BIBLIOGRAPHY Armendariz de Aghion, Beatriz and Jonathan

advertisement
BIBLIOGRAPHY
Armendariz de Aghion, Beatriz and Jonathan Morduch. 2000. Microfinance beyond Group
Lending. Economics of Transition 8, 401-420.
Adjei, Joseph, Thankom Arun and Farhad Hossain. 2009. The Role of Microfinance in AssetBuilding and Poverty Reduction: The Case of Sinapi Aba Trust of Ghana. University of
Manchester Brooks World Poverty Institute, Working Paper 87.
Arun, Thankom and David Hulme. 2011. What’s wrong and right with microfinance – missing
an angle on responsible finance? University of Manchester Brooks World Poverty
Institute, Working Paper 155.
Athmer, Gabrielle and Fion de Vletter. 2006. The microfinance market in Maputo,
Mozambique: supply, demand and impact – A case study of Novobanco, Socremo and
Tchuma. Commissioned by the Netherlands Platform for Microfinance.
Bakhtiari, Sadegh. 2006. Microfinance and Poverty Reduction: Some International Evidence.
International Business and Economics Research Journal, Volume 5, No. 12.
Banerjee, Abhijit, Esther Duflo, Rachel Glennerster and Cynthia Kinnan. 2013. The miracle of
microfinance? Evidence from a Randomized Evaluation. National Bureau of Economic
Research, Working Paper No. 18950.
Bateman, Milford and Ha-Joon Chang. 2009. The Microfinance Illusion. Paper accessed at
http://www.econ.cam.ac.uk/faculty/chang/pubs/Microfinance.pdf on 15 July 2013.
Biggar, Nigel. 2009. Measuring Poverty Outreach: How Two Different Microfinance
Institutions Used the Progress out of Poverty Index. Enterprise Development and
Microfinance Vol. 20 No. 3 pp. 177-187.
Charitonenko, Stephanie. 2003. Commercialization of Microfinance: The Philippines. Manila:
Asian Development Bank.
Chaudhury, Iftekhar and Imran Matin. 2002. Dimensions and dynamics of microfinance
membership overlap — a micro study from Bangladesh. Small Enterprise Development,
Vol. 13 No. 2.
Chen, Greg, Stephen Rasmussen and Xavier Reille. 2010. Growth and Vulnerabilities in
Microfinance. Focus Note 61. Washington D.C.: Consultative Group to Assist the
Poor.
Chowdhury, Anis. 2009. Microfinance as a Poverty Reduction Tool – A Critical Assessment.
United Nations Department of Economic and Social Affairs Working Paper no. 89.
Chua, Ronald, Simon Gregorio, Marcia Miranda, Marie Apostol and Daniel Del Rosario. 1999.
Risk, vulnerability, assets and the role of financial services in reducing vulnerability: A
75
study of the women clients of CARD Bay Laguna, Philippines. Report prepared for the
2000/2001 World Development Report.
Chua, Ronald, Asuncion Sebastian and Andrea Silva. 2012. Poverty outreach of selected
microfinance institutions in the Philippines. Washington, D.C.: Grameen Foundation.
Cohen, Monique. 2001. Conceptual Framework for Assessing the Impacts of Microenterprise
Services. Assessing the Impact of Microenterprise Services (AIMS) Synthesis Study
submitted to the USAID Office of Microenterprise Development. Washington, D.C.:
Management Systems International.
Coleman, Brett. 1999. The impact of group lending in Northeast Thailand.
Development Economics Vol. 60, pp. 105-141.
Journal of
Coleman, Brett. 2006. Microfinance in Northeast Thailand: Who Benefits and How Much?
World Development Vol. 34, No. 9 , pp. 1612-1638.
Collins, Daryl, Jonathan Morduch, Stuart Rutherford and Orlanda Ruthven. 2009. Portfolios of
the Poor: How the World’s Poor Live on $2 a Day. Princeton, NJ: Princeton University
Press.
Cull, Robert, Asli Demirguc-Kunt and Jonathan Morduch. 2007. Financial performance and
outreach: a global analysis of leading microbanks. Economic Journal 117, 107-133.
Cull, Robert, Asli Demirguc-Kunt and Jonathan Morduch. 2009.
Market. Journal of Economic Perspectives, 23, 167-192.
Microfinance Meets the
Daley, Stephen and Jocelyn Badiola. 2003. Assessing microfinance and the USAID MABS
program in the Philippines under the microscope of the New Institutional Economics.
Paper presented as part of the Forum Series on “The Role of Institutions in Promoting
Economic Growth” of the Mercatus Center, George Mason University.
Duvendack, Maren, Richard Palmer-Jones, James Copestake, Lee Hooper, Yoon Loke and Nitya
Rao. 2011. What is the evidence of the impact of microfinance on the well-being of
poor people? London: EPPI-Centre, Social Science Research Unit, Institute of
Education, University of London.
El-Zoghbi, Mayada, Barbara Gahwiler and Kate Lauer. 2011. Cross-border Funding of
Microfinance. Washington DC: Microcredit Summit Campaign.
Fernando, Nimal. 2007. Low-Income Households’ Access to Financial Services: International
Experience, Measures for Improvement, and the Future. Manila: Asian Development
Bank.
Fischer, Greg and Maitreesh Ghatak. 2011. Spanning the Chasm: Uniting Theory and Empirics
in Microfinance Research. In Handbook of Microfinance, Beatriz Armendariz and Marc
Labie, editors. Singapore: World Scientific Publishing Co.
76
Geron, Ma. Piedad. 2010. Microfinance industry report: Philippines. Manila: Microfinance
Council of the Philippines.
Ghatak, Maitreesh and Timothy Guinnane. 1999. The Economics of Lending with Joint
Liability: Theory and Practice. Journal of Development Economics 60, 195-228.
Gibbons, David and Jennifer Meehan. 2000. The Microcredit Summit’s Challenge: Working
Towards Institutional Financial Self-Sufficiency while Maintaining a Commitment to
Serving the Poorest Families. Updated version of the paper presented at the 1997
Microcredit Summit.
Gine, Xavier and Dean Karlan. 2008. Peer Monitoring and Enforcement: Long Term Evidence
from Microcredit Lending Groups with and without Group Liability. Paper accessed at
http://karlan.yale.edu/p/bulak.pdf on 15 July 2013.
Godquin, Marie. 2004. Microfinance Repayment Performance in Bangladesh: How to Improve
the Allocation of Loans by MFIs. World Development Vol. 32, No. 11, pp. 1909-1926.
Gonzalez, Adrian. 2008. Microfinance Incentives to Repay and Overindebtedness: Evidence
from a Household Survey in Bolivia. Ph. D. dissertation. Graduate School of The Ohio
State University.
Accessed at http://etd.ohiolink.edu/send-pdf.cgi/Gonzalez%20
Adrian.pdf?osu1211556326 on 15 July 2013.
Grameen Foundation, 2014. 2014 Global Report on Poverty Measurement with the Progress out
of Poverty Index. Washington, D.C.: Grameen Foundation.
Hoddinott John and Agnes Quisumbing. 2003. Methods for microeconometric risk and
vulnerability assessments. World Bank, Human Development Network Social
Protection Unit, Discussion Paper No. 324.
Hulme, David and Karen Moore. 2006. Why has Microfinance been a Policy Success in
Bangladesh (and Beyond)? Economic and Social Research Council (ESRC) Global
Poverty Research Group Working Papers Series No. 041.
Karlan, Dean and Jonathan Morduch. 2013. Access to Finance. Chapter 2 of Handbook of
Development Economics, Volume 5, Dani Rodrik and Mark Rosenzweig, eds. Amsterdam:
Elsevier B.V.
Kondo, Toshio, Aniceto Orbeta Jr., Clarence Dingcong and Christine Infantado. 2008. Impact of
Microfinance on Rural Households in the Philippines. PIDS Discussion Paper Series
No. 2008-05.
Khandker, Shahidur. 2005. Microfinance and Poverty: Evidence Using Panel Data from
Bangladesh. The World Bank Economic Review, Vol. 19, No. 2, pp. 263-286.
Kiiru, Joy. 2007. The impact of microfinance on rural poor households’ income and
vulnerability to poverty: Case study of Makueni District, Kenya. Ph.D. dissertation.
Bonn University.
77
Lamberte, Mario. 2001. Expanding Banking Services to Micro, Small and Medium Enterprises
and Poor Households in the Philippines. PIDS Discussin Paper Series No. 2001-24.
Llanto, Gilberto. 2004a. Microfinance in the Philippines: Status, Issues and Challenges.
Philippine Institute for Development Studies Policy Notes, No. 2004-10.
Llanto, Gilberto. 2004b. Rural Finance and Developments in Philippine Rural Financial
Markets: Issues and Policy Research Challenges. Philippine Institute for Development
Studies Discussion Paper Series No. 2004-18.
Llanto, Gilberto and Ryu Fukui. 2003. Innovations in Microfinance in Southeast Asia.
Philippine Institute for Development Studies Discussion Paper Series No. 2003-11.
Marr, Ana. 2003. A Challenge to the Orthodoxy Concerning Microfinance and Poverty
Reduction. Journal of Microfinance, Volume 5, No. 2, 7-42.
Matin, Imran, David Hulme and Stuart Rutherford. 2002. Finance for the Poor: From
Microcredit to Microfinancial Services. Journal of International Development 14, 273294.
Matul, Michal, 2009. Financial Behaviours and Vulnerability to Poverty in Low-Income
Households in Transition Context. Paper accessed at http://www.microfinancegateway.
org/p/site/m/template.rc/1.9.40255/ on 15 July 2013.
McIntosh, Craig, Alain de Janvry and Elisabeth Sadoulet. 2005. How Rising Competition
among Microfinance Institutions Affects Incumbent Lenders. Economic Journal 115,
987-1004.
Montgomery, Heather and John Weiss. 2005. Great Expectations: Microfinance and Poverty
Reduction in Asia and Latin America. Asian Development Bank Institute Research
Paper Series No. 63.
Morduch, Jonathan. 1998. Does Microfinance Really Help the Poor? New Evidence on Flagship
Programs in Bangladesh. Princeton University, Woodrow Wilson School of Public and
Internatinal Affairs, Working Paper 198.
Morduch, Jonathan. 1999. The Microfinance Promise. Journal of Economic Literature, Vol. 37,
pp. 1569-1614.
Morduch, Jonathan and Barbara Haley. 2002. Analysis of the Effects of Microfinance on Poverty
Reduction. NY Wagner Working Paper No. 1014.
Mosley, Paul and David Hulme. 1998. Microenterprise finance: Is there a conflict between
growth and poverty alleviation? World Development, Vol. 26(5), pp. 783-790.
Navajas, Sergio, Jonathan Conning and Claudio Gonzalez-Vega. 2003. Lending Technologies,
Competition and Consolidation in the Markets for Microfinance in Bolivia. Journal of
International Development 15, 747-770.
78
Navajas, Sergio, Mark Schreiner, Richard Meyer, Claudio Gonzalez-vega, Jorge Rodriguez-Meza.
2000. Microcredit and the Poorest of the Poor: Theory and Evidence from Bolivia.
World Development, Vol. 28(2), pp. 333-346.
Odell, Kathleen. 2010. Measuring the Impact of Microfinance: Taking Another Look.
Washington, D.C.: Grameen Foundation.
Pitt, Mark and Shahidur Khandker. 1998. The Impact of Group-Based Credit Programs on Poor
Households in Bangladesh: Does the Gender of Participants Matter? Journal of
Political Economy, Vol. 106, No. 5, 958-996.
Priyadarshee, Anurag and Asad Ghalib. 2011. The Andhra Pradesh microfinance crisis in India:
manifestation, causal analysis and regulatory response. University of Manchester,
Brooks World Poverty Institute, Working Paper 157.
Reed, Larry. 2011. State of the Microcredit Summit Campaign Report 2011. Washington DC:
Microcredit Summit Campaign.
Roodman, David and Jonathan Morduch. 2009. The impact of microcredit on the poor in
Bangladesh: Revisiting the evidence. Working Paper No. 174. Washington, D.C.:
Center for Global Development.
Roodman, David and Uzma Qureshi. 2006. Microfinance as Business. Working Paper No. 101.
Washington, D.C.: Center for Global Development.
Rosenberg, Richard. 2010. Does Microcredit Really Help Poor People?
Washington D.C.: Consultative Group to Assist the Poor.
Focus Note 59.
Rutherford, Stuart, 1999. The Poor and Their Money: An Essay about Financial Services for
Poor People. Paper accessed at http://www.microfinancegateway. org/gm/document1.9.28437/1737_01737.pdf on 15 July 2013.
Schicks, Jessica. 2010. Microfinance Over-indebtedness: Understanding Its Drivers and
Challenging the Common Myths. Solvay Brussels School of Economics and
Management, Centre Emile Bernheim Working Paper No. 10/048.
Schicks, Jessica. 2011. Over-indebtedness of microborrowers in Ghana: An empirical study
from a customer protection perspective. Publication No. 15. Washington, D.C.: Center
for Financial Inclusion.
Schicks, Jessica and Richard Rosenberg. 2011. Too much microcredit? A survey of the
evidence on over-indebtedness. Occasional Paper No. 19. Washington D.C.:
Consultative Group to Assist the Poor.
Schreiner, Mark. 2009. Progress out of Poverty Index: A Simple Poverty Scorecard for the
Philippines. Washington DC: Grameen Foundation.
79
Sebstad, Jennefer, and Monique Cohen. 2000. Microfinance, Risk Management, and Poverty.
Assessing the Impact of Microenterprise Services (AIMS) Synthesis Study submitted to
the USAID Office of Microenterprise Development. Washington, D.C.: Management
Systems International.
Sharma, Manohar and Manfred Zeller. 1997. Repayment Performance in Group-Based Credit
Programs in Bangladesh: An Empirical Analysis. World Development, Vol. 25, No. 10,
pp. 1731-1742.
Siliki, Anne-Claire. 2011. Why people dropout from Microfinance institutions? Case study of
an MFI in Mali (Nyesigiso). Paper accessed at http://www.rug.nl/gsg/Research/
Conferences/EUmicrofinconf2011/ Papers/1new.15A.Siliki.pdf on 15 July 2013.
Simanowitz, Anton and Alice Walter. 2002. Ensuring Impact: Reaching the Poorest while
Building Financially Self-Sufficient Institutions, and Showing Improvement in the Lives
of the Poorest Women and their Families in Pathways out of Poverty: Innovations in
Microfinance for the Poorest Families by Sam Daley-Harris. Bloomfield CT: Kumarian
Press.
Simtowe, Franklin, Manfred Zeller and Alexander Phiri. 2006. Determinants of Moral Hazard in
Microfinance: Empirical Evidence from Joint Liability Lending Programs in Malawi.
Paper presented at the International Association of Agricultural Economists Conference,
Gold Coast, Australia, August 12-18.
Stewart, Ruth, Carina van Rooyen, Kelly Dickson, Mabolaeng Majoro, Thea de Wet. 2010.
What is the impact of microfinance on poor people? A systematic review of evidence
from sub-Saharan Africa. London: EPPI-Centre, Social Science Research Unit,
Institute of Education, University of London.
Stuart, Guy, Michael Ferguson and Monique Cohen. 2011. Managing Vulnerability: Using
Financial Diaries to Inform Innovative Products for the Poor. Report prepared by the
Financial Services Assessment project of the IRIS Center at the University of Maryland
and its partner, Microfinance Opportunities.
Sulaiman, Munshi and Imran Matin. 2008. Making Microfinance Work for the Extreme Poor.
Finance for the Poor, Volume 9, Number 1. Manila: Asian Development Bank.
Swain, Ranjula Bali and Maria Floro. 2008. Effect of Microfinance on Vulnerability, Poverty
and Risk in Low Income Households. American University, Department of Economics,
Working Paper Series No. 2008-02.
Todd, Helen. 2000. Poverty Reduced through Microfinance: The Impact of ASHI in the
Philippines. An ASHI-CASHPOR-PHILNET impact assessment paper accessed at
http://www.microfinancegateway. org/gm/document-1.9.25891/19274_N_087.pdf on
15 July 2013.
80
Vigenina, Denitsa and Alexander Kritikos. 2004. The Individual Micro-Lending Contract: Is It a
Better Design than Joint-Liability? Evidence from Georgia. Economic Systems 8, 155176.
Vogelgesang, Ulrike. 2003. Microfinance in Times of Crisis: The Effects of Competition,
Rising Indebtedness and Economic Crisis on Repayment Behavior. World Development
Vol. 31, No. 12, pp. 2085-2114.
Wydick, Bruce. 2001. Group Lending under Dynamic Incentives as a Borrower Discipline
Device. Review of Development Economics Volume 5, Issue 3, pp. 406-420.
Zaman, Hassan. 2000. Assessing the Poverty and Vulnerability Impact of Micro-credit in
Bangladesh: A Case Study of BRAC. Policy Research Working Paper No. 2145.
Washington, DC: World Bank.
81
APPENDIX 2.1
In the model of Ghatak and Guinnane (1999), it is assumed that borrowers are
risk-neutral and that they are of two types, safe (a) and risky (b). Given a project undertaken by a
borrower of type i, output now takes two values YiH and 0, with the probability of high output
being pi, i = a, b. It is assumed pa > pb. If the lender does not know the type of a borrower and
standard screening instruments like collateral cannot be used, the lender will offer loans with the
same nominal interest to all borrowers. In this situation, safe borrowers will cross-subsidize risky
borrowers because safe borrowers succeed more often (even as both repay the same amount when
they succeed). The presence of enough risky borrowers, however, can raise the equilibrium
interest rate high enough so that safe borrowers leave the market. Alternatively, safe borrowers
may continue to subsidize some undeserving risky projects.
Assuming borrowers know each other’s types, a joint-liability contract can
improve efficiency. With a joint liability credit contract, a borrower must repay the amount r if
her project yields a high return. She must also pay an extra amount c if the project of her partner
yields a low return. Thus, the expected payoff of a borrower of type i when her partner is of type
j in a joint-liability arrangement is the following:
EUij (r, c) = pi pj ( YH – r ) + pi ( 1 – pj ) ( YH – r – c )
(A2.1.1)
In general, all borrowers would prefer a safe partner. But the safer a borrower is,
the more she would want to have a safe partner. Consequently, safe borrowers would tend to
group together. In theory, a risky borrower could pay a safe borrower so that the latter would
agree to become her partner. The calculations below, however, indicate that this payment would
have to be larger than that which the risky borrower would be willing to make. In this regard,
82
given Equation (2.1), note that the net expected return of a risky borrower when she has a safe
partner is the following:
EUba (r, c) – EUbb (r, c)
= pb ( pa – pb ) c
(A2.1.2)
At the same time, the net expected loss of a safe borrower when she has a risky partner is the
following:
EUaa (r, c) – EUab (r, c)
= pa ( pa – pb ) c
(A2.1.3)
If c > 0 and given pa > pb, then the latter expression ( = the compensation a safe borrower would
require for making a risky borrower her partner) is greater than the former ( = the amount the
risky borrower would be willing to pay). All the foregoing, therefore, implies that safe borrowers
will end up grouping together, and risky borrowers by default will do the same.
Given the assortative matching result described above, a lender can screen
borrowers by offering two contracts. If the first contract has high joint liability and a low interest
rate while the second has low joint liability with a high interest rate, then safe borrowers will
choose the former while risky borrowers will prefer the latter. Consequently, repayment rates
and efficiency improve as joint-liability contracts are able to take advantage of a useful resource
untapped by conventional individual-liability contracts: the information borrowers have of each
other.
To show how joint liability addresses the moral hazard problem, Ghatak and
Guinnane (1999) assume that borrowers are risk-neutral as before, but this time the actions of the
borrower determine the probability of project success. In this regard, borrowers choose a specific
level of effort p ( 0 ≤ p ≤ 1 ) which imposes on them a disutility cost equal to (1/ 2) γ p2 (with γ >
0). As a result of the effort level choice of the borrower—a choice which is assumed to be
unobservable to the lender—output takes on two values: it is YH with probability p and 0
83
otherwise. Given the foregoing, social surplus p YH – (1/ 2) γ p2 is maximized if p = p* = YH /
γ. With regard to the latter, it is assumed that:
YH < γ
(A2.1.4)
to ensure an interior solution. Under these circumstances, if the lender had perfect information, it
could specify that all borrowers choose level of effort p = p* and consequently charge an interest
rate r = ρ / p*. Given asymmetric information, however, with the choice of p subject to moral
hazard, an individual borrower will choose p so as to maximize her individual gain. Given a
particular interest rate r, she will solve the following maximization problem as shown below:
Max p ( r ) = Max { p ( YH – r ) – (1/ 2) γ p2 } = pi ( r ) = ( YH – r ) / γ
p
p
(A2.1.5)
Given the above, we have p* = pi ( 0 ) > pi ( r ) for r > 0, and the higher the interest rate, the
lower is pi. In other words, the interest rate is like a tax on success, and the higher is the interest
rate, the lower is the resulting level of effort of the borrower.
The zero-profit condition of the lender is the following:
pr=ρ
(A2.1.6)
Substituting pi ( r ) = ( YH – r ) / γ into the above, we obtain:
γ ( pi ) 2 – YH pi + ρ = 0
(A2.1.7)
Solving for pi and assuming that the equilibrium with the higher value of pi is chosen (since the
lender is indifferent while the borrower is strictly better off), we have:
pi = { YH + [ (YH )2 – 4 ρ γ ]1/2 } / 2 γ
(A2.1.8)
With joint liability, a borrower once again pays the amount c when the project of
her partner fails. Given that her partner chooses the action p’, the borrower maximizes her
individual payoff by solving the following maximization problem:
Max { p ( YH – r ) – c p ( 1 – p’ ) – (1/ 2) γ p2 }
p
84
(A2.1.9)
Given the above, the best response function of each borrower, pg1 (r), is the following:
pg1 = [ ( YH – r – c ) / γ ] + [ ( c / γ ) p’ ]
(A2.1.10)
If borrowers who are partners make decisions about project choice non-cooperatively, then we
obtain the following symmetric Nash equilibrium:
pg1 = p’ = ( YH – r – c ) / ( γ – c )
(A2.1.11)
The zero-profit condition of the lender under joint liability is the following:
rp+cp(1–p) =ρ
(A2.1.12)
Inserting the value of r in Equation (A2.1.11) into Equation (A2.1.12), we obtain the following:
γ ( pg1 ) 2 – YH pg1 + ρ = 0
(A2.1.13)
Equation (2.13) is identical to Equation (2.7) and thus implies that the equilibrium
project choice of a borrower in a non-cooperative group liability situation is the same as the
choice of a borrower under individual lending: joint liability by itself does not seem to alleviate
the moral hazard problem. This result, however, arises because it is assumed that each borrower
does not consider how her actions affect the choice of action of her partner. If, on the other hand,
we assume that borrowers decide on project choice in a cooperative manner—with each borrower
now fully internalizing the effect of her actions on the choice of action of her partner—the
maximization problem of an individual borrower is the following:
Max { p ( YH – r ) – c p ( 1 – p ) – (1/ 2) γ p2 }
p
(A2.1.14)
The solution to the above maximization problem, pg2 ( r, c ), is the following:
pg2 = ( YH – r – c ) / ( γ – 2c )
(A2.1.15)
Substituting pg2 for pg1 in Equation (A2.1.11) and then inserting the value of r in the transformed
equation into Equation (2.15), we obtain the following:
( γ – c ) ( pg2 ) 2 – YH pg2 + ρ = 0
85
(A2.1.16)
Solving for pg2 above as before, we have:
pg2 = { YH + [ (YH )2 – 4 ρ ( γ – c ) ]1/2 } / 2 ( γ – c )
(A2.1.17)
Given that γ > YH by Equation (2.4) and the fact that a borrower cannot be asked to pay more
than what her project yields, it must be true that γ > c. Consequently, for 0 ≤ c ≤ γ, we have:
pg2 > pi
(A2.1.18)
The above tells us that the equilibrium value of p, and therefore the repayment rate, is higher
under joint-liability group lending when borrowers decide on p cooperatively, as compared to its
value under individual-liability lending
The discussion of joint liability above has thus far assumed that borrowers can
contract on p among themselves. As part of this assumption, it is understood that borrowers can
observe the actions of one another perfectly and without cost, as well enforce agreements among
themselves. In an extension of the model as elaborated on above, Ghatak and Guinnane show
that, even when monitoring is costly, joint liability lending can still improve repayment rates
through peer monitoring. This requires, however, that monitoring costs are not too high and/ or
that associated social sanctions are effective enough.
In the final theoretical section of their research, Ghatak and Guinnane (1999)
discuss the problem of enforcement—when the project of a borrower is successful but she still
refuses to repay. The latter does not result from informational asymmetries but arises rather from
the limited ability of the lender to impose sanctions on a delinquent borrower. In this context,
joint liability lending has two opposing effects on repayment rates. On one hand, group lending
allows a borrower whose project has realized very high returns to repay the loan of a partner
whose project has failed. On the other hand, a borrower who is only moderately successful may
decide to default on her own repayment because of the burden of having to repay the loan of her
partner. It is argued that the net effect of these two movements is positive as long as social ties
86
among borrowers are sufficiently strong. This is because borrowers who willfully default would
receive sanctions from both the lender and the community. In other words, with sufficient social
capital, the repayment rate would be higher under group lending compared to what it would be
under individual lending.
The following simple model illustrates this point. We assume that borrowers are
risk-averse and that the only departure from first-best happens when borrowers default even when
they are capable of repaying. The punishment that the lender imposes on a delinquent borrower
is to never lend to her again. In the context of individual lending, if the project of a borrower
produces output Y ≥ r (i.e. the borrower is capable of repaying her loan), she will repay only if
the benefit she derives from the additional income as a result of defaulting ( = the interest cost) is
less than B, the present value of the net benefit to the borrower of having continued access to
credit from the lender. In notational form, the borrower will repay if:
u (Y) − u (Y − r) ≤ B
(A2.1.19)
The above implies that, for a given r, there will be some critical Y (r) such that the borrower will
repay only if Y ≥ Y (r). As project returns get smaller (i.e. Y gets lower), there is an increasing
motivation for a borrower not to repay. This is because repayment gets more costly as the
marginal utility of income increases.
Under joint-liability group lending, a borrower is considered to be in default if her
partner does not repay. Under these circumstances, that borrower will choose to repay her loan
and that of her defaulting partner (assuming she has income Y with Y ≥ 2r) if:
u (Y) − u (Y − 2r) ≤ B
(A2.1.20)
Once again there is a critical level of income Y (2r) such that if Y ≥ Y (2r), then the borrower
will repay her own liability and that of her partner. It is to be noted, moreover, that Y (2 r) > Y
(r). This is because paying off two debt obligations (that of the borrower and her partner) is more
87
burdensome for the borrower than just paying of her own debt obligation, and thus income would
need to be higher for her to decide to do this.
If we assume for simplicity that Y (r) > 2r and that two borrowers bound together
by a joint liability repay if each has an income Y > Y (r), then the following two distinct cases
emerge with group lending contracts: (1) One borrower is unable or unwilling to repay [has
income Y1 with Y1 ≤ Y (r) ] while the other is willing to repay both her own obligation and that
of her partner (has income Y2 with Y2 ≥ Y (2r). In this case, repayment rates are higher under
group lending than they are under individual lending. (2) One borrower is again unable or
unwilling to repay [ Y1 ≤ Y (r) ] while the other is willing to repay her own debt but not her debt
and that of her partner together [ Y (r) < Y < Y (2r) ]. In this case, as far as repayment rates are
concerned, individual lending outperforms group lending. Whether case 1 or 2 is more likely to
occur depends on the probability distribution of output. Whichever case is more prevalent,
however, social sanctions would alter significantly decisions made by borrowers in such a joint
liability setting. Social sanctions would reduce the payoff stream of borrowers who default
intentionally [those with Y such that r < Y < Y (r) ] and those who are willing to repay their own
loan but not that of their partner also [ those with Y such that Y (r) < Y < Y (2r) ]. In such a
context, it is argued, repayment rates would definitely be higher with group lending contracts.
88
APPENDIX 2.2
In the model of Armendariz and Morduch (2000), the focus is on the individual
debt contract between a MFI and a borrower, with the assumption that the MFI has all the
bargaining power. A simple two-period model is considered in which a loan of size D is
extended by the MFI to the borrower at the beginning of each period. The borrower uses the loan
extended for the period to invest in a project which yields a total return π with a probability p,
and a total return of zero with probability (1 − p). Initially, p is assumed to be exogenous,
meaning there is no moral hazard problem with regard to the effort exerted by the borrower in
undertaking the project. The moral hazard problem is only in the repayment stage at which point
the client can “take the money and run” after returns on investment have been realized (also
referred to as the ex post moral hazard problem).
To prevent borrowers from doing the above, the MFI can threaten to not extend a
new loan in case of default, in which case it is assumed that the borrower will be unable to
finance a second-period investment.
Given the foregoing, the sequence of events is the
following: In period t = 1, the borrower is given a loan of size D and this is invested by the
borrower to gain a first-period investment return. She then makes a decision on whether or not
she will default on her first-period debt obligation. In period t = 2, the MFI decides on whether
or not to extend a new loan to the borrower. If a new loan is extended, this is invested by the
borrower to generate a second-period return.
If projects are sure to succeed (that is, assuming p = 1), the total net pay-off of the
borrower if she defaults is the following:
π + δvπ
(A2.2.1)
89
where δ is the discount factor and v is the probability of the borrower being given a new loan by
the MFI (0 ≤ v ≤ 1). On the other hand, the total net pay-off of the borrower if she decides to
repay is the following
π − R + δπ
(A2.2.2)
where R is the debt obligation of the borrower. The repayment R is deducted from the firstperiod return π, after which the MFI automatically grants a second-period loan (that is, we set
v = 1) to the borrower to reward her for her good conduct. For both Equation (2.21) and (2.22), it
is assumed that the borrower defaults with certainty on her second-period debt obligation, after
she realizes her second-period return. This is because the bank, in the present limited horizon
model, can no longer reward the borrower with a new loan if she repays during the second period.
Comparing the two prospective total net pay-offs, the borrower will decide to
repay the MFI in the first period if:
π + δvπ ≤ π − R + δπ
(A2.2.3)
The above is referred to as the incentive compatibility constraint and it specifies that the MFI
must ensure that the pay-off of the borrower is at least as large when she repays as when she
defaults. It is binding not only when p = 1 but for any value of p. Moreover, if the MFI credibly
carries out the threat to not finance a second-period loan in case of first-period default, v = 0 in
Equations (2.21) and (2.23). This in turn implies, based on Equation (2.23), that the maximum
interest rate that the bank can charge the borrower is R = δ π. In other words, the value δ π is the
opportunity cost of the borrower in not repaying her first-period obligation, and she cannot be
asked to repay more than this.
The MFI maximizes R subject to: (1) the incentive compatibility constraint as
defined by Equation (2.21); and (2) the individual rationality constraint of the borrower as
defined by the following equation:
90
p (π − R + δπ) ≥ 0
(A2.2.4)
This latter constraint means that it must be profitable for a non-delinquent borrower to secure
loans over the two periods from the MFI—her total net pay-off must be positive.
The optimal solution for the MFI is to always carry out the threat of not extending
new loans to delinquent borrowers (i.e. to set v = 0) and to set R = δ π. Alternatively, it may opt
to share surpluses (by setting R < δ π) in accordance with its social objectives.
In order to lessen the probability of default, an MFI can impose collateral
requirements in the case of individual lending programs or induce social sanctions in the case of
group lending programs.
Both can be introduced into the model and represented by the
additional sanctions variable W. Given W and assuming once again that the MFI follows the
optimal strategy of setting v = 0, the incentive compatibility constraint of the borrower is the
following:
π − W ≤ π − R + δπ
(A2.2.5)
which simplifies to
− W ≤ − R + δπ
(A2.2.6)
The latter in turn implies that the optimal R of the MFI, R*, is the following:
R* = δ π + W
(A2.2.7)
It is assumed that R* is less than π because limited liability means that the borrower cannot be
forced to repay the MFI more than the value of her investment. When there are no additional
sanctions (W = 0), R* would be equal to δ π. On the other hand, additional sanctions (W > 0)
would allow the MFI to charge a higher R* without inducing a higher probability of borrower
default.
91
The variable W can be re-interpreted as representing not sanctions but positive
inducements for repayment. If the MFI, for example, establishes a reputation of extending
progressively larger loans to borrowers who pay their obligations, Equation (2.21) becomes the
following:
π ≤ π − R + δ π2
(A2.2.8)
where π2 > π. Under these circumstances, we have:
R* = δ π2 = δ π + δ ( π2 − π )
(A2.2.9)
In the last equation, we can set W = δ ( π2 − π ) and the variable W would now represent the net
present value of future loans of increasing size over and above the net present value of loans that
remain constant at the initial size.
Apart from representing sanctions or incentives to repayment, the variable W can
also be used as a proxy for the probability of a borrower being re-financed by a rival lender. In
this regard, we assume that the borrower can secure a second loan from another lender with
probability v2. As a result, the incentive compatibility constraint (assuming as before that v = 0)
becomes the following:
π + δ v2 π ≤ π − R + δ π
(A2.2.10)
According to this last equation, as the likelihood v2 of a borrower being refinanced by a second
lender increases, the incentive to repay the first lender decreases, and as a consequence, the
maximum repayment R that can be extracted by the first lender goes down as well.
Thus far, the model of Armendariz and Morduch (2000) has focused on incentives
to repay loans after projects have been undertaken successfully. This framework, however, can
be extended to cases in which the probability of success is endogenous. We introduce the latter
by assuming that the borrower can choose the level of p. In doing so, we get the following
sequence of events: The MFI first proposes a debt contract to the borrower, with a specified
92
repayment schedule R.
The borrower then makes her effort choice, which is modeled as
choosing a particular level of p. Returns are then realized from projects that are undertaken and
repayments are made. The borrower, of course, can still choose to default on her debt obligation.
If she does, she is not able to secure a second-period loan with probability (1 − v) and has to bear
the additional social sanction W.
The non-monetary cost of effort for the borrower is assumed to be the following:
c (p) = k ( p2 / 2 )
(A2.2.11)
where k is a fixed cost factor and the increasing marginal cost of effort is captured by quadratic
form of the cost function. The borrower chooses p so as to solve the following maximization
problem:
Max p ( π − R + δ π ) + ( 1 − p ) ( δ v π − W ) − c ( p )
p
(A2.2.12)
The first-order condition for maximization of the above is:
π−R + (1−v) δπ + W = pk
(A2.2.13)
The equilibrium probability p* corresponding to the equilibrium level of effort of the borrower is
thus the following:
p* = [ π − R + ( 1 − v ) δ π + W ] / k
(A2.2.14)
Based on this last equation, it can be inferred that p* is: (1) decreasing as the debt obligation R of
the borrower increases; (2) increasing as the social sanction W increases; and (3) decreasing as
the probability of gaining access to a second-period loan increases.
Anticipating the equilibrium effort response of the borrower as described above,
the MFI will choose ex ante to set v = 0 (to not extend a new loan to a borrower who defaults)
and offer a repayment schedule R that will maximize its expected repayment revenue.
accordance with the latter, the MFI will solve the following maximization problem:
93
In
Max p (R) R = [ ( π − R + δ π + W ) / k ] R
R
(A2.2.15)
subject to R ≤ π. The first-order condition for maximization of the above is:
π−R + δπ + W = R
(A2.2.16)
which yields the following optimal repayment schedule:
R* = [ ( 1 + δ ) π + W ] / 2
(A2.2.17)
with the assumption that R* < π. The last equation indicates that R* is an increasing function of
social sanctions W, project returns π and the discount factor δ.
As a final theoretical discussion, Armendariz and Morduch (1999) analyze what
was—at the time they wrote—one of the least remarked upon but most unusual features of
microfinance credit contracts.
This was the requirement that repayments must start almost
immediately after disbursement of a loan and then proceed regularly thereafter. In the Grameenstyle MFIs surveyed by the authors, the repayment schedule for a year-long loan was determined
by summing up the total principal and interest due, dividing by 50, and then beginning weekly
collections a couple of weeks after loan disbursement. Such a system of weekly repayments
meant that these MFIs were in effect lending partly against streams of household income realized
outside of projects financed by their loans.
The desirability of regular repayment schedules from the point of view of MFIs is
modeled by the Armendariz and Morduch (2000) as follows. It is assumed that households, after
the purchase of household necessities, have disposable income X every week. This amount is
generated from “outside” sources and does not come from the household enterprise that is
financed by the MFI.
This income, moreover, is diverted into miscellaneous consumption
expenses and decays at a discount factor d per week. It is assumed that the mentioned expenses
94
do not provide the household any utility—an assumption, however, that can be released with
affecting the main argument.
Assuming that a loan has a one-year duration, the MFI must determine the number
of installments in a year ( = n ) by which the loan is to be repaid. Given that T = the length of a
period separating successive loan repayments (in terms of number of weeks), then n = 52 / T.
The loan may be paid in a one-time installment, for example, in which n = 1 and T = 52.
Alternatively, it may be paid monthly (n = 12, T = 52 / 12) or weekly (n = 52, T = 1). The total
of all the principal and interest payments made by the borrower to the MFI is denoted by L, while
the transaction costs per installment payment—which is assumed to be borne wholly by the
borrower—is denoted by γ. Assuming that the preferences of the borrower with respect to
income are linear and that the optimal loan size is not bigger than that which can be supported by
outside revenues of the borrower, the MFI chooses the value of T that will maximize the size of L.
This maximization problem is the following:
Max L = { ( d + d2 + d3 + … + dT ) ( 52 X / T ) − ( 52 γ / T) }
T
(A2.2.18)
Based on the maximization problem above, what is the optimal value of T for the
MFI? If d is close to one and γ is large, the optimal T will tend toward 52. However, as is more
likely with poorer households, if d is low (because income gets channeled to miscellaneous
expenses and mechanisms to enforce financial discipline are relatively limited) and γ is also low
(because the opportunity cost, for example, of time is relatively low), then the optimal T will tend
toward 1. This result is reinforced by the fact that micro-enterprises funded by MFIs (such as
those engaged in petty trading) usually generate a flow of revenue on a daily or weekly basis,
making frequent collections desirable given the lack of satisfactory savings facilities.
95
As demonstrated in theoretical terms above, regular repayment schedules make the
largest amount of household income available for repayment to MFIs.
It is also argued,
moreover, that they help screen out undisciplined borrowers, as well as provide MFIs with a
regular flow of information on borrowers. Finally, various authors have noted that regular
repayment schedules make microfinance credit contracts resemble arrangements for saving.
They provide a substitute for imperfect savings vehicles. By committing to make small, regular
payments to an MFI, borrowers are able to access a usefully large amount of money, in a way not
different from what would happen through a regular saving plan.
96
APPENDIX 2.3
In the model of Vogelgesang (2003), the borrower obtains a loan of size L and has
non-business income V and wealth W. The borrower invests L and V into her business and the
latter yields a return equal to g (V + L, A) when successful. In the preceding function, A
represents the idiosyncratic characteristics that determine productivity and g is assumed to be
increasing in all arguments with decreasing returns in the first argument. Business output is
defined further by a binary variable π whose value is one if the business is successful and zero
otherwise. It is assumed that the borrower has to repay the total loan amount plus interest at the
end of the period, and thus the total repayment is equal to (1 + r) L where r is the interest rate.
Alternatively, the borrower has to pay a penalty P if the loan is not paid on time. This penalty
consists of higher interest rates if the borrower pays late or of collateral seizure if eventually no
payment is made at all. Furthermore, repeat loans from the same lender can only be secured by
the borrower if she has paid back the first loan on time. Otherwise, a new loan is not granted by
the lender and the borrower can be penalized with a bad credit record.
The future benefits from timely repayment are denoted by B. the factors that
determine the magnitude of B are assumed to include the following: (1) the degree of availability
of credit records, denoted by δ, (2) the extent of competition among MFIs, denoted by ζ, (3) the
borrower’s possession of idiosyncratic characteristics that increase her access to alternative
borrowing possibilities, denoted by R, and (4) the degree of leverage of the business of the
borrower, denoted by λ. A greater availability of credit records increases the barriers to obtaining
a future loan after default and is thus assumed to increase B. On the other hand, greater MFI
competition (which increases the supply of loans and the possibility of a borrower obtaining a
loan from another lender after default) is assumed to decrease B. Likewise, greater possession of
97
favorable idiosyncratic characteristics (associated with gender, business sector and location) is
also assumed to decrease B. Finally, greater leverage of the business of the borrower—as well as
a bigger loan size L—increases the borrower’s need for a subsequent loan to maintain the scale of
her business and thus increases B.
For simplicity of notation, all relevant values in the analysis that follows are
denoted in terms of second period utility and there is no discounting. Total utility of the
borrower with repayment, UR, and with default, UD, are specified, respectively, by the following
equations:
UR = v { W + π g [V + L, A] – (1 + r) L }
(A2.3.1)
+ B ( W, R, δ, ζ, λ, L )
UD = v { W + π g [V + L, A] – P }
(A2.3.2)
Based on the Equations (2.39) and (2.40) above, the following variables used in the analysis are
defined:
vSR = v { W + π g [V + L, A] – (1 + r) L }
(A2.3.3)
vFR = v { W – (1 + r) L }
(A2.3.4)
vSD = v { W + π g [V + L, A] – P }
(A2.3.5)
vFD = ( W – P )
(A2.3.6)
B* = B* ( W, V, P, r, L, A ) = vSD – vSR
(A2.3.7)
B^ = B^ ( W, P, r, L ) = vFD – vFR
(A2.3.8)
As defined in the equations above, vSR is the total utility of the borrower excluding
B when her business is successful ( π = 1 ) and she repays. vFR is the total utility of the borrower
excluding B when her business fails ( π = 0 ) and she repays. vSD is the total utility of the
borrower when her business is successful ( π = 1 ) and she defaults. vFD is the total utility of the
98
borrower when her business fails ( π = 1 ) and she defaults. B* is the gain in total utility of the
borrower as a result of not paying her loan when her business succeeds, with her loss of future
benefits B not being taken into consideration. Finally, B^ is the gain in total utility of the
borrower as a result of not paying her loan when her business fails, with her loss of future
benefits B not being taken into consideration. It is to be noted that both B^ and B* are positive
because it is assumed (based on the actual practice of MFIs) that (1 + r) L > P. Furthermore, B^
> B* given the standard assumption that the utility of the borrower increases with income but at a
decreasing rate.
Based on all of the foregoing, borrowers can be divided into three client groups
based on their optimal repayment behavior. The latter is derived by comparing the borrower’s
total utility with repayment to her total utility with default, including now in the comparison how
the borrower would or would not enjoy future benefits B in the respective scenarios. The first
group of clients is constituted by those whose optimal policy is always to repay (whether their
business succeeds or fails) because for them the value of future benefits B is big enough such that
B ≥ B^ > B*. The second group of clients is constituted by those whose optimal policy is to
repay if their business succeeds and to default if it fails. For this second group, B^ > B ≥ B*.
With the last group of clients, the optimal policy is never to repay because B > B* ≥ B.
Who are the borrowers who are more likely to repay? First, borrowers who derive
greater future benefits B are more likely to repay. Secondly, borrowers with lower B^ (= smaller
gains from not repaying when their business fails) are more likely to repay after a business failure.
Third, borrowers with lower B* (= smaller gains from not repaying when their business succeeds)
are more likely to repay when their business is successful. Finally, borrowers with higher W, V,
A and π are more likely to have incomes that are sufficiently large to repay.
99
Based on the model, therefore, individual borrower characteristics would affect
repayment as follows. Higher borrower productivity increases the likelihood of repayment.
Higher borrower nonbusiness income and/ or wealth have the same effect. On the other hand,
better alternative borrowing possibilities decreases the probability of repayment. With respect to
loan terms, low interest rates and high penalties increase the likelihood of repayment. Finally,
the environment also significantly influences repayment.
The availability of credit records
increases the probability of repayment and so does a positive economic environment. More
competition among MFIs and the consequent greater supply of loans in the market decreases the
likelihood of repayment.
100
APPENDIX 2.4
In the model of Gonzalez (2008), each household is assumed to live for only two
periods. In each period, it has a maximum endowment of labor emax. It also has an initial
endowment of productive assets (capital) a0 with no depreciation. Assets can be accumulated or
they can be liquidated to be transformed into consumption. The price of assets in period one is
L1 while its price in period two is L2 with L1 < L2 < 1.
In each period t, net output yt is a stochastic function such that:
zt Y ( kt , et ) with probability p
yt = {
(A2.4.1)
} for t = 1, 2
0 wih probability (1 – p)
In the specification above, the first component zt is a non-tradable factor input consisting of the
entrepreneur’s skill or ability in production, which is assumed to be unevenly distributed in the
population. The second component represents a production function with two inputs, capital kt
and labor et with Yk > 0, Ykk < 0, Ye > 0 and Yee < 0. The second component, moreover, is
assumed to be zero when the project fails with a probability (1 – p).
Households can store output produced in period one to be used as capital and later
consumed in period two. As such, household consumption in each period ct is specified by the
following equation:
ct = yt – pt – ( at – at -1 ) Lt
(A2.4.2)
where ct is consumption, pt is debt service payments, ( at – at -1 ) is asset accumulation (savings),
with all variables being end of period totals. Equation (2.48) represents the budget constraint that
must be satisfied in every period. It is assumed, moreover, that there is a minimum level of
consumption cMin > 0 below which the household does not survive.
The household maximizes expected utility W specified as:
101
W = E0 [ U1 + β U2 ]
(A2.4.3)
where β is the discount factor with 0 < β < 1. β is equal to 1 / (1 + ρ ) where ρ is the rate of time
preference. U is a utility function with Uc > 0, Ucc < 0, Ue > 0 and Uee < 0. Et is the expectations
operator conditional on the information/ expectations available to the individual at time t. For
simplicity, it is assumed that output yt is not equal to zero and the expectations operator is
removed and Equations (2.47) and (2.49) respectively become:
yt = zt Y ( kt , et )
(A2.4.4)
W = [ U1 + β U2 ]
(A2.4.5)
It is assumed that each household does not have sufficient resources to take full
advantage of its investment opportunities and thus borrowing is welfare-improving. Furthermore,
each borrower establishes a credit relationship with only one lender and all funds accessed from
borrowing are used to purchase capital at the beginning of each period. Debt matures in one
period and the repayment function is the following:
dt = ( 1 + rt ) bt
(A2.4.6)
where dt is the total debt service obligation at the end of the period, rt is the interest rate and bt is
the amount borrowed for the period with bt ∈ { 0, Bt ]. Bt is the maximum loan size available to
the household in each period. B1 can be based on characteristics of the borrower household (that
are associated with its repayment capacity) so that for period one we have the following:
B1 = B1 (a0 , emax, z1)
(A2.4.7)
The maximum loan size in period two is a function of the maximum loan size and
the repayment performance of the household in period one:
B2 = B2 ( B1, T)
(A2.4.8)
102
where T ∈ { D or Default, R or Repayment }. When default—defined as any situation where
pt < bt —occurs in period one, the household is denied access to a loan in period two. In other
words, we have:
B2 ( B1, D ) = 0
(A2.4.9)
for any value of B1. Under the preceding full repayment assumption, the household will choose
p1 = d1 if it decides to pay and p1 = 0 if it decides not to pay its debt in period one.
The household has an incentive to repay its loan in period one because this is
required in order to access a welfare-enhancing loan in period two. There is no such incentive for
loan repayment in period two, however, because the model assumes that the household lives for
only two periods. Consequently, default is always observed in the second period. Furthermore, a
no bequest condition is assumed (the household does not pass on wealth to a future generation)
and thus a2 = 0.
Apart from assets acquired through loaned funds, the household owns assets which
are available for production in each period. Consequently, capital kt is the following:
kt = bt + at - 1
(A2.4.10)
The household maximizes its intertemporal utility by choosing the optimal
amounts of effort, consumption, savings and borrowing/ capital in periods one and two. The
optimization problem of the household is given by the following:
Max
W = [ U1 ( c1 , e1 ) + β U2 ( c2 , e2)]
c1, c2, e1, e2, k1, a1
(A2.4.11)
where Uc > 0, Ucc < 0, Ue > 0, Uee < 0, Yk > 0, Ykk < 0, Ye > 0 and Yee < 0, 0 < β < 1 and
β = 1 / (1 + ρ )
subject to:
a0 ≥ 0
(A2.4.12)
103
et ≤ emax
(A2.4.13)
ct ≥ cMin
(A2.4.14)
ct = yt – pt – ( at – at -1 ) Lt
(A2.4.15)
yt = zt Y ( kt , et )
(A2.4.16)
kt = bt + at - 1
(A2.4.17)
dt = ( 1 + rt ) bt
(A2.4.18)
a1 ≥ 0 (maximum asset drawdown in period one condition)
(A2.4.19)
a2 = 0 (no bequest motive condition)
(A2.4.20)
0 ≤ b1 ≤ B1 (a0 , emax, z1)
(A2.4.21)
b2 = B2 ( B1, T) with B2 ( B1, D) = 0 if default
(A2.4.22)
and B2 ( B1, R) > B1 if repayment
W ≥ W0 = sup W ( b1 = b2 = 0 )
(A2.4.23)
If full repayment of debt in period one is assumed and with some substitutions, the
following Kuhn-Tucker Lagrangian associated with the household optimization problem is
obtained:
Γ = [ U1 ( c1 , e1 ) + β U2 ( c2 , e2)]
(A2.4.24)
+ λ1 { z1 Y ( k1 , e1 ) – ( a1 – a0 ) L1 – c1 – ( 1 + r1 ) ( k1 – a0 ) }
+ λ2 { z2 Y ( a1 + b2 , e2 ) + a1 L2 – c2 }
– μ (e1) ( e1 – emax ) – μ (e2) ( e2 – emax )
+ μ (c1) ( c1 – cMin ) + μ (c2) ( c2 – cMin )
– μ (B1) ( b1 – B1) + μ (b1) b1 + μ (a1) a1
The first-order conditions (FOCs) for the intertemporal maximization of utility with loan
repayment are derived from the equation above and are the following:
∂ Γ/ ∂ c1 = ∂ U1 (c1 ,e1) / ∂ c1 – λ1 + μ (c1) = 0
(A2.4.25)
∂ Γ/ ∂ c2 = ∂ U2 (c2 ,e2) / ∂ c2 – λ2 + μ (c2) = 0
(A2.4.26)
∂ Γ/ ∂ e1 = ∂ U1 (c1 ,e1) / ∂ e1 + λ1 z1 [ ∂ Y (k1, e1) / ∂ e1 ] – μ (e1) = 0
(A2.4.27)
∂ Γ/ ∂ e2 = ∂ U2 (c2 ,e2) / ∂ e2 + λ2 z2 [ ∂ Y (k2, e2) / ∂ e2 ] – μ (e2) = 0
(A2.4.28)
104
∂ Γ/ ∂ k1 = λ1 [ z2 (∂ Y (k2, e2) / ∂ e2) – (1 + r1) ] – μ (B1) + μ (b1) = 0
(A2.4.29)
∂ Γ/ ∂ a1 = – λ1L1 + λ2 [ z2 (∂ Y (k2, e2) / ∂ k2) + L2] + μ (a1) = 0
(A2.4.30)
∂ Γ/ ∂ λ1 = { z1Y (k1 , e1) – (a1 – a0) L1 – c1 – (1 + r1 )(k1 – a0)} = 0
(A2.4.31)
∂ Γ/ ∂ λ2 = { z2 Y (a1 + b2 , e2) + a1 L2 – c2 } = 0
(A2.4.32)
μ (e1) (e1 – emax) = 0 ,
μ (e2) (e2 – emax) = 0
(A2.4.33)
μ (c1) (c1 – cMin) = 0 ,
μ (c2) (c2 – cMin) = 0
(A2.4.34)
μ (B1) (b1 – B1) = 0 ,
μ (b1) b1 = 0
(A2.4.35)
μ (e1) , μ (e2) , μ (c1) , μ (c2) , μ (B1) , μ (b1) ≥ 0
(A2.4.36)
( e1 – emax ) ≤ 0 , ( e2 – emax ) ≤ 0
(A2.4.37)
( c1 – cMin ) ≥ 0 , ( c2 – cMin ) ≥ 0
(A2.4.38)
( b1 – B1) ≤ 0 ,
(A2.4.39)
b1 ≥ 0 ,
b2 = B2
μ (a1) a1 = 0
μ (a1) ≥ 0 ,
(A2.4.40)
a1 ≥ 0
(A2.4.41)
It is assumed that a solution to the maximization problem exists and that this solution is the
following:
Ψ* = ( c1*, c2*, e1*, e2*, k1*, a1*)
(A2.4.43)
The above solution defines the optimal levels of utility in period one and two, U 1* and U2*,
respectively, as well as the maximum level of utility (value function) W*, with repayment.
If the household defaults on loan repayment in period one, it loses access to credit
in period two. The household can choose to do this even when it generates sufficient capacity to
repay its loan in period one. The household chooses to default when the maximum level of utility
of the household under default is greater than its maximum level of utility under repayment. In
the default scenario, the household decides on the optimal level of the decision variables,
105
constrained by the fixed level of capital k1* set in the beginning of period one, with k1* equal to
the initial level of assets a0 plus the loan funds b1 obtained for period one. The Kuhn-Tucker
Lagrangian associated with the optimization problem of the household under default in the first
period is thus the following:
Γ
= [ U1 ( c1 , e1 ) + β U2 ( c2 , e2)]
(A2.4.44)
k1 = k1*
+ λ1 { z1 Y ( k1*, e1 ) – ( a1 – a0 ) L1 – c1 }
+ λ2 { z2 Y ( a1 , e2 ) + a1 L2 – c2 }
– μ (e1) ( e1 – emax ) – μ (e2) ( e2 – emax )
+ μ (c1) ( c1 – cMin ) + μ (c2) ( c2 – cMin )
+ μ (a1) a1
The FOCs for the intertemporal maximization of utility under default are derived from the
equation above. It is assumed that a solution to the maximization problem exists and that this
solution is the following:
ΨD = ( c1D, c2D, e1D, e2D, a1D )
(A2.4.45)
The above solution defines the optimal levels of utility in period one and two, U1D and U2D,
respectively, as well as the maximum level of utility WD, under default.
The maximum value of the objective function for different levels of b1 given the
decision to repay the loan accessed in period one is the following:
WR (b1) = sup
{ U1 [ (z1 Y ( b1 + a0 , e1 ) – (a1 – a0) L1 – b1 (1 +
r1) , e1 ]
(A2.4.46)
b1 ≤ B1
+ β U2 [ (z1 Y (b2 + a1 , e2) + a1 L2 , e2 ]
}
On the other hand, the maximum value of the objective function for different levels of b1 given
the decision to default on the loan accessed in period one is the following:
WD (b1) = sup
{ U1 [
(z1 Y (b1 + a0 , e1) – (a1 – a0) L1 , e1 ]
b1 ≤ B1
+ β U2 [ (z1 Y (a1 , e2) + a1 L2 , e2 ]
106
}
(A2.4.47)
Let b1R be the amount of assets purchased with the use of loaned funds by the household in
period one at which its maximum utility under repayment is equal to its maximum utility under
default. In other words, b1R is defined by the following equation:
WR ( b1R ) = WD ( b1R )
(A2.4.48)
It is shown that b1R > 0 may not exist. Moreover, it is also be shown that when b1R > 0, for any
b1' < b1R and b1'' > b1R, respectively, we have:
WR (b1' ) > WD (b1' )
(A2.4.49)
WR (b1'' ) < WD (b1'' )
(A2.4.50)
The above equations imply the following: (1) For any loan b1' smaller than b1R, WR > WD and
the household repays its loan. (2) For any loan b1'' larger than b1R, WR < WD and the household
defaults on its loan obligation. The preceding points relate to the importance of an MFI being
able to correctly estimate the entrepreneurial ability of a borrower household, particularly in
terms of not over-estimating the capacity of the latter to use a particular amount of loaned funds
in a productive manner.
The analysis thus far has not modeled the impact of unexpected adverse shocks on
both the repayment capacity of the household and its decision to repay its loan, respectively.
Adverse economic shocks, however, can have a significant effect on household welfare. The
analysis that follows now takes this into account and models household repayment behavior with
the occurrence of an unexpected adverse shock in period one. In this regard, it is assumed that
the household experiences this shock early in period one, just after assets have been bought with
the loan. The latter investment is assumed to be irreversible (that is, the household cannot
disinvest and return the loaned funds to the lender). The shock reduces the level of z in the
production function of the household so its value after the shock, z1E, is lower than z1. It is
assumed further that, after the shock has occurred, there is no longer uncertainty about the
107
production function of period one. In particular, the household becomes certain that the effective
level of output after the shock for period one equal to z1E YE ( k1*, e1* ) will be lower than the
previously planned level of output z1 Y ( k1*, e1* ), given the previously set optimal amount of
capital k1* and planned level of effort e1*.
The lower post-shock level of output makes the effective repayment capacity of
the household inadequate to meet its debt obligation. The latter situation is described by the
following equation:
(1 + r1) (k1* – a0) > ztE YE ( k1*, e1*) – c1* – (a1* – a0) L
(A2.4.51)
Given the above, the household has only two options, either to default on the loan or to repay by
undertaking “costly actions.” There are three types of “costly actions” that the household can
engage in, all involving a modification of an initially planned level of (optimal) action in period
one: (1) reduce consumption, (2) increase effort and (3) reduce savings (sell assets). The first
two costly actions reduce the level of household utility in period one while the third action
reduces the level of household utility in period two. Furthermore, the shock may change the
production function of period two causing a reduction in the level of output of the household in
period two.
Let G1 represent the additional repayment capacity necessary for a household to
repay its loan after the occurrence of an adverse economic shock. The value of G1 is given by the
following equation:
G1 = ztE YE (k1*, e1*) – {c1* + (a1* – a0) L1 + (1 + r1) (k1* – a0)} < 0
(A2.4.52)
A larger negative value of G1 implies that the household must engage in greater magnitudes of
costly action in order to repay its loans. This would involve one or two or all of the following:
more reduction in consumption, a larger asset drawdown and greater increase in effort.
108
If the household chooses to repay its loan after a shock, its new optimization
problem involves generating the additional repayment capacity it needs with a minimum loss of
utility. This is equivalent to the optimization problem in which the objective function is still the
one specified in Equation (2.57) and but the constraints now include the requirement that G 1 = 0
with k1 = k1* and b1 = b1*.
The Kuhn-Tucker Lagrangian associated with this optimization
problem is the following:
Γ
= [ U1 ( c1 , e1 ) + β U2 ( c2 , e2)]
(A2.4.53)
k1 = k1*, G1 = 0
+ λ1 { z1E YE ( k1* , e1 ) – ( a1 – a0 ) L1 – c1 – ( 1 + rt ) ( k1* – a0 ) }
+ λ2 { z2 Y ( a1 + b2 , e2 ) + a1 L2 – c2 }
– μ (e1) ( e1 – emax ) – μ (e2) ( e2 – emax )
+ μ (c1) ( c1 – cMin ) + μ (c2) ( c2 – cMin ) + μ (a1) a1
The corresponding FOC for the intertemporal maximization of utility are derived from the
equation above. It is assumed that a solution to the maximization problem exists and that this
solution is the following:
ΨRS = ( c1RS, c2RS, e1RS, e2RS, a1RS )
(A2.4.54)
The above solution defines the optimal levels of utility in period one and two, U1RS and U2RS,
respectively, as well as the maximum level of utility WRS, with repayment after an adverse
economic shock.
If the household defaults on its loan in period one, it loses access to credit in
period two. It will, however, avoid engaging in costly actions in period one. The Kuhn-Tucker
Lagrangian associated with the optimization problem of the household if it chooses to default on
its loan obligation after an adverse economic shock is the following:
109
Γ
= [ U1 ( c1 , e1 ) + β U2 ( c2 , e2)]
(A2.4.55)
k1 = k1*
+ λ1 { z1E YE ( k1* , e1 ) – ( a1 – a0 ) L1 – c1 }
+ λ2 { z2 Y ( k2 , e2 ) + a1 L2 – c2 }
– μ (e1) ( e1 – emax ) – μ (e2) ( e2 – emax )
+ μ (c1) ( c1 – cMin ) + μ (c2) ( c2 – cMin ) + μ (a1) a1
The corresponding FOC for the intertemporal maximization of utility are derived from the
equation above. It is assumed that a solution exists and that this solution is the following:
ΨDS = ( c1DS, c2DS, e1DS, e2DS, a1DS )
(A2.4.56)
The above solution defines the optimal levels of utility in period one and two, U1DS and U2DS,
respectively, and the maximum level of utility WDS, with default after an adverse economic shock.
The household will choose to repay its loan after a shock if WRS > WDS, and
conversely, will default after a shock if WDS > WRS.
110
APPENDIX 2.5
Sharma and Zeller (1997) survey the group-based credit programs of three
Bangladeshi MFIs—the Association for Social Advancement (ASA), the Bangladesh Rural
Advancement Committee (BRAC), and the Rangpur Dinajpur Rural Service (RDRS). They
examine the factors that affect the repayment performance of 128 borrower groups belonging to
the said institutions. Their research uses the following repayment function:
DELIQ = f (LNSIZE, X, Z, M)
(A2.5.1)
where DELIQ is equal to the delinquency rate defined as the proportion of the total loan amount
in arrears at the date when complete repayment was promised, LNSIZE is the loan size, X is a
vector of group characteristics, and M is a vector of lender characteristics. Z is a vector of
community characteristics. The function is specified such that
Lim
DELIQ = 0
(A2.5.2)
LNSIZE  0
It is argued that the preceding specification is a reasonable assumption.
It,
moreover, implies that the effects of X, Z, M on the default rate are conditional on the loan size.
Consequently, when Equation (2) is a linear function, the repayment function interacts X, Z, M
with LNSIZE as follows:
DELIQi* = β1 (LNAMT) + (LNAMT) X β2 + (LNAMT) Z β3
(A2.5.3)
+ (LNAMT) M β2 + ei
where DELIQ* = 0 if DELIQi* ≤ 0 and DELIQi = DELIQi* if DELIQi* > 0. DELIQi* is
thus a latent variable that is observable only when it takes a positive value. Equation (A2.5.3) is
estimated by using the TOBIT maximum likelihood technique.
Based on the results of the estimation, the factors that lower the borrowing group
delinquency rate include the following: (1) the mean level of land owned by the borrowing group,
111
(2) a higher group-wise mean dependency ratio, (3) a greater percentage of group members who
are female and (4) the fact that the borrowing group had formed itself on its own. The factors
that increase the borrowing group delinquency rate include the following: (1) an increasing size
of the loan to the group, (2) a greater proportion of members in the group that are related to each
other, (3) a greater proportion of group members reporting agricultural production as the
principal occupation, (4) a greater number of informal mutual self-help and insurance groups in
the village and (5) the presence of a food-for-work program in the village.
112
APPENDIX 2.6
Gonzalez (2008) studies the over-indebtedness of borrower households in Bolivia
in the 1997-2001 period. The research defines “over-indebtedness" as occurring in the following
three situations: (1) when the borrower is not willing to repay the loan, even if she has the ability
to do so, and consequently, default occurs; (2) when the borrower has to undertake costly
“extraordinary” actions in order to repay the loan, beyond those anticipated at the time when
agreement for the transaction was completed, and (3) when the borrower is willing to repay the
loan, but does not have the ability to do so in full and when agreed, and consequently, arrears,
partial repayment or full default are observed.
1,282 lending relationships with formal lenders in the 1997-2001 period (and
involving 959 borrower households) are the focus of the research analysis. Each of the lending
relationships is classified into one of the following three mutually exclusive groups, according to
associated levels of arrears: (1) those with a perfect repayment record or 0 days of arrears for all
loans in the period, (2) those with arrears of less than 30 days at least once in the period, and (3)
those with arrears of 30 or more days at least once in the period. Furthermore, each lending
relationship is also classified into one of the following two groups, according to whether or not
the concerned borrower household had to engage in costly actions (involving the increase of
labor, the reduction consumption and/ or the sale of assets) in the process of repaying its loans:
(1) those with an active borrower household (which engaged in some costly actions at least once
in the period) and (2) those with an inactive borrower household (which did not engage in any
costly action). By putting together the two ways of classifying lending relationships explained
above, the following six mutually exclusive categories of lending relationships are observed: (A)
those with 0 days of arrears and no costly actions, (B) those with 0 days of arrears and costly
113
actions, (C) those with less than 30 days of arrears and costly actions, (D) those with less than 30
days of arrears and no costly actions, (E) those with 30 or more days of arrears and costly actions,
and (F) those with 30 or more days of arrears and no costly actions. It is to be noted that the only
borrower households who are not over-indebted—according to the research definition of overindebtedness—are those involved in lending relationships of category (A).
Given all the foregoing, two sets of logistic regressions are performed. The first
set analyzes factors that are associated with the borrower household being over-indebted, in terms
of its being willing to repay but lacking sufficient capacity to do so without arrears and/ or
engaging in costly actions in repaying. The dependent variable in these regressions is a dummy
variable that is equal to 1 when the lending relationship belongs to either category (B), (C), (D) or
(E) and 0 when the lending relationship belongs to category (A), with lending relationships
belonging to category (F) excluded from the analysis.
The second set of regressions analyzes
the factors associated with over-indebted borrower households having less willingness to engage
in costly actions in order to repay their loans and being less able to generate extraordinary
repayment capacity when they do. The dependent variable in these regressions is a dummy
variable that is equal to 1 when the lending relationship belongs to category (D) or (E) and 0
when the lending relationship belongs to category (B) or (C), with lending relationships
belonging to category (A) and (F) excluded from the analysis.
The independent variables for both sets of regressions are grouped into the
following four categories:
(1) household experience of shocks, expectations and timing of
events; (2) lender and loan characteristics; (3) household experience with lenders and incentives
to repay and (4) household repayment capacity. Table A2.6.1 below contains a description of the
independent variables that turned out to be significant in the regressions.
114
With reference to Table A2.6.1, the main significant results of the first set of
regressions are the following: (1) Being located in particular cities of the study increases the
probability of over-indebtedness (reflecting the effect of location-specific shocks as well as the
different degrees of competition among MFIs in different locations).
(2) The borrower
household’s experience of an adverse shock increases the probability that it is over-indebted. (3)
An increase in the number of years in which a borrower household receives at least one new loan
from a particular lender decreases the probability that it is over-indebted. (4) An increase in the
number of years to maturity of the last loan provided by a particular lender increases the
probability of over-indebtedness but its effect is small. (5) The borrower household having an
outstanding loan with a commercial lender increases the probability that it is over-indebted. (6)
The only socio-economic variable associated with household repayment capacity that is
significantly related to over-indebtedness is the level of education of the main person in the
household, with an increase in the latter reducing the probability of over-indebtedness.
The main significant results of the second set of regressions are the following: (1)
The identity of the lender in a lending relationship is related to the probability that the concerned
borrower household belongs to the “lower quality” group of over-indebted households. Borrower
households in the “lower quality” group (as viewed from the financial management perspective
of the MFI) are less willing to engage in costly actions and/ or less able to generate extraordinary
repayment capacity when they do. Village banks (which are not assigned a dummy variable in
the regressions) had the lowest probability of having over-indebted households in the “lower
quality” group. Having a relationship with a group lender, individual lender and consumption
lender, respectively, significantly increases, in ascending order, the probability that the overindebted household belongs to the “lower quality” group. (2) The location of the borrower
115
Table A2.6.1
Description of Significant Independent Variables
Used in the Empirical Models of Gonzalez (2008)
Household experience of shocks
Regional dummies
Shocks
Dummies for sector of
main household economic
activity
Dummy variable = 1 if the borrower household is located in a particular
city and 0 otherwise (NOTE: A dummy variable is assigned, respectively,
for four of the five cities in which borrower households of the sample are
located.)
Dummy variable = 1 if the household experienced a shock and there was
an outstanding formal loan during any year in the 1997-2001 period and 0
otherwise
Dummy variable = 1 if the main economic activity in the 1997-2001
period belongs to a particular sector and 0 if the main economic activity
was not stable or belonged to some other sector (NOTE: The three sectors
assigned dummy variables are manufacturing, commerce and services,
respectively.)
Lender and loan characteristics
Individual lender
Group lender
Consumption lender
Commercial lender
Loan term
Dummy = 1 if lending relationship is with one of the four MFIs without a
banking license that operate on the basis of individual lending and 0
otherwise
Dummy = 1 if lending relationship is with one of two regulated MFIs that
operate on the basis of group lending and 0 otherwise
Dummy = 1 if lending relationship is with a consumption lender and 0
otherwise
Dummy = 1 if borrower household has an outstanding loan with a
commercial lender in the 1997-2001 period (for purchase of goods at
particular stores, like hardware stores, warehouses, etc.)
The number of years to maturity of the last loan provided by a particular
lender
Household experience with lenders
Cohort 1997 – cohort 2001
Years with new loans
Default first
Five dummy variables defined on the basis of the year that the household
received its first formal loan in the 1997-2001 period.
The total number of years in which the borrower received at least one new
loan from a particular lender in the 1997-2001 period
Dummy = 1 if the borrower household indicates that, if in trouble, they
would repay the particular lender last and 0 otherwise
Household repayment capacity
Education main person
Measure of the education of the household head or household member that
generated the most household income
Source of basic data: Gonzalez (2008)
household in particular cities of the study increases the probability that it belongs to the “lower
quality” group of over-indebted households.
116
APPENDIX 3.1
The following analysis of household decision-making when there is no economic
shock proceeds from the exposition of the Gonzalez (2008) model in Appendix 2.4 and Chapter 3.
It is assumed in what follows that the optimal level of effort of each household under the four
different repayment/ default scenarios of the Gonzalez (2008) model are all close to the
maximum level of effort emax if not equal to it. Consequently, for each household, e1* is not
significantly different from e1D, so that the simplifying assumption can be made that:
e1* = e1D = e1**
(A3.1.1)
Likewise, e2* is not significantly different from e2D, so that the simplifying assumption can be
made that:
e2* = e2D = e2**
(A3.1.2)
Under the preceding assumptions, differences in the levels of consumption c1*, c2*, c1D and c2D,
respectively, of the household become a function of its accessing a second period loan equal to b 2
if it repays and its setting of the levels of a1* and a1D, respectively.
Let A be the change in the level of the capital stock from period one to two that
maximizes the utility of a household under repayment when there is no economic shock, such
that:
a1* – a0 = A
(A3.1.3)
Given the above, if the household chooses to repay when there is no economic shock, its optimal
levels of consumption in period one and two, respectively, are the following:
c1* = z1 Y ( b1* + a0 , e1** ) – A L1 – b1* ( 1 + r1 )
(A3.1.4)
c2* = z2 Y ( b2 + a0 + A, e2**) + (a0 + A) L2
(A3.1.5)
117
Consequently, the maximum level of intertemporal utility of the household, WR, when there is no
adverse economic shock and it repays its period one loan is:
WR = U1 [ c1* , e1** ] + β U2 [ c2* , e2** ]
(A3.1.6)
where c1* and c2* are defined by Equations (A3.4) and (A3.5), respectively.
Consider that the household, in determining how to maximize its utility if it
chooses to default, contemplates a preliminary level of capital stock at the beginning of period
two equal to a1DA where:
a1DA – a0 = b2 + A
(A3.1.7)
a1DA makes the total amount of productive capital in period two when the household defaults
equal to its optimal total amount when the household repays ( both are equal to [b2 + a0 + A] ). If
the household sets a1 = a1DA, its levels of consumption in period one and two, respectively, would
be the following:
c1DA = z1 Y ( b1* + a0 , e1** ) – ( b2 + A) L1
(A3.1.8)
c2DA = z2 Y ( b2 + a0 + A, e2**) + (a0 + b2 + A) L2
(A3.1.9)
Let the level of intertemporal utility of the household when there is no adverse economic shock
and it defaults on its loan with its level of capital stock at the beginning of period two set at to
a1DA be equal to WDA where:
WDA = U1 [ c1DA , e1**] + β U2 [ c2DA , e2**]
(A3.1.10)
In choosing to repay or default, the household can compare WDA with WR— even
without yet determining a1D, its optimal level of capital stock at the beginning of period two if it
defaults, and WD, its maximum level of intertemporal utility when there is no adverse economic
shock and it chooses to default. Given that WD ≥ WDA, if WDA > WR, or alternatively, if
WDA – WR > 0, the household chooses to default.
118
Let Vc1* ( k ) be the change in household utility as a result of adding k to the value
of c1* and Vc2* ( m ) be the change in household utility as a result of adding m to the value of
c2* such that
Vc1* ( k ) = U1 [ c1* + k , e1** ] − U1 [ c1* , e1** ]
(A3.1.11)
Vc2* ( m ) = U2 [ c2* + m , e2** ] − U2 [ c2* , e2** ]
(A3.1.12)
Based on the foregoing definitions and on Equations (A3.4) to (A3.6) and Equations (A3.8) to
(A3.10), we have the following:
WDA – WR =
{ U1 [ c1DA, e1**]
−
+ β U2 [ c2DA, e2**] }
{ U1 [ c1*, e1**]
(A3.1.13)
+ β U2 [ c2*, e2**] }
= U1 [ c1DA, e1**] − U1 [ c1*, e1**]
β
{ U2 [ c2DA, e2**] −
U2 [ c2*, e2**] }
= U1 [ c1* + b1* (1 + r1) − b2 L1 , e1**] − U1 [ c1*, e1**]
+ β
{ U2 [ c2* + b2 L2 , e2**]
− β U2 [ c2*, e2**] }
= Vc1* [ b1* (1 + r1) − b2 L1 ] + β Vc2* [ b2 L2 ]
The household chooses to default on its loan if:
WDA – WR = Vc1* [b1* (1 + r1) − b2 L1] + β Vc2* [ b2 L2] > 0
(A3.1.14)
Let B be the change in the level of the capital stock from period one to two that
maximizes the utility of a household under default when there is no economic shock, such that:
a1D – a0 = B
(A3.1.15)
Given the above, if the household chooses to default when there is no economic shock, its
optimal levels of consumption in period one and two, respectively, are the following:
c1D = z1 Y ( b1* + a0 , e1**) – B L1
(A3.1.16)
c2D = z2 Y ( a0 + B, e2**) + (a0 + B) L2
(A3.1.17)
119
Consequently, the maximum level of intertemporal utility of the household, WD, when there is
no adverse economic shock and it defaults on its period one loan is:
WD = U1 [ c1D , e1** ] + β U2 [ c2D , e2** ]
(A3.1.18)
where c1D and c2D are defined by Equations (A3.16) and (A3.17), respectively.
Consider that the household, in determining how to maximize its utility if it
chooses to repay, contemplates a preliminary level of capital stock at the beginning of period two
equal to a1*A where:
a1*A – a0 = – b2 + B
(A3.1.19)
a1*A makes the total amount of productive capital in period two when the household repays equal
to its optimal total amount when the household defaults ( both are equal to [a0 + B] ). If the
household sets a1 = a1*A, its levels of consumption in period one and two, respectively, would be
the following:
c1*A = z1 Y ( b1* + a0 , e1** ) – ( – b2 + B ) L1 – b1* ( 1 + r1 )
(A3.1.20)
c2*A = z2 Y ( a0 + B , e2** ) + ( a0 – b2 + B ) L2
(A3.1.21)
Let the level of intertemporal utility of the household when there is no adverse economic shock
and it repays its loan with its level of capital stock at the beginning of period two set at to a1*A be
equal to WRA where:
WRA = U1 [ c1*A , e1**] + β U2 [ c2*A , e2**]
(A3.1.22)
In choosing to repay or default, the household can compare W*A with WD—even
without yet determining a1*, its optimal level of capital stock at the beginning of period two if it
repays, and WR, its maximum level of intertemporal utility when there is no adverse economic
shock and it chooses to repay.
Given that WRA ≤ WR, if WD < WRA, or alternatively, if
WD – WRA < 0, the household chooses to repay.
120
Let Vc1D ( p ) be the change in household utility as a result of subtracting p from
the value of c1D and Vc2D ( q ) be the change in household utility as a result of subtracting m
from the value of c2D such that:
Vc1D ( p ) = U1 [ c1D , e1** ] − U1 [ c1D – p , e1** ]
(A3.1.23)
Vc2D ( q ) = U2 [ c2D, e2** ] − U2 [ c2D – q , e2** ]
(A3.1.24)
Based on the foregoing definitions and on Equations (A3.16) to (A3.18) and Equations (A3.20)
to (A3.22), we have the following:
WD – WRA =
{ U1 [ c1D, e1**]
−
+ β U2 [ c2D, e2**] }
{ U1 [ c1*A, e1**]
(A3.1.25)
+ β U2 [ c2*A, e2**] }
= U1 [ c1D, e1**] − U1 [c1*A, e1**]
β
{ U2 [ c2D, e2**] −
U2 [c2*A, e2**] }
= U1 [ c1D, e1**] − U1 [ c1D − b1* (1 + r1) − b2 L1 , e1**]
+ β
{ U2 [ c2D , e2**]
− U2 [ c2D − b2 L2 , e2**] }
= Vc1D [ b1* (1 + r1) − b2 L1 ] + β Vc2D [ b2 L2 ]
The household chooses to repay its loan if:
WD – WRA = Vc1D [b1* (1 + r1) − b2 L1] + β Vc2D [ b2 L2] < 0
(A3.1.26)
Taken together, Equation (A3.14) and (A3.26) tell us that decision of the
household to repay or default on its loan—while being a function of the values of c1*, c2*, c1D
and c2D—can be determined theoretically by the value of the following parameters: b1*, r1, b2 ,
L1, β and L2. In this regard, when b1* and r1 decrease in value, the decision to repay becomes
more likely. When the value of L1 decreases so that the value of k = L1 / L2 (with 0 < L1 < L2 ≤ 1)
decreases, repayment becomes more likely. When the value of β increases, repayment becomes
121
more likely. An increase in the value of b2 either increases or decreases the probability or
repayment, depending on the relative values k and β.
A theoretical scenario where all households repay their respective loan is
possible—no matter what the level of productivity z and asset endowment a0 of each particular
household may be. This happens when the values of b2 and L1 are high enough while the value of
β is low enough to make WD – WRA < 0 no matter what the value of c1D and c2D are. Conversely,
another theoretical scenario where all households default on their respective loan is also
possible—once again, no matter what the level of productivity z and asset endowment a0 of each
particular household may be. This happens when the values of b2 and L1 are low enough while
the value of β is high enough to make WDA – WR > 0 no matter what the value of c1* and c2* are.
The first scenario would include the case of a household with very low
productivity, which—instead of choosing to default—pays off its period one loan and sells
relatively high-priced assets to augment its consumption in period one. It then uses its second
period loan to restore the level of its capital stock for period two. The second scenario would
include the case of a household with very high productivity, which—instead of choosing to
repay—defaults on its period one loan. It then uses the additional income generated by its first
period loan to buy low-priced assets to augment its capital stock, and hence, its income, in the
second period.
122
APPENDIX 3.2 (A)
PROGRESS OUT OF POVERTY INDEX (PPI) SCORECARD
Figure 1: A simple poverty scorecard for the Philippines
Entity
Member:
Loan officer:
Name
ID
Date (DD/MM/YY)
Joined:
Today:
Household
size:
Branch:
Indicator
1. How many people in the family are aged 0
to 14?
Value
Points
0
4
9
15
20
26
A. Five or more
B. Four
C. Three
D. Two
E. One
F. None
Total
Progress out of Poverty
IndexTM
2. Do all children in the family of ages 6 to
14 go to school?
A. No
B. Yes
C. No children ages 6 to 14
0
2
4
A. Graduate primary or less
B. First- to fourth-year secondary
C. Graduate secondary
D. First-year college or higher, or no
female head/spouse
0
3
6
A Simple Poverty Scorecard for Bolivia
3. What is the education level of the female
head/spouse?
4. Do any family members have salaried
employment?
A. No
B. Yes
5. What are the house’s outer walls made of?
A. Light materials (cogon, nipa, or
sawali, bamboo, anahaw)
B. Strong materials (iron, aluminum,
tile, concrete, brick, stone,
wood, asbestos)
6. What is the house’s roof made of?
11
0
5
A. Light materials (Salvaged,
makeshift, cogon, nipa, or
anahaw)
B. Strong materials (Galvanized iron,
aluminum tile, concrete, brick,
stone, or asbestos)
0
4
0
2
7. What kind of toilet facility does the family
have?
A. None, open pit, closed pit, or other
B. Water sealed
0
7
8. Does the family own a refrigerator?
A. No
B. Yes
0
10
9. How many television sets does the family
own?
A. None
B. One
C. Two or more
0
6
21
10. Does the family own a washing machine?
A. No
0
B. Yes
10
Microfinance Risk Management, L.L.C., http://www.microfinance.com
123
Total score:
APPENDIX 3.2 (B)
TABLE OF ESTIMATED POVERTY LIKELIHOODS
ASSOCIATED WITH PPI SCORES
Figure 4 (National poverty line): Estimated poverty likelihoods
associated with scores
. . . t hen t he l ik eli hood ( %) of being
below t he pov er t y l i ne is:
0–4
96.6
5–9
93.7
10–14
91.5
15–19
87.8
20–24
80.9
25–29
68.5
30–34
59.6
35–39
48.9
40–44
36.8
45–49
21.1
50–54
14.8
55–59
7.2
60–64
5.0
65–69
3.2
70–74
1.4
75–79
1.4
80–84
0.0
85–89
0.0
90–94
1.5
95–100
0.0
Surveyed cases weighted to represent households in t he Philippines.
Based on the 2004 APIS.
I f a househol d' s scor e is . . .
124
For more information visit www.progressoutofpoverty.org
APPENDIX 3.3
According to Grameen Foundation (2014), there are currently more than 200
organizations worldwide that use Grameen Foundation’s Progress out of Poverty Index (PPI).
These organizations—nonprofit organizations, for-profit businesses, investors, networks and
rating agencies, in countries across Africa, Asia, Latin America and the Middle East—work with
poor clients and communities in diverse ways, including providing financial services, providing
health care and conducting research that can benefit the wider anti-poverty community. For them,
the PPI is a statistically-sound yet simple tool used to measure poverty outreach, to assess the
performance of interventions among the poor and poorest and to track poverty levels over time.
There are now 46 country PPI scorecards that have been constructed, and if an expert-based
scorecard for China—constructed using an alternative methodology due to data restrictions—is
included, the PPI now covers countries that are home to 90 % of the world’s poorest people.
In the Philippines, as in many other countries, the PPI is associated primarily with
the microfinance industry. In this regard, Grameen Foundation (2011) describes how the Center
for Agriculture and Development (CARD) Bank is moving towards collecting the PPI score of
every one of its nearly 580,000 clients. CARD Bank in particular hopes to use the PPI to help
generate marketing strategies that promote micro-savings among its clients.
Biggar (2009)
reports on how the Negros Women for Tomorrow Foundation (NWTF)—with its more than
70,000 clients spread across its 37 branches in the Visayas region—has increasingly used the PPI
to target new clients and to adjust its products to serve those clients more effectively.
As explained by Schreiner (2009), the Philippine PPI scorecard is based on data
from the 2004 Annual Poverty Indicators Survey (APIS) conducted by the National Statistics
125
Office (NSO). The 2002 APIS is also used for testing the accuracy of changes in poverty rates
from 2002 to 2004 estimated with the use of the scorecard.
Schreiner (2009) describes how the procedure for making the Philippine PPI
scorecard involved randomly dividing the 42,789 households sampled by the 2004 APIS into
three sub-samples, with every household designated to be part of either: (1) the construction subsample (to be used for selecting poverty indicators and associated scores), (2) the calibration subsample (to be used for associating scores with poverty likelihoods), and (3) the validation subsample (to be used for testing scorecard accuracy with data not used in construction or
calibration). Table A3.3.1 below lists down the sample sizes of construction, calibration and
validation sub-samples, respectively, and associated poverty rates by sub-sample and poverty line.
In the whole PPI scorecard construction process, it is household income (rather
than consumption) that is used as the basis for classifying the household as being below the
poverty line or not. Poverty rates, moreover, are calculated at the household-level, that is, each
household is counted as if it had only one person, regardless of true household size, so that all
households are counted equally. The Philippine PPI scorecard is “calibrated” to eight different
poverty lines. Sample and sub-sample poverty rates in reference to six of the eight poverty lines
are shown in the Table A3.3.1.
Altogether, the eight poverty lines are the following:
(1)
National, (2) Food, (3) USAID “extreme,” (4) USD 1.25/day 2005 PPP, (5) USD 2.50/day 2005
PPP, (6) USD 3.75/day 2005 PPP, (7) USD 5.00/day 2005 PPP and (8) USD 4.32/day 1993 PPP.
For the national and food poverty lines, respectively, the 2004 lines set by the National Statistics
Coordination Board (NSCB) based on the 2003 Family Income and Expenditures Survey (FIES)
were used.
The first phase of constructing the Philippine PPI scorecard began with 60
potential poverty indicators—covering the areas of family composition, education, housing and
126
Table A3.3.1
PPI scorecard construction based on 2004 APIS:
Sample sizes of construction, calibration and validation sub-samples,
and household poverty rates by sub-sample and poverty line
______________________________________________________________________________
% with income below designated poverty line
______________________________________________________________________________
Source: Schreiner (2009)
ownership of durable goods—being screened with the use of the construction sub-sample.
Screening involved calculating for each poverty indicator an “uncertainty coefficient” that
measured how well it predicted poverty on its own (with reference to the national poverty line).
In this regard, Table A3.3.2 lists the poverty indicators with the highest “uncertainty coefficients.”
After the screening process, the scorecard itself was built using the national poverty line and logit
regression on the construction sub-sample. Poverty indicators were selected based on both
statistics and judgment. As outlined by Schreiner (2009), the steps followed were the following:
(1) Logit regression was used to build a scorecard for each candidate indicator. (2) One of the
one-indicator scorecards was chosen based on factors that included improvement in accuracy,
likelihood of acceptance by users, sensitivity to changes in poverty status, variety among
indicators and verifiability. (3) A series of two-indicator scorecards were then built by adding a
second candidate indicator to the one-indicator scorecard that was previously chosen. (4) The
127
Table A3.3.2
PPI scorecard construction based on 2004 APIS:
Poverty indicators with the highest uncertainty coefficients
based on use of construction sub-sample
Source: Schreiner (2009)
best two-indicator was then chosen, again based on statistics and judgment. (5) The preceding
steps were repeated until the scorecard had 10 indicators.
128
The second phase in the construction of the Philippine PPI scorecard used the
calibration sub-sample.
In this phase, a given score was non-parametrically associated
(“calibrated”) with a poverty likelihood by defining the latter as the share of households in the
calibration sub-sample which had the particular score while also having income below a given
poverty line. This method was used to “calibrate” scores with estimated poverty likelihoods for
all the poverty lines. Table A3.3.3 shows the end result of the whole “calibration process” in
terms of the distribution of household poverty likelihoods across income ranges demarcated by
poverty lines. Schreiner (2009) emphasizes how the use of judgment in the construction of the
Philippine PPI scorecard in no way compromises the objectivity of the poverty likelihoods that it
generates. This is because the “calibration” process ensures that the aforementioned poverty
likelihoods are derived from objective survey data and quantitative poverty lines.
The final phase in the construction of the Philippine PPI scorecard was concerned
with testing the accuracy of its estimated poverty rates. To do this, the scorecard was applied to
1,000 bootstrap samples of size n = 16,384 using the validation sub-sample. In each bootstrap
sample, differences in actual vis-à-vis estimated poverty rates across the various poverty lines
and for different sample sizes was calculated. With reference to the national poverty line,
Table A3.3.4 shows the average difference of estimated poverty rates from their respective true
values as these were recorded for different sample sizes with each bootstrap sample. The
associated confidence intervals of the estimated poverty rates for different sample sizes are also
shown.
An important question that can be raised with regard to the use of the Philippine
PPI scorecard in the present research is the following: Beyond the legitimacy it has earned from
its acceptance and wide use by numerous socially-oriented organizations like ASHI, is the
scorecard a valid measure of the poverty likelihood of ASHI borrower households? In this regard,
129
Table A3.3.3
2004 poverty scorecard based on 2004 APIS:
Distribution of household poverty likelihoods across income ranges
demarcated by poverty lines based on the calibration sub-sample
______________________________________________________________________________
______________________________________________________________________________
Source: Schreiner (2009)
a central issue is whether the poverty likelihood estimates of ASHI borrower households
produced by scorecard are statistically “unbiased” (in the sense that in repeated samples from the
same population, the average estimate matches the true poverty likelihood). Schreiner (1990)
notes that as long as the scorecard is applied to the same population from which it was
constructed—so that the relationship between poverty indicators and the poverty situation does
not change—poverty likelihood estimates remain unbiased. But this is precisely the “problem”
that arises with the use of the Philippine PPI scorecard to estimate the poverty likelihood of ASHI
130
Table A3.3.4
Poverty scorecard based on 2004 APIS:
Average difference from the true value and precision of differences
of bootstrapped estimates of poverty rates
(with reference to the national poverty line)
for groups of households at a point in time by sample size
________________________________________________________
Source: Schreiner (2009)
borrower households (as the present research does):
it involves crucial changes in the
relationship between the poverty indicators and the poverty situation in terms of: (1) time and (2)
the population surveyed. In other words, while the Philippine PPI scorecard was constructed
based on data collected from the whole Philippine population in the year 2004, it is being used to
generate estimates for a small sub-group of the Philippine population (= ASHI borrower
households) in the year 2013.
The above “problem” is treated as a data limitation by the present research. It is
argued here that the PPI scores generated by ASHI and other socially-oriented MFIs—despite
their inherent limitations as pointed out above—remain a valid tool for estimating the poverty
likelihood of microfinance borrower households. In the present research, the use of PPI scores is
131
considered a valid empirical methodology for—at very the least—segmenting ASHI borrower
households into broad groups on the basis of differences in average poverty likelihood that are
statistically significant. In other words, PPI scores can be used to identify a “poorer” group of
ASHI borrower households (e.g. those with a PPI score of 24 or lower), whose members can be
considered to have a statistically significant greater average likelihood of being officially poor
when compared to ASHI borrower households belonging to a “more well-off” group (e.g. those
with a PPI score of 45 or more).
132
APPENDIX 3.4
DISTRIBUTION OF ASHI BORROWERS BY PPI SCORE, LOCATION
AND MEMBERSHIP DURATION
(ASHI Borrowers Surveyed Between January and June 2011)
133
APPENDIX 3.5
Data-gathering Procedures
(A) 3-step process in choice of ASHI borrowers for inclusion in the research
sample of borrowers: First, data on the number of members in each of all the borrower centers
of a concerned branch was obtained from the ASHI branch manager, with the said borrower
centers classified by the branch manager into the following three mutually exclusive types of
“very good,” “problem,” and “crisis.” Second, particular borrower centers of each branch were
chosen for inclusion in the data-gathering process, based on (1) the total number of each of three
aforementioned types of centers in the branch, (2) prioritizing centers which had the most
borrower households with data available from the ASHI MIS database, and (3) including centers
of all the ASHI branch loan officers in the data-gathering process. Third, particular borrowers
from each of the chosen centers were selected based on the availability of their relevant
member’s assessment record as these were accessed from the files of the branch. Borrowers on
their first general loan cycle were chosen only if they received their first general loan by October
2013.
(B) 2-stage process with regard to collecting data on the repayment
performance of a borrower household:
The first stage involved accessing the relevant
member’s assessment form from the records of the ASHI branch. In this way, information was
obtained on the repayment performance of a borrower household over a one-year period ending
on the month in the year 2012 or 2013 when its second to the last general loan matured or was
fully paid. The foregoing information was obtained only for borrower households that had
received at least two general loans in the years 2012-2013 and only for months in the
aforementioned one-year period when the borrower household had to make loan obligation
134
payments. The second stage involved interviewing the concerned ASHI loan officer. Through
the latter, information was obtained on the repayment performance of the concerned borrower
household over the period consisting of all the months in the six-month period from June to
November 2013 in which the borrower had to make loan obligation payments excluding all the
months already covered by the first stage.
(C) 2012-2013 repayment observation period of the borrower household: All
the months in which information on the repayment performance of a borrower household was
obtained as explained above constitutes the 2012-2013 repayment observation period of the said
borrower household.
(D) Division of the 2012-2013 repayment observation period of the borrower
household into segments:
The 2012-2013 repayment observation period of the borrower
household is divided into segments (consisting of a one to six-month periods each) defined as
follows.
The first segment consists of all the months in the six-month period from June to
November 2013 in which the borrower had to make loan obligation payments. The first segment
excludes all months that are part of the second segment as the latter is defined below.
If the borrower has received at least two general loans in the 2012-2013 period, the
second segment is the six-month period that ends on the month when the second to the last
general loan of the borrower matured (i.e. was fully paid) and begins on the 5 th month before the
aforementioned month. If the borrower has received only one general loan, there is no second
segment.
135
There is a third segment only if there is a second segment. The third segment
consists of all the months in which the borrower had to make loan obligation payments in the six
month period immediately preceding the second segment.
(E) Calculation of the “group_particip” variable: For each borrower center of
the branch, data on borrower member attendance in center meetings was provided by the branch
manager. For the Rizal East branch (REB), data for the months of November 2013, May 2013
and November 2012 was provided. For the Rizal Southwest branch (RSW), data for the months
from September to October 2013 was provided. For the Rizal West branch (RWB), data for the
months of September, October and November 2013 was provided. The “group_particip” variable
is equal to the average value—over the months for which data was provided—of the “attendance
rating” given by ASHI branch managers to the respective borrower centers in their branch. As
computed by the branch managers, the borrower center “attendance rating” is equal to the
percentage value for a particular month of the ratio AA / AE. AA is “actual borrower attendance in
center meetings” computed by adding together the actual number of borrowers present in each
borrower center weekly meeting over a particular month. AE is “expected borrower attendance in
center meetings” and is equal to the total number of borrower members of the center multiplied
by the number of meetings held by the center in a particular month.
136
APPENDIX 4
Table A4.1
Logistic regression results for borrowers of three ASHI Rizal branches
(Results with use of the “rpcap_34” variable)
Explanatory
variable
Coefficient
rpcap_34
lend_memyrs
luse_maxloan
luse_minratio
shock_yagri
shock_yemploy
shock_agri
shock_retoth
shock_ms
group_particip
obs_months
cons
.2523431
-.0734679
-.0001352
-.072566
-.2689482
.2051053
-1.142298
-.347866
-.4901079
-.0219163
.111405
-.2611304
*
*
**
***
Std. Err.
z
P>z
.3037911
.0441744
.0001411
.4702976
.7686782
.5190063
.6023307
.3193497
.453583
.0098379
.0339946
1.076235
0.83
-1.66
-0.96
-0.15
-0.35
0.40
-1.90
-1.09
-1.08
-2.23
3.28
-0.24
0.406
0.096
0.338
0.877
0.726
0.693
0.058
0.276
0.280
0.026
0.001
0.808
[95% Conf.Interval]
-.3430765
-.1600482
-.0004117
-.9943324
-1.77553
-.8121284
-2.322844
-.9737799
-1.379114
-.0411982
.0447768
-2.370512
.8477627
.0131123
.0001413
.8492004
1.237633
1.222339
.0382485
.278048
.3988984
-.0026345
.1780332
1.848251
Number of obs. = 404, Dependent variable = repaydifrizal, Log likelihood = - 168.77369
LR chi2(12) = 28.64, Prob > chi2 = 0.0026, Pseudo R2 = 0.0782
*, ** and *** denote significance at the 10%, 5% and 1% level, respectively
Table A4.2
Logistic regression results for borrowers of three ASHI Rizal branches
(Results with use of the “rpcap_29” variable)
Explanatory
variable
Coefficient
rpcap_29
lend_memyrs
luse_maxloan
luse_minratio
shock_yagri
shock_yemploy
shock_agri
shock_retoth
shock_ms
group_particip
obs_months
_cons
.4142346
-.0747183
-.0001403
-.0770574
-.2375406
.2118918
-1.191643
-.3561575
-.5168958
-.0226625
.1139904
-.2048578
*
*
**
***
Std. Err.
z
P>z
.3565393
.0442394
.0001405
.4698131
.7739433
.5192421
.6093894
.3192553
.4552158
.0098903
.0341808
1.065022
1.16
-1.69
-1.00
-0.16
-0.31
0.41
-1.96
-1.12
-1.14
-2.29
3.33
-0.19
0.245
0.091
0.318
0.870
0.759
0.683
0.051
0.265
0.256
0.022
0.001
0.847
[95% Conf.Interval]
-.2845696
-.1614259
-.0004157
-.9978742
-1.754442
0-.805804
-2.386024
-.9818865
-1.409102
-.0420471
.0469973
-2.292262
Number of obs. = 404, Dependent variable = repaydifrizal, Log likelihood = - 168.46339
LR chi2(12) = 29.26, Prob > chi2 = 0.0021, Pseudo R2 = 0.0799
*, ** and *** denote significance at the 10%, 5% and 1% level, respectively
137
1.113039
.0119893
.0001351
.8437593
1.279361
1.229587
.0027382
.2695714
.3753108
-.0032778
.1809835
1.882546
Table A4.3
Summary statistics for data of borrowers of ASHI Rizal Southwest branch (RSW)
Variable
No. of
Obs
Mean
Std. Dev.
Min.
Max.
repaydif_rsw
rpcap_ppi
rpcap_39
rpcap_34
rpcap_29
lend_memyrs
167
167
167
167
167
167
.1736527
41.98204
.4850299
.3113772
.1976048
4.994012
.3799498
14.13382
.5012789
.4644493
.3993899
4.349902
0
9
0
0
0
1
1
75
1
1
1
16
luse_maxloan
luse_minratio
shock_yagri
shock_yemploy
shock_agri
167
167
167
167
167
1811.737
.6774251
.0958084
.0838323
.1796407
1057.089
.3777469
.2952135
.2779697
.3850425
417
0
0
0
0
5917
1
1
1
1
shock_retoth
shock_ms
group_particip
obs_months
167
167
167
167
.3353293
.1916168
74.83832
11.21557
.4735253
.3947568
10.93977
4.899127
0
0
54
1
1
1
93
18
Table A4.4
Logistic regression results for borrowers of ASHI Rizal Southwest Branch (RSW)
(Estimation using “rpcap_ppi” variable)
Explanatory
variable
Coefficient
rpcap_ppi
lend_memyrs
luse_maxloan
luse_minratio
shock_yagri
shock_yemploy
shock_agri
shock_retoth
shock_ms
group_particip
obs_months
_cons
-.0617506
-.0823039
-.0006562
-.3543814
8788217
-1.20652
-1.991295
-.5987587
.1405514
.0262755
.1750937
-.9079004
***
**
*
***
Std. Err.
z
P>z
.0205368
.0659847
.000326
.7064962
1.238408
.9693794
1.10357
.5624062
.6537812
.0209903
.063314
1.963299
-3.01
-1.25
-2.01
-0.50
0.71
-1.24
-1.80
-1.06
0.21
1.25
2.77
-0.46
0.003
0.212
0.044
0.616
0.478
0.213
0.071
0.287
0.830
0.211
0.006
0.644
[95% Conf.Interval]
-.102002
-.2116316
-.0012952
-1.739089
-1.548413
-3.106469
-4.154253
-1.701055
-1.140836
-.0148647
.0510005
-4.755895
-.0214992
.0470237
-.0000172
1.030326
3.306057
.6934282
.1716627
.5035373
1.421939
.0674157
.2991869
2.940094
Number of obs. = 167, Dependent variable = repaydifrsw, Log likelihood = - 64.741292
LR chi2(11) = 24.70, Prob > chi2 = 0.0101, Pseudo R2 = 0.1602
*, ** and *** denote significance at the 10%, 5% and 1% level, respectively
138
Table A4.5
Logistic regression marginal effects for borrowers of ASHI Rizal Southwest Branch (RSW)
(Estimation using “rpcap_39” variable)
Explanatory
variable
dy/dx
rpcap_39
lend_memyrs
luse_maxloan
luse_minratio
shock_yagri
shock_yemploy
shock_agri
shock_retoth
shock_ms
group_particip
obs_months
.1512351
-.0092712
-.0000714
-.0512947
.0729934
-.1186717
-.1984245
-.0660273
.0010577
.0034884
.0168907
Std. Err.
***
.0541271
.0073654
**
.000035
.0787577
.1340436
.1085803
*
.1151886
.063097
.0737423
.0023303
*** .0065257
z
P>z
2.79
-1.26
-2.04
-0.65
0.54
-1.09
-1.72
-1.05
0.01
1.50
2.59
.005
0.208
0.042
0.515
0.586
0.274
0.085
0.295
0.989
0.134
0.010
Mean
value
[95% Conf.Interval]
.0451479
-.0237071
-.0001401
-.205657
-.1897273
-.3314853
-.4241899
-.1896952
-.1434746
-.0010789
.0041006
.2573223
.0051646
-2.74e-06
.1030675
.3357141
.0941418
.027341
.0576405
.1455901
.0080556
.0296808
.4850299
4.994012
1811.737
.6774251
.0958084
.0838323
.1796407
.3353293
.1916168
74.83832
11.21557
*, ** and *** denote significance at the 10%, 5% and 1% level, respectively
Table A4.6
Summary statistics for data of borrowers of ASHI Rizal East branch (REB)
Variable
No. of
Obs
Mean
Std. Dev.
Min.
Max.
repaydif_reb
repay_ppi
rpcap_39
rpcap_34
rpcap_29
lend_memyrs
131
131
131
131
131
131
.1145038
42.54198
.4427481
.3282443
.1832061
4.687023
.3196445
15.50943
.4986182
.4713768
.3883204
4.591102
0
13
0
0
0
1
1
82
1
1
1
20
luse_maxloan
luse_minratio
shock_yagri
shock_yemploy
shock_agri
131
131
131
131
131
2074.099
.6651145
.221374
.1832061
.2900763
1460.374
.3585191
.4167655
.3883204
.4555394
813
0
0
0
0
9250
1
1
1
1
shock_retoth
shock_ms
group_particip
obs_months
131
131
131
131
.3740458
.1526718
89.72519
11.28244
.485733
.3610515
8.522417
5.226081
0
0
65
1
1
1
100
18
139
Table A4.7
Logistic regression results for borrowers of ASHI Rizal East Branch (REB)
(Estimation with use of the “rpcap_ppi” variable)
Explanatory
variable
Coefficient
Std. Err.
z
P>z
rpcap_ppi
lend_memyrs
luse_maxloan
luse_minratio
shock_yagri
shock_yemploy
shock_agri
shock_retoth
shock_ms
group_particip
obs_months
cons
-.0181295
-.1386463
.0004032
.3670393
-.6712285
-.1149379
-.7887043
1.125968
-.0174146
-.0842715
.1703742
3.005108
.0220442
.1124307
.0002916
1.107465
1.356741
1.014926
1.444965
.8422808
1.256041
.0398484
.0914913
3.796636
-0.82
-1.23
1.38
0.33
-0.49
-0.11
-0.55
1.34
-0.01
-2.11
1.86
0.79
0.411
0.218
0.167
0.740
0.621
0.910
0.585
0.181
0.989
0.034
0.063
0.429
*
**
*
[95% Conf.Interval]
-.0613354
-.3590065
-.0001683
-1.803552
-3.330391
-2.104156
-3.620785
-.524872
-2.47921
-.162373
-.0089454
-4.436162
.0250763
.0817138
.0009747
2.53763
1.987934
1.87428
2.043376
2.776808
2.44438
-.0061699
.3496939
10.44638
Number of obs. = 404, Dependent variable = repaydifrizal, Log likelihood = - 33.868755
LR chi2(11) = 25.49, Prob > chi2 = 0.0077, Pseudo R2 = 0.2734
*, ** and *** denote significance at the 10%, 5% and 1% level, respectively
Table A4.8
Logistic regression results for borrowers of ASHI Rizal East Branch (REB)
(Estimation with use of the “rpcap_29” variable)
Explanatory
variable
Coefficient
rpcap_29
lend_memyrs
luse_maxloan
luse_minratio
shock_yagri
shock_yemploy
shock_agri
shock_retoth
shock_ms
group_particip
obs_months
_cons
1.83179
-.1777691
.0005125
.1015673
-.6653914
-.073344
-1.055823
1.204669
-.6781252
-.0997816
.1656133
3.28397
**
*
**
*
Std. Err.
z
P>z
.8386123
.1275171
.0003067
1.161112
1.413899
1.017783
1.527549
.8769315
1.465616
.0426585
.0970763
3.930856
2.18
-1.39
1.67
0.09
-0.47
-0.07
-0.69
1.37
-0.46
-2.34
1.71
0.84
0.029
0.163
0.095
0.930
0.638
0.943
0.489
0.170
0.644
0.019
0.088
0.403
[95% Conf.Interval]
.18814
-.4276981
-.0000885
-2.17417
-3.436582
-2.068163
-4.049763
-.5140848
-3.55068
-.1833906
-.0246527
-4.420366
3.47544
.0721598
.0011136
2.377305
2.105799
1.921475
1.938117
2.923423
2.194429
-.0161725
.3558793
10.98831
Number of obs. = 131, Dependent variable = repaydifreb, Log likelihood = - 31.684102
LR chi2(12) = 29.86, Prob > chi2 = 0.0017, Pseudo R2 = 0.3203
* and ** denote significance at the 10% and 5% level, respectively
140
Table A4.9
Summary statistics for data of borrowers of ASHI Rizal West branch (RWB)
Variable
No. of
Obs
Mean
Std. Dev.
Min.
Max.
repaydif_rwb
repay_ppi
rpcap_39
rpcap_34
rpcap_29
lend_memyrs
106
106
106
106
106
106
.2264151
42.84906
.4339623
.3113208
.1886792
4.179245
.420499
16.93337
.4979743
.4652333
.3931123
3.161667
0
4
0
0
0
1
1
89
1
1
1
13
luse_maxloan
luse_minratio
shock_yagri
shock_yemploy
shock_agri
106
106
106
106
106
2249.208
.7398113
0
.0471698
.0377358
1642.477
.3026297
0
.2130091
.191462
500
0
0
0
0
7894
1
0
1
1
shock_retoth
shock_ms
group_particip
obs_months
106
106
106
106
.5188679
.1037736
74.56604
10.75472
.5020175
.3064154
17.92525
6.266785
0
0
44
1
1
1
100
18
Table A4.10
Logistic regression results for borrowers of ASHI Rizal West Branch (RWB)
Explanatory
variable
Coefficient
Std. Err.
z
P>z
rpcap_ppi
lend_memyrs
luse_maxloan
luse_minratio
shock_yemploy
shock_agri
shock_retoth
shock_ms
group_particip
obs_months
cons
.0169532
.1075203
.0002556
1.037623
1.072727
1.342438
.5666399
1.187151
.0170929
.0591289
2.0725
0.27
0.10
-1.73
-0.35
1.60
0.26
-1.38
-1.34
-1.90
1.80
0.63
0.786
0.921
0.084
0.728
0.109
0.797
0.167
0.181
0.057
0.071
0.532
-.0286231
-.2000309
-.0009421
-2.394593
-.3809274
-2.28569
-1.894347
-3.914142
-.0660598
-.0093048
-2.765378
*
*
*
[95% Conf.Interval]
.0378324
.2214411
.0000598
1.672813
3.824083
2.97657
.3268411
.7394026
.000943
.2224763
5.358673
0.27
0.10
-1.73
-0.35
1.60
0.26
-1.38
-1.34
-1.90
1.80
0.63
Number of obs. = 106, Dependent variable = repaydifrwb, Log likelihood = - 48.597805
LR chi2(10) = 16.20, Prob > chi2 = 0.0939, Pseudo R2 = 0.1429
* denotes significance at the 10% level
141
Download