ifpri-new technologies for better insurance: picture based crop insurance-berber kramer
TRANSCRIPT
New technologies for better insurance: Picture Based Crop InsuranceBerber Kramer ([email protected])
IFPRI Research team: Miguel Robles ([email protected]) Francisco Ceballos ([email protected])
New Delhi, December 21 2016
Why do farmers lack crop insurance?
• Farmers too remote and too small for indemnity insurance
• Administrative costs
• Costs of loss verification
• Monitoring of cheating / Moral hazard
• Index insurance
• Basis risk
• Beyond the control of the farmer
• Technological innovations:
• Satellites: Difficult to understand, cloud cover
• Drones: Better resolution but high operational cost
• Generally: High biodiversity and small scale
• How to bundle the best of these worlds?
Opportunity: Rise of Smartphones
Source: Global Attitudes Spring 2015 & 2014, PEW Research Center
In 2015, there were 220 million unique smartphone users in India
Greenness index estimationO
nlin
e se
rver
GCC reference curve
GCC curve
Machine learning
Low-cost lossassessment
Farmers take
pictures
Additional information
CCE
A hybrid insurance approach
We can test ways to limit moral hazard
Remote sensing from above to detect anomalous behavior?
Use nearby farmers as a benchmark?
Pictures, data, sensorsExperts
/Agronomists
- Weather stations
- Satellite images…
Agro-advise
Use agro-advisories?
Bundled with other services
• Use pictures to provide agro-advisory services
• This is a natural complementarity
• PBI is already collecting rich field data
• Data can be analyzed by experts to provide advise
• Incentive to report truthfully and not cheat the system
Pictures, data, sensorsExperts
/Agronomists
- Weather stations
- Satellite images…
Agro-advise
What do we need to get there?
• Pictures and data for machine learning
• Not only pictures of damage: Algorithms will need both damaged and non-damaged pictures
• Not only data on yields: Farmers’ perceived damage, causes of damage, practices, etc.
• High-frequency at first in order to estimate optimal frequency and need for standardization
• Farmers’ interest in such products
• Not only their willingness to pay
• Also impacts on behavior: investments and cheating
IFPRI’s Picture Based Crop Insurance
• States: Punjab & Haryana
• 6 districts
• 50 villages
• 750 farmers
• Wheatcrop (winter/rabi)
Wheat Cam
First step towards new crop insurance approach
• Collect pictures and data with high frequency
• Include perceived damage and cause of damage
• Practices and input use (seed and soil type, fertilizer, etc.)
• Yields at the end of the season (crop cut)
• Assess practical feasibility
• Can farmers use the app and submit pictures? How often?
• Can we predict damage (or even yield) on the basis of the pictures?
• Assess farmers’ interest in, and response to, such products
Smartphone app
• Android app
• User friendly
• Facilitates taking pictures at exactly same location • Relies on GPS coordinates
• Short survey after taking picture
• Important for later analysis
• Input use and practices
• Currently standardized procedures with reference poles and auxiliary poles
Wheat Cam
Auxiliary Pole
Reference Pole
Aplicación Android
Adding sites…
Capturing initial picture…
Choose site
Capturing repeat pictures…
Picture can be taken only between
10AM and 2PM…
Capturing repeat pictures…
Initial picture shown as ghost image in the background…
Capturing repeat pictures…
Capturing repeat pictures…
Afterwards, the farmeranswers a few questions…
(data for machine learning)
N = 1317
As of now, 290 (22.02%) pictures where farmer selected ‘Yes’.- 3 fully damaged- 33 partially damaged- 257 slightly damaged
No visible damage in 11.7% of these pictures.
Take-up of the technology
Example of good pictures
Example of good pictures
Example of good pictures
Example of bad pictures
Example of bad pictures
Example of bad pictures
Committee of experts
Reviews pictures +
additional information
On
line
se
rver
Loss assessment(or field visit)
Farmers take regular
pictures
This year
Machine learning
Advantages and Disadvantages
If smartphone pictures accurately capture damage, then picture-based crop insurance can contribute:
Lower basis risk (going back to indemnity product)
Easy to understand, easy to relate to (farmer is at the center)
Complements weather index-based products
Leverages increasing use of smart-phones
Feasible for the insurance companies
Main risk: Potential moral hazard
To what extent is there moral hazard? (this year)
How to design the product to limit potential hazard?
THANK YOUQuestions: [email protected]