Microfinance: Group Lending versus Individual contracts

Microfinance: Group Lending versus Individual contracts

Master thesis Hendrien Damman1

June 2007

Abstract:

This paper examines what causes microfinance institutions to decide to subscribe either individual or group contracts to their clients; 177 microfinance institutions are included that subscribe either group or individual contracts. Microfinance institutions that have higher loans, are older, have a high financial revenue ratio and are located in South or East Asia, the Pacific, Latin America or the Caribbean are more likely to subscribe individual contracts. Financial revenue has the strongest impact on the probability that banks provide individual contracts. Microfinance institutions with a high profit margin, a large part of their portfolio at risk, a high percentage of female borrowers and if grants are part of the main funding are more likely to subscribe individual contracts. The profit margin, the percentage female borrowers, the portfolio at risk and the indicator whether grants are part of the main funding have the strongest impact on the probability that banks provide group contracts.

JEL- Classification: D82, G21, O16

Supervisor: Prof. dr. E. Sterken Rijksuniversiteit Groningen

1 _{Hendrien Damman (s1322257), Boermandestraat 12, 9723 DS Groningen, Phone: 050-5733777, email:}

1. Introduction

Microfinance is in general small finance for poor people. They are not able to get a loan in the regular loans market. The costs of screening and monitoring are too high compared to the loan size. This is where microfinance institutions come in. Traditionally, they grant a loan to groups of very poor people. Group contracts reduce the monitoring and screening costs. Usually, these loans are used to set up their own business. Microfinance could thus be an effective way to solve poverty in less developed countries.

Group lending has been developed many years ago. In the nineteenth century there were people in rural Ireland that undertook group loans. Joint liability contracts got attention by the success of the Grameen Bank in Bangladesh. In 2006, the Nobel Peace Prize was awarded to the Grameen Bank and its founder, Muhammad Yunus, for their joint efforts to create economic and social development for the poor. Recently, large microfinance bank shift their attention to individual loans.

2. Individual versus Group Contracts

There are numerous explanations why group lending systems perform better than individual loan contracts. The theoretical explanations for the high repayment probability range from peer pressure to mutual insurance and the dynamic structure of joint liability contracts. However, group finance has its limitations. In this section I give a summary of the literature that explains the success of group lending systems compared to individual contracts and what the limitations of group lending systems are compared to individual contracts. I start with the differences in repayment probabilities followed by the disadvantages of group lending systems. Then, I discuss the relation between microfinance and poverty followed by the effect of increased competition in the market of microfinance. Finally, I discuss some empirical evidence in the field of microfinance.

2.1 Repayment rates

Stiglitz (1990) presents a model wherein borrowers can either choose to invest in a risky project that has a return of Yr

( )

L , where L is the loan size, or in a safe project with a

return ofYs

( )

L . If the project fails, the investor has a return of zero. The probability of

success for the risky project isp and for the safe project_r p , with_s p_s > p_r. The returns for both projects are given by an increasing function of scale. The fixed costs associated with the projects, L , are larger for the risky project, than for the safe project, so Lr >Ls,

further ' ' s

r Y

Y > ; the marginal return of the risky project is higher than for the safe project.

However, the safe project has a higher return than the risky project, when one takes the probability of success into account:

(1)

where r is the rate of interest on the loan. So the safe project has a higher return when taking the probability of success into account, but the return when successful is lower. An individual who invests his own funds would thus always choose the safe project. Expected utility Vifrom investing in project i is:

( )

L p

(

r

)

L Y L p

(

r

)

L

( )

L,r U

[

Y

( ) (

L 1 r

)

L

]

p v(e(L))

Vi = i − + i − (2)

where i is either s or r, U(Y)is the utility of income, U'>0,U ''<0and U

( )

0 =0. The last term, v

( )

e

( )

L is the disutility of effort, e ; v'>0, v''>0. Effort is included in the model so that the utility from investing in project does not increase always as the loan size increases. After a certain point, the disutility of effort becomes larger than the utility of investing. So there is an optimal loan size and thus project scale2_{, the relation between} the loan size and the return of the project is also shown in figure 1. The level of effort increases as the project size increases; e'

( )

L >0.

The ‘switch line’ is defined as those combinations of (L, r) where the individual is indifferent between the two projects, thus if the following condition holds:Vs

( )

L,r =Vr

( )

L,r . Ignoring the level of effort, the expected utility from the safe

and risky project are respectively: Vs(r,L)=U

[

Ys(L)−

(

1+r

)

L

]

ps and

(

)

[

r

]

r

r r L UY L r L p

V ( , )= ( )− 1+ . There is assumed that the returns to scale are higher for the risky project than for the safe project. An increase in L keeping r fixed makes the risky project more attractive. Thus in the relevant region

(

L>L_r

)

:

s s s r r r Y r p U Y r p U'( '₋(1₊ )) _> '( '₋(1₊ )) _. This can also be written as:

L V L Vr s ∂ ∂ > ∂ ∂ ₍₃₎

The total differential of the switch line is:

(4)

2 _{Figure 1 shows the relation between the loan size and the utility of a project. In the further analysis, the}

(4) is smaller that zero since according to (3) L V L Vr s ∂ ∂ > ∂ ∂ _{and since} s r p p < (denoted above). Therefore, the switch line has a negative slope. This is shown in figure 1.

Figure 1: Influence of Loan Size and interest rate in Selection of Safe and Risky Projects

At the market equilibrium the borrower is credit constrained. The lender knows that if he subscribes a larger loan amount the borrower has the incentive to invest in the risky project. The expected loss is limited for the borrower since he does not invest with his own funds, he will therefore choose the risky project that has a higher outcome if successful. The lender cannot monitor the borrower and not influence the project chosen. This is the moral hazard problem; the behaviour of the lender is influenced by the terms of the loan contract.

and his co-signer’s project succeeds; (b) if his project succeeds but that of his co-signer fails and (c) if his own project fails. The utility in the three different states is respectively:

[

]

[

]

( )

0 0 ) 1 ( ) ( ) 1 ( ) ( = − + − ≡ + − ≡ U qL L r L Y U U L r L Y U U i iq i i

Expected utility in a symmetric equilibrium, where both the borrower and his co-signer decide to choose r or s, is:

) 1 ( 2 i i iq i i i U p U p p V = + − (5)

This means that the switch line also changes. The switch line under peer monitoring is given by3:

( ) (

)

[

Ys L r L

]

ps U

[

Ys L

(

r

)

L qL

] (

ps ps

)

U − 1+ 2 + ( )− 1+ − 1−

( ) (

)

[

Yr L r L

]

pr U

[

Yr

( ) (

L r

)

L qL

] (

pr pr

)

U − + + − + − − = 1 2 1 1 ₍₆₎

The first term represents the expected pay-off if both borrowers succeed; the second term represents the pay-off if one borrower succeeds and the other fails. In figure 1 the switch line is drawn. At the line, the investor is indifferent between the two projects. Above the line, it chooses the risky projects and under the line, it chooses the safe projects. Above the switch line, the risky project has the highest return, below the line the safe project. The group contract causes the switch line move to the right (under the condition that q is equal to zero). This means that the area where the risky project is chosen decreases.

Peer monitoring shifts up the ‘switch line’. The constraint on (L, r) is relaxed, so people are less constrained in their loan size in order to ensure them to undertake the safe project. Further, the shift of the ‘switch line’ is larger than needed to keep the borrower at the

3 _{This expression is slightly different from Stiglitz (1990), pag. 361, however there is a typo in the article, it}

same utility level; the increase in L that is needed to pursue people to take on the extra risk is larger than required for the compensation. In sum, group contracts reduce moral hazard; lenders have the incentive to pick the risky project when they invest with borrowed money.

Banjerlee et al. (1994) also focus on moral hazard that is present under individual contracts and the role monitoring by other agents plays in solving this problem. The difference with the previous model is that only one member of the group has the possibility to invest and monitoring costs are included in the model.

The authors base their model on the German cooperatives in the nineteenth century. Each cooperative consists of two members and each member has two assets; land and money. At the start of the period, only one of the coop members has the ability to make his land more productive. This requires an investment ofK+ ; each coop k

member is endowed with a positive level of wealth, k. Moreover, K > k; so the amount of money available in the cooperation, (i.e. the sum of money of both coop members), is insufficient to finance the investment. The member that has no opportunity to invest receives a deterministic return of θ on his land.

The coop borrows b from outside to undertake the investment project. The interest rate paid on outside funds is denoted by R and on inside funds by r . This inside fund does not have a safe return, since the borrower can default. The non-borrowing member of the cooperation serves three different functions. In the first place, he is a lender, second, he might stand liable for the borrower, this amount is equal to l, withl≤bR, third, he could monitor the borrower. When the borrower receives his funds, the other member of the coop chooses the level of monitoring to influence the borrower’s project choice. If the borrower receives sufficient funds from the project, the borrower repays the monitor and the outside lender, if he defaults the monitor has to pay out l to the outside lender.

ρ and a net return ofρ− whereδ δ can be either positive of negative4. The return on the inside opportunity has to be higher than the outside opportunity, since the non-borrower has to be compensated for the fact that the borrower might default. So r has to be as high as the outside opportunity compensated for the risk of default.

The choice of the project depends on

(

b,l,r

)

. The borrower chooses the investment project, but the other member of the coop can monitor and influence the decision. Projects are categorized by a success probabilityπ ∈

[ ]

π ,1 . A project has a return with probability π and a zero return if the project fails. The expected return of a project isE

( )

π ≡πφ

( )

π , where φ

( )

π the return of the projects is. Further there is assumed that E'

( )

π >0andφ'

( )

π <0. This means that projects with a higher expected return are safer although the return when successful is lower for the safe project.

In a competitive credit market the expected return on the inside and the outside funds is equal, this means that the following condition must hold for the lender:

b l

Rb π ρ

π +(1− ) = (7)

With probabilityπ the loan is repaid, and with probability (1−π)the lender receives l from the non-borrowing member. The costs of funds is bρ , that is the amount that non-borrowing members receive if they decide to invest in the outside fund. Denote r as the total interest rate to be paid on the outside and inside funds, r is defined as:

r b K bR

r ≡ +( − ) (8)

Solve (7) for R this gives the interest rate charged by the lender:

b l

R = ρ /π − ((1− π ) )/π . Replace this equation in (8) and one obtains the total interest payment on any project:

π π π ρ (1 ) ( ) )/ ( b l K b r r = − − + − (9)

Once the borrower has obtained the funds, there is an incentive to invest in the most risky project. This project has the lowest probability of a return, π, but the return that it creates is the highest. This behaviour is inconsistent with the interest rate charged by the lender; the rate has to be higher compared to the risk of the loan.

This is where the monitor comes in, he can impose a penalty c on the borrower if he chooses the project with probabilityπ. This penalty must be high enough so that choosing the more safe project πand not paying the penalty is preferred to choosing π

and paying c . So the following incentive compatibility constraint must hold:

( )

(

φ π −r

)

≥π

(

φ

( )

π −r

)

−c

π (10)

So monitoring and the availability of punishments make the investor to choose the safer project. Peer monitoring solves the moral hazard problem and will result in higher repayment rates than under individual contracts.

( )

r

r ≤ φ . (11)

Groups consists of two borrowers, who are ex-ante equal, referred to as borrower 1 and 2. The group is granted a loan of two units of capital, one for each borrower. Both borrower 1 and 2 invest in a different project, which returns are independent. The loan and interest must be repaid at the end of the period; the total amount to be repaid is thus 2r. The repayment decision is an all or nothing decision. The bank will impose penalties on the borrowers if they do not repay. If the whole group defaults while the two borrowers receive respectively, θ1 andθ2, the bank will impose penalties p

( )

θ1 andp

( )

θ2 .

After the returns of the projects are realized, borrowers have the following strategies to follow. First, borrower 1 and 2 decide simultaneously whether to contribute the individual share of the total repayment, r. If both borrowers decide to pay their share, the loan is repaid and the pay-offs are for borrower 1 and 2 respectively, θ1−r

andθ2 −r. If both borrowers decide not to repay and the bank imposes its penalties the

pay-offs are: θ₁− p

( )

θ₁ andθ₂ −p

( )

θ₂ . If one borrower decides to repay, while the other does not, it has to decide at the second stage to repay (which has pay-offs:θ1−2r,θ2) or

default (pay-offs: θ1−p

( )

θ1 , θ2 − p

( )

θ2 ).

Besley and Coate (1995) show that the loan is repaid if at least one of the borrowers receives a return in excess of φ

( )

2r . It might be repaid if both borrowers have a return between φ

( )

r andφ

( )

2r . The loan is not repaid if both borrowers have a return lower thanφ

( )

r . When one compares the group lending system with the individual system, the main drawback is that the group lending system has a higher repayment when interest rates are low. If the interest rates increase, the individual lending scheme has higher repayment rates.

reputation in society, etc. These penalties are increasing in the harm caused upon the group member. Further, they depend on the reasonableness of the decision not to support. The social penalty function is denoted by s .

( )

⋅

The difference compared with the result without social sanctions is when one borrower receives a return that lies between φ

( )

r and φ

( )

2r while the other receives a return that is less than φ

( )

r . Suppose that borrower 1 has a return between φ

( )

r and

( )

2r

φ . Borrower 2 has a return less than φ

( )

r . Borrower 2 expects that borrower 1 contributes his share for the repayment of the loan. Borrower 2 must now compare the following pay-offs: θ2 −r and θ2−p

( )

θ2 −s

(

p

( )

θ1 −r,θ2

)

. Borrower 2 will pay his share

if r<

( ( )

pθ₂ +s

(

p

( )

θ₁ −r,θ₂

) )

; if the repayment is smaller than the sum of social and bank penalties. So, if the social penalties are severe enough, the group lending system will result in higher repayment rates than under individual contracts.

Ghosh and Ray (1994) state that there not necessarily have to be social ties between group members to benefit from efficient cooperation. They focus on the costs that have to be made to form a new borrowing group. These costs are very high, so existing borrowing groups are not likely to break. Ghosh and Ray (1994) study social interactions in a society where information about past actions of people is absent, or very limited. People have, however, the option to form a group with the same partner in subsequent periods. If one finds a reliable partner, that does not default strategically or takes on too risky projects, one will continue to form a group with this partner. The costs of breaking the relationship become very high. One loses the gain from working together in the future, one has to find a new reliable partner and build a new relationship. So, even if people do not have information about each other, there can still be efficient cooperation.

The result that lenders with similar risk types group together is also mentioned by Ghatak (1999). Every group is formed of two people. There are two types of potential investors, one risky and one safe. The risky investors have a probability of success of p and a net r

return of R . The safe investor has a probability of success of r p and net returns R . If the s

s

r R

R > and prRr = psRs ≡R. This means that the projects have the same expected

return. Suppose that a group contract is formed that has the following terms, the borrower pays a fixed r if the project is successful. It has to pay c if the other group member defaults. The expected return of a safe type grouped with a risky type

(

ER is: _sr

)

(12) (13)

The expected return of a safe type grouped with a safe type

(

ER is: ss

)

(14) (15)

and so ERss <ERsr since pr < ps.

In order to make a safe type willing to form a group with a risky type, he must at least be compensated for the lower return he is expecting. When comparing (13) and (15) one can see that this amount must be equal to or greater than: ps

(

ps − pr

)

c. The amount that the

risky type is willing to pay is equal to the extra revenue it gets by forming a group with the more risky type. The expected revenue from the investment project for the risky type when he forms a group with a risky type

(

ER is: rr

)

(16)

If the forms a group with a safe type the expected return is:

(17)

When comparing (16) and (17) it is clear that the risky type would not prefer to pay more than his expected gain in revenue to form a group with a safe type. This gain is equal to:

(

p p

)

c

pr s − r . Since pr < ps, the maximum amount that the borrower is willing to pay

to form a group with a safe type is smaller than the amount a safe type would like to have to form a group with a risky type. This means that a safe type and a risky type will never form a group together; people would only like to form groups with people that have the same risk characteristics. For a microfinance institution (MFI) this means they have only to screen one of the potential borrowers to establish the type of the group. Further, the group forming solves the adverse selection problem. Suppose that the bank offers two different kinds of contracts, one with low joint liability and high interest rates and one with high joint liability and low interest rates. The safe borrowers will choose the contract with high joint liability and the risky borrowers will choose the contract with the low joint liability. Compared with the situation under individual contracts, repayment rates and efficiency are higher under the joint liability contract, because the joint liability contract exploits the information that the borrowers have about each other and the bank does not have.

extend any loans further, the penalty of not repaying has disappeared. The borrower decides to repay if:

δπ π

π δ

π + v ≤ − R + (18)

This means that the bank should make the borrower’s pay-off equal in both situations so she has no the incentive to default. If the bank sets v=0the maximum interest rate it can charge is: R=δπ, otherwise the borrower will always default in the first period. The borrower will enter in an arrangement with the bank if:

(

π −R+δπ

)

≥0

p (19)

The optimal solution for the bank is set to fully carry out the threat of non-refinancing in the second period if the borrower defaults and setR=δπ . In this case the bank deals with the fear of the ex post moral hazard problem and maximizes income.

One other feature of group loans that result in higher repayment rates in the insurance the group members provide to each other. According to Varian (1990) group members have the incentive to insure each other in different states of nature. If the agents insure each other in states that are not observable by the principal this will make the principal better of compared to no insurance. In the group lending contract the group members also insure each other. Members will stand by in times of need. This feature of group lending contracts is not present under individual contracts. This makes that the group lending system is likely to have higher repayment rates; insurance is not present under standard individual debt contracts.

Second, the bank implicitly requires that the borrower has other sources of income against which the bank is effectively lending to. Finally, these repayment schedules monitor undisciplined borrowers; they give an early warning about emerging problems. An example of a collateral substitute is the contribution to an emergency fund by every group member that provides insurance in cases of default.

According to the literature there are several factors that explain high repayment rates of group contracts compared to individual contracts, for example peer pressure and peer monitoring between different lending groups. This does not only hold group members from undertaking more risky investment projects, there is also less change of ex-ante moral hazard. Further, groups are formed of people that have the same characteristics. This results in a decrease of monitoring costs for the bank. Finally, the insurance incentive that is present in a borrowing group and the dynamic character of group contracts all result in higher repayment rates compared with individual contracts. There are also a couple of limitations to group lending systems; I will discuss them in the following section.

2.2 Disadvantages of Group Lending Systems

Group contracts transfers tasks from lender to their customers. They are responsible for screening potential borrowers and monitoring the outcomes of their investment projects. In return, customers get loans that would otherwise not be available, or only with a very high interest rate. Not in every situation are group loans the best solution. Attending group meeting and monitoring group members can be very costly if group members are not living very close to each other. (Armendáriz de Aghion and Morduch, 2005)

Two other limitations that are present in group finance are the possibility of colluding against the bank and that lending is too costly to implement. Besley and Coate (1995) describe the first drawback. In their model, one of the equilibrium solutions is that groups decide to collude against the bank. Once they have obtained their loan they default together. The second drawback is that most microfinance institutions are not profitable (Morduch, 1999). They rely on outside funds or subsidies too fully cover their costs. This suggests that micro credit is not an efficient way of financing; the lending is too costly to implement.

2.3 Costs of monitoring and screening potential borrowers

Group contracts and individual contracts do not only differ in their repayment rates, the monitoring costs between the two contracts are also different. Conning (1999) compares the differences in costs of monitoring performed by peer groups or monitoring performed by delegated staff monitors of a microfinance institution (MFI). There is a population of risk neutral micro-entrepreneurs. Each of them needs inputs for an investment project. Each entrepreneur is characterized by its skills z, an amount of a tradable factor I, and a chosen level of effort in the project. Further, each entrepreneur has an amount of cash K and assets that can be used as collateral A. The vector v=(z,K,A) summarizes the characteristics of the entrepreneur.

An entrepreneur borrows I – K to invest in his project. If the borrower puts al his effort in the project, he would use the loan together with his own cash K to carry out the project at scale I. The project has then the probability of success πand the expected project return is πzf

( )

I where f(I) is a twice differentiable and concave production function.

twice differentiable; B’(c)<0 and B’’(c)>0. The higher the amount spend on monitoring, the fewer the rewards for the borrower from diverting resources for private uses.

The optimal contract must determine the optimal investment scale; I_v =I

(

z,K,A

)

and a rule for dividing the project returns, x , (where i = s, f) between repayments Ri i to

the MFI, payments wi to a delegated monitor and si are residual returns for the

entrepreneur; si = xi – Ri – wi.

The following conditions must hold to make the contract feasible:

( )

R |

(

I K

)

c0 E _i π ≥γ − + (20)

( )

s E

(

s

) ( )(

Bc I K

)

E i |π ≥ i |π + − (21)

( )

wi |π c E

(

wi |π

)

E − ≥ (22) A x w Ri + i ≤ i + (23)

c0 are the fixed costs of handling loans, c are the monitoring costs.

Inequality (20) gives the break-even condition of the bank, where is equal to one plus the interest rate on bank deposits, (21) is the incentive compatibility constraint of the borrower, (22) is the delegated monitor’s incentive compatibility constraint and (23) is the limited liability constraint for the borrower.

A borrower that receives a loan of the size (I-K) will only choose the less risky investment project if the incentive compatibility constraint in (21) holds. Given that

where ∆π =

( )

π −π . Expression (24) makes it clear that the borrower must be rewarded enough for successful outcomes than for failures to be sure that the borrower acts diligent. There is assumed that the borrower has a zero return in the case of failure, if that occurs the lender receives the value of the collateral, A. This is all the borrower loses if a bad outcome occurs. To ensure that the borrower puts all his effort in the project, successful outcomes need to be rewarded, a large share of the outcomes of a project need to go to the borrower, especially for the poor borrower, to pursue him to make the right investment choice. In sum, maximum repayment in the failure state is Rf = and A

A

sf =− . Substitute this conditions in (24) one can rewrite the condition;

A K I c B ss = _∆ ( − )− ) (

π . For a given loan size (I-K) a borrower must earn at least, to be

diligent, the following return:

( )

s B c I K A E i = _∆ ( − )− ) ( | π π π (25)

A loan will be only made if the net project returns are higher than the borrowing enforcement costs (25), the monitoring costs (c) and the fixed costs for loan handling, c . ₀

So if the following condition holds:

( )(

)

0 ) (I I Bc I K A K c c zf − − − + + ∆ ≥ − γ π π γ π (26) K

γ are the opportunity costs of the borrowers’ cash.

( )

f s

(

)

f

s w c w w

w π π π

π +1− − ≥ + 1−

This can be arranged to:

π

∆ ≥ −w c

w_{s f} (27)

In order to carry out the optimal level of monitoring, the monitor must expect to earn as least as much from monitoring as from not monitoring. The monitor expected minimum pay for monitoring each borrower is:

π π

∆

c ₍₂₈₎

With delegated monitoring the lender only makes a loan if the net projects returns are higher than the enforcement rent of the borrower (25), the delegation enforcement rent (28) plus the fixed costs, so only if the following condition holds:

( )

( )(

)

c0 c K A K I c B I I zf + ∆ + − − − ∆ ≥ − π π γ π π γ π (29)

Most MFI’s use peer monitoring in order to reduce the high monitoring costs compared to the size of the loan. In the model of Conning (1999) there are symmetric groups that consist of two members. First, a contract is chosen and then the borrowers decide on the optimal level of monitoring. The optimal way to provide the right effort in their projects is to make the borrowers jointly liable and reward them heavily if both projects succeed and make than maximum liable if one of the projects fail. Reward the outcome that both borrowers are successful both improves the incentive for borrowers to choose the right project and monitor the other borrower. So the following condition must hold:

(

)

A c s

(

)

A B c I K c sss −1−ππ − ≥ππ ss −1−ππ + ( )( − )− π

This condition states that each borrower needs to earn as least as much from being diligent given that the other borrower behaves diligent than behaving non-diligent. Further, each borrower is monitoring each other at the optimal level c. Solving (30) for

ss

s gives: sss =−A+B(c)(I−K)/π∆π. The enforcement rent turns out to be exactly the

same as under the individual contract given in equation (25). The lender will only subscribe a loan to a group of two people if the following condition holds:

( )(

)

0 ) (I I Bc I K A K c zf − − − + ∆ ≥ − γ π π γ π (31)

The only thing where this condition differs from the individual lending contract given in equation (26) are the fixed costs of monitoring. The monitoring costs are not longer born by the MFI, but the group takes over these costs. So the monitoring costs are the smallest for the peer monitored lending; compare equation (26) and (29) with (31). Further, the enforcement rent plus the monitoring costs are also the smallest for the peer group lending system. Joint liability contracts have the smallest costs for banks. One drawback of the contracts is as the loan amount increases the probability of collusion among lenders increase. Borrowers thus remain credit constraint if only joint liability contracts are offered.

2.4 Microfinance and Poverty

Morduch (1998) investigates nearly 1800 households in Bangladesh, to test the impacts of microfinance on solving poverty. There are three different microfinance programs active in the area under consideration. Almost all households have only access to one of the three programs. The survey also includes a control group that has no access to microfinance. Morduch (1998) finds that the consumption and labour supply variability is much lower for the group that uses microfinance. Households that use the microfinance programs have the ability to smooth consumption over time. However, there is no significant difference between the level of consumption between the households that use the microfinance program and the control group. Further, there is no difference in the school attainment rate between the microfinance group and the control group.

A complete different result is obtained by Khandker (2005) and Mosley (2001). Khandker (2005) also investigates data from Bangladesh, obtained from two household surveys at two moments of time, in 1991-1992 and 1998-1999. The survey shows that loans that are obtained by females have a significant effect on the household per capita consumption. By obtaining a loan women can create their own income and have a positive influence on the assets of a household. This means that they can also decide on the consumption plans of a family. Further, there are also spill over effects present to the rest of the village from microfinance. Not only have the residents that receive a loan benefited, but also the other villagers. Khandker (2005) also estimates the contribution of microfinance in reducing poverty in Bangladesh. In the investigated period, poverty has declined 3 percent per annum. Microfinance reduces poverty for participants in the project for people that are living in moderate poverty by 1.6 percent and for people that were living in extreme poverty by 2.2 percent. Mosley (2001) investigates the relation between microfinance and poverty in Bolivia. He concludes that microfinance can solve poverty, but does not solve extreme poverty. People are solved from poverty in both having more income and more assets.

on the decisions concerning the household budget. Directing resources to women will thus have a stronger impact on development. Further, women make more prudent investment decisions and are less likely to misuse the loan.

2.5 Microfinance and Competition

Increased competition has two different effects in the market for microfinance. On the one hand, people that were excluded from loans might now be able to get loan. So the outreach of microfinance is likely to increase. On the other hand, the threat of no more access to future credit disappears if one can get a loan from another bank. This problem can be solved if bank share information about ‘bad risks’. What also can be a problem is that persons take on loans from different banks at the same time; this imposes them to too much risk.

Morduch (1999) states that increased competition has caused for BankoSol (Bolivia) a larger part of the portfolio to be at risk. At the end of 1997 there was 2.03 percent of the portfolio at risk; at the end of 1998 this was increased to 4.89 percent. The success of BankoSol causes other (non-bank) institutions and formal banks to open for poor clients. This caused not only a rapid increase in the credit supply, the repayment rates declined.

McIntosh et al. (2005) give the consequences from increased competition in the market for microfinance in Uganda. That market started will few institutions offering microfinance, all institutions had local monopolies. They decided not to compete on each other’s markets. Attracted by the success of the microfinance market, other players flooded the market. The main hypothesis of the article is if this increased competition had an influence on the repayment rate and the dropout rate among borrowers. It turns out that this depends crucially on the level of information sharing between institutions that provide microfinance.

does have an influence on the repayment rate and the deposited savings by the incumbent lender. Most microfinance institutions require that one has to deposit their savings by the institution when obtaining a loan. A decreased savings performance and a worsening of the repayment rate indicates that the borrowers that already had a loan from the incumbent lender take on new loans from the new competitors.

2.6 Microfinance and empirical studies

Wydick (1999) is one of the first that gives empirical results for the relation between performance and group lending and three different types of social cohesion; peer monitoring, social ties and borrowing group pressure. The paper analysis the effect of these on the provision of intra-group insurance, the mitigation of moral hazard within borrowing groups and overall group repayment performance. Wydick (1999) uses data of 137 borrowing groups in Guatemala. Loan applicants form groups varying from 3 to 8 persons. They apply for a loan together which is initially quite small, ranging from 200 to 500 US Dollar. The loan is divided between the group members and each invests the amount in their own project.

A logit model is used to test the relation between repayment rates and the level peer monitoring, social ties and borrowing group pressure. From these three measures of social cohesion, peer monitoring has the most effect on repayment rates. It reduces moral hazard and gives group insurance. The effect of group pressure on repayment rates is also present, but this is more limited. Social within groups do not have a significant effect on repayment rates.

one from their own pool of savings. Interest rates on both loans are 3 percent per month. The group members have to make every week an instalment on their loan and open a savings deposit such that at the end of the loan term they have at least 20% of the amount they borrowed. If a group member defaults on her loan, the savings deposit is used to pay back the loan. FINCA has perfect repayment on their group loans, individuals might default, but groups as a whole never default.

The analysis estimates loan default, savings and attrition of group members on geographic and cultural dispersion. The geographic and cultural dispersion measure the social connectedness of the group members. The dataset is restricted to persons that joint the group and were uninvited by any of the other group members. The group formation of 30 persons is thus an exogenous process. If one of the group members leaves the group, the gap is usually filled by someone that is invited by one of the group members. By restricting the analysis to only uninvited members, the peer selection process is excluded as a possible factor for the success of group lending systems.

The basic model that is estimated has the following form:

i i i

i X Z

Y =α +β1 +β2 +ε (32)

where Yi is a financial outcome (either default, savings or dropout), Xi is one of the social

connections measure (either geographic proximity or cultural similarity) and Zi is a

matrix of neighbourhood and cultural dummies and other demographic information. The results show that both cultural similarity and geographic concentration lead to lower default. Since default are covered by the savings of the group, lower default imply a higher return on savings. According to the results, geographic concentration has a positive influence on the level of savings. Finally, the dropout decision is examined. Higher levels of social connections cause less default and dropout of the group simultaneously.

and on the character of the joint liability contract; group lending might succeed whether or not it is implemented among high levels of social capital.

There are microfinance experiments carried out on pools of subjects that reflect the characteristics of actual recipients of microfinance in Cape Town, South Africa and Berd, Armenia. The subjects were women, eighteen years or older, either employed or available for work. The women filled out demographic questionnaire and they performed in a ‘trust game’. The women were randomly selected in pairs and either assigned the role of receiver or sender. They were each endowed with an amount of money. The sender has to decide how much of this amount it wants to transfer to the receiver. This amount is multiplied by three by the experimenter and passed to the receiver. In the following round, the receiver has the opportunity to give some of the received amount back to the sender.

The experiment carried out is based on the model of Abbink et al. (2006). This model captures the dynamic aspect of group lending. A group of 6 individuals received in South Africa a loan of 30 rand and in Berd a loan of 3000 drams. All group members are jointly liable for repayment. Each group member can invest 5 rand in an investment project, if the project succeeds he receives 12 rand, if it fails zero. The probability of success is 5/6. After the outcomes of the projects are known, the loan and interest have to be paid back. There is assumed that the interest is 20%, the group thus has to pay 36 rand. There have to be at least three successful group members, to fulfil this obligation. If the loan and interest is paid back, the group can proceed in another investment round. Information on the individual success or failure of a project is private. The main problem of the experiment of Abbink et al. (2006) is that group members know that the game is played a finite number of rounds. This causes in the final period non-contribution of the members. Cassar et al. (2007) only utilise data of the first six rounds in their analysis.

members does not have an influence of repayment rates. Social homogeneity does improve repayment rates of borrowing groups compared with heterogeneous groups. In sum, the different aspects of social capital have a different influence on repayment of group borrowers. The relation social capital between members is according to this survey the most important factor in determining the repayment rates of group liability contracts.

Ahlin and Townsend (2007) use empirical data to test the theories link that joint liability and repayment rates. They test if repayment rates differ between group with different characteristics. The data used in the investigation is form the Townsend Thai data base, a large cross section of 192 villages in two different regions of Thailand. In each village as many borrowing groups for of the Bank for Agriculture and Agricultural Cooperatives (BAAC) as possible were interviewed, up to two. The authors required data on 262 groups, 62 of them are the only group present in the village. This is called the BAAC survey, each group has chosen a leader that responded to the questions on behalf of the group. The BAAC is a development bank operated by the government. It receives a subsidy from the government. Smaller loans can be backed by social collateral in the form of joint liability. Larger loans need to be back by collateral. The fraction of joint liability depends thus on the number of people in a group that have no land.

The dependent variable in the model is a binary dummy from the BAAC survey that is equal to zero if the bank ever raised interest rates for the group due to late payment. The variable is one otherwise. There are several control variables included, log-age of group, size of a group, a measure of village-wide risk, a measure of village-average household wealth, and two measures of village-wide non-BAAC credit options that measure respectively the village-wide prevalence of commercial bank membership and production credit group membership. Further, average group landholdings and average education are included as productivity shifters. The degree of joint liability is included; this is measured by the percentage of the group that is landless. Other independent variables are the cooperation in a group, cost of monitoring, screening official and unofficial penalties.

village, this is a proxy for the peer monitoring capacity of a group, also has a positive impact on repayment rates. However, if there are relatives present in the borrowing group, this has a negative effect on repayment rates. This could be because imposing penalties is harder among relatives.

Cull et al. (2007) explore patterns of profitability, loan repayment and cost reductions of 124 institutions in 49 developing countries. The data set identifies three different kinds of institutions, namely 20 village banks, 56 individual-based lenders and 48 group-based lenders. The data set is obtained from the Microfinance Information Exchange (MIX), a not-for-profit private organisation that aims to promote information exchange in the microfinance industry. The dataset contains one observation per institution, ranging from 1999 to 2002; 70% of the observations are from 2002.

The key dependent variable in the analysis is the financial self-sufficiency (FSS) ratio; this measures the ability of an institution to generate sufficient revenue to cover its costs. This is the adjusted financial revenue divided by the sum of adjusted financial expenses, adjusted net loan loss provision expenses and adjusted operating expenses. It indicates the ability of an institution to operate without ongoing subsidy, including soft loans and grants. The variables operation self-sufficiency5 _{(OSS) and adjusted return on} assets (ROA)6 are also used as a dependent variable.

The village banks that are included in the survey charge the highest average interest rates and face the highest average costs. Only for the individual-based lenders, the ROA is positive. Village banks have the largest share of funding through subsidies.

The aim of the model by Cull et al. (2007) is to understand why some microfinance institutions are more profitable than others. They focus mainly on the role of costs and interest rates that are charged on the loans. These factors vary with the different kinds of lending type. The following reduced form equation is used:

5 _{The OSS-ratio is financial revenue divided by the sum of financial expenses, net loan loss provision}

expenses and operating expenses.

6 _{The ROA is measured as adjusted net operating income (net of taxes) divided by adjusted average total}

i i i i i i i i i i i i i i i Region n Orientatio MFIHistory LeningType e LendingTyp t CapitalCos t CapitalCos LeningType LabourCost LabourCost e LendingTyp Yield Yield FSS + + + + + × + + × + + × + + = 10 9 8 7 6 5 4 3 2 1 (33)

Yield is the gross portfolio yield, LendingType is the distinction between the three types

of institutions (mentioned above) and the matrix MFIHistory includes two variables, the age of the institutions and the size, measured by total assets. Labourcost are the personnel expenses while CaptitalCost are the rents, transportation, depreciation, office and other costs. Orientation contains three variables that describe the business practice of the institution, the ratio of loans to assets, the average loan size relative to GNP per capita and a dummy variable indicating the institution’s formal profit status. Region is a matrix of dummy variables for each main region of the developing world. As mentioned above, there are three different types of profitability measures used as dependent variable. Not only FSS is estimated in this way, but also OSS and ROA.

The aim is to find an answer on the following three questions: does raising interest rates exacerbate agency problems as detected by lower repayment rates and less profitability? Is there evidence between the dept of outreach to the poor and the pursuit of profitability? Have microfinance institutions moved away from serving poorer clients in pursuit of commercial viability?

The basic results show that only for the individual based lender, an increase in the interest rate is associated with increased financial performance. The capital costs have only for the individual borrower a negative influence on the profitability ratio. This means that for the banks that lend to individuals, capital costs are important in determining profitability. Labour costs are not significant is explaining profitability. Further, neither the village bank dummy nor the solidarity group dummy is significant in explaining financial performance. Regional dummies do explain some variation in financial performance. The age and size of an institution does have a positive influence of the financial performance of the institution.

worsen problems of moral hazard and adverse selection. If this assumption is true, microfinance institutions that charge higher interest rates should expect to have lower repayment rates and lower profitability. The results indicate that for the lenders that subscribe loans to individuals, an increase in interest rates increases profitability but only up to a certain point. For most of the solidarity group lenders, an opposite patterns holds. Further, there is a significant positive relation between the portfolio at risk and the real yield on the portfolio for the individual based lender. These are consistent with the agency problem.

The second question is the relation between the depth of outreach to the poor and profitability. This question is answered by regressing the average loan size on the profitability. Smaller loan sizes are supposed to be made to the poorest people. The authors do not find that smaller loans sizes mean that MFI’s are less profitable. However, they do find that larger loan sizes are associated with lower average costs.

The last question, if microfinance institutions moved away from serving their poorest clients is difficult to answer, since the dataset measured the variables only at one moment in time. The clients of a successful microfinance institution will become richer over time; they are able to hand in collateral and can receive a larger loan. It is possible that the outreach to the poorest clients is sacrificed to serving richer clients. In the regression, the outreach to the poorest clients is measured by the loan size. Financially self-sustaining individual based lenders tend to have smaller loan sizes and lend relatively more to women. This indicates a bank can still be profitable when lending to the poorest. However, banks that lend to individuals and exist for a longer period of time extend larger loans. This is consistent with the idea that maturing banks focus increasingly on wealthier clients.

members. In sum, one would expect that the more risky projects that have higher returns are funded with individual contracts. In the next section I will describe the empirical model to explain the differences between banks that provide group or individual contracts.

3. Group or individual contracts

I use an empirical framework to examine which factors determine whether banks subscribe group or individual contracts to their lenders. The following variables explain which type of contract a bank offers: the average loan per borrower, total assets of a bank, the repayment rate, the size of the portfolio that is at risk, the percentage female borrowers, the average loan as a fraction of gross national income (GNI), financial revenue, operation self-sufficiency, the profit margin, the cost per borrower, the age of the bank, the region where it is located, the profit status (non-profit or profit) and an indicator if grants are part of the main funding.

According to the literature, the amounts of group contracts are on average, per borrower, lower than individual contracts. Individual contracts are more offered to wealthier borrowers that are able to hand in collateral. I expect from the empirical investigation that the loan size per borrower is higher for the individual contract than within the group contract. This also suggests that the size of the total assets is higher for banks that provide individual contracts, since the assets of microfinance institutions mainly consist of their loan portfolio.

Both the percentage female borrowers and the fraction of the loan to GNI measure the outreach to the poor. These variables are common to measure the outreach (Cull et. al., 2007). Khandker (2005) also mentions the relation between loans to women and the effect on poverty. I expect that banks that provide individual contracts have a smaller outreach to the poor. So, they serve less female borrowers than banks that provide group contracts and the average loan in relation to gross national income for group contracts higher than for individual contracts.

The financial revenue ratio is the financial revenue divided by total assets. The assets of microfinance institutions are mainly the loan portfolio; a small part consists of other assets. The revenues from the loan portfolio are the interest revenues paid on the loan. I expect that financial revenue is higher for the banks that subscribe group contracts since the repayment rate is higher for group contracts compared to individual loans.

The reason that banks started with microfinance and offer group contracts to poor people is that the costs of monitoring and screening potential borrowers are too high compared to the loan size. One would thus expect that banks that offer individual contracts have higher costs per borrower than the banks that subscribe individual contracts.

The operation self-sufficiency (OSS) is a measure of the bank financial revenues are high enough to cover the financial costs of a bank. Group contracts are more efficient (Ghatak, 1999) and have lower costs (Conning, 1999). I expect that the OSS is higher for the banks that provide group contracts. As a consequence, I also expect that the profit margin is higher for banks that provide group contracts.

The region where the bank is located is a dummy variable. I will distinguish between the following regions of the world: Africa, East Asia and the Pacific, Eastern Europe and Central Asia, Latin America and the Caribbean, Middle East and North Africa, South Asia.

To summarize, I expect that banks that provide individual contracts have a higher average loan, a lower repayment rate, a higher fraction of the portfolio that is at risk, a lower percentage female borrowers, a higher average loan as a fraction of gross national income, a lower financial revenue, a higher operation self-sufficiency, a lower profit margin, higher costs per borrower, that the bank is older, has a lower change to be non-profit and a relies less on grants as main funding compared to banks that provide individual contracts.

To test the theoretical predictions between the two different types of contracts, I will estimate the following model:

[

] [ ]

[

] [ ]

[ ]

[

] [ ] [

]

[

]

[ ]

[

]

[

]

[

]

[

]

i i grants us profitstat region s totalasset age costs profit OSS evenue financialr loan GNI women risk repayment e Lendingtyp ε β β β β β β β β β β β β β β β + + + + + + + + + + + + + + + = 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 ) (log (34)

where i is the lending type of bank i, either an individual or a group contract. i (i=1, 2,

costs measures the operation costs per borrower. Totalassets are the total assets of a bank, from this variable the natural logarithm is taken, to level out the extreme values. The variable age gives the age of a bank, measured in years. Region, profit status and grants

are all dummy variables. Region gives the region were the MFI is located, profitstatus

indicates if the bank is a non-profit institution, grants indicates whether a large part of the main funding of a bank consists of grants. The last term, i is the standard independent

and identically distributed (i.i.d) error term.

The dependent variable in the model is the lending type of a bank. A bank subscribes either group contracts or individual contracts. This variable is either zero or one (zero if the bank has group contracts, one if the bank has individual contracts). I will thus estimate a binary choice model. In most binary choice models, the number of periods is limited, and the sample size is large. This is here also the case. Further, the banks do not change the type of contract they offer in the selected period and the three periods considered are subsequent years. If one averages the observations over the three periods one gets more efficient and consistent estimators. Further, there is less missing data. Therefore, I will average the data over the three periods.

I will estimate both a probit and a logit model. The probit model has a standard normal distribution, while the logit model has a standard logistic distribution; the logit model has a variance of _π2/3 _{instead of 1. Both models usually give very similar results in} empirical work, when one corrects for this difference in scale.

4. Data

I obtained data from The Microfinance Information eXchange (MIX)7: this is an organisation that gives information about Microfinance Institutions to the public and actors in the sector of microfinance. The sample is restricted to institutions where microfinance is their core business; it contains more than 90 percent of their operations. I obtained data from 177 institutions8, 111 of them provide contracts to individual

7 _{www.mixmarket.org}

borrowers and 66 of them provide group contracts. The banks are selected on the quality of their data, so it is not an exact representation of all microfinance institutions around the world. However, the selected sample represent almost 25 percent of the MFI’s that are listed on the Mixmarket and more than 50 percent of all microfinance borrowers. The data ranges from 1996-2005. However, most data is available for the last three periods; so from 2003 to 2005. Therefore, I will restrict my analysis to this period.

A description of the data can be found in table 1 to 9 of the appendix. The descriptive statistics when divided between individual and group contracts (table 1, appendix) show that the data has the same characteristics as predicted by the literature. Individual contracts have a lower repayment rate than group contracts, the percentage of the portfolio at risk is higher and the cost per borrower is higher. Financial revenue is higher for banks that subscribe group contracts than individual loans. The outreach of banks that provide group contracts is higher, as predicted. Group loans are more subscribed to women, further the average loan as a fraction of GNI is lower. More, the gross loan portfolio and the total assets are higher for banks that provide individual contracts than banks that subscribe group contracts, although banks that provide group contracts have on average more borrowers. This means that the average loan per borrower is smaller for banks that provide group contracts.

Other characteristics that distinguish banks that provide group and individual contracts are the profit status of the banks. A larger part of the banks that provide group contracts are non-profit institutions (table 3, appendix). Moreover, banks that provide group contracts rely more on grants as the main funding source (table 5, appendix). Further, it turns out that a very large share of the banks that provide individual contracts are located in Latin America and the Caribbean (table 4, appendix).

There are two variables that differ from the ex ante expectation, when comparing the medians of the samples, it turns out that the individual contracts have a higher profit margin and a higher operation self-sufficiency than the group contracts.

the age of a bank, the cost per borrower, the average loan as a fraction of GNI, the operation self-sufficiency, the profit margin, the portfolio at risk, the profit status, the gross loan portfolio, the total assets, and the dummy variable for Eastern Europe and Central Asia, Latin America and the Caribbean, Middle East and North Africa. The correlation coefficient is small for the operation self-sufficiency, the portfolio at risk and the dummy for Latin America and the Caribbean. The estimation needs to point out if the relation between the variables and the type of contract is significant. Further, there is a negative correlation between the type of contract and the following variables: the financial revenue, the repayment rate, the percentage female borrowers, if grants are part of the main funding of a bank, the number of borrowers, the dummy for Africa, East Asia and the Pacific and South Asia. The correlation between the type of contract and repayment is small, this holds also for the dummy for East Asia and the Pacific.

I expected a negative correlation between the type of contract and the operation self-sufficiency and the profit margin, because group loans are more efficient. Further, there is a positive, but small correlation between the type of contract and the portfolio at risk. I expected the opposite, since group contracts are safer for a bank. In the following section, I will estimate the model, so one can see witch variables have a significant influence in determining the type of contract a bank subscribes.

5. Results

The results of the estimated probit and logit model can be found in table 11 and 14 of the appendix. In table 11 the results of the probit model are given, in table 14 the results of the logit model. All coefficients have the same sign across both models. Since the logit distribution has a variance of _π2/3 _{all the coefficients that are estimated with the logit} model are approximately a factor π/ 3 larger than the estimates that are obtained from the probit model.

revenue ratio and is either located in South or East Asia, the Pacific, Latin America or the Caribbean is more likely to subscribe individual contracts. The profit margin of a bank, the portfolio at risk, the percentage women borrowers and whether or not grants are part of the main funding all have a significant negative influence on the probability that a bank subscribes individual contracts. So a bank with a high profit margin, a higher part of the portfolio at risk, a higher percentage of women borrowers and if their main funding consists of grants is more likely to subscribe individual contracts.

There are two results that are different from expected; the financial revenue ratio of a bank has a positive influence on the probability that a bank subscribes individual group contracts, the portfolio at risk has a negative influence of the probability that a bank provides individual contracts. So, a bank that has a high financial revenue ratio is more likely to subscribe individual contracts. I expected the opposite since group contracts have a higher repayment rate. Further, a bank with a high level of the portfolio that is at risk is more likely to subscribe group contracts. I expected the opposite, since group contracts are assumed to be safer.

For the logit model, the results are similar. All variables that have a significant positive impact on the probability that a bank provides individual contracts in the probit model are also statistically significant in the logit model. The only difference lies in the variables that have a negative influence. The portfolio at risk has a significant negative influence in the probit model, but the result is not significant in the logit model.

than for the logit model. In sum, the logit model performs better when explaining whether the bank subscribes individual or group contracts.

More interesting is to compute the marginal effects of the different variables. The results of the regression indicate that the financial revenue has a positive influence on the probability that a bank subscribes individual contracts, but the coefficient does not indicate how large this effect is. The coefficient specifies only if the variable has a positive or negative influence. In order to calculate the marginal effects I used a different software package. The results of the regression of the full probit and logit model are given in table 17 and 19 of the appendix; they are very similar to the results from the first regression, given in table 11 and 14. The logit model has a slightly higher R2 than the probit model; this was also the case in the first regression. In table 18 and 20 of the appendix, one can find respectively the marginal effects after the probit and logit model. The marginal effect gives the derivative of the function with respect to one explanatory variable; it gives the change in the probability that a bank subscribes individual contracts if the explanatory variable changes. The marginal effect is evaluated at the mean of the explanatory variables.

The results of the probit and logit model are very similar; the coefficients are of the same sign, and thus, in this case, the marginal effects also. I will first discuss the results of the logit model. First, the variables that has a positive effect on the probability that a bank offers individual contracts. An increase in the financial revenue of a bank has a very large influence on the probability that a bank offers individual contracts. If the financial revenue increases with one percent, the probability that a bank offers individual contracts increases with 0.34. The average loan and the age of an institution have only a small positive effect on the probability. If the age of a bank increases with ten years, the probability that a bank subscribes individual contracts only increases with 0.03. If the size of the average loan increases with 100 US dollar, the probability that a bank subscribes individual contracts increases with 0.01.