Microfinance at the margin: Experimental evidence from Bosnia í Herzegovina

Microfinance at the margin: Experimental evidence from Bosnia í Herzegovina Britta Augsburg (IFS), Ralph De Haas (EBRD), Heike Hamgart (EBRD) and Costas Meghir (Yale, UCL & IFS) London, 3ie seminar, 25 January 2012

Research Questions and Motivation Can microfinance be extended to a new/poorer client group and improve welfare? => Evidence on effectiveness of microfinance in general not conclusive and in particular not conclusive for very poor. Can microfinance be extended to a new/poorer client group and still be profitable? => Unresolved (political) discussion ofhigh interest rates, discussionof 'mission drift' of MFIs Can it achieve both aims together? => Microfinance crisis, which questions the whole existence of microfinance (for sure in India) - partly driven by questioned practices of profit- seeking MFIs

The aims of the project clients... A marginal client: they would consider extending loans if they were asked to take up (Note: no credit scoring system) (2)...and measure the effect of extending such a loan product. Characterise the implications of the programme for the household. Characterise the implications of the programme for the institution

Related Literature (1/3) Evidence on the poverty impact of Microfinance RELAXING LIQUIDITY CONSTRAINTS: Banerjee, Duflo, Glennerster & Kinnan (2010), Attanasio, Augsburg, DeHaas, Fitzsimons & Harmgart (2011), Karlan & Zinman (2010) POSITIVE effects (on consumption): Pitt &Khandker (1998), Kaboski & Townsend (2005), Attanasio et al(2011) NO IMPACT: Morduch (1998), Morduch & Roodman (2009), Attanasio et al (2011), Karlan & Zinman (2009) NEGATIVE: Cotler & Woodruff (2008)

Related Literature (2/3) Empirical literature on impact of MF in BiH RELAXING LIQUIDITY CONSTRAINTS: Hartarska & Nadolnyak (2007) presence of MF induces investment in business Demirgüc- Kunt, Klapper & Panos (2011) wealthier households more likely to be self- employed & MF helped to make switch from informal to formal business

Related Literature (3/3) Does finance matter for human capital investment? Most empirical and theoretical work focuses on the negative effects of market failures on HCI. Evidence of positive effects of credit access on HCI: Karlan & Zinman (2010) Peru higher school attendance Jacoby & Scoufias (1997) India, Beegle et al (2003) Tanzania credit allows to smooth consumption other than through child labour Bruce Wydick (1999) shows theoretically that credit access may affect schooling decision negatively or positively. Peru: positive dominates No or negative effects: Kind (2004) Philippines: no change in child labour, footwear industry (despite preference of parents) Menon (2005) Pakistan: credit may reduce likelihood of schooling

Outline remaining talk 1. The aim of the project and related literature 2. Background and Overview of the experiment The setting, The implementing institution, The experiment, The loan product The average marginal client 3. Treatment- Control balance & Attrition 4. The impact on the marginal clients and their household: A simple interpretative model Methodology Results 5. Commercial viability of the programme 6. Summary & Recommendations

Background & Overview of the experiment (1/5) The setting: Bosnia í Herzegovina pop: ~4.7m economy that still faces difficulties Affected by crisis -

Background & Overview of the experiment (2/5) The Implementing Institution: EKI (since 1996) (MF introduced in 90s) ~55,000 active borrowers (at the time of baseline), individual loans Total loan portfolio: ~150mUSD Double bottom line: aims at being financially sustainable while contributing to poverty reduction

Background & Overview of the experiment (3/5) The experiment: 1. Identify marginal client 2. Institution follows normal loan appraisal procedures with slightly relaxed criteria. If approved by loan committee 3. Loan officer informed client about experiment. If agreed to participate 4. Interview 5. Randomization (T: 637, C: 569) 6. Loan disbursement if allocated to treatment group

Background & Overview of the experiment (4/5) The loan product: Individual loan, ~1,500 USD (~12% of yearly hh income) 11 installments (regularly 9), interest rate: 22% Main use: purchase of livestock

The average marginal client: 37-38 years, Male (60%) Married (61%) Highest grade completed: Secondary (62%) (33% primary) Employed (57%), Works 47hrs a week 3.5 household members, 1.1 members employed Younger, less likely to be married, lower education Less full- time employed, more unemployed More household members, lower income

Treatment- Control balance & Attrition (1/1) Treatment control balance: Baseline sample (T: 637, C: 569) balanced Chi2 for joint significance: 39.68, p- value: 0.62 Attrition: Follow- up sample (T: 551, C: 443) 17% attrition Attrition reduced through incentives to participate Attrition significantly higher for control Constrain baseline sample to follow- up sample, repeat above analysis: sample remains balanced

Outline remaining talk 1. The aim of the project and related literature 2. Background and Overview of the experiment The setting, The implementing institution, The experiment, The loan product The average marginal client 3. Treatment- Control balance & Attrition 4. The impact on the marginal clients and their household 5. Commercial viability of the programme 6. Summary & Recommendations

A simple interpretative model We construct an economic model to understand the mechanisms behind our findings. The model gives one possible interpretation. In this model: 1. Households chose to set- up an enterprise or not, and to hire labour for this enterprise or not 2. Labour can be from outside or can be household members/teenagers The impact on the marginal clients and their households A simple interpretative model (1/4) 3. Outside labour is more expensive, but household members/teenagers forego going to school if they work on the enterprise 4. Households need a certain amount of capital to invest, which has to be invested upfront, i.e. before getting a return from the enterprise

A simple interpretative model The impact on the marginal clients and their households A simple interpretative model (2/4) 1. If the household is offered a microfinance loan and business opportunity is taken up, it is wealthier. 2. Consumption will grow because of relaxed liquidity constraints 3. Because of the threshold effect, it may be that c 1 declines to below the situation where the household was not able to borrow. This will happen for low income individuals (which might include business owners).

The impact on the marginal clients and their households A simple interpretative model (3/4) In the model, when moving from no investment to investment: 1. Consumption should increase for higher income households who had a business as the liquidity constraint is alleviated. 2. Consumption may decrease for lower income households (crowd in resources by reducing consumption and invest) 3. Savings may decrease to top up capital requirements 4. Youth labour may increase particularly for lower return/ability kids as they are withdrawn from school to work in the home business

Do borrowers make lumpy investments? The impact on the marginal clients and their households A simple interpretative model (4/4) EKI loan used for on average 1.32 purposes (std.dev. Of 0.59) 30% of loans used for one purpose only Typical combination: livestock and investment in auxiliary agricultural inputs

The impact on the marginal clients and their households Estimating Programme Effects - Methodology Main outcomes we consider: Business creation and development Consumption expenditures Savings Labour supply School attendance Estimation strategy - simple comparison of means: We allow for 2 types of heterogeneity: 1. By whether the household had a business at baseline or not 2. By education level (interpreted as proxy for income)

The impact on the marginal clients and their households Impact results (1/7) Impact results- Business creation & Loan use EKI Loans meant to finance investment in enterprises (working capital or fixed assets)

Impact results- Consumption The impact on the marginal clients and their households Impact results (2/7) Prediction: (1) Consumption should increase for those who had a business (2) Consumption may decrease for lower income households (with a start- up) (crowd in resources by reducing consumption and invest)

Impact results- Savings Prediction: (3) Savings may decrease to top up capital requirements The impact on the marginal clients and their households Impact results (3/7) Households not less or more likely to save in general Households reduce weekly/yearly contributions to savings (driven by business owners and highly educated)

Impact results Labour supply The impact on the marginal clients and their households Impact results (4/7) Prediction: (4) Youth labour may increase, particularly for lower return/ability kids

Impact results School attendance The impact on the marginal clients and their households Impact results (5/7) Are increased working hours accompanied by a decrease in school attendance?

Impact results by Gender The impact on the marginal clients and their households Impact results (6/7) Female marginal clients: Household more likely to get income from self- employment (8%) Marginal client more likely to own a business (8%) 7% more likely to own inventory Male marginal clients: Reduction in consumption (cigarettes & alcohol, non- durables) Reduction in savings Increase in working hours, teenagers 16-19 Decrease in likelihood of schooling, teenagers 16-19

Impact results (7/7)

Outline remaining talk 1. The aim of the project and related literature 2. Background and Overview of the experiment The setting, The implementing institution, The experiment, The loan product, The average marginal client 3. Treatment- Control balance & Attrition 4. The impact on the marginal clients and their household: A simple interpretative model Methodology Results 5. Commercial viability of the programme 6. Summary & Recommendations

Commercial Viability of the programme (1/5) Analysis of profitability/commercial viability Information from EKIs Management Information System on all regular and marginal clients over the time of the experiment Compare two groups of loans: marginal clients and first- time regular borrowers Look at: 1. Repayment performance 2. NPV analysis rates of return

Commercial Viability of the programme (2/5) Repayment performance Marginal clients: significantly more likely to be late at least once in repayment (38 versus 19%) Significantly more likely to be written off (19 versus 3%)

Commercial Viability of the programme (3/5) NPV - Rates of returns Calculation: Sum up all discounted outgoing (loan disbursements) and incoming (repayments, fee income, interest revenue) cash flows Discount rate: EKIs weighted (by size of individual outstanding liabilities) average cost of debt funding in March 2011. 3 rates are used: 1. (weighted cost) of EKIs commercial funding 2. (weighted cost) of EKIs concessional funding 3. (weighted cost) of all of EKIs funding Rates of return: Divide these NPVs by total amount of loans granted under the programme IRR: the discount rate at which the NPV of the sum of all cash flows equals zero

Commercial Viability of the programme (4/5) NPV - Rates of returns

Commercial Viability of the programme (5/5) NPV - Rates of returns Programme was not commercially viable We calculate that EKI paid an implicit subsidy to the average marginal borrower of BAM 194 (USD 135)

Outline remaining talk 1. The aim of the project and related literature 2. Background and Overview of the experiment The setting, The implementing institution, The experiment, The loan product, The average marginal client 3. Treatment- Control balance & Attrition 4. The impact on the marginal clients and their household: A simple interpretative model Methodology Results 5. Commercial viability of the programme 6. Summary and Recommendations

Summary Summary and Recommendations We find that the marginal/poorer clients do invest in business Certain type of clients reduce their consumption and savings => Is this an effective strategy that will lead to higher household income in the long- run? Certain households take their children age 16-19 out of school to work on the business. => Are children taken out of school who would have higher return from staying in school than working on the business? These loans are riskier for the institution and not commercially viable => What are the longer run welfare effects when a subsidy is involved? Are incentives being distorted?

Recommendations Summary and Recommendations Recommendations (1/3) Current screening process of EKI works well Extending loan product to marginal clients as done during the experiment currently NOT advisable: (Highly) loss- making Potentially negative for the clients Nevertheless, financing demand by marginal clients is there and we found evidence that they do invest in businesses => If EKI does not want to exclude this potential niche, what are the options (if any)?

Summary and Recommendations Recommendations (2/3) Possible actions for the implementing institution: Option 1: Tighter screening of marginal clients? Certain characteristics correlate with late repayment/default: Purpose of loan Age of client Business ownership... Option 4: Changing loan conditions? Lower interest rates (attract poor and female), higher interest rates, increased maturity (higher demand)... Results are inconclusive and difficult to extrapolate to our setting. Questionable whether they would be effective.

Summary and Recommendations Recommendations (3/3) Possible actions for the implementing institution: Option 3: Offering additional services Generally, low levels of financial literacy observed, which lead to poor financial decisions Attendance usually problematic, but some evidence that especially simple trainings are beneficial It might be worth considering making financial literacy training mandatory for marginal clients Option 4: Additional analysis Second follow- up planned, to concentrate on long- term impacts and include a module focussing on teenagers

Thank You!

Interview response baseline and follow- up:

A simple interpretative model Consider a two period model. In the first period production is set up with hiring of labour and capital. Income from production accrues in the second period. Suppose the production function for a small business is (L+ 1 ) 1-1 is youth labour. 1 =0 K L 1

Hiring inside labour is the cost of foregone returns to education over the lifecycle Hiring outside labour is more expensive by (=wedge between the cost of hiring internal and external labour (i.e. tax rates and other regulatory costs) Internal youth labour will be used if r <1+ (return to schooling<labour cost) r= interest rate hence, low- return to education kids may be withdrawn from school Liquidity constraints are exacerbated by the fact that both, capital and external labour have to be financed before production by currently available resources

The first period budget constraint is Y + B c 1 K (1+ )wl=0 Y= Income, constant in each period B= borrowing (possibly with ceiling) c 1 = consumption in period 1 If the household does not invest, then c 1 =c 2 =Y

If the household is offered a microfinance loan and business opportunity is taken up, it is wealthier. Consumption will grow because of liquidity constraints, so that c 2 = (1 + G) {- 1/p} c 1 where G depends on the wage, the returns to education, preference and technology parameters. Because of the threshold effect, it may be that c 1 declines to below non- - include business owners).

Moving from no investment to investment 1. Consumption may decrease for lower income households with a start- up (crowd in resources by reducing consumption and invest) 2. Consumption should increase for those who had a business as the liquidity constraint is alleviated. 3. Savings may decrease to top up capital requirements 4. Youth labour may increase particularly for lower return/ability kids as they are withdrawn from school to work in the home business

Descriptive Statistics

Descriptive Stats - Business

Descriptive Stats - Consumption

Descriptive Stats - Savings

Descriptive Stats Labor supply

Descriptive Stats School attendance

Coef StdErr P> t female - 0.023 0.046 0.612 age 0.004 0.002 0.046 Not married 0.031 0.072 0.663 Married - 0.056 0.056 0.326 Employed 0.009 0.083 0.913 Unemployed - 0.045 0.078 0.565 Student - 0.404 0.155 0.009 primary school 0.027 0.093 0.770 secondary school 0.029 0.089 0.745 currently attending school 0.179 0.110 0.103 hrs worked per week (resp) - 0.003 0.001 0.025 hrs worked on business (resp) 0.001 0.001 0.438 # household members 0.017 0.033 0.613 # children 0-5 - 0.007 0.046 0.886 # children 6-10 0.010 0.052 0.853 # children 11-16 0.012 0.049 0.804 # elders (>64) - 0.086 0.048 0.075 hh mem attending school 0.003 0.041 0.946 hh mem employed - 0.036 0.051 0.475 hh mem employed (female) 0.020 0.046 0.674 hh mem unemployed - 0.011 0.037 0.765 hh mem retired 0.028 0.046 0.542 hrs worked by hh 0.001 0.001 0.250 hrs worked by hh (on business) 0.000 0.001 0.845 constant 0.464 0.183 0.012