1

Microfinance

2

Overall motivation for microfinance

• Lack of access to financial instruments (savings, credit) is a key obstacle to poor families seeking to improve their own lives

• Many investments that are good for households' long-run prospects require large up-front costs– e.g., tuition for education, capitalization of small

enterprises

• But it is often difficult to pay such up-front costs – Savings mechanisms are inefficient or nonexistent– Credit mechanisms poor

• Microfinance institutions seek to fill this gap, by bringing financial services to the poor and previously unserved

3

Microfinance: common elements

• Focus on providing financial services to those excluded from the formal banking sector– Most common: credit– More recently: savings– New frontier: insurance

• Credit mostly intended to finance self-employment activities

• Provide small loans (as small as $75), to be repaid over several months to a year

• Many dispense with collateral requirements– Key for poor households with few assets– How are microfinance lenders able to do this?

Asymmetric information problems

• Adverse selection: individuals who know they are likely to default select into borrowing pool, raising default rates and interest rates for everyone– “Hidden type”, “hidden information”

• Moral hazard– Individuals exert less effort than the lender would desire,

raising default rates and interest rates for all– Ex-ante: less effort is exerted to make the “project”

succeed– Ex-post: even if project succeeds, may voluntarily default– “Hidden action”

4

5

Group liability lending

• Widely-publicized mechanism of Grameen Bank for dispensing with collateral requirements– a.k.a. joint liability

• Idea: make everyone in a group of ~5 borrowers jointly liable for repaying each of the loans to group members

• If group doesn't repay each loan, no-one in group gets subsequent loans

• Not the same as “group lending”

6

Group liability in theory

• Why does it work?

• Helps solve asymmetric information problem that usually exists between lenders and borrowers (and that is costly for conventional lenders to deal with)– Adverse selection – Moral hazard

• Reduces adverse selection– groups will only form if all have confidence in individuals'

repayment– people generally know each other beforehand, members will be

“selected” for their reliability as borrowers

• Reduces moral hazard– creates incentives for within-group monitoring and enforcement

• In the end, key is to reduce transaction costs for lenders, allowing them to serve borrowers with very small loans

7

Group liability drawbacks

• What drawbacks might group liability have?

• Increases tension among members– Leads to voluntary dropouts– Can harm social capital among members

• More costly for clients who are good risks, because they are more likely to pay off loans of their peers– Bad clients can “free ride” off good ones– Makes it more difficult to attract and retain good clients

• As groups mature, loan sizes typically diverge– Smaller clients may not want to guarantee larger loans of

other group members

• Overall: group liability’s beneficial effect on repayment may reduce client base (and poor’s overall access to finance)– Also: bank profitability may be lower

8

Gine and Karlan (2006)

• Field experiment in the Philippines

• Microfinance bank where borrowers (all women) organized into joint-liability groups of 20

• 169 pre-existing centers randomized into:– Treatment: converted to individual-liability centers– Control: no change from joint liability

• Findings: – No impact on repayment rate– Attracts new clients to individual-liability centers

• Caveats– Groups were still formed under joint liability (so still benefit from

joint structure’s impact on adverse selection)• Only moral hazard is affected by experiment

– Next step: form groups under individual liability, and test effects

Karlan and Zinman (2007): Motivation

• Credit markets thought to be imperfect due to asymmetric information problems– Adverse selection– Moral hazard

• Policy responses– Microlending: resolve adverse selection, moral hazard via

joint liability, group lending– Subsidies for lenders if moral hazard, asymmetric

information problems make private sector lending unprofitable for the poorest sectors

• Appropriate policies depend on understanding extent of these asymmetric information problems

9

Karlan and Zinman (2007)

• Goal: quantify importance of various information asymmetries in a credit market– Adverse selection– Repayment burden– Moral hazard

• Typically assumed to be unobservable

• Experiment with a consumer lender to the working poor in South Africa

• Randomization used to separately identify these effects– “offer interest rate” identifies adverse selection– “contract interest rate” identifies repayment burden– dynamic repayment incentive identifies moral hazard

10

Experimental design

11

Findings

• Evidence of moral hazard– Dynamic repayment incentive has significant effects on

default

• No evidence of adverse selection or repayment burden overall– But analysis by gender reveals:

• adverse selection for females• Repayment burden for males

– See Tables 4, 5

12

Table 4

13

Table 5

14

Discussion items

1. Why is it important for loan supply decision to be “blind” to the experimental offer rates? – And was it, in fact?

2. Why present results for the standardized index of three default measures? (Kling, Liebman, and Katz 2007)– Ditto the seemingly unrelated regression (SUR)

3. How to interpret results by gender?

15

16

Savings

17

The role of savings

• Transform a series of small payments into a usably large lump sum (Rutherford 1999)

• For investment

• As buffer stock (self-insurance)

• Less costly than credit: no need to pay for lender’s risk

Barriers to savings

• Problems with self-discipline– While understanding the need to save for the future,

individuals can’t resist the temptation to spend now 

• Strong social pressures to share accumulated assets with others who have immediate needs– Reflective of informal insurance/risk-sharing arrangements

• High transactional or informational costs– Distance to branches, unfamiliarity with formal financial

institutions, difficulty filling out forms, etc.– A barrier to formal savings

18

19

Informal savings

• In the absence of formal savings mechanisms, households in poor countries have developed a variety of informal means to save

• Cash savings at home– But vulnerable to temptation, theft, and pressure to share with

others

• Asset accumulation and decumulation– E.g., livestock

• Czukas, Fafchamps and Udry (1998)• Rosenzweig and Wolpin (1993)

– But this comes at an efficiency cost

• ROSCAs

• … but some innovative MFIs are starting to offer formal savings

20

SafeSave

• Helping overcome transactional cost barriers to savings

• New program in Dhaka, Bangladesh• Deposit collectors visit people in

their homes• Clients may deposit as little as one

taka ($0.015) when the collector calls at their house each day

• Accounts with balances above 1,000 taka ($15) earn 6% interest.

• Clients may withdraw up to 500 taka per day ($7.50) at their doorstep, or up to 5,000 taka per day ($75) at the branch office

• 22,000 clients, with average savings balance of $22

Source: http://www.safesave.org/

21

Commitment savings

• Do people need help with self-control, with committing to savings?

• Ashraf, Karlan, and Yin (2006), “Tying Odysseus to the Mast”

• Randomized offer of commitment savings product to customers of a rural bank in Philippines

• Customers pre-commit to save a certain amount or for a certain time period before withdrawal– Withdrawals not allowed before pre-committed amount or

time period, except for emergencies

Table 1

22

23

Table 2

24

Table 3

25

Table 5

26

Table 6

Comments on Ashraf, Karlan, and Yin

• Effects may be due to helping overcome:– Self-discipline problems– But also: helps resist pressure to share with others

• More research needed on whether this is substitution from other forms of saving (other banks, or physical asset holdings)

• A follow-up paper indicates that savings do not seem to be sustained in longer term

27

28

Risk and insurance

29

Agenda

• Risk-coping mechanisms

• Townsend (1994), Udry (1994), and related literature

• A field experiment in Malawi: insurance, credit, and technology adoption

Micro-level responses to risk

• How do households cope with risk?

• In rich countries, people have insurance– Fire insurance, home insurance, auto insurance, life insurance,

medical insurance– These insulate people from the potentially ruinous effects of

catastrophic shocks

• In poor countries, formal insurance markets tend not to exist or to be very limited– The poor have to rely on informal insurance– A vast literature in development economics illustrates the

ingenious ways poor households insure themselves from adverse shocks

• A theme: idiosyncratic risk is easier to cope with than aggregate risk

30

31

Poverty and vulnerability: a vicious circle

VulnerabilityPoverty

Ways to cope with risk

• Ex ante: smooth income

• Ex post: smooth consumption

32

Smoothing income

• Choose a safe production technology: farm a food crop like cassava rather than a cash crop like coffee

• Avoid risky new investments, transitions to different technologies (Malawi example)

• Diversify income sources

• Diversify farming plots spatially

• References: Morduch (1992, 1995, 1999)

• Note all of these are costly (reduce average income, even while making income more stable)

33

Smoothing consumption

• Reciprocal transfers (informal insurance)– Coate and Ravallion (1993), Townsend (1994), Udry (1994),

Ligon (1998), Banerjee and Newman (1993)

• Credit: Udry (1994)

• Asset sales: Rosenzweig and Wolpin (1993)

• Savings: Paxson (1992)

• Labor supply: Kochar (1999)

• Migration by family members: Rosenzweig and Stark (1989)

• Remittances: Yang (2008), Yang and Choi (2007)

34

35

Theory: risk-sharing between households

• Basic result: if there is a Pareto-efficient allocation of risk across households, one households’s consumption should not depend on idiosyncratic shocks

• 2 households, indexed by i=1,2• Uncertain income, separable utility• Pareto efficient allocation of risk between households 1 and 2

implies:

• Any two households’ marginal utilities are proportional – consumption moves in tandem

• If utility is CARA:

U1 cst

1

U2 cst

2 2

1, for all st, and t

csti c st

ln i 12ln 1 ln 2

36

Empirical implication

• Consumption depends only on mean village income (and household’s weight in the Pareto program), and not on idiosyncratic shocks

• Consumption should comove within villages

• Empirical test: regress household consumption on idiosyncratic shocks, controlling for village income (or village fixed effects in panel setting), and idiosyncratic shocks should not have effect

• Townsend (1994), Ravallion and Chaudhuri (1997) find high degree of comovement in consumption across Indian ICRISAT households, even with substantial idiosyncratic income variation– But can reject full risk-sharing (idiosyncratic shocks do have

some effect)

csti c st

ln i 12ln 1 ln 2

37

Xavier Gine World Bank

Dean YangUniversity of Michigan

Insurance, Credit and Technology Adoption:

Field Experimental Evidence from Malawi

A technology adoption puzzle

38

• Green Revolution high-yield crop varieties have led to significant increases in agricultural productivity worldwide

• But there is enormous variation in the extent to which households have adopted these new technologies– In Malawi, hybrid maize adoption has lagged behind

Kenya, Zambia, and Zimbabwe– Need to look beyond credit constraints: even when credit

offered, only 33% of farmers took up a loan for improved seeds

Credit or insurance as the key barrier?

• In observational data, the relative importance of credit constraints and imperfect insurance may be confounded

• Example: widely-observed correlation between wealth and adoption of new technology– May be because wealthier farmers have better access to

credit– But wealthier households may also have better access to

(formal and informal) insurance mechanisms

• Disentangling the two explanations is crucial to good policymaking

• Needed: exogenous variation in insurance

Technology adoption, risk, and credit

• Key question: Does risk inhibit adoption of new technologies?

• High-yielding varieties have higher yields but may also be riskier– So households unwilling to bear fluctuations in their

consumption may decide not to adopt– Downside risk of adoption may be exacerbated when

adoption requires credit• Failure of crop is compounded by the consequences of

default

• Problem: absent or imperfect insurance markets

This paper

• A field experiment where insurance was allocated randomly

• Question of interest:Does providing insurance against a major source of risk increase farmers’ willingness to take out a loan to adopt a new technology?

• Adoption decision: whether or not to take out a loan for improved groundnut and maize seeds

Harvesting groundnuts

42

Weather insurance and loan take-up in theory

• Risk-averse farmers choose between traditional seeds, and taking out loan for improved seeds – Improved seeds have higher mean yield, but are riskier– Consider attractiveness of bundling loan with weather insurance (at

actuarially fair rate)

• Loans subject to limited liability: in case of default, lender can only seize the value of production

• Under certain conditions, farmers might take the uninsured loan if offered, but prefer the status quo (traditional seeds) to the insured loan– Basic idea: limited liability provides implicit insurance– Insurance premium may exceed benefit from insurance

• Rosenzweig and Wolpin (1993): welfare gain from actuarially-fair weather insurance is minimal

43

Simple model

• Output from traditional seeds: YT

• Output from improved seeds: YH, YL with probabilities p, 1-p

– Output positively covaries with rainfall

• Farmers are offered loans to purchase improved seeds (repayment R); lender can only confiscate production, but cannot seize assets (so there is a consumption floor)

• CRRA utility: u(c) = c1-/(1-)

• Farmers are heterogeneous in risk aversion (i) and low-state income from improved seeds (YL,i)

• Some farmers offered loan bundled with actuarially fair rainfall insurance policy (loan forgiven if low state occurs)

• Does rainfall insurance raise loan take-up?

44

What farmers take up the loan?

• Find coefficient of relative risk aversion TU(YL) such that farmer whose i =TU is indifferent between traditional seeds and uninsured loan for hybrid seeds

– Farmer takes up the uninsured loan if i <TU

• Find analogous cutoff for insured loan, TI(YL)

• Cutoffs will be function of income from improved seeds in low state, YL

• See Figure 1

45

Figure 1

46

Key partners in project

• Rural lenders– Malawi Rural Finance Company (MRFC)– Opportunity International Bank of Malawi (OIBM)

• National Smallholder Farmers Association of Malawi (NASFAM)– Contact with farmers

• Insurance Association of Malawi– Underwrites insurance

• World Bank / University of Michigan– Technical advice on design of insurance policy– Design of randomized evaluation

Experimental design

• Joint liability loans for “clubs” of 10-15 farmers– Participation is individual farmer decision

• Randomization across 32 localities

• Treatment: farmers offered hybrid seed loan with insurance against poor rainfall– 393 farmers

• Control: farmers offered hybrid seed loan only (no insurance)– 394 farmers

Loan details

• Farmers given option to purchase either groundnut package only, or both groundnut and maize– Seeds and fertilizer for planting 1 acre (groundnut) or ½ acre

(maize)– Initial deposit of 12.5% of principal– Repayment due in 10 months– 27.5% interest rate (33% annual interest rate x 10/12)

• Maize repayment:– Uninsured: $36– Insured: $40-$43

• Groundnut repayment:– Uninsured: $34– Insured: $36-$38

49

Weather insurance policy

• Farmers insured against poor rainfall as measured at nearest weather station

• Paid continuous amount depending on shortfall below “1st trigger”, up to maximum amount for rainfall at or below “2nd trigger”

• Insurance premium = actuarially fair price + 17.5% surtax

50

Insurance payout structure

payout

1st trigger2nd trigger(corresponds to crop failure)

rainfall during phase

Project locations

Orientation meeting, October 2006

Simple treatment-control comparison

• Take-up rate for uninsured loan: 33.0%

• Take-up rate for insured loan: 17.6%

Regression specification

• For farmer i in group j:

Yij = + Ij + Xij + εij

– Yij = takeup indicator

– Ij = treatment indicator

– Xij = vector of control variables (collected at baseline)

• Standard errors reported:– clustered at locality level– bootstrapped

Table 3: Impact of insurance on take-up of loan for hybrid seeds(Ordinary least-squares estimates)

Dependent variable: Respondent took up loan for November 2006 planting season

(1) (2) (3) (4)

Treatment indicator -0.154 -0.141 -0.132 -0.128[0.109] [0.082]* [0.082] [0.074]*

Clustered s.e. p-value: 0.155 0.085 0.107 0.082Bootstrapped p-value: 0.198 0.116 0.140 0.120

Region fixed effects Y Y YLinear control variables Y

Indicators for 5-year age categories YLand quintile indicators YIncome quintile indicators YEducation quintile indicators Y

Mean dependent variable 0.253 0.253 0.253 0.253Observations 787 787 787 787R-squared 0.03 0.13 0.15 0.17

* significant at 10%; ** significant at 5%; *** significant at 1%

Impact of insurance on take-up

56

Implied interest rate elasticity

• For a farmer placing zero value on insurance, effective annual interest rates for groundnut loan were:– 27.5% for uninsured loan– 37.8% to 44.4% for insured loan (varied according to

location)

• The 13-percentage-point decline in take-up (from baseline 33.0%) a 39.4% decline

• Increase in effective interest rate due to insurance: 37.5% to 61.3%

• Implied interest rate elasticity of credit demand ranging from 0.64 to 1.05

57

Additional testable predictions from theory

• Take-up rates for insured vs. uninsured loan suggest that sample tends to have lower levels of YL

• In this range of YL, there is another theoretical implication to test:

– YL should be positively correlated with take-up of insured loan

– But not correlated with take-up of uninsured loan

• But how to measure YL?

– Assume farmers with higher socio-economic status have higher YL

• Regress take-up on education, income, and wealth– Separately for farmers offered insured and uninsured loans

58

Figure 1

59

Other determinants of take-up

60

Table 5: Determinants of take-up in treatment and control groups(Ordinary least-squares estimates)

Dependent variable: Respondent took up loan for November 2006 planting season

Treatment group (insured loan) Control group (uninsured loan)(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

Years of schooling 0.014** 0.011* -0.001 -0.002[0.005] [0.005] [0.008] [0.009]

Net income (MK 100,000) 0.098 0.075 0.004 0.003[0.059] [0.053] [0.010] [0.010]

House quality 0.041 0.027 0.011 0.011[0.027] [0.030] [0.022] [0.022]

Land owned 0.001 0.001 0.001 0.001[0.003] [0.003] [0.002] [0.002]

Risk aversion (self-reported) -0.008 -0.008 -0.015*** -0.015***[0.006] [0.006] [0.004] [0.004]

Region fixed effects Y Y Y Y Y Y Y Y Y Y Y Y

Mean dependent variable 0.176 0.176 0.176 0.176 0.176 1.176 0.330 0.330 0.330 0.330 0.330 0.330Observations 393 393 393 393 393 393 394 394 394 394 394 394R-squared 0.074 0.078 0.073 0.058 0.061 0.101 0.274 0.274 0.274 0.274 0.286 0.287

F-stat: Joint signif. of first 4 indep. variables: 3.446 F-stat: Joint signif. of first 4 indep. variables: 0.113P-value: 0.03 P-value: 0.98

* significant at 10%; ** significant at 5%; *** significant at 1%Notes -- Standard errors clustered by localities in square brackets. Dependent variable equal to 1 if respondent took up loan for November 2006 planting season, and 0 otherwise. Omitted region indicator is for Kasungu. See Appendix for variable definitions.

Other potential explanations

• Complexity

• Risk priming

• Differential default cost perceptions

61

In sum

• Take-up is lower for loans bundled with insurance against poor rainfall (priced actuarially fairly)– Compared with identical loans that are uninsured

• Potential explanation: – Farmers already implicitly insured by limited liability

inherent in loan contract– Reduces value of the formal, explicit insurance

• Among farmers offered the insured loan, take-up is higher among farmers with higher education, income, and wealth – But not among farmers offered the uninsured loan– Perhaps because higher-status farmers have higher

default costs

62

Ongoing related research

• Current projects continue to examine the nature of constraints that Malawian farmers face in financial markets

• Credit– How important are difficulties in enforcement in limiting

credit supply in rural areas?– In particular, can improvements in identification

technology raise loan repayment rates?

• Savings– How important are imperfect savings mechanisms in

explaining low input use on farms?– How important are transactions costs in explaining low

utilization of formal savings mechanisms?