AI summary by TwinMind
Lecture 25: Credit (Part 2)
MIT OpenCourseWare
1h 15m 1s
Course Recap and Introduction to Microfinance
- Reviewed previous models: adverse selection, moral hazard, monitoring models for credit
- Today's focus: microfinance/microcredit, impact evaluation, heterogeneous effects
- Emphasis on microfinance as bundle of different components requiring research to unpack
Background and History of Microfinance
- Modern microfinance model originated with Grameen Bank, Muhammad Yunus (Nobel Prize winner)
- Made small loans to women in Bangladesh
- Scale: 25+ billion dollars, hundreds of millions of clients globally (numbers likely outdated/higher)
- Typically high repayment rates, some institutions profitable, others not
- Loan sizes: $50-$1000 (small loans for small borrowers)
- Theory: helps break credit constraints for very poor who can't borrow anything/very little
Key Questions for Evaluating Microfinance Claims
- Need to measure net access to credit (not just gross borrowing from MFIs)
- Important outcomes to measure:
- Business profits and consumption
- Return to capital (F'(K)) vs level of capital (F(K))
- If credit constrained: expect F'(K) >> R (marginal return much greater than interest rate)
- If unconstrained: expect F'(K) = R
- Distribution of effects matters, not just averages:
- Quantile treatment effects important
- World where some benefit, none harmed vs some benefit greatly, some harmed substantially
- Predictability of heterogeneity - can people/banks predict who benefits?
- Take-up rates consideration
- Comparison to payday loans in US:
- Both target low-income individuals
- Payday loans collateralized against future paychecks, microfinance typically uncollateralized
- Different uses: payday loans for consumption smoothing vs microfinance for investment
Challenges in Evaluating Microfinance
Institutional Resistance
- MFIs may resist evaluation if profitable/breaking even
- Don't want rigorous impact evaluation for outcomes beyond profits
- Risk of negative findings
Design Challenges
- Multiple outcomes with long timelines
- General equilibrium effects:
- Individual-level randomization problematic due to spillovers
- Product market competition (e.g., fried rice vendors competing for limited demand)
- Demand effects and information spillovers
- Solution: village/cluster-level randomization better than individual-level
Low Take-up Rates
- Typical take-up ~10% of population
- Creates power problems due to square root of n issue
- Need much larger sample sizes to detect effects
- Solution: predict take-up from baseline survey, restrict analysis to high predicted take-up subsample
- Can achieve 80%+ take-up on restricted sample while maintaining randomization at village level
Major RCT Evidence
Six-Study Synthesis
- Special issue in American Economic Journal: Applied Economics
- Studies from India, Ethiopia, Mexico, Mongolia, Bosnia, Morocco
- Two main experimental designs:
- Place-based randomization (village/area level)
- Randomize order of MFI rollout
- Exploits MFI indifference on timing/order
- "On the bubble" randomization
- MFI scores applicants, randomizes marginal cases
- Exploits MFI indifference on borderline applicants
- Place-based randomization (village/area level)
Meta-Analysis Results (Meager)
- Hierarchical Bayesian framework to combine studies
- Main findings: not massive average impacts on consumption
- Larger impacts for people with existing businesses
- Heterogeneity important - different effects for different populations
Hassan, Narayanan & Roth Paper on Predicting Returns
Background and Motivation
- Built on targeting paper concepts - if communities know income levels, might also know marginal returns
- Intellectual arbitrage from poverty targeting to microfinance context
- Originally attempted loan study, switched to grants due to MFI unwillingness to randomize loans
Methodology
- Asked entrepreneurs in peri-urban Maharashtra to rank peers (groups of 4-6) on business profitability and growth potential
- Randomly distributed $100 cash grants to 1/3 of entrepreneurs
- Key insight: needed random assignment to measure returns (F'(K)) not levels (F(K))
- Grants vs loans trade-offs discussed:
- Selection effects (everyone takes grants vs selective loan uptake)
- Timing/repayment constraints may affect investment choices
- Risk aversion considerations
Key Regression Framework
- Main equation: Outcome = α + β₁(Cash grant) + β₂(Predicted returns) + β₃(Predicted returns × Cash grant)
- Also compared community predictions vs machine learning predictions of heterogeneous returns
- Two approaches: separate regressions vs horse race regression
Results
- Community members have substantial information about peers:
- Income, profits, assets, digit span (numeracy test)
- Key finding: predictable heterogeneity in marginal returns
- Higher predicted returns correlated with higher treatment effects from grants
- People invest in more assets, work harder, hire more labor
- Community predictions had additional predictive power beyond observable characteristics
- Suggests local information valuable for targeting
Unbundling Microfinance Components
Traditional Grameen Bank Model Bundle
- Lending only to women
- Weekly repayment schedule starting immediately after loan
- Group lending with group liability
- Regular meetings and monitoring
- Dynamic incentives (start small, grow with good repayment)
- Credit officers spending significant time monitoring
- High interest rates (20%+ annually, often much higher)
Research Program to Decompose Bundle
- Series of papers systematically studying individual components
- Goal: understand which elements actually matter vs unnecessary
Gender Targeting Study (de Mel, McKenzie, Woodruff)
- Sri Lanka study: 400 households with small businesses
- Randomized cash/asset grants of $100-200
- Main finding: 60% annual returns (very high, consistent with credit constraints)
- But: no effects for female-owned businesses
- Follow-up study (Bernhardt et al.):
- Shouldn't look at business in isolation - need household perspective
- When women get grants, household income rises even if her business doesn't grow
- Suggests intra-household reallocation to husband's business
- Female-only households show returns similar to men's businesses
Grace Periods Study (Field et al.)
- Standard microfinance: start repaying immediately (seems suboptimal for investment)
- Experiment: normal contract vs 2-month grace period
- Expected effects: higher returns but potentially higher default rates
- Results:
- More business investment, less household expenditure with grace period
- Higher business income and profits
- But also higher default rates
- Illustrates tension between bank interests (repayment) vs borrower welfare (income)
Additional Topics Mentioned
Indigenous Credit Institutions
- Rotating Savings and Credit Associations (ROSCAs)
- Groups pool money monthly, take turns receiving pot
- Functions as informal credit system
- Very common throughout developing world
Future Research Directions
- Demand for credit
- Banking intermediaries
- Credit constraints for larger firms
- Digital lending and fintech applications
