segments of herd size
distribution (trimming poor, immobile HH and rich
ones), in 5 villages.
• GPS collar data:
• In order to study impacts on movement, track 3
cows per household (so 60 collars in total) using
satellite-based GPS collars.
• August, 2011 to present (some gaps due to collar
issues/failure).
• Record exact location at 5 minute intervals.
Introduction | Setting | Design | Data | Results | Conclusion
Data: GPS Collaring
Globular
Movement Pattern
Each Globule
Centered Around
a Watering Point.
Linear, Trailing Features
Several Globules of
Concentrated Use
High Potential for
Environmental Impact
Limited Foraging Extent
Very Extensive
Movement Patterns
Daily
Watering
Foraging Much
More Dispersed.
Foraging Loops
Generally Less Impactful.
Introduction | Setting | Design | Data | Results | Conclusion
Data: Checking Baseline Balance
Treatment
Control
Mean
SD
Mean
SD
Variable name
Age HH head
49.55
9.55
49.95
16.72
Sex HH head
1.00
0.00
1.10
0.31
Num HH member
7.90
3.16
8.40
5.07
Num subherds
3.65
1.04
2.75
1.02
TLU whole herd
32.15
41.21
26.40
19.63
TLU cattle only
36.75
41.89
29.75
22.42
IBLI know. Index (/8)
5.40
0.82
5.50
0.83
Concern 1
1.57
0.87
1.50
1.05
Concern 2
1.26
0.91
1.84
1.69
Concern 3
0.83
0.68
0.76
0.61
Concern 4
0.58
0.56
0.62
0.61
Concern 5
0.68
0.67
0.71
0.92
Concern 6
0.71
0.86
0.71
0.75
Concern 7
0.75
0.87
0.78
0.84
Concern 8
0.56
0.52
0.64
0.67
Concern 9
0.93
0.87
0.89
0.89
Concern 10
0.69
0.72
0.78
0.95
N
20
20
Note: *** p<0.01, ** p<0.05, * p<0.1
Note: TLU = Tropical Livestock Units (1 TLU = 1 cow = 0.7 camels = 0.1 sheep or goats)
1 TLU = 1 adult cow
Difference
0.93
0.15
0.71
0.01 ***
0.58
0.51
0.70
0.81
0.19
0.72
0.86
0.91
0.99
0.91
0.66
0.88
0.74
Introduction | Setting | Design | Data | Results | Conclusion
Econometric Approach
• Use a diff-in-diff estimator on the IBLI
randomization:
𝑦𝑖𝑗𝑡
= 𝛽1 𝑇𝑟𝑒𝑎𝑡𝑖𝑗𝑡 + 𝛽2 𝑃𝑜𝑠𝑡𝑖𝑗𝑡 + 𝛽3 𝑇𝑟𝑒𝑎𝑡𝑖𝑗𝑡 ∗ 𝑃𝑜𝑠𝑡𝑖𝑗𝑡 + 𝐹𝐸𝑖 + 𝛿𝑋𝑖𝑗𝑡
+ 𝜀𝑖𝑗𝑡 ,
where 𝑖 indexes households, 𝑗 indexes villages, 𝑦𝑖𝑗𝑡 is
an outcome, 𝐹𝐸𝑖 is an individual-specific fixed effect,
and 𝑋𝑖𝑗𝑡 represents other time-varying controls.
• Also implement a triple-difference estimator to
expose variation along understanding of the IBLI
contract (index from 8 questions).
Introduction | Setting | Design | Data | Results | Conclusion
Results: Impacts on Asset Accumulation: TLU cattle
Treatment
Post
Treament*Post
Age of HH Head
Age of HH head squared
HH members
(1)
-154.319**
-68.823
-22.114**
-9.595
11.167*
-5.711
28.638***
-10.536
-0.117
-0.079
-109.285***
-31.342
Post*IBLI index
Treatment*Post*IBLI index
R-squared
0.939
Adj. R-squared
0.919
N
100
* p<0.10, ** p<0.05, *** p<0.01
Note: All regresssion control for HH fixed effects.
(3)
-64.221
-66.392
-57.955
-79.872
36.459
-60.818
30.249**
-13.788
-0.133
-0.11
-122.133***
-41.562
6.581
-14.454
-4.631
-10.874
0.939
0.917
100
At baseline: mean(TLU cattle) = 29.3, sd(TLU cattle) = 32
IBLI increases herd
accumulation by
about 11 TLU, about
1/3 of mean herd size
and 1/3 of the sd.
Introduction | Setting | Design | Data | Results | Conclusion
Results: Impacts on Asset Accumulation: TLU cattle
Treatment
Post
Treament*Post
Age of HH Head
Age of HH head squared
HH members
(1)
-154.319**
-68.823
-22.114**
-9.595
11.167*
-5.711
28.638***
-10.536
-0.117
-0.079
-109.285***
-31.342
Post*IBLI index
Treatment*Post*IBLI index
R-squared
0.939
Adj. R-squared
0.919
N
100
* p<0.10, ** p<0.05, *** p<0.01
Note: All regresssion control for HH fixed effects.
(3)
-64.221
-66.392
-57.955
-79.872
36.459
-60.818
30.249**
-13.788
-0.133
-0.11
-122.133***
-41.562
6.581
-14.454
-4.631
-10.874
0.939
0.917
100
At baseline: mean(TLU cattle) = 29.3, sd(TLU cattle) = 32
The mean herd size
result does not vary
in knowledge of IBLI
Introduction | Setting | Design | Data | Results | Conclusion
Results: Impacts on Max Distance Travelled
(1)
Treatment
Post
Treament*Post
4.350***
-0.099
-0.064
-0.052
0.101
-0.084
Age of HH Head
Age of HH head squared
HH members
No significant average
effect of IBLI on max
distance travelled.
(2)
-3.867
-3.919
-0.113
-0.146
0.097
-0.084
0.552***
-0.169
-0.005***
-0.001
-0.651*
-0.353
Number IBLI Q correct (/8)
Post*IBLI index
Treatment*IBLI index
Treatment*Post*IBLI index
R-squared
0.729
Adj. R-squared
0.729
N
17569
* p<0.10, ** p<0.05, *** p<0.01
Note: All regresssion control for HH fixed effects.
0.73
0.729
17569
(3)
.
.
-1.159***
-0.378
5.103***
-0.591
0.217
-0.177
-0.003**
-0.001
-0.163
-0.527
0.588
-2.21
0.205***
-0.065
-0.252
-0.896
-0.942***
-0.109
0.731
0.731
17569
At baseline: mean(Max. Dist. Traveled) = 4.2 km, sd(Max. Dist. Traveled) = 3 km
Introduction | Setting | Design | Data | Results | Conclusion
Results: Impacts on Max Distance Travelled
(1)
Treatment
Post
Treament*Post
4.350***
-0.099
-0.064
-0.052
0.101
-0.084
Age of HH Head
Age of HH head squared
HH members
Number IBLI Q correct (/8)
Post*IBLI index
Treatment*IBLI index
Max distance
decreases by ~1 km
(25% of mean) per
day per correct
question (out of 8)
(2)
-3.867
-3.919
-0.113
-0.146
0.097
-0.084
0.552***
-0.169
-0.005***
-0.001
-0.651*
-0.353
Treatment*Post*IBLI index
R-squared
0.729
Adj. R-squared
0.729
N
17569
* p<0.10, ** p<0.05, *** p<0.01
Note: All regresssion control for HH fixed effects.
0.73
0.729
17569
(3)
.
.
-1.159***
-0.378
5.103***
-0.591
0.217
-0.177
-0.003**
-0.001
-0.163
-0.527
0.588
-2.21
0.205***
-0.065
-0.252
-0.896
-0.942***
-0.109
0.731
0.731
17569
At baseline: mean(Max. Dist. Traveled) = 4.2 km, sd(Max. Dist. Traveled) = 3 km
Introduction | Setting | Design | Data | Results | Conclusion
Results: Impacts on Average Speed
Treatment
Post
Treament*Post
(1)
0.617***
-0.008
0.005
-0.004
0.020***
-0.007
Age of HH Head
Age of HH head squared
HH members
IBLI increases average
speed (economically
small effect, 3.1% of
mean, 6.7% of sd).
(2)
0.691**
-0.343
0.036***
-0.013
0.019***
-0.007
0.016
-0.015
-0.000***
0.00
0.038
-0.031
Number IBLI Q correct (/8)
Post*IBLI index
Treatment*IBLI index
Treatment*Post*IBLI index
R-squared
Adj. R-squared
N
0.893
0.893
17569
* p<0.10, ** p<0.05, *** p<0.01
Note: All regresssions control for HH fixed effects.
0.893
0.893
17569
(3)
.
.
-0.071**
-0.032
0.476***
-0.048
-0.015
-0.015
-0.000*
0.00
-0.140***
-0.045
0.645***
-0.192
0.021***
-0.006
0.202***
-0.079
-0.086***
-0.009
0.894
0.894
17569
At baseline: mean(Avg. Speed) = 0.63
km/hr, sd(Avg. Speed) = 0.3 km/hr
Introduction | Setting | Design | Data | Results | Conclusion
Results: Impacts on Average Speed
Treatment
Post
Treament*Post
(1)
0.617***
-0.008
0.005
-0.004
0.020***
-0.007
Age of HH Head
Age of HH head squared
Average speed
significantly
HH members
decreases in
Number IBLI Q correct (/8) knowledge of IBLI
(14% of mean, 29% of
Post*IBLI index
sd per correct
question)
Treatment*IBLI index
(2)
0.691**
-0.343
0.036***
-0.013
0.019***
-0.007
0.016
-0.015
-0.000***
0.00
0.038
-0.031
Treatment*Post*IBLI index
R-squared
Adj. R-squared
N
0.893
0.893
17569
* p<0.10, ** p<0.05, *** p<0.01
Note: All regresssion control for HH fixed effects.
0.893
0.893
17569
(3)
.
.
-0.071**
-0.032
0.476***
-0.048
-0.015
-0.015
-0.000*
0.00
-0.140***
-0.045
0.645***
-0.192
0.021***
-0.006
0.202***
-0.079
-0.086***
-0.009
0.894
0.894
17569
At baseline: mean(Avg. Speed) = 0.63
km/hr, sd(Avg. Speed) = 0.3 km/hr
Introduction | Setting | Design | Data | Results | Conclusion
Conclusion and Discussion
• Evidence of an economically-significant impact
of IBLI on asset accumulation.
 Consistent with fears about excess accumulation.
• Interesting interaction in IBLI knowledge –
potentially increasing grazing intensity.
Future Work
• Richer measures of movement.
• More fine-grained analysis (sub-herd level).
• Bring in data from large HH survey:
o
Potential for IV on insurance uptake from larger sample
(using cruder movement data on large HH survey).
Download