Transcript Prototype lab

Supporting the development of inclusive financial markets in Kenya
Using Alternative Data for Financial
Decisions:
Lending to Farmers Using Nontraditional Data
Farmers needs are beyond agriculture operations financing
Flexible financial instruments might be required to meet their needs
REASONS WHY PEOPLE BORROW CREDIT PER LIVELIHOOD
CATEGORY
Employed
Own business
Dependent
Casual
~34% access
some form of
credit. Largely
informal
OTH E R
7%
9%
4%
4%
6%
2%
3%
2%
3%
3%
PAY OF F DE B T
V E H ICL E
1%
2%
2%
1%
1%
9%
4%
9%
11%
39%
FinAccess, 2015
H OU SE / L AN D
EM E RGE N CY
12%
11%
11%
10%
19%
23%
35%
16%
18%
14%
E DU CATION
AGRICU LTU RE
19%
12%
7%
6%
8%
42%
6%
3%
BU SIN E SS
P U RP OSE S
DAY- TO- DAY
N E E DS
9%
10%
56%
50%
48%
74%
69%
Agriculture
~ 3% of
national bank
credit portfolio
goes to
agriculture
Informal sources are the main type of ag credit
SOURCES OF CREDIT FOR ALL TYPES OF NEEDS IN
AGRICULTURE LIVELIHOOD CATEGORY
Informal Moneylender
House/ Land Bank
SOURCES OF AGRICULTURE CAPITAL
ASCA /
SACCO, 0.8%
CHAMA/
ROSCA, 4.1%
0.1%
or… 0.2%
Buyer Credit
Credit Card
Government e.g. HELB, Youth…
MFI
Banks (savings, current accounts)
CHAMA / ROSCA
0.2%
Family /
Friend /
Neighbour,
11.3%
1.0%
1.2%
2.0%
2.5%
3.4%
SACCO
3.4%
Shopkeeper
Own savings/
Last harvest’s
surplus, 86.7%
5.0%
Family / Friend / Neighbour
ASCA
Local shop goods credit
Agro-dealers
inputs credit ,
0.30%
Buyer
Credit,
12.8%
3.0%
Mobile banking
Banks (savings, current
accounts), 0.30%
5.7%
6.1%
10.7%
0.0% 2.0% 4.0% 6.0% 8.0% 10.0%12.0%
FinAccess, 2015
Challenges in agriculture credit scoring
• Ag actors largely lack sufficient and reliable conventional data
– The ag produce markets are largely informal – over 90% hence weak
documentation
• Weak financial transactions trail - much of the transaction are
cash based, 97% of all national transactions
• Weak methodologies for use of alternative data
Emerging credit scoring data in the agriculture sector
(general credit needs & ag capital)
1. Collateralisation of
receivables
5. Machine learning and
social media
2. Assets and performance score
3. Inputs purchase / character
scoring / peer scoring
4. Airtime and mobile money
transactions
6. Data mining
7. Credit reference bureau
data
+ Collateralisation of receivables
• Supply chains financing: mainly delivery invoices from blue chips buyers
– Amount of credit is based on a % of delivered value and as per the tripartite agreement
– The buyer undertakes to make remittances through the seller account held in the financial
institutions.
– Credit period averages 30 to 90 days
– In events of prolonged payment delays, debts could be converted into short term loans
– National movable assets registry underway
– Example: http://mobipay.co.ke/ (IT platform, banks, milk processors, farmers)
• Warehouse receipts
– At infant stages.
– Tested for wheat and maize
– National regulation underway
+ Assets and performance score
• Use of business assets and performance as a measure of
credit worthiness
– Platforms that digitise assets and business performance e.g.
livestock details for use in credit, insurance and input financing
decisions
– Details include valuation of the animal, output records,
geotagging, agronomic programmes e.g. vaccination tracking etc.
– Example: http://www.agrilife.co.ke/
+ Inputs purchase / character / peer scoring
• Agro-dealer led credit scoring
– Agro-dealer acts as first line credit score unit based on their client
knowledge
– Example: http://farmshop.co.ke/ testing an agro-dealer led credit
through its franchisees.
• Group based scoring by FSPs
– Most common is group lending or guaranteed lending
– Individual lending for groups members coming up (requires further
innovations – mobile based systems emerging)
• Mining group and individual financial profiles in ASCA / ROSCA / Chamas
+ Airtime and mobile money transactions
• Algorithm using a combination of mobile money
transactions, airtime usage and lock-in savings account
details
– Case study: http://fsdkenya.org/publication/how-m-shwariworks-the-story-so-far/
+ Machine learning and social media
• Machine learning aided algorithms: Mobile based Apps that use various
data ranging from
•
•
•
•
•
Mpesa data (mobile money wallet)
Social networks data – e.g. Facebook
Calls, SMS and Mpesa logs - transactions messages on phone (accessed via App).
Geotagging
Handsets data
– FSPs examples Tala https://tala.co.ke, branch https://branch.co, Pesazetu,
• Data service providers examples https://farmdrive.co.ke
– Datasets: individual, social, agronomic, environmental, economic, satellite data
+ Data mining (the gold within)
• Much of the collected data during application and appraisal stage is
hardly utilised by financial institutions in making lending decisions
• Understanding correlated drivers to good or bad loan repayment using
existing data (application forms, call reports, credit reference data,
cashflows, appraisal, ROSCA & ASCA data) etc.
– Digitisation of physical credit related data with high correlation to repayment
behaviour
– Identification of correlated repayment drivers
– Developing scores based on the good predicators and segmentation
Challenges in alternative data for credit scoring
• Consumer protection
• Data privacy and access
– Consumer consent
• Data availability and authentication
• High cost of credit by emerging alternative credit providers (short
tenure, high interest products)
• Outside the regulation scope
Michael Mbaka
Senior Innovations Specialist
FSD Kenya www.fsdkenya.org
[email protected]
+254 721 83 93 83