The Promise of Microfinance (and the

difficulty of proving it)

Craig McIntosh, UCSD/IRPS‘Microfinance in San Diego’

CourseApril 19, 2011

Group Lending springs the Poverty Trap:

Basic problem of the poor: you have to have money to make money!

Absence of collateral means lack of credit, absence of credit means inability to run efficient businesses, lack of formal-sector jobs means that the poor are trapped in low-return occupations indefinitely.

Question: how to get credit to those without collateral?

The key:Social Capital.

The poor may have few material possessions, but:

1. Poor communities usually have excellent information about each other

2. Lots of social capital because you have to rely on it

3. Profits may be very high due to lack of competition

So, can you replace physical collateral with social collateral?

How do you make somebody without collateral

pay you back?

Joint Liability: if I don’t repay my loans, other members of my group must. Therefore, not only do group members screen each other (I don’t admit anyone into my group I don’t trust) but they monitor each other (make sure that your group members are working diligently) and insure each other (I’ll repay your loan without enacting social sanctions if I know you missed a payment unavoidably).

Dynamic Incentives: future access to capital and future increases in loan size are predicated on reliable repayment today. This way, not only does the lender screen out impatient/unreliable borrowers early on with small loan sizes, but the future value of the relationship with the lender is kept higher than the benefit from defaulting on the current loan.

Collateralization of Savings: Most MF institutions require clients to save some percentage of their repayments in a personal or group savings account. These savings later come to serve as collateral, in the sense that under default the group and/or individual’s savings are confiscated.

What MF is supposed to be:

Small loans given for business investment to the entrepreneurial poor.

Use of ‘joint liability’: clients must repay loans not made by fellow group members, so triggers screening & monitoring.

Sustainable in the long run, because interest rates cover costs of lending.

Usually woman-focused; both because of discrimination and higher female willingness to spend income on kids.

Basic conundrum: you need money to make money, so what are those without collateral for loans supposed to do?

Some of the problems with MF:

Loans are often used for consumption, not investment. This may be optimal short-term, but doesn’t usually create long-term mobility.

Interest rates can be very high indeed; between 20% and 100% depending on country and clientele.

Impact of MF on client welfare still not very clear.

MF has been prone to collapses of repayment under certain circumstances (Quito, Bolivia).

MF may struggle to succeed in countries with well-developed credit markets.

So Does MF Work?Answer is not clear. The better papers

have found: Decrease in volatility of income

(smoothing) Larger income increases for women

than men Frustratingly few MF clients really

climb the ‘ladder of credit’ to start larger businesses.

Lots of reselling, not much manufacturing.

Groups can help to foster social capital when all goes well, can be focus points for tension otherwise.

What makes evaluation in MF hard?

Regression- or matching-based impact evaluations can only control for differences that are observable.

‘entrepreneurship’, ‘risk tolerance’, ‘quality of business ideas’, ‘opportunity’: these words describe the key determinants of success and they are partially or wholly unobservable.

Therefore a comparison of the average MF client to the average individual is likely stacked in favor of finding impacts: the type of person who takes a MF loan would have had more rapidly improving outcomes in the absence of credit.

How is Evaluation Done?

We’d like to randomize across villages, establish T3 in the control villages, and compare T3 across treatment and control (TET).

Problem: T3 not observable in control, and uptake of MF usually too low to compare T2 & T3 across treatment and control.

not accepted accepted not accepted acceptedT3=0 T3=1 T3=? T3=?

Control Village, T1=0

T1=T2=1

T1=T2=T3=0ineligibleeligible

T1=0, T2=1

T1=1, T2=T3=0ineligible

Treatment Village T1=1

eligible

Thought experiment:

Rational investors will take loans only if their profits are higher than the interest rate. Credit markets therefore select the highest-return investors. If you can’t perfectly predict counterfactual profits, you can’t do an unbiased ‘observational’ impact study.

Interest Rate

People who take loans

People who don’t take loans

Distribution of Profits in the absence of Credit:

What has been tried in terms of impact (1)?

Pitt & Khandker (98): use of eligibility rule as ‘instrumental variable’, find consumption increases by 18% (women) and 11% (men) of amount borrowed.

Morduch (98): shows eligibility rule isn’t followed; uses a simpler triple-difference and finds no impact on levels, but a decrease in variance.

Coleman (99): Comparison of new groups to older groups, finds little impact of MF. This is the AIMS/SEEP recommended method, but suffers from attrition bias.

What has been tried in terms of impact (2)?

Karlan & Zinman (07): randomized provision of credit to those just barely denied access by a scoring lender. Find positive impacts.

McIntosh et al (08): retrospective survey data on ‘unforgettable events’. Find increased probability to improve housing stock in year after first receipt of MF loan.

Several ongoing field studies: Randomized promotion of credit to microentrepreneurs; measures the ‘Local Average Treatment Effect’

Granger versus actual causality:

The new state of the art: Banerjee et al 2010, “Miracle of

MF?”. Randomize access to new credit from

Spandana across 104 slums neighborhoods in Hyderabad, India.

Some credit exists even in the control groups, so they distinguish impact of credit from one more lender

Find treatment effects that are different across three strata:

Non-business owners: increase consumption.

Already business owners: increase investment.

Business starters: increase investment, start businesses, and decrease consumption.

No effect on health, education, or women’s empowerment.

Banerjee et al 2010:

New business starts go up by a third, business outcomes are very noisy and no significant effects are detected.

Banerjee et al 2010:

Banerjee et al 2010:What should we take away from the

paper? Has MF failed?

No, it just hasn’t achieved a transformative effect on people’s lives. But then, this is high-interest debt that has to be paid back. Why did we think it would?

Has MF been oversold? Many MFIs make grand claims to donors

to raise money. Would people support MF if they thought it led to a small increase in the rate of business ownership? Should they?

Impacts summarized: MF ‘works’ for lenders. Default very

low. MF does not have a transformative

effect on the average entrepreneur who receives a loan. Average benefits, if any, are small.

MF decreases the variance of consumption; suggests it is better thought of as an insurance product than some kind of venture capital.

MF increases the variance of outcomes across individuals; the poor are similar because they make few choices; credit is power.

Perhaps the average effect is not the interesting question?

Discussion Break 1.

20

Now a segue to some of my own research:

MF markets work by the use of joint liability and dynamic incentives to prevent default.

What, then, will happen when lenders compete?

Competition may have negative consequences in MF markets because:

1. There is no collateral on the line so default is easy

2. Lenders were using rents from richer borrowers to cross-subsidize the poor. The upshot:

Credit bureaus should prove to be a critical institution for the long-term health of competitive MF markets.

21

Three approaches to measuring impacts: Natural experiment, randomized

trial, and a laboratory experiment.

1. Our collaborating microfinance institution (Genesis Empresarial) enters the bureau in a staggered fashion across its administrative branches, without telling clients.

2. We conduct a large randomized educational campaign a year later, informing clients how the bureau works, and how they can take advantage.

3. We create an experimental game in order to understand how microfinance borrowers trade off group versus individual reputation.

This structure allows us to tease out impact from lenders’ side and impact from borrowers’ side separately.

22

Part 1. Natural experiment: rollout of bureau.

MFI rolls its branches into the bureau over 18 months without telling clients

Bureau reveals borrower quality to lenders with only minimal change in borrower knowledge .

Staggered fashion of the rollout of the bureau allows you to construct a difference-in-differences by comparing branches already in to those not yet in.

Impacts: Large increase in number of new clients selected

(screening in) and current clients dropped out (screening out).

Screening out of members of solidarity groups, leading to large drop in group size.

Change in composition of clientele, toward more women among individual borrowers.

Improved selection of new clients. Large increases in efficiency of credit officers.

23

Precipitous fall in size of Solidarity Groups:

24

A spike in the number of new individual loans:

25

Training of groups to inform borrower on how the bureau works (163 CBs and 207 SGs trained,~5,000 clients): Cost of bad reputation: General loss of access

to future loans Benefit from good public reputation: Access to

outside loans

Impacts: For SG borrowers, bureau is used intensively

and so is a real check on MH. These clients improve repayment performance inside and out, do not increase total borrowing.

Bureau rarely used to screen ongoing CB borrowers, but is used on new clients. Not much restraint on MH is imposed. Little improvement in inside performance, big increase in outside borrowing with differentiated effects.

Part 2. Randomized Experiment.

26

27

Timelines:

600

800

1000

1200

US

$

2000 2001 2002 2003 2004 2005Year

Timeline of Loan Sizes0

510

15P

erc

ent

2000 2001 2002 2003 2004Year

Loan never repaid Loan > 2 mos delinquent

Timeline of Default and Delinquency

28

Question of the game: Most microfinance bureaus share information about

group performance. This would appear to be inefficient (noisy signal of individual performance). Seems that moving to full individual bureaus would be a good idea.

But is it? Individualizing incentives can undermine the group, and

MF depends first and foremost on group incentives. Net effect could be negative.

Game is a new variant on ‘Public Goods’ game, tests for willingness to think long-term and maintain collective action. We vary rule that re-assigns players to game as follows:

No information game reassigns them randomly Group information game reassigns those from the best groups Individual information game reassigns those with the best individual

play.

Results confirm that AS effects of group rules are stronger, do not find expected MH effects of individual rules.

Upshot is that the expensive transfer to individual bureaus not worth it.

Part 3. Laboratory Experiment.

29

Larger loans, expansion of individual lending, decrease in the size of solidarity groups. The ‘individualization of microfinance’. Screening effect: almost 1/3d of the portfolio ejected the

first time a new loan is applied for after the implementation of the system.

Incentive effect: once clients are informed as to the working of the system, some decrease in delinquency and an expansion of applications to outside lenders.

Credit Expansion effect: large increase in loan sizes to good ongoing borrowers. This expansion in loan size leads indirectly to some increases in the default rate for these clients.

Overall default significantly lower with the use of the bureau.

But social differentiation at cost of weaker borrowers Women expelled from credit groups after information Inexperienced CB clients increase indebtedness and worsen repayment

performance. We see no evidence that shift to individual information is

warranted.

Overall Summary of Impacts:

30

Incentive problems for firms: In markets with too few dominant lenders, large incumbents

may perceive that they are giving away more than they are getting and be unwilling to join.

Everybody wants to check in a bureau, nobody wants to put their data into it.

Lenders see a huge benefit from the bureau initially as it allows them to ‘cleanse’ the portfolio, but ongoing benefits may be more limited.

There is a role for public actors in the formation of bureaus:

Pay for fixed costs of establishment of system Help to overcome agency problems in establishment of

bureau: ‘natural monopoly’ Guarantee clients’ privacy protection, ability to correct

mistakes, statute of limitations.

The Post-Mortem: Genesis withdraws from Crediref:

31

Should we be thinking about trying to create stable, competitive marketplaces, or should be be focusing on providing services to the poorest of the poor?

Who is MF really there to help; the entrepreneurial poor (who are likely to be more entrepreneurial than poor), or the needy?

Is credit a mechanism suited to the poor at all, or do they have needs that are more basic than finance?

The Ethical Arguments:

32

Discussion Break 2

Some new developments:

1. Micro-savings.

2. Microinsurance.

3. M-banking.

4. P2P MF: (Kiva, MicroPlace, Prosper).

34

Background: savings from the supply side:

Lenders worldwide were hit by sudden lack of access to lending capital during financial crisis, focus on currency mismatches in MF.

Recent CGAP paper on MF & the Financial Crisis says: “deposit-taking MFIs are well-insulated from refinancing risks” “MFIs accelerating the move to become licensed to mobilize deposits” “most deposit-taking MFIs mobilize larger deposits from nonpoor customers, and these

may be more sensitive to the economic downturn”

However, many MF lenders have only recently acquired ability to take savings, so mechanisms for increasing savings balances are poorly understood.

Our intervention compares three types of ‘nudges’ in Credito Hypotecario Nacional, Guatemala’s largest public-sector bank, intended to help micro-entrepreneurs to achieve a higher savings trajectory.

35

Background: why is it difficult to save?Behavioral finance has recently uncovered a number of decisionmaking

quirks that prevent people from meeting their own savings trajectories: ‘time inconsistency’: I fundamentally value consumption today differently than

consumption tomorrow (Laibson, 2009) Intrahousehold conflict: husbands and wives have different preferences, women

cannot control money inside the household (Ashraf et al., 2010). ‘temptation goods’: there are certain goods (luxuries, alcohol) that I value

today but do not value today whether I consume tomorrow (Banerjee & Mullainathan, 2009).

Any of these can lead to people being unable to stick to their own intended savings path.

In the developed world, many tools exist to overcome these problems (periodic paychecks, automatic deposit, retirement plans, life insurance, etc).

For micro-entrepreneurs, however, They operate mostly off of ‘cash in hand’. They lack any formal instruments to allow them to commit to save.

36

Savings as the new microfinancial service: Institutions want to build savings in order to decrease

currency risk, liquidity mismatches, lower cost of capital.

Donors want institutions to provide a broader suite of financial services (microfinance, not microcredit).

Borrowers want to save, even if the interest rate on borrowing is lower than the interest rate on savings!

This appears to be a new win-win in financial service development.

37

Our interventions:

Take advantage of the long loan cycles (12-36 months) within CHN and the regular repayment periods (1 month) to experiment with commercial commitment savings products.

Products based on recent literature showing the importance of Commitment in increasing savings amounts Default options in driving behavior

Three treatments:1. Basic: just savings promotion.2. Open: opportunity to commit any amount.3. Default: suggested savings of 10% of loan payment.

38

Description of the treatments: “Basic” Borrowers applying

for new loans are given a promotion on the value of savings, including a brochure.

At the time of disbursement of loans (2 wks later), they are asked if they want to open an account, all savings voluntary from there.

4% interest paid.

39

Description of the treatments: “Open”

Same promotion & brochure as the ‘Free’ treatment. In addition, borrowers are then permitted to create a

savings account linked to the loan account with a commitment savings amount stipulated by themselves.

Borrowers fill out paperwork to create savings account which reinforces commitment amount (written on loan agreement).

Commitment amounts are currency amounts and therefore very likely to be round figures (20, 50, 100 Quetzales).

Reinforced that these savings are NOT collateralized; savers have full access to accounts.

If account is created, borrowers are prompted by tellers at each loan payment to make committed savings amount, but no penalty for not doing so.

Delinquent borrowers not prompted to save.

40

Description of the treatments: “Default” Same promotion & brochure as the ‘Free’ treatment. In addition, borrowers are told that lender has new

policy under which savings accounts will be automatically created for new borrowers & 10% added on to loan payments to be deposited into these accounts, unless borrower opts out.

Commitment amounts in percentages, therefore rounding effects come from rounded loan sizes, not rounded savings.

Reinforced that these savings are NOT collateralized; savers have full access to accounts.

If account is created, borrowers are prompted by tellers at each loan payment to make committed savings amount, but no penalty for not doing so.

Delinquent borrowers not prompted to save.

41

Creation of new Savings accounts, by treatment:

42

Average savings balances, 12-month loans:

43

Conclusion: Tradeoffs between borrowers and banks:

Borrowers with savings appear to be better repayers. Savings deposits are not coming from loan

repayment. In the short term, then, savings products are improving the loan

portfolio. In the longer term, some evidence that savings

(particularly Default treatment) are allowing borrowers to escape debt-financed trajectories.

In the long term, use of this product may erode borrower base, while building savings in the institution and improving client profits.

Savings will likely enhance the stability of micro-finance lenders by providing domestic liquidity and improved repayment, but may also damage profits by allowing clients to finance investment from savings rather than debt.

44

Micro-Insurance: Insurance is the third financial service often

lacking for the poor. Risk is a major feature in the lives of the poor,

and the poor may be more sensitive to risk.

Basic problems in insurance: how can I insulate you from the unavoidable risks

that you face without creating moral hazard? How can I select the client base so that I don’t end

up facing adverse selection in my pool of beneficiaries?

45

Micro-Insurance: New solution to this problem: INDEX

INSURANCE: build an index off of weather, or aggregated area yields.

Need to pick an index that is closely correlated with farmer yields to minimize basis risk, but

Should also be uncorrelated with farmer decisions so I don’t create moral hazard.

46

Micro-Insurance: How far can villagers get with informal insurance?

Answer: they do very well for risks that are idiosyncratic; not well with correlated risks

47

My EPIICA Study in Ethiopia: Motivation.

Rain-fed agriculture exposes farmers to huge risks in the purchase of inputs: I pay for fertilizer today, will it rain tomorrow? Risk is a commonly given reason for low input use in

Ethiopian agriculture (Dercon and Christiaensen, 2009).

Most farmers need credit in order to be able to make the purchase of fertilizer + seeds in the leanest season. Research from Kenya indicating that many farmers indicate at

harvest time they would like to use fertilizer in the next season, but then don’t.

The large correlated risks from weather make agricultural lending extremely risky. Most developing countries have very thin rural credit

markets, rely on government subsidies and guarantees.

48

The interlocking puzzle of input use in

agriculture:Implication:The presence of large correlated risks prevent:

banks from lending to agriculture. farmers from using inputs.

Since the core source of correlated risk is weather, index insurance seems to provide a natural way to resolve this problem: Provision of insurance to lenders means that they can take on

the risk of lending to agriculture. Provision of insurance to farmers means that they can afford

to take on the risk of using and borrowing for inputs. Simultaneous provision of credit and insurance allows us to

create ‘state-contingent loans’: Receive inputs on credit, if the weather is bad you pay nothing

back, if the weather is good you pay loan + premium + interest on both.

49

Obstacles to Credit Provision on the Supply Side:

Banks in most developing countries very reluctant to lend to agriculture: Correlated shocks mean that even if average

default probability is low, portfolio risk from agriculture to lenders is huge.

Predominant source of correlated risks is weather, rainfall.

Pressure to forgive loans to farmers when default is caused by weather may be irresistible.

Consequence: private capital to ag very scarce even in countries where agriculture provides the best avenue for export-driven growth.

50

Obstacles to Insurance uptake on Demand Side:

Recent research: Demand for index insurance products is typically quite low,

even though they seem to solve a problem in a very natural way. Why?

Trust? Is a new institution credible when asking for money now in return for future promises of payouts?

Time inconsistency? Difficult to ask poor people to pay up front for a service whose benefits will not be realized immediately?

Credit constraints? The poor simply can’t afford the premia?

In addition, Duflo, Kremer, & Robinson (2010) show that:

Time inconsistency is a major problem in the demand for fertilizer: farmers understand that yields are higher with fertilizer, but

the time gap between costs and benefits makes purchase hard.

So, on the demand side as well, linking credit and insurance may overcome the behavioral problems that are barriers to the uptake of index insurance products.

51

The Interlinking solution: Provide loans to farmers that are

explicitly weather-contingent: Farmers take loans to purchase inputs, insurance premium is

added on to the loan amount and paid immediately to the insurer.

The beneficiary of the insurance policy is the bank itself, so if the weather index triggers the bank is paid with certainty (no intermediaries between bank and insurer).

The Cooperative Unions sit between the financial institutions and the borrowers and serve several critical roles:

First, they aggregate transactions and decrease the fixed costs of making loans.

Second, they are entities with the legal authority to contract with banks, much easier for formal financial institutions to deal with than smallholder farmers.

Third, they can use their extensive relationships with primary cooperative and farmers to serve as enforcers of the loan contracts, minimizing default risks.

Credit contracts written with Unions.

52

Our research partners: Nyala Insurance:

Provide rainfall & frost-based index insurance to farmers in Northern Shoa, North & South Wollo, and Gojam.

Insurance is intended to cover the inputs to production, not the output of the farm.

Dashen Bank: Will provide credit to farmers that will be backed up by

the Nyala product; serves as a form of collateral substitute in ag lending.

Contracting is done through Cooperative Unions, who recruit farmers through Kebele-level cooperatives. No loan contracts with farmers.

This means that Dashen can contract with only a few, financially sound and legally well-founded intermediaries, who in turn use their relationships with farmers to enforce contracts.

NISCO’s Agronomist discussing with farmers

We Care We Protect 53

54

The research design: Randomized controlled trial to provide simple,

statistically robust measures of impact. Two arm trial:

A control group receives no insurance and no credit. A ‘standalone’ arm receives only the index insurance

product; we don’t prevent the use of credit but we also don’t provide any explicit form of interlinking.

The ‘interlinked’ arm receives state-contingent loans. The study will then be conducted by comparing

each of the two treatment arms to the control, and to each other. Provides a simple, transparent measure of the impact of

insurance, the impact of interlinked insurance, and the impact of the interlinking itself.

Three years of household surveys to track farmer behavior.

55

The research design:

Credit users at baseline

Non-credit users at baseline

Survey experiment randomized at household level. For each Kebele:

6 coop households survey only 18 coop household surveys

6 coop households survey + insurance promotion 2 non-coop households

6 coop households survey + promotion + price voucher

2 non-coop households

Control (N=40)

Stand-alone Insurance (N=40)

T1a

120 Kebeles selected by Nyala

Random assignment

T1b

CbControl

T2b

CaControl

Interlinked Credit & Insurance (N=40)

T2a

Subsidy to price of insurance randomized at Kebele level

56

Longer-term question on supply side: Can the provision of index insurance crowd in

private sector credit to agricultural markets? Long history of government ‘amnesties’ on agricultural

loans when drought occurs. Historically, virtually all credit to ag has been provided

or backed by the government. Government is now interested in trying to have the

private sector take over more of this role, but a viable commercial model has yet to emerge.

The empirical strategy: Track over the course of time as index insurance is switched on in new parts of the country: Use institutional data from Dashen to track the spatial

coverage of agricultural lending to see the extent to which they expand credit in the places that the insurance will cover them.

57

Concluding Points: Financial service demand is interlinked:

In the rich world, we typically borrow and save at the same time, but the poor often do not have financial institutions that let them do this.

When individuals can be provided with financial tools that let them save to meet their needs this will be preferable to debt by the interest rate margin.

Current global microfinance movement has been too focused on debt and has not paid enough attention to savings.

Do MF borrowers really just use debt as a discipline device? If so, provide them with commitment savings instead.

Do MF borrowers really just use debt to insure themselves against shocks? If so, provide them with insurance instead.

Overall, need to be more serious in thinking about how to design products, institutions, and a regulatory environment that succeed in providing a suite of sustainable financial ser vices to the poor.