Presentation Transcript
Does Microfinance Make $ense? Experimental Approaches: Does Microfinance Make $ense? Experimental Approaches
IFC M&E Conference
May 9, 2006
Jonathan Zinman
Dartmouth College
Plan for Talk: Plan for Talk Evaluating Impacts of Microcredit Access Using Randomized Credit Supply Decisions
Design
Implementations
Some Results
Beyond Risk Assessment: Evaluating “Access” Interventions Broadly Defined
Other efficiency and strategy interventions
Enforcement
Pricing
Other contract terms: (maturity, loan size)
Savings takeup
Product presentation (marketing, mental accounting)
Product development (reminders)
Distribution Channels (“impulse savings”)
Beyond Impact Evaluation: Experimentation and Innovation
The importance of measuring why interventions (don’t) work
The possibility of transforming organizations into learning laboratories
Evaluating Impacts of Microcredit Access Using Randomized Credit Supply Decisions : Evaluating Impacts of Microcredit Access Using Randomized Credit Supply Decisions Ongoing work with Dean Karlan (Yale)
Our Methodology:
Lender randomizes credit supply decisions:
Randomized-control design: social science gold standard
Subject pool of marginal applicants (“grey area”)
Some in grey area randomly treated (“derationed”)
Remaining in grey area control group (“rationed”)
We follow-up with household and/or business surveys:
Measure investments, broadly defined
Measure impacts, broadly defined
On borrowing/credit access
On various measures of well-being
Measuring Impacts Using Derationing: Measuring Impacts Using Derationing Impact= the difference in an outcome of interest in derationed and rationed groups:
Examples of outcomes:
lender’s profits
applicant borrowing (do rationed get credit elsewhere?)
applicant revenues
applicant consumption smoothness
NOT needed to measure impacts using this method:
No baseline survey needed
No perfect compliance with treatment assignment needed: workable if some derationed borrowers get loans, or vice versa
Can use statistical technique called “Intent to Treat” to measure impacts based on remaining random variation
Measurement Strategy: Measurement Strategy Formally:
(1) Yi = a + bderationedi + driski + fmonthi + ei
Y is an outcome from admin or survey data
derationed is randomly assigned by Lender
risk conditions the randomization (“reversal”) probability on the Lender’s assessment of how close to creditworthy
month partials out aggregate shocks in the time series
Derationing Implementations: Derationing Implementations Completed in South African consumer loan market
Underway in Filipino microenterprise loan market
Planning in Peruvian microenterprise loan market
Market Settings: Market Settings Microenterprise credit market in Metro Manila
For-profit lender
Individual liability
Partly secured
Primarily small grocery/convenience stores
No targeting
Market Settings: Market Settings Consumer loan market in South Africa
For-profit lender regulated by Microfinance Regulatory Council
Unsecured
Individual liability
High-risk
Short-term (4 months), fixed repayments
Expensive (11.75% monthly, simple)
Untargeted, “working poor” clientele
Implementation Details: Engineering Randomness: Implementation Details: Engineering Randomness South Africa: derationing by random reversal (or not) of rejections in grey area
Metro Manila: derationing via implementation of new credit scoring model with random component in grey area
What’s in it for the Lenders?: What’s in it for the Lenders? Improve profitability by careful identification of the profitability frontier
What does the marginal profitable/break-even applicant look like
“Pilot approach”
Systematic and gradual changes
Improve efficiency by process innovation
Introduction of credit scoring
Experimentation and the learning organization
Democratization of approach used by sophisticated firms
ICIC, Green Bank
Preliminary Results from South African Implementation: Preliminary Results from South African Implementation Derationing does increase borrowing over the 6-12 months following the experiment
Some positive impacts 6-12 months out:
Derationed households have less hunger
Derationed households more likely to maintain formal employment
No negative impacts on households
But power issues: small sample, so imprecise estimation of null effects
Derationed loans did have substantially worse repayment.
Profitability?
Beyond Risk Assessment: Access Broadly Defined: Beyond Risk Assessment: Access Broadly Defined Several other aspects of financial product delivery affect access:
Loan pricing: targeted groups may have different takeup elasticities
Dehijia et al vs. Karlan-Zinman
Maturity & loan amount elasticities may dwarf price elasticities for constrained borrowers
Karlan-Zinman; Attanasio et al
Access Broadly Defined: Access Broadly Defined Efficiency-Sustainability-Access nexus:
Risk assessment (credit scoring)
Enforcement & monitoring experiment in Peru (Karlan, Mullainathan,and Zinman)
Access Broadly Defined: Savings: Access Broadly Defined: Savings Do consumers have difficulty saving?
Self-control; Household control
Other motivation and follow-through problems
Then savings takeup decision critical: what drives it?
Product presentation:
Mental accounting (KMZ puzzles experiment)
Marketing and framing a la BKMSZ on loans
Product features (reminders, SMART, SEED)
Distribution channels: “Impulse Savings”
Beyond Evaluation: Why?: Beyond Evaluation: Why? Interventions: how do we know what to try in the first place?
Intuition
Theory
Anecdata
Past Evaluations
Presence or absence underlying market failures interventions are designed to solve
Beyond Evaluation: Why?: Beyond Evaluation: Why? Scientific evidence on empirical relevance of specific market failures also rare
Important to build into evaluations, experimentation
Example: measuring adverse selection and moral hazard
Most important theoretical motivations for microcredit
Little clean evidence on importance of either friction
Beyond Evaluation: Identifying Market Failures: Beyond Evaluation: Identifying Market Failures Karlan-Zinman pricing experiment in South Africa (2005a, 2005b)
Derive profit-maximizing interest rate by randomizing interest rates
This requires one dimension of interest rate variation
Also measure why optimal interest rate is where it is
Demand elasticities
Repayment elasticities due to separate effects of adverse selection and moral hazard
Requires three dimensions of interest rate variation
Why invest in the why of interventions?: Why invest in the why of interventions? Policy
E.g.: adverse selection and moral hazard have different remedies
Practice:
Investments in screening?
Investments in enforcement?
Design of future interventions
Ongoing experimentation as process innovation
Experimentation & the Learning Organization: A Virtuous Cycle: Experimentation & the Learning Organization: A Virtuous Cycle