Microfinance Impact Assessment
- Introduced by
- Dean Karlan, Jonathan Zinman; Banerjee, Duflo, Glennerster & Kinnan; J-PAL
- Year
- 2010
- Type
- Programme impact evaluation
- Family
- Financial-services impact evaluation
What does it do?
Microfinance impact assessment is the set of methods used to measure the causal effects of small loans, savings, and related financial services — long promoted as a tool against poverty — on borrowers' income, business activity, consumption, and empowerment.
After two decades in which observational studies reported large gains, a wave of randomized evaluations from around 2010 onwards, exemplified by Banerjee, Duflo, Glennerster, and Kinnan's Hyderabad study with Spandana and Karlan and Zinman's randomised credit-scoring work, delivered a more sober and credible verdict.
The intuition behind the method
The headline difficulty is selection: people who borrow differ from those who do not — they may be more entrepreneurial, more optimistic, or more desperate — so comparing borrowers to non-borrowers conflates the loan with the borrower. The same logic infects lender-side variation, since microfinance institutions choose where to expand. Credible assessment therefore engineers a comparison the borrower does not control: randomly opening branches in some neighbourhoods, randomising marginal loan approvals through credit scoring, or randomising an encouragement to borrow and using it as an instrument for actual take-up.
History
Microfinance rose to global prominence with Muhammad Yunus and the Grameen Bank, whose 2006 Nobel Peace Prize crowned a narrative of credit as a near-miraculous escape from poverty. That narrative rested largely on observational evidence vulnerable to selection bias. From around 2008–2010 a cluster of randomized evaluations — Banerjee, Duflo, Glennerster, and Kinnan in Hyderabad with the lender Spandana, Karlan and Zinman's randomised credit scoring in South Africa and the Philippines, and parallel trials — found modest, heterogeneous effects: more business investment but little average change in consumption or empowerment. A 2015 American Economic Journal special issue synthesised six such studies, recalibrating expectations toward financial inclusion rather than transformation.
When to use it
Use microfinance impact assessment when evaluating whether access to credit, savings, or insurance products changes the economic lives of poor households and microenterprises, and when a source of exogenous variation — randomised branch placement, credit-score thresholds, or encouragement — is available. It is well suited to testing specific product designs (group versus individual liability, repayment schedules, grace periods). It is less suited where credit markets are already saturated, where general-equilibrium effects on local enterprise dominate, or where only long-run transformational effects matter and short follow-up windows cannot detect them.
Limitations and considerations
- Instrumental-variable estimates identify a local effect on compliers — those induced to borrow — which may not represent the average borrower.
- Short evaluation horizons may miss slow-building transformational effects on business growth and human capital.
- General-equilibrium effects — displacement of non-borrowing competitors, changes in local wages and prices — are hard to capture at the household level.
- Heterogeneous, often-skewed business returns mean average effects can mask large gains for a minority of high-return entrepreneurs.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Randomised and instrumental-variable designs break the selection bias that made earlier observational microfinance studies unreliable.
- Coordinated multi-country randomized evaluations allow comparison of effects across very different credit markets.
- The methods can isolate the effect of specific product features (liability structure, repayment timing) on take-up and outcomes.
- Credible null and modest results have reshaped policy expectations, replacing hype with evidence-based product design.
Applications
- Estimating the effect of expanding group-lending microcredit branches on business creation and household consumption.
- Testing how loan terms — grace periods, repayment frequency, individual versus joint liability — affect default and investment.
- Evaluating microsavings and commitment-savings products on asset accumulation and resilience to shocks.
- Assessing the empowerment effects of lending targeted at women on bargaining power and enterprise.
- Using randomised credit scoring to measure the marginal effect of approving otherwise-rejected loan applicants.
How it works
How it works
The evaluator specifies the outcomes implied by microfinance's theory of change: business creation and investment, profits and revenues, household income and consumption, asset accumulation, and softer outcomes such as women's decision-making power, schooling, and health. Distinguishing first-order effects on borrowing and business from downstream effects on living standards is essential, because microcredit may change how money is used without raising aggregate consumption.
The core threat is that borrowers self-select and lenders place branches non-randomly, so observed correlations between credit and prosperity overstate impact. The assessment confronts this head-on, rejecting naive borrower-versus-non-borrower comparisons in favour of a source of variation in credit access that is independent of borrower characteristics — the foundation for any causal claim.
A credible design is built: randomised expansion of microfinance branches across communities; randomised approval of marginal applicants via a credit-score threshold (a regression-discontinuity or experimental margin); or a randomised encouragement to take up a loan. Each creates exogenous variation in access that supports either a direct treatment-effect estimate or an instrument for endogenous borrowing.
Effects are estimated as intention-to-treat impacts of expanded access, or, where the design randomises an offer or uses a credit-score instrument, via two-stage least squares (instrumental variables) that recovers the local effect on those induced to borrow. The first stage confirms the instrument shifts take-up; the second stage estimates the effect on outcomes, with inference clustered at the level of randomisation and heterogeneity examined across borrower types.
Formula
\hat{\beta}_{IV} = (Z'X)^{-1}Z'YCommon pitfalls
- Comparing borrowers to non-borrowers without exogenous variation, conflating the loan's effect with the characteristics that led people to borrow.
- Using a weak instrument for take-up, which produces biased and imprecise two-stage least squares estimates.
- Measuring only loan take-up or business investment and inferring welfare gains without checking household consumption.
- Generalising a local average treatment effect on marginal borrowers to the entire population or to a different credit market.
Frequently asked questions
Why can't we just compare microfinance borrowers with non-borrowers?
Because borrowing is a choice. People who take loans differ systematically from those who do not — in entrepreneurial drive, optimism, existing business opportunities, or distress — and lenders choose where to operate. A simple comparison therefore measures the combined effect of the loan and of whatever made the person borrow, overstating (or occasionally understating) the true causal impact. Credible assessment requires variation in credit access that is independent of borrower traits, such as randomisation.
What did the randomized evaluations actually find about microcredit?
The cluster of trials synthesised around 2015 found that expanding microcredit access modestly increased business investment and self-employment activity and gave households more financial flexibility, but produced little or no average effect on household consumption, women's empowerment, health, or education within the evaluation windows. Effects were heterogeneous, with larger gains for some existing entrepreneurs. The verdict reframed microcredit as a useful financial-inclusion tool rather than a transformational anti-poverty cure.
How is instrumental variables used in microfinance impact assessment?
When researchers cannot force people to borrow, they randomise something that shifts borrowing — an encouragement, a marketing offer, or a credit-score-based approval at the margin — and use it as an instrument for actual loan take-up. Two-stage least squares first predicts take-up from the random instrument, then estimates the effect of predicted take-up on outcomes. This recovers the local average treatment effect on compliers — those who borrowed because of the instrument — provided the instrument is strong and affects outcomes only through borrowing.
Related methods
Sources
Banerjee, A., Duflo, E., Glennerster, R., & Kinnan, C. (2015). The Miracle of Microfinance? Evidence from a Randomized Evaluation. American Economic Journal: Applied Economics, 7(1), 22–53.
Open DOI referenceCopied.Karlan, D., & Zinman, J. (2011). Microcredit in Theory and Practice: Using Randomized Credit Scoring for Impact Evaluation. Science, 332(6035), 1278–1284.
Open DOI referenceCopied.