theses.frAIX-MARSEILLE UNIVERSITÉ aFculté des Sciences Economiques et de Gestion Ecole Doctorale...

AIX-MARSEILLE UNIVERSITÉ

Faculté des Sciences Economiques et de Gestion

Ecole Doctorale de Sciences Economiques et de Gestion d'Aix-Marseille n�372

Année 2014 Numéro attribué par la bibliothèque

| | | | | | | | | | | |

Thèse pour le Doctorat en Sciences Economiques

Présentée et soutenue publiquement

le 18 Septembre 2014

Anastasia Cozarenco Lock

��

Essays on Microfinance in Developed Countries:

The Role of Business Training, Information, and

Regulation��

Directeurs de Thèse

M. Dominique Henriet

Professeur à l'Ecole Centrale Marseille

M. Renaud Bourlès

Maître de Conférence à l'Ecole Centrale Marseille

Jury

Rapporteurs

M. Robert Lensink Professeur à l'Université de Groningen et à

l'Université de Wageningen

M. Thierry Magnac Professeur à Université Toulouse 1

Examinateurs

Mme. Habiba Djebbari Professeur associé à l'Université d'Aix Marseille

Mme. Ariane Szafarz Professeur à l'Université Libre de Bruxelles

L'Université d'Aix-Marseille n'entend ni approuver, ni désapprouver les opinions particulièresdu candidat: ces opinions doivent être considérées comme propres à leur auteur.

To my parents, Svetlana and Valeriu

Acknowledgments

Accomplishing this PhD would not be possible without the intellectual guidance and support of

my PhD advisors Renaud Bourlés and Dominique Henriet. I have learnt many things from them

and it was a privilege for me to do this research with them. I am deeply indebted to Renaud

Bourés for his scienti�c and moral support and for the time he has dedicated to guide me during

my thesis.

My deep gratitude goes to the members of my thesis committee - Habiba Djebbari, Robert

Lensink, Thierry Magnac, and Ariane Szafarz - for accepting being part of this committee in-

spite of their busy agenda. I strongly appreciate their constructive feedback and comments that

contributed to the improvement of my thesis.

I would like to express my sincere gratitude to Ariane Szafarz for the opportunity to do a research

visit at the Centre for European Research in Micro�nance (CERMi), in Brussels. Working with

her was a nourishing experience that has inspired my research.

I am profoundly indebted to Xavier Joutard for his technical and intellectual support in the

econometric exercises of this thesis. I was very fortunate to be able to improve my econometric

skills working with him.

My thoughts also go to Gilbert Cette who has introduced me to research during my internship

at the Bank of France. I thank him for giving me this unique chance that has contributed to my

decision to do a PhD.

I thank Mohamed Belhaj, Sebastian Bervoets, Yan Bramoullé, Olivier Chanel, Pierre Philippe

Combes, Bruno Decreuse, Fréderic Deroian, Mathieu Faure, Cecilia Garcia-Penalosa, Patrick

Pintus, Juliette Rouchier, Marc Sangnier, Christian Schluter, Tanguy Van Ypersele for valuable

discussions and advices they gave me during various stages of my thesis.

I thank the direction of AMSE, GREQAM and Doctoral School N372 for the �nancial support

they provided to me for participation in academic conferences, summer schools and my research

visit at CERMi. I also thank the administrative and technical team of AMSE and GREQAM -

in particular, Agnès, Aziza, Bernadette, Carole, Grégory, Gérald, Isabelle, Mathilde, Yves - for

the logistic support and help they were providing to me throughout these four years. A special

thought goes to Bernadette for her backup during the �nal administrative stage of this thesis.

I would like to thank my colleagues and friends at the GREQAM (in alphabetical order): Amal,

Anne-Sarah, Antoine (Bonlieu), Antoine (Le Riche), Anwar, Aziz, Bilel, Camila, Clémentine,

Daria, Emma, Eric, Florent, Gilles, Jean-Baptiste, Jian, Joao, Kalila, Kadija, Karine, Lise,

Maame Esi, Manel, Morgane, Moustapha, Nariné, Natasha, Nicholas (Sheard), Nicolas (Abad),

Nicolas (Caudal), Martha, Maty, Marion, Pauline, Qays, Sandrine, Sarra, Thanh, Thomas, Vin-

cent, Vivien, Waqar, and Zakaria. I wish them all the best in their future work.

VII

I have a special thought for my friend Natasha for her moral support during this experience

that we have started together. I was very lucky to be surrounded by my friends Karine, Marion,

and Martha during these four years. Their presence, help, and good mood made this experience

particularly joyful.

I thank my mother for her priceless help and encouragement especially during these last six

months.

Last but not least, I deeply thank my husband's support and patience during these four years.

He has always been present for me, by encouraging me and giving me strength to persevere. I

have a loving thought for Matthias who �lls my days with joy and motivation since his birth.

Abstract

This thesis is organized in four chapters dealing with micro�nance in the developed countries.

Chapters one, two and three address this topic through the lens of theoretical modeling. Chap-

ters two, three and four provide empirical assessment.

Chapter 1 analyses how various forms of state intervention can impact micro�nance institutions'

lending behavior. Using a simple model where entrepreneurs receive individual uncollateralized

loans, we show that, not surprisingly, state intervention through the loan guarantee increases

the number of entrepreneurs receiving a loan. However, after modeling business development

services provided by the micro�nance institution, we show that the loan guarantee can have a

counterproductive e�ect by reducing the number of entrepreneurs bene�ting from such services.

We therefore analyze an alternative policy: business development services subsidization. We

show that if business development services are e�cient enough and are targeted toward less per-

forming borrowers then - for �xed government expenditures - such subsidies do better in terms of

�nancial inclusion than the loan guarantee. Moreover, we argue that - under similar conditions

- business development services subsidization alone does better in terms of �nancial inclusion

than a mix of policies.

Chapter 2 analyses how decisions of a micro�nance institution (MFI) on business training (or

help) provision can impact borrowers' behavior. In the theoretical model we assume a situation

where the MFI - through help - and the borrower - through e�ort - can act on the probability of

borrower's project to succeed. We show that, in contexts where under symmetric information the

MFI optimally helps more the borrower with lower probability of success, superior information

(when the MFI has better information on probability of success than the borrower) can lead the

MFI not to help (or to help less) riskier borrowers. In this last case, because of a "looking-glass

self" e�ect, MFI's choice of help impacts borrower's belief about his risk. We then test this pre-

diction using data from a French MFI. By means of empirical models, taking into account both

the credit-granting and the training-granting processes, we analyse how training programs are

assigned to di�erent borrowers. Con�rming our theoretical reasoning, we �nd a non-monotonic

relationship between the MFI's decision to help and the risk of micro-borrowers. The probability

to be helped appears to increase with risk for low-risk borrowers and to decrease with risk for

high-risk borrowers.

IX

In most developed countries, regulators have imposed loan ceilings to subsidized micro�nance

institutions (MFIs). Micro-entrepreneurs in need of above-ceiling loans are left with the co-

�nancing option, which means securing the above-ceiling share of the loan with a regular bank,

and getting a ceiling-high loan from the MFI. Co-�nancing is attractive to MFIs because it

allows them to free-ride on the regular banks' screening process. Therefore, loan ceilings can

have the perverse e�ect of facilitating the co-�nancing of large projects at the expense of micro-

entrepreneurs who need below-ceiling loans only. This is the gist of our theoretical model in

chapter 3. We test the predictions of this model by exploiting the natural experiment of a French

MFI that became subject to the French EUR 10,000 loan ceiling in April 2009. Di�erence-in-

di�erences probit estimations con�rm that imposing loan ceilings to MFIs can have unexpected

and socially harmful consequences.

Chapter 4 compares the loans granted to male and female entrepreneurs by a French micro�-

nance institution (MFI). We use data before and after the MFI implemented France's regulatory

EUR 10,000 loan ceiling. In the �rst period, the MFI granted loans without bank co-�nancing,

and we �nd that the MFI selected women with larger requested amounts, corresponding to

more ambitious projects. In the second period, the institution started co-�nancing above-ceiling

projects with mainstream banks. Under ceiling enforcement the MFI no longer selected women

with larger requested amounts. Our �ndings suggest that co-�nancing has led an originally

positively oriented MFI to give up some control of its loan allocation decision to mainstream

banks.

Key words: Micro�nance, microcredit, business development services, regulation, mission drift,

developed countries

Résumé

Cette thèse est composée de quatre chapitres sur la micro�nance dans les pays développés. Les

chapitres un, deux et trois présentent des contributions théoriques. Les chapitres deux, trois et

quatre constituent des contributions empiriques.

Le premier chapitre analyse comment les di�érentes formes d'intervention publique impactent

le processus d'octroi des microcrédits. Dans un modèle théorique simple nous considérons des

entrepreneurs qui reçoivent des prêts individuels sans exigence de collatéral. Nous montrons que

l'intervention publique via la garantie des prêts augmente le nombre d'entrepreneurs �nancés.

Néanmoins, après l'introduction de l'accompagnement professionnel par l'institution de micro-

�nance (IMF), nous montrons que la garantie des prêts peut avoir un e�et contreproductif en

réduisant le nombre d'entrepreneurs béné�ciant de l'accompagnement. Par conséquent, nous

nous intéressons à une autre forme d'intervention publique consistant dans le subventionnement

de l'accompagnement professionnel. Nos résultats suggèrent que si l'accompagnement est e�cace

et cible les entrepreneurs les moins performants alors, à dépense publique égale, le subvention-

nement de l'accompagnement est un meilleur outil contre l'exclusion �nancière comparé à la

garantie des prêts.

Le deuxième chapitre étudie comment les décisions d'une IMF concernant l'accompagnement

peuvent impacter le comportement des emprunteurs. Premièrement, dans un modèle théorique,

nous considérons une situation dans laquelle une IMF à travers l'accompagnement et l'emprunteur

à travers son e�ort peuvent in�uencer la probabilité de succès du projet. Nous considérons

le cas où en présence d'information symétrique (situation où l'IMF et l'emprunteur connais-

sent toute l'information pertinente) l'IMF accompagne plus les emprunteurs les plus risqués.

Nous montrons que, dans ce contexte, lorsque l'IMF possède une meilleure information que

l'emprunteur sur son risque de défaut, elle peut ne pas accompagner les entrepreneurs les plus

risqués. Dans cette dernière situation, la décision d'accompagnement de l'IMF impacte les croy-

ances de l'entrepreneur sur son risque via l'e�et "miroir" (looking glass-self). Nous testons cette

prédiction du modèle théorique en utilisant les données d'une IMF française. Les résultats em-

piriques con�rment les intuitions théoriques : nous trouvons une relation non-monotone entre la

décision de l'IMF d'accompagner un micro-entrepreneur et son risque de défaut. La probabilité

de recevoir un accompagnement est croissante avec le risque de l'entrepreneur pour les individus

les moins risqués et décroissante avec le risque de l'entrepreneur pour les individus les plus risqués.

XI

Dans la majorité des pays développés, les régulateurs ont imposé aux IMFs des seuils sur le mon-

tant maximal de microcrédit. Les micro-entrepreneurs ayant besoin d'un �nancement supérieur

à ce seuil peuvent demander un co-�nancement bancaire. Le co-�nancement est pro�table pour

les IMFs car il leur permet de béné�cier du processus de sélection bancaire. Par conséquent,

les seuils peuvent avoir des e�ets pervers facilitant le �nancement des projets plus importants

au détriment des petits projets qui ne nécessitent pas de co-�nancement. C'est l'idée principale

de notre modèle théorique dans le troisième chapitre. Nous testons les prédictions de notre

modèle en exploitant une expérience naturelle d'une IMF française qui est devenue réglementée

en Avril 2009. Nos estimations probit di�érence-en-di�érences con�rment que l'introduction du

seuil de 10,000 EUR a des conséquences inattendues et socialement indésirables.

Le quatrième chapitre compare les prêts octroyés aux entrepreneurs hommes et femmes par

une IMF française. La période d'observation couvre deux périodes : avant et après l'introduction

du seuil réglementaire de 10,000 EUR sur les microcrédits octroyés par l'IMF. Dans la première

période, l'IMF ne co-�nance pas les projets avec les banques commerciales et l'IMF choisit les

femmes avec les demandes plus élevées qui correspondent aux projets plus ambitieux. Dans la

seconde période, l'IMF co-�nance les projets au-dessus du seuil réglementaire avec les banques

commerciales. Avec l'introduction du seuil l'IMF abandonne le choix des femmes avec les deman-

des les plus élevées. Nos résultats suggèrent que le co-�nancement a conduit l'IMF initialement

orientée positivement envers les femmes de céder une partie de sa décision d'allocation des crédits

aux banques classiques.

Mots clés: Micro�nance, microcrédit, accompagnement entrepreneurial, régulation, éloignement

de la mission, pays développés

Introduction in French

La micro�nance regroupe un ensemble de services �nanciers tels que le crédit, l'épargne, l'assu-

rance, les transferts d'argent, dediÃ c©s aux populations n'ayant pas accès aux services bancaires

classiques. Ces services �nanciers peuvent être complétés par des outils non-�nanciers consistant

en l'accompagnement ou le training des microemprunteurs sur le plan entrepreneurial ou person-

nel. L'objectif de cette thèse est d'analyser le secteur de la micro�nance dans les pays développés

qui, à contrario des pays en développement, reste peu étudiée. Cela s'explique notamment par

la jeunesse de ce secteur et par un accès réduit aux données. Cette thèse, composée de quatre

chapitres, présente une contribution innovante à cette question de recherche en se focalisant plus

particulièrement sur le rôle de l'accompagnement entrepreneurial, de l'information sur le marché

du microcrédit et de la régulation de ce jeune secteur �nancier.

Avant de se focaliser sur le continu des quatre chapitres, il est utile de décrire le contexte de la

micro�nance dans les pays développés. Ce secteur occupe une position modeste en termes du

nombre des clients face à l'impressionnante échelle caractérisant les pays plus pauvres. Notam-

ment, à la �n de l'année 2011, cette industrie comptait 189.5 millions clients dans les pays en

développement et 5.5 millions de clients dans les pays industrialisés selon Reed (2013). Malgré

cette importante di�érence de taille, l'objectif général de la micro�nance dans les deux types

d'économies reste la réduction de la pauvreté et l'inclusion sociale. Toutefois, les moyens em-

ployés pour parvenir à ces �ns di�èrent de plusieurs points de vu.

Le microcrédit et l'accompagnement des microentrepreneurs sont deux services piliers de la mi-

XIII

XIV Introduction in French

cro�nance européenne.1 Tandis que le microcrédit sera scruté sous ses di�érents aspects dans tous

les chapitres de cette thèse, les chapitres un et deux se focaliseront plus particulièrement sur

l'accompagnement entrepreneurial. Le microcrédit et l'accompagnement présentent également

deux moyens pour signaler l'e�ort d'un auto-entrepreneur. Dans le cas de l'accompagnement,

l'e�ort est signalé par la présence pendant les séances de training et l'avancement réussi d'une

séance à l'autre. Dans le cas du microcrédit l'e�ort est signalé par le remboursement à temps des

versements dus (Schreiner and Morduch 2001). Dans ce qui suit, nous analysons davantage ces

deux outils ainsi que leurs particularités dans les di�érents types d'économies.

La Commission Européenne donne la dé�nition suivante du microcrédit :

Le microcredit en Union Européenne (UE) est un prêt d'un montant inférieur à 25,000 euros.

Il est destiné aux micro-entreprises avec moins de 10 salariés (91% des toutes les entreprises

européennes), aux individus au chômage ou inactifs qui souhaitent créer leur propre emploi mais

n'ont pas d'accès au crédit bancaire traditionnel (European Commission 2009).

Jusqu'à récemment les données concernant le microcrédit en Europe étaient très peu disponibles.

La contribution de Bendig et al. (2012) est particulièrement appréciée. Selon les auteurs le nombre

des IMFs en Europe (autres que les unions de crédit et les banques commerciales) se situe entre

500 et 700. Leur étude analyse les réponses de 154 IMFs provenant de 32 pays Européens et

fournit une image de la maturité du secteur de micro�nance. En 2011 les IMFs ayant participé à

l'étude ont octroyé 204,080 crédits (122,370 dans les états membres de l'Union Européenne). A

titre de comparaison, Mix Market2 récence 1,421 IMFs dans le reste du monde avec 96 353 872

emprunteurs en 2011. Le montant moyen octroyé par emprunteur s'élevait à 5,135 euros (7,129

euros pour les pays membres de l'UE). L'encours moyen s'élevant à 1,879 dollars en 2011 dans

le reste du monde.

Concernant les méthodes déployées, le modèle de microcrédit dans les pays en développement a

connu une très forte popularité grâce à la méthodologie d'octroie de crédit collectif. Cependant,

1Selon Bendig et al. (2012) la microassurance et la microépargne sont présentes à la marge en Europe.2MIX est une source d'informations objectives et véri�ées sur les IMFs partout dans le monde. Pour

plus d'informations voir mixmarket.org.

Introduction in French XV

cette méthode de prêt est peu adaptée aux les pays plus riches. Cela s'explique par la faiblesse

du capital social dans ce type d'économies, où les liens sociaux sont plus faibles et les niveaux

d'individualisation plus forts, et par les coûts d'opportunité de participation plus élevés (par

exemple, en termes du temps dédié aux rencontres collectives). Par conséquent dans cette thèse

nous modélisons le crédit individuel.

Outre le microcrédit, l'accompagnement, ou le training des micro-emprunteurs constitue un ser-

vice essentiel de la micro�nance dans les pays développés. En e�et, le microcrédit en soi n'est

pas une source unique du succès entrepreneurial (De Mel et al. 2008). L'accompagnement des

emprunteurs fait référence à une o�re très variée des services non-�nanciers qui complètent le

dispositif de microcrédit. Dans les pays en développement, le training est mis en place d'une

façon moins formelle. Il consiste souvent en programmes de développement social di�usant des

informations sur la santé, les responsabilités civiles et droits, les règles de fonctionnement d'une

banque (McKernan 2002). Dans les pays développés, l'accompagnement se focalise le plus sou-

vent sur l'aspect entrepreneurial tel que le développement d'un projet d'entreprise (l'étude de

pro�tabilité, la dé�nition d'une stratégie commerciale et besoins de �nancement, l'aide admi-

nistrative), l'information et l'aide pour l'obtention du �nancement, le training en comptabilité,

gestion, marketing, partie juridique et le suivi du projet. En e�et, les contraintes créées par un

faible capital humain, c'est-à-dire le manque des connaissances appropriées pour commencer ou

développer une micro-entreprise, peuvent être plus di�cilement surmontables pour les entrepre-

neurs pauvres (Berge et al. 2011). Schreiner and Morduch (2001) soulignent que le problème

d'un faible capital humain est plus exacerbé dans les pays développés par rapport au problème

d'accès au crédit. L'accompagnement entrepreneurial qui complète un microcrédit peut relâcher

ces contraintes. Néanmoins, la littérature traitant ce sujet est sceptique en ce qui concerne l'e�et

de l'accompagnement entrepreneurial.

A notre connaissance, dans les pays développés l'évaluation formelle de l'accompagnement en-

trepreneurial n'a pas encore été réalisée. Ce manque est particulièrement problématique dans le

contexte français où l'accompagnement entrepreneurial est une composante cruciale de la micro-

XVI Introduction in French

�nance (Camdessus 2010). Cependant, Balkenhol et al. (2013) a�rment que l'impact des services

d'accompagnement est di�cilement dissociable du service de microcrédit. La réalisation de telles

études reste di�cile, car on se heurte au problème d'insu�sance ou d'un manque de détail concer-

nant la micro�nance européenne. En outre, dans l'UE, la plus part des micro-entrepreneurs ne

ressent pas le besoin d'être accompagnés par l'IMF et le substituent par des sources de training

informelles provenant de la famille, des amis ou les média (Lammermann et al. 2007). Finalement,

Lammermann et al. (2007) soulignent que le challenge principal de la micro�nance européenne

demeure l'amélioration de l'e�cacité de l'accompagnement.

Toutefois, la micro�nance dans les pays développés et en développement ne di�ère pas uni-

quement en ce qui concerne les services proposés. Les deux modèles économiques présentent

également des di�érences signi�catives en termes d'équilibre poursuivi (Johnson 1998). Plus gé-

néralement, on attribue au secteur de la micro�nance la poursuite d'un double objectif, social

et �nancier. Dans les pays "du Nord", la micro�nance s'impose à traves un système bancaire

avec une orientation sociale, alors que dans les pays "du Sud" beaucoup d'institutions de mi-

cro�nance (IMFs) se focalisent sur l'atteinte de l'équilibre �nancier. En analysant le modèle de

la micro�nance aux Etats-Unis, Schreiner and Morduch (2001) expliquent que la majorité des

IMFs ne visent pas l'équilibre �nancier dans un futur proche, et certaines IMFs ne croient pas

l'atteindre un jour. Néanmoins, Evers et al. (2007) défendent la place de la micro�nance dans le

paysage économique de l'Europe de l'Ouest. Selon ces auteurs, le coût moyen du support d'un

micro-entrepreneur est plus faible comparé au coût moyen des aides sociales nécessaires dans

un scénario de chômage. Les avis les plus récents mettent en valeur l'objectif �nancier des IMF

européennes visé au moins à moyen terme (Kraemer-Eis et al. 2013).

Ce double objectif ajoute une dimension supplémentaire dans la modélisation des IMFs, la ren-

dant di�érente de celles des �rmes qui maximisent leurs pro�ts selon la théorie économique

traditionnelle. Dans cette thèse, nous avons intégré dans la modélisation théorique cette dimen-

sion sociale présente lors la décision d'octroi de microcrédit.

Le double objectif est présent dans la modélisation de l'IMF dans le premier chapitre de cette

Introduction in French XVII

thèse, où nous étudions les di�érents outils permettant à l'IMF de maximiser le nombre de clients

accédant à ses services, dans le contexte d'un pro�t espéré nul. Plus particulièrement, nous ana-

lysons comment les di�érentes formes d'intervention publique peuvent impacter l'allocation des

crédits d'une IMF.

L'intervention publique sur le marché du crédit a été utilisée historiquement comme un moyen de

transfert de ressources aux segments de population défavorisés. Malgré un lien historique entre

l'Etat et le secteur de la micro�nance, ce sujet reste toujours d'actualité. Il y a plusieurs expli-

cations à la résistance de ce lien. Par exemple, l'asymétrie d'information est relativement plus

forte sur le marché du microcrédit. Cela s'explique, d'une part, par un ciblage des créations de

micro-entreprise (plutôt que développement ou reprise d'entreprises) se révélant par dé�t être des

acteurs qui n'ont pas d'historique ou de relation bancaire qui pourraient réduire le gap informa-

tionnel (Bruhn-Leon et al. 2012). D'autre part, les micro-emprunteurs n'ont pas de collatéral qui

traditionnellement représente un signal de la qualité du projet. Plus généralement, l'asymétrie

d'information est décroissante avec la taille de l'entreprise (Bruhn-Leon et al. 2012).

Cette asymétrie d'information peut générer des défaillances de marché comme par exemple le

rationnement du crédit. Ce phénomène se produit lorsque parmi les demandeurs de crédit iden-

tiques certains reçoivent un crédit alors que d'autres n'en reçoivent pas ou, encore, lorsqu'un

groupe d'individus ne reçoit pas de crédit alors qu'ils l'auraient reçu si l'o�re de crédit était

plus élevée (Stiglitz and Weiss 1981). En réalité, le rationnement de crédit est di�cile à mesurer

car l'o�re et de la demande de crédit ne sont pas observables (Aubier and Cherbonnier 2007).

L'intervention publique, à travers des garanties de crédit par exemple, peut être utilisée pour

atténuer le problème de rationnement du crédit (Aubier and Cherbonnier 2007; Gale 1990). Dans

le premier chapitre de cette thèse, nous illustrons ce mécanisme à l'aide d'un modèle théorique

inspiré de Tirole (2006).

La littérature qui étudie le rôle des subventions dans la micro�nance reste relativement rare,

car les données concernant ces ressources sont di�cilement accessibles. L'article par Hudon and

Traça (2011) en est une exception. Les auteurs montrent que les subventions augmentent généra-

XVIII Introduction in French

lement l'e�cacité des IMFs. Cela peut être relié au concept de "subventions intelligentes" dé�ni

par Armendariz et Morduch (2010, pp. 333) comme des interventions conçues attentivement a�n

de minimiser les distorsions, le mauvais ciblage des clients et les ine�cacités, tout en maximisant

les bienfaits sociaux. Mieno and Kai (2012) sont également en faveur de l'utilisation de ce type

de subventions. Ils montrent que les subventions reçues par les IMFs à l'étape de leur création

diminuent la pression des coûts et permettent de réaliser des économies d'échèle. Finalement,

Armendariz et al. (2013) a�rment que les subventions sont e�caces lorsqu'il n'y a pas d'incer-

titude concernant leur montant et le moment de leur versement.

Nous étudions l'impact des subventions sur la performance sociale des IMFs dans le premier

chapitre de cette thèse. Nous analysons, plus précisement, comment les di�érentes formes d'in-

tervention publique impactent le processus d'octroi des microcrédits. Dans un modèle théorique

simple, nous considérons des entrepreneurs qui reçoivent des prêts individuels sans exigence

de collatéral. Nous montrons que l'intervention publique via la garantie des prêts augmente le

nombre d'entrepreneurs �nancés. Nous introduisons, en suite, l'accompagnement entrepreneurial.

En suivant Tirole (2006) et Barry and Bruno (2008), nous modélisons le training comme une

action de la part de l'IMF qui augmente la probabilité de succès d'un projet. En outre, l'octroi

des services non-�nanciers est couteux pour l'IMF. Ces coûts ne peuvent pas être intégrés dans le

prix du microcrédit payé par les emprunteurs. Cela s'explique d'une part par les lois sur l'usure,

et d'autre part par un pool des clients plus risqué acceptant un taux d'intérêt plus élevé. Bien

entendu, ces coûts doivent être pris en compte dans la modélisation théorique du comportement

de l'IMF. Par conséquent dans les chapitres 1 et 2, les coûts de l'accompagnement entrent

directement dans le pro�t espéré de l'IMF. Néanmoins, après l'introduction de l'accompagne-

ment professionnel par l'institution de micro�nance, nous montrons que la garantie des prêts

peut avoir un e�et contreproductif en réduisant le nombre d'entrepreneurs béné�ciant de l'ac-

compagnement. Par conséquent, nous nous sommes intéressés à une autre forme d'intervention

publique consistant dans le subventionnement de l'accompagnement professionnel. Nos résultats

suggèrent que si l'accompagnement est e�cace et cible les entrepreneurs les moins performants

Introduction in French XIX

alors, à dépense publique égale, le subventionnement de l'accompagnement est un meilleur outil

contre l'exclusion �nancière comparé à la garantie des prêts. En outre, sous des conditions si-

milaires, nous prouvons que le subventionnement de l'accompagnement est plus e�cace dans la

lutte contre l'exclusion �nancière comparé à un mélange de deux di�érentes politiques.

Dans le premier chapitre, nous avons montré que l'accompagnement entrepreneurial peut amé-

liorer l'accès à la �nance à travers un e�et direct positif sur la probabilité de succès d'un projet.

Cependant, l'accompagnement peut en outre impacter la probabilité de succès à travers le com-

portement de l'emprunteur. Cet e�et indirect apparait quand les emprunteurs croient que l'IMF

a une information supérieure concernant leur type. Cette situation se prête bien au marché de

travail (Ishida 2006), à l'école et dans les contextes familiaux (Benabou and Tirole 2003a) et

elle convient particulièrement bien au marché du microcrédit. En e�et, d'une part, les micro-

emprunteurs manquent de l'expérience en créant une entreprise pour la première fois. D'autre

part, l'IMF a une information supérieure concernant le micro-entrepreneuriat à travers son ex-

périence passée.

Le deuxième chapitre présente une contribution à la littérature sur l'impact de l'accompagne-

ment. Dans ce chapitre nous réalisons une analyse formelle de l'e�et du training entrepreneurial

sur la performance des prêts en termes de l'historique de remboursements. En dépit d'un e�et

mitigé sur la probabilité de défaut d'un prêt, nous trouvons que l'accompagnement a un impact

positif sur le temps de survie des prêts.

Dans le deuxième chapitre nous étudions comment les décisions d'une IMF concernant l'ac-

compagnement peuvent impacter le comportement des emprunteurs. Cette contribution inter-

disciplinaire se focalise sur les concepts sociologiques comme le "soi-miroir" (ou le looking-glass

self) et la con�ance en soi. L'e�et soi-miroir apparait lorsque l'environnement social tente de

manipuler la perception de soi d'un individu. Ce phénomène a été largement étudié dans la lit-

térature sociologique et plus récemment dans la littérature économique. Le terme soi-miroir est

pour la première fois introduit par Cooley (1902), qui a�rme que les individus se dé�nissent à

travers l'observation des attitudes et du comportement de leur entourage envers eux. Dans le

XX Introduction in French

modèle théorique, nous considérons la situation dans laquelle une IMF, à travers l'accompagne-

ment, et l'emprunteur, à travers son e�ort, peuvent in�uencer la probabilité de succès du projet.

L'accompagnement peut être interprété comme l'e�ort de l'IMF. En ligne avec la littérature sur

le double aléa moral, nous modélisons les deux types d'e�ort (celui par l'entrepreneur et celui

par l'IMF) comme des substituts parfaits (Casamatta 2003).

D'abord, nous étudions le contexte où, sous information symétrique, l'IMF accompagne plus les

emprunteurs les plus risqués. Nous montrons que, dans ce contexte, lorsque l'IMF possède une

meilleure information que l'emprunteur, elle peut ne pas aider les entrepreneurs les plus risqués.

Dans cette dernière situation, la décision d'aider de l'IMF impacte les croyances de l'entrepre-

neur sur son risque via l'e�et "soi-miroir". Nous testons cette prédiction du modèle théorique

en utilisant les données d'une IMF française. Cette base de données originale a été collectée

manuellement et contient des informations sur le processus d'octroie des microcrédits, le proces-

sus d'attribution de l'aide, ainsi que l'historique de remboursements par les micro-entrepreneurs.

Les résultats empiriques con�rment les intuitions théoriques : nous trouvons une relation non-

monotone entre la décision de l'IMF d'accompagner et le risque du micro-entrepreneur. La pro-

babilité de recevoir un accompagnement est croissante avec le risque de l'entrepreneur pour les

individus les moins risqués et décroissante avec le risque de l'entrepreneur pour les individus les

plus risqués.

Le second chapitre fournit des résultats intéressants sur comment les décisions d'une IMF

concernant l'accompagnement entrepreneurial pourraient démotiver les emprunteurs à faire l'ef-

fort. Dans les chapitres 3 et 4, nous nous intéressons aux conséquences d'une autre déci-

sion faite par l'IMF, notamment la mise en conformité avec la régulation. Nous montrons que

cette décision a également des implications importantes sur les micro-entrepreneurs. Cependant,

ces conséquences ne découlent plus de l'asymétrie d'information comme c'était le cas dans le

deuxième chapitre. Nous montrons que la mise en conformité avec la régulation peut limiter

l'accès à la �nance de certains groups de demandeurs.

Le marché de la micro�nance en Europe se caractérise par une forte hétérogénéité en ce qui

Introduction in French XXI

concerne l'environnement économique, les cadres légaux et judiciaires, ainsi que les approches

politiques dans la gestion des problèmes socio-économiques. La micro�nance est encadrée par la

régulation dans les pays développés au moins pour deux raisons. Premièrement, les IMFs o�rent

des services �nanciers. La supervision est nécessaire pour protéger les clients contre les consé-

quences de l'asymétrie d'information, les situations de monopole, et les externalités négatives

(Freixas and Rochet 1997). Deuxièmement, les IMFs béné�cient de subventions publiques et de

donations. Les régulateurs sont par conséquent concernés par la possibilité de détournement des

fonds. Cependant, en dépit d'un accord général concernant la nécessité de superviser les IMFs

subventionnées, la conception de la régulation adaptée reste un dé�. Les règles adéquates doivent

se traduire par un meilleur contrôle sans mettre en di�culté les IMFs dans l'atteinte de leur mis-

sion sociale, qui consiste en l'octroi des crédits aux micro-entrepreneurs pauvres.

Concernant l'impact de la régulation sur la performance des IMFs, la littérature existante four-

nie des résultats controversés. Armendariz and Morduch (2010) a�rment que les régulations

existantes sont mal adaptées à cette jeune industrie. En outre, Demirguc-Kunt et al. (2008)

expliquent que les régulations prudentielles visant la stabilité �nancière peut endommager la

capacité des banques de �nancer les petits projets. En utilisant les données pour 114 IMFs

provenant de 62 pays di�érents, Hartarska and Nadolnyak (2007) montrent que la régulation

n'impacte pas directement l'autosu�sance opérationnelle et la performance sociale. Cull et al.

(2009) soulignent que la mise en conformité avec la régulation est couteuse pour les IMFs et

peut conduire à l'exclusion des emprunteurs potentiels. Les avantages et les inconvénients des

seuils sur le montant du crédit sont présentés dans le rapport (CGAP and World Bank 2012).

Selon ce rapport, les seuils contraignent les IMFs de se focaliser sur les porteurs des projets les

plus pauvres et en même temps découragent les porteurs des projets les plus gros d'accéder aux

services �nanciers. En outre, les seuils réduisent les opportunités de l'inter�nancement. Nous

étudions l'impact de la régulation des seuils de crédit dans les chapitres 3 et 4 . Le souhait

d'atteindre l'autosu�sante �nancière peut conduire à l'éloignement de la mission initiale. Par

conséquent, la tendance vers une plus forte commercialisation est directement liée à ce phéno-

XXII Introduction in French

mène. Cull et al. (2007) dé�nissent l'éloignement de la mission initiale comme une situation où

les microbanques abandonnent le �nancement des plus pauvres dans la poursuite d'une viabilité

commerciale, ou, encore, comme une situation ou les IMFs servent les clients les plus riches au

détriment des clients les plus pauvres (Armendariz et al. 2013). En pratique, les données sur les

revenus des emprunteurs sont rarement disponibles. Par conséquent, les chercheurs utilisent la

taille moyenne des prêts pour avoir une mesure de cette variable, les prêts les plus larges étant

associés à des clients plus riches. Cependant, Christen (2001) a�rme que les prêts plus larges

n'impliquent pas forcement l'éloignement de la mission. D'abord, les IMFs qui sont autosu�santes

�nancièrement pourraient cibler un segment de la population di�érent ou pourraient choisir une

fonction objective di�érente. Deuxièmement, les prêts peuvent augmenter avec la maturité du

portefeuille des clients ou avec un environnement économique favorisant le développement des

micro-entreprises faisant face à une demande plus élevée. Finalement, il est souvent di�cile de

dissocier l'éloignement de la mission de l'inter�nancement (Armendariz and Szafarz 2011).

Dans la deuxième partie de cette thèse, nous nous focalisons sur l'éloignement de la mission

initiale ayant lieu après un changement réglementaire. Dans la majorité des pays développés,

les régulateurs ont imposé aux IMFs des seuils sur le montant maximal de microcrédit. Les

micro-entrepreneurs ayant besoin d'un �nancement supérieur à ce seuil peuvent demander un

co-�nancement bancaire. Le co-�nancement est pro�table pour les IMFs car il leur permet de

béné�cier du processus de screening bancaire. Par conséquent, les seuils peuvent avoir des e�ets

pervers facilitant le �nancement des projets plus importants au détriment des petits projets qui

ne nécessitent pas de co-�nancement. C'est l'idée principale de notre modèle théorique dans le

troisième chapitre. Selon ce modèle, l'IMF peut modi�er son allocation des crédits suite à

l'introduction du seuil.

Dans la logique d'intégration de la dimension sociale dans la modélisation du comportement d'une

IMF, nous considérons une IMF qui maximise le nombre de clients �nancés, sous la condition

d'un budget équilibré. En outre, nous faisons l'hypothèse que la totalité des ressources de l'IMF

consistent en subventions en ligne avec Bendig et al. (2012). Selon ces auteurs les subventions

Introduction in French XXIII

restent un élément essentiel de la micro�nance en Europe. Notre modèle théorique montre que

l'IMF préfère �nancer les petits projets sans le seuil et les projets plus larges après l'introduction

du seuil, sous certaines conditions sur les coûts de �nancement de ces deux types de projets. Nous

testons les prédictions de notre modèle théorique en exploitant une expérience naturelle d'une

IMF française qui en Avril 2009 est devenue réglementée par le seuil de 10,000 euros �xé par le

régulateur en France, malgré le seuil de 25,000 euros proposé par la Commission Européenne.

Nos estimations probit di�érence-en-di�érences con�rment que l'introduction du seuil de 10,000

EUR a des conséquences inattendues et socialement indésirables. Nous montrons que les petits

projets ont une probabilité plus élevée d'être choisis avant l'introduction du seuil et les projets

les plus larges ont une plus forte probabilité d'être acceptés en présence du seuil. Ce chapitre

montre que la mise en conformité avec la régulation peut conduire à l'éloignement de la mission

sociale de l'IMF.

Dans le troisième chapitre nous avons utilisé la taille des projets comme une mesure de la

performance sociale. Dans le quatrième chapitre nous allons un pas plus loin dans l'analyse

des e�ets pervers des seuils de crédit.

Une particularité du microcrédit dans les pays industrialisés consiste en la population ciblée.

Dans les pays en développement les femmes représentent la cible principale des IMFs, tandis

que dans les économies industrialisées la cible principale est composée des individus n'ayant pas

d'accès à la �nance classique, le plus souvent représentée par des chômeurs de longue durée,

quelque soit leur genre.3 De ce point de vue, la micro�nance en Europe ciblant la réduction de la

pauvreté et l'exclusion sociale peut être assimilée à un dispositif visant á la réduction du chômage

de longue durée (Armendariz 2009). Incontestablement, les disparités de genre identi�ées dans

les pays en développement sont également présentes sur le marché européen (voir Brana 2013

pour la France).

En outre, les régulations neutres peuvent parfois générer des résultats sensibles du point de vue

3Selon Reed (2013), 36% des clients des IMFs provenant de l'Amérique du Nord et l'Europe de l'Ouestétaient des femmes comparé aux 77% pour les pays en développement en 2011.

du genre (Johnson and Nino-Zarazua 2011). Cull et al. (2011) montrent que les IMFs avec une

orientation vers la performance �nancière répondent à la régulation en servant moins de femmes

a�n de maintenir leur niveau des pro�ts. Nous contribuons à cette littérature sur la micro�nance

et l'autonomisation des femmes dans le quatrième chapitre de cette thèse.

Dans cette partie empirique, nous utilisons une mesure di�érente de la performance sociale qui

est la taille des microcrédits octroyés aux femmes. Plus particulièrement, nous nous concentrons

sur les di�érences en termes de taille du crédit entre les entrepreneurs hommes et femmes avant

et après l'introduction du seuil.

Ce quatrième chapitre fournit des résultats additionnels qui corroborent l'éloignement de la

mission initiale. Nous montrons que l'IMF orientée positivement vers le �nancement des micro-

entrepreneurs femmes avant le seuil, abandonne cette stratégie après l'introduction du seuil. Nous

expliquons que l'éloignement de la mission initiale pourrait survenir en réponse aux schémas de

co-�nancement entre l'IMF et les banques classiques. Précisons que, nous n'identi�ons aucun

impact de genre sur la probabilité d'être sélectionné par l'IMF. Néanmoins, dans l'ensemble, la

situation des femmes en termes de montants reçus de la part de l'IMF se dégrade après l'intro-

duction du seuil. Par conséquent, nous con�rmons que les régulations de microcrédit semblant

avoir une orientation sociale, peuvent a�ecter l'accès au crédit des femmes entrepreneurs. Nous

fournissons également une explication possible à ce résultat. Un seuil bas conduit au renforce-

ment des schémas de co-�nancement entre l'IMF et les banques classiques. Par conséquent, les

IMFs sont a�ectées par les biais potentiels existant sur le marché �nancier classique.

Les deux derniers chapitres contribuent au débat sur l'impact de la régulation sur la perfor-

mance sociale de l'IMF. NÃ c©anmoins, la mise en conformité réglementaire ouvre les portes vers

la commercialisation. Nous montrons que l'éloignement de la mission initiale présente, en e�et,

une menace crédible dans le processus de commercialisation.

Contents

Introduction in French XIII

General Introduction 1

1 State intervention and the microcredit market: The role of business develop-

ment services 29

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

1.2 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

1.3 The introduction of the loan guarantee . . . . . . . . . . . . . . . . . . . . . . . . 38

1.4 Modeling Business Development Services (BDS) . . . . . . . . . . . . . . . . . . . 39

1.4.1 In the laissez-faire case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

1.4.2 In the presence of the state guarantee . . . . . . . . . . . . . . . . . . . . 41

1.5 An alternative policy: business development services subsidization . . . . . . . . 44

1.6 Mixing policies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

1.7 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

2 Informed Principal and the Microcredit Market: Should Business Training be

Targeted towards the Least Able Borrowers? 55

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

2.2 Theoretical model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

2.2.1 General framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

2.2.2 A discrete model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

2.2.3 The continuous model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

2.3 Institutional context of the MFI . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

2.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

2.5 Econometric model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

2.6 Econometric results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

XXV

Contents

2.7 Robustness checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

2.7.1 Correcting for selection bias . . . . . . . . . . . . . . . . . . . . . . . . . . 89

2.7.2 An alternative measure of risk: the inverse of the survival time . . . . . . 93

2.8 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

3 Microcredit in Developed Countries: Unexpected Consequences of Loan Ceil-

ings 101

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

3.2 Theoretical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

3.2.1 Loan Allocation without Ceiling . . . . . . . . . . . . . . . . . . . . . . . 106

3.2.2 Loan Allocation with Ceiling . . . . . . . . . . . . . . . . . . . . . . . . . 108

3.2.3 Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

3.3 Data and Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

3.4 Regression Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

3.5 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

3.5.1 Speci�cation of Project Size . . . . . . . . . . . . . . . . . . . . . . . . . . 132

3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

4 Loan Ceilings and Women's Access to Credit in France 139

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

4.2 Data and Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

4.3 Regression Analysis: Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

4.4 Regression Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

4.5 Robustness check: Heckman selection model . . . . . . . . . . . . . . . . . . . . . 161

4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

General Discussion 167

Bibliography 176

XXVI

List of Figures

2.1 Timing of contracting under symmetric information . . . . . . . . . . . . . . . . . 64

2.2 Timing of contracting under asymmetric information . . . . . . . . . . . . . . . . 65

2.3 Bivariate Probit Model Estimations . . . . . . . . . . . . . . . . . . . . . . . . . . 87

3.1 Applicants' project �nancing without and with ceiling . . . . . . . . . . . . . . . 120

3.2 Loan Size as a Function of Project Size . . . . . . . . . . . . . . . . . . . . . . . . 120

4.1 Loan Allocation Process in the First Period . . . . . . . . . . . . . . . . . . . . . 145

4.2 Loan Allocation Process in the Second Period . . . . . . . . . . . . . . . . . . . . 145

XXVII

List of Figures

XXVIII

List of Tables

0.1 Data for MFIs reporting to Mix Market in 2011 . . . . . . . . . . . . . . . . . . . 11

2.1 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

2.2 Descriptive statistics on the predicted intrinsic risk of the applicants . . . . . . . 76

2.3 Descriptive statistics for survival time . . . . . . . . . . . . . . . . . . . . . . . . 79

2.4 Determinants of Business Training and Default Processes . . . . . . . . . . . . . 86

2.5 Determinants of Approval, Business Training and Default Processes . . . . . . . . 91

2.6 Determinants of Business Training and the Inverse of the Survival Time . . . . . 95

3.1 Comparison of optimal loan allocations without and with ceiling . . . . . . . . . 111

3.2 Descriptive Statistics: Characteristics of Applicants and Borrowersa . . . . . . . . 117

3.3 Descriptive Statistics: Bank Loan and Approval Rate, with Ceiling only . . . . . 122

3.4 Descriptive Statistics: Project Sizes and Approval Rates without and with Ceiling 123

3.5 Hypotheses to be Tested . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

3.6 Probability of Approval . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

3.7 Probability of Approval: Di�erent Speci�cations for Project Size . . . . . . . . . 132

3.8 Descriptive Statistics: Reduced Observation Perioda . . . . . . . . . . . . . . . . 134

3.9 Probability of Approval: Reduced Period . . . . . . . . . . . . . . . . . . . . . . . 135

4.1 Descriptive Statistics on Applicants . . . . . . . . . . . . . . . . . . . . . . . . . . 146

4.2 Descriptive Statistics on Borrowers . . . . . . . . . . . . . . . . . . . . . . . . . . 147

4.3 First Correlation Matrice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

4.4 Second Correlation Matrice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

4.5 Regression results for Project Size and Bank Loan . . . . . . . . . . . . . . . . . 156

4.6 Regression results for the Requested Amount and Loan Size . . . . . . . . . . . . 158

4.7 Robustness check: Heckman Selection Model . . . . . . . . . . . . . . . . . . . . 164

XXIX

General Introduction

Micro�nance is a relatively new industry in developed countries. Its implementation and growth

since 1980s has been triggered by a spectacular development of micro�nance programs in the

developing world. One of the most famous micro�nance programs is Grameen Bank settled

in 1976 in Bangladesh by the professor of economics Muhammad Yunus. Many European and

North American programs have been inspired from Grameen Bank and its peer programs in

Latin America and South Asia.4

Interestingly, micro�nance initiatives in developing countries originate in their turn from Euro-

pean credit cooperatives. These initiatives were designed to serve groups of low-income indi-

viduals through saving and credit facilities. Credit cooperatives started by Frederick Rai�eisen

served 1.4 million people in Germany by 1910 (Morduch 1999b). German model has further been

replicated by Ireland, Northern Italy, and by South India under British authority and further

spread to Bengali State which today partly belongs to Bangladesh. The success of these pro-

grams has additionally inspired Edward Filene, an American merchant who has imported the

cooperative model in the United States.

Credit cooperatives have since lost popularity in Bangladesh. Nevertheless, the idea of �nancial

services provided to groups of individuals has been preserved and improved through experimen-

tation and modi�cations. As a result, it has successfully been implemented by a large number

of micro�nance institutions (MFIs). Today, micro�nance models continue to evolve and many

MFIs have integrated or shifted toward individual lending.

4For example, Adie, the largest French MFI was designed according to the Grameen Bank model.

1


Micro�nance industry has followed di�erent development paths in poor and industrialized economies

due to di�erent social and economic contexts. For instance, �nancial sustainability is targeted

by most of the institutions operating in poor countries where the sector has reached maturity.

Some particularly successful MFIs report having reached the double bottom-line objective com-

bining both social outreach and �nancial performance. Concerning scale, the industry served

189.5 millions of people in the developing world, by the end of 2011 (Reed 2013), through the

provision of a variety of services, such as credit, savings, insurance, training, and remittances.

In the developed countries the situation is considerably di�erent. The actual breadth of outreach

of MFIs in industrialized economies is comparably low (5.5 millions of people by the end of 2011

according to Reed (2013)). Nevertheless, in general the micro�nance market in Europe follows

trends towards e�ciency and self-sustainability. Its main di�culty remains the access to stable

funding (Kraemer-Eis et al. 2013). Moreover, the two main services o�ered by MFIs in richer

economies are microcredit and business training.

The scope of this thesis is to analyze micro�nance in the developed countries. This topic is

under-researched and poorly documented. Our four essays present an innovative contribution in

this domain with a particular focus on the role of business training, information and regulation.

In the following of this introduction we will �rst analyze two main issues faced by micro�nance

worldwide, that is reaching the double bottom-line objective and the threat of mission drift.

Second, we will discuss the particularities of micro�nance in Europe. Third, we will turn to

micro�nance subsidization which plays an important role in micro�nance's implementation, es-

pecially in developed economies. Finally, we will focus on how to model MFIs. This last section

will allow us to introduce the main theoretical ingredients of the modeling developed in this

thesis.

2


1. Micro�nance institutions and the double bottom line

(a) Financial versus social performance

Micro�nance has the potential to explore a win-win situation, where it can both

contribute to poverty alleviation and be pro�table for investors. This view has been

built on several success stories, such as BancoSol in Bolivia, Compartamos in Mexico,

or Bank Rakyat in Indonesia. More precisely, this debate has been initiated in the

1980s by the policymakers and it relies on several arguments such as: the elasticity

of demand of microcredit with respect to its price is considered to be very low, sub-

sidies can trigger dependency and impede innovation, past government subsidized

programs have resulted in excessive defaults and diverted funds.

Poverty alleviation is measured by social performance indicators or outreach. The

outreach has several dimensions. The breadth of outreach is measured in the number

of poor clients, usually people living bellow the poverty line. The depth of outreach

measures how poor are the clients. Some papers are also interested in the out-

reach in terms of the proportion of women, individuals in rural areas, immigrants,

or unemployed. Unfortunately, studies assessing the depth of outreach of MFIs re-

main anecdotal or based on case studies (Hermes and Lensink 2007). Improving the

outreach can be very costly. First granting very small credits is expensive due to

transaction costs including administration, screening, monitoring, reaching the poor

costs.

Financial performance has become an important issue since the more recent trend

of commercialization5 of micro�nance. An institution is considered to be �nancially

5According to Christen (2001), three main constitutes of a commercial approach to micro�nance arepro�tability, competition and regulation.

3


sustainable if it does not need subsidized resources in order to operate (Morduch

1999a). Indeed, access to donations and subsidies is not as reliable as access to

commercial funds.6 While subsidies' e�ciency has been largely recognized in the

short-run (Hudon and Traça 2011), at the start-up phase of the MFIs, in the long-

run their uncertainty can hamper the functioning of the MFIs (Armendariz et al.

2013), create dependency, or decrease the e�ciency of MFIs. At present, most of

the MFI's are not �nancially sustainable and require (direct or indirect) subsidies in

order to meet their high costs.

The de�nition of the outreach and of �nancial sustainability raises an obvious con-

cern about the compatibility of these two objectives. Morduch (2000) argues that

the win-win vision where �nancial and social performance are positively correlated

is neither logical nor supported by empirical evidence. Is increased sustainability an

impediment to poverty alleviation, the raison d'être of the MFIs? On other words

are these objectives complementary or subject to a trade-o�? The answer to this

question is not obvious as two competing scenarios seem to be plausible. On one

hand, increased access to stable commercial funds can expand the number of loans

granted to poor individuals and for a longer period of time. Moreover, commercial-

ization together with increased competition can contribute to increased e�ciency and

innovation, and therefore positively impact MFI's capacity to reach poor individuals.

On the other hand, improved �nancial performance can go at the expense of the

poor, as providing �nancial services to the poorest of the poor can be very costly. In

this scenario the desire to achieve sustainability can reduce the depth of the outreach

as the MFI can be tempted to grant larger loans and �nance wealthier individuals.

This outcome is termed "mission drift" in micro�nance literature and it consists in

a deviation from the initial social mission of the MFI.

6Commercial banks have incentives to invest in micro�nance programs as beyond pro�tability, theseinvestments allow them to illustrate their corporate social responsibility.

4


Furthermore, rigorous existing literature exploring the complementarity or the trade-

o� of these two objectives remains scarce, lacks generality and does not reach consen-

sus. One �rst formal analysis of the compatibility of �nancial and social performance

is Cull et al. (2007) suggesting that MFIs granting individual loans tend to �nance

wealthier clients as they grow larger. Interestingly, Caudill et al. (2009) suggest that

MFIs receiving less subsidies or o�ering deposits tend to become more cost e�cient

over time. In a subsequent paper, Cull et al. (2009) argue that commercial MFIs do

worse in terms of outreach compared to nonpro�ts, however, they do better in terms

of innovation and adoption of new technologies. Hermes et al. (2011) use a set of 435

MFIs in a stochastic frontier analysis and conclude that there is a negative correlation

between e�ciency and outreach. In contrast, Quayes (2012) demonstrates a comple-

mentary positive relationship between depth of outreach and �nancial performance

for high-disclosure MFIs, i.e. MFIs providing data of better reliability. To reconcile

both views, Schreiner (1999) argues that �nancial sustainability and outreach con�ict

in the short-run, however they may be complements in a long-run.

The existence of the double bottom line speci�c to the micro�nance sector makes

the theoretical modeling of the micro�nance sector di�erent from standard economic

models. Indeed, the social objective as a key characteristics of the sector has to be

taken into account. In the �rst and the third chapters of this thesis we focus on the

loan granting process of an MFI, by assuming that the MFI receives subsidies and

is concerned about outreach maximization. Due to the presence of the subsidies, the

�nancial sustainability is not directly included in the model. We make this choice

as in the developed countries most of the MFIs receive either subsidies or donations.

Nevertheless, we rule out the possibility of (expected) losses.

Further, we focus on the risk of mission drift which can undermine the double-bottom

line objective set by MFIs.

5


(b) Mission drift: do higher pro�ts lead to lower outreach?

The desire to compel with the �nancial sustainability objective can result in mis-

sion drift. Hence, the trend toward commercialization is directly linked to this phe-

nomenon. Cull et al. (2007) de�ne mission drift as a situation where "microbanks

moved away from serving their poorer clients in pursuit of commercial viability" or,

conversely, a situation where MFIs serve wealthier clients at the expense of poor

clients (Armendariz and Szafarz 2011). In practice, data on borrowers' wealth is

rarely available. Therefore, existing literature proxies it by the average loan size, the

larger the loan the wealthier the client. However, Christen (2001) argues that larger

loans do not necessarily imply mission drift. First, �nancially sustainable MFIs might

serve a di�erent segment of population and might have chosen a di�erent objective

function. Second, loans may increase as the portfolio of the MFI matures or the

economic environment favors microenterprises which in turn have higher demands.

Finally, it is not always easy to delimit mission drift from cross-subsidization (Ar-

mendariz and Szafarz 2011). Due to limited assessment of mission drift phenomenon,

the debate is dominated by ideological arguments (Bruck 2006).

Ghosh and Van Tassel (2008) contribute to the debate on mission drift by providing

the �rst formalization of the issue. The authors show that the presence of pro�t

maximizing donors can indeed result in MFIs �nancing larger loans, despite MFI's

objective of poverty minimization. Armendariz et al. (2013) argue that subsidy un-

certainty can trigger mission drift, as MFIs facing volatility on subsidies commitment

will be forced to realize precautionary savings and focus on wealthier clients. Their

empirical results unveil a positive relationship between subsidy uncertainty and the

interest rate charged to borrowers, suggesting that poorer clients are crowded-out

from the market. One of the �rst studies to answer the question whether and how

6


mission drift occurs is Mersland and Strøm (2010). Using a GMM model, the au-

thors �nd no evidence for mission drift in general. However, there is indeed a positive

relationship between higher pro�ts and costs and higher average loan size.

In this thesis we discuss the mission drift occurring due to regulation change. We

have access to data from an MFI which has chosen to comply with new regulation in

order to access commercial funds from mainstream banks and the European Union

(EU) funds. This original data set has been hand collected and presents very valuable

information both on loan acceptance and on the reimbursement history for accepted

clients. Despite their commercial character, the loanable funds were granted in form

of soft loans at very low interest rates. Compliance with regulation resulted in the

introduction of a EUR 10,000 ceiling on granted microcredits. In the third and forth

chapter of this thesis we study how MFI's loan allocation was impacted by the in-

troduction of the loan ceiling. In the third chapter we build a theoretical model

and show that there exists a scenario where the MFI shifts its loan allocation from

small projects to larger projects due to ceiling enforcement. The econometric analy-

sis con�rms the occurrence of this case. We �nd that small projects are more likely

to be �nanced in the period without ceiling and larger projects are more likely to

be accepted under ceiling enforcement. In the forth chapter of this thesis we study

how loan allocation to women has changed after the introduction of the ceiling. We

�nd that there is no gender impact in the approval process, however women receive

smaller loans from the MFI with ceiling enforcement. These two chapters contribute

to the debate on the impact of the regulation (or commercialization more generally,

as compliance with regulation has opened the access to commercialization) on the

outreach of microborrowers. We �nd that, indeed, mission drift presents a credible

threat in MFIs' commercialization process.

7


2. Micro�nance in Europe

Small and medium sized enterprises (SMEs)7 represent the lion's share in the total number

of European �rms. Their development is associated with higher employment and economic

growth. Access to �nance is crucial for SMEs. According to European Commission (2013),

access to �nance remains second most important concern of SMEs in Europe (after �nding

customers). Banks remain the main source of �nancing of SMEs in Europe as other sources

of �nance such as issuance of bonds or debt securities are not available for them. European

Commission (2013) reports that 85% of all SMEs who have borrowed in the two previous

years received a loan from a bank. Whilst, 5% borrowed from a family member or a friend

and 9% from another source such as government or micro�nance institutions. Yet, bank

�nancing has dramatically decreased during the past years due to the �nancial crisis. The

net tightening of credit standards has been applied more to SMEs as compared to large

�rms (Kraemer-Eis et al. 2013). Moreover, larger interest rate spreads between large and

small loans seem to reveal "some degree of discrimination by banks against small �rms"

(European Central Bank 2012). Therefore, the recent �nancial crisis acted as a catalyst

to micro�nance development during the last �ve years.

Despite a dynamic growth, micro�nance in Europe is still far from reaching the scale re-

ported in the developing countries. As argued by Schreiner and Morduch (2001) for the

United States, but the argument applies equally for Europe, in developed countries set-

tling a business is considerably more di�cult than in the developing countries. This is,

among others, due to the presence of operational safety nets which present an alternative

to self-employment, competition from larger �rms, competition from commercial banks,

regulation, etc. Regulation can be cumbersome both for MFIs (interest-rate caps limit,

loan ceilings, impossibility to collect deposits) and micro-enterprises (administrative and

7European Commission (2013) de�nes an SME as a �rm with 1 to 149 employees.

8


�scal burden).

Remarkably, the micro�nance model in itself is very di�erent in developed economies.

While in both types of economies micro�nance aims at poverty alleviation and social in-

clusion, the means they use to reach these objectives are signi�cantly di�erent (Johnson

1998). "North" countries target a socially-oriented banking system, while in South �nan-

cial sustainability is generally the driving goal. By the same token, Schreiner and Morduch

(2001) argue that in the United States, "very few MFIs intend to become self-sustainable

soon or even believe that self-sustainability is possible." Finally, Evers et al. (2007) argue

that despite limited chances of reaching pro�tability, micro�nance in Western Europe eco-

nomically makes sense. Indeed, the average costs of supporting a micro-borrower seem to

be weaker than the cost for one year of support in the traditional social welfare system.

Despite this agnostic view concerning the �nancial bottom-line some of the European MFIs

aim at �nancial sustainability at least in the medium run (Kraemer-Eis et al. 2013).8

Further, concerning lending practices, the largely praised group-lending methodology is

not appropriate to richer economies due to the lack of social capital (i.e. weak social ties,

higher levels of individualization) and high opportunity costs of participating (ex. time

to attend group meetings). Importantly, the lack of the human capital, i.e. appropriate

knowledge to start, manage and develop a business, is a more binding constraint in devel-

oped countries as compared to the lack of �nancial capital (Schreiner and Morduch 2001).

Therefore, the two main micro�nance tools in developed countries consist in microcredit

and business training. Microcredit and business training also represent two ways to signal

self-employment e�ort. Concerning business training, the signal is attendance and suc-

cessful progress. Concerning microcredit, the signal is on time repayment (Schreiner and

Morduch 2001). We further analyze the particularities of these services in Europe and

compare them to their equivalents in developing countries.

8For instance, Hannam and Cheng (2012) report that "Fair �nance", an MFI in London expects tobecome �nancially sustainable by 2015.

9


(a) Microcredit

The European Commission provides the following de�nition of microcredit:

"Microcredit in the European Union (EU) means loans under EUR 25,000. It is

tailored for micro-enterprises, employing less than 10 people (91% of all European

enterprises), and unemployed or inactive people who want to go into self-employment

but do not have access to traditional banking services" (European Commission 2009).

Until recently data on microcredit in Europe was particularly scarce. Bendig et al.

(2012) represent a salient contribution �lling this gap. The authors provide a survey

of 154 responding MFIs within 32 European countries covered by the European Mi-

cro�nance Network (EMN) Overview. They estimate that the number of institutions

providing microcredit in Europe ranges between 500 and 700 entities (without credit

unions and commercial banks). According to the authors, in 2011 the MFIs covered

by the survey disbursed 204,080 loans (122,370 in the EU member states). The av-

erage amount per borrower was EUR 5,135 (EUR 7,129 for EU member states).

In contrast to �gures observed in developing economies, 92% of disbursed loans were

individual loans, as compared to group loans.9

Finally, the interest rate ranged from 4% in countries like Austria, France and Italy

to around 20% in Balkan states (18% in Albania, 24% in Bosnia and 35% in Serbia).

MFIs generally charge �xed interest rates across borrowers, akin to the practices in

the developing countries.

For the sake of comparison, we provide similar data for other geographical regions

9For studies on group lending in North America see Conlin (1999) and Gomez and Santor (2003).

10


in Table 0.1. We note considerable discrepancies between �gures for Europe and the

rest of the world. As expected, the average loan balance is signi�cantly lower in

other regions of the world, which mainly regroup developing countries. The interest

rates, if we exclude the Balkan countries, are particularly favorable in Europe (given

the risk of the micro�nance industry). Finally, the scale of micro�nance in Europe is

very small as compared to other regions in number of served borrowers and operating

MFIs.

Table 0.1: Data for MFIs reporting to Mix Market in 2011

Average loan balance Average yield on Number of Number of

per borrower (in USD) gross portfolio (%) active borrowers reporting MFIs

Africa 846 38% 7 770 072 305

East Asia and the Paci�c 2715 29% 13 461 470 203

Eastern Europe and Central Asia 4704 32% 2 694 353 207

Latin America and The Caribbean 1917 38% 18 378 875 399

Middle East and North Africa 1082 30% 2 384 784 59

South Asia 239 25% 51 664 318 248

Total 1879 33% 96 353 872 1 421

Source: Author's calculations based on Mix Market data.

Nevertheless, there exist at least two rationales for these considerable gaps. On one

hand, the number of poor individuals is considerably larger in developing countries.

On the other hand, micro�nance focuses on the informal sector which is importantly

higher in developing economies.

Another particularity of microcredit in industrialized countries consists in its target

population. In developing countries women are the main focus of MFIs. Whereas,

in industrialized economies the focus is mainly on unbanked individuals, often long-

term unemployed, regardless their gender.10 From this perspective, micro�nance in

10According to Reed (2013), 36% of the clients of reporting MFIs from North America and WesternEurope were women in contrast to 77% for the developing world in 2011.

11


Europe aiming at poverty and social exclusion alleviation can be viewed as a mean

to reduce long term unemployment (Armendariz 2009). Remarkably, gender gaps

reported in micro�nance in developing economies are equally present in micro�nance

in developed countries (see Brana 2013 for France). We contribute to this strand of

literature on micro�nance and women empowerment in Chapter 4 of this thesis.

In line with this analysis we model microcredit under individual lending and �xed

interest rates. In addition, microcredit is often jointly provided with a business

training which should be taken into account in the theoretical modeling. Armendariz

(2009) terms this novel �nancial service a "guided" microcredit. We discuss the

e�ciency of business training in the following section.

(b) Business Training

Access to �nancial capital is not a panacea for successful business (De Mel et al.

2008). Indeed, human capital constraints may be more binding for poor entrepreneurs

(Berge et al. 2011; Schreiner and Morduch 2001). Business training attached to

micro�nance programs is used to release such constraints. However, the existing lit-

erature is agnostic about its e�ciency on business outcomes.

There is a growing literature of the impacts of micro�nance trainings in develop-

ing economies. Bjorvatn and Tungodden (2010) use a randomized controlled trial

(RCT) to assess a business training in Tanzania. They �nd a positive e�ect on

business knowledge. They, moreover, �nd that training is the most e�ective for less

educated entrepreneurs. Berge et al. (2011) �nd a signi�cant positive impact of

business training on the pro�ts of male entrepreneurs and no signi�cant treatment

e�ects for female entrepreneurs. By the same token, Giné and Mansuri (2014) �nd

that the e�ects of business training are mainly concentrated among men. Authors

12


suggest that these discrepancies by gender may be explained by social norms re-

stricting women's empowerment. These �ndings are in line with Karlan and Valdivia

(2011) results within a Peruvian institution targeting exclusively women. Karlan and

Valdivia (2011) study the e�ect of business training in FINCA-Peru using an RCT.

They �nd a signi�cant impact of training on client retention in the MFI, business

knowledge improvement but little evidence on the pro�t or revenues increase. In

addition, De Mel et al. (2014) �nd, using a randomized experiment among women

in Sri Lanka, that training results in some changes in business practices but does not

impact pro�ts, sales or capital stock. In contrast, Van Velzen (2014) �nd a positive

e�ect of training on business pro�tability for women in Vietnam, despite a non sig-

ni�cant impact on sales. Lensink et al. (2011a) use data for MFIs in 61 emerging

and developing countries and show that MFIs providing both �nance and business

development services have similar performance as MFIs providing no "plus" services.

Finally, McKernan (2002) �nds that noncredit aspects of micro�nance have positive

e�ects on borrowers' pro�ts. However, these noncredit aspects are composed of group

cohesion, joint liability and social development programs which are absent in devel-

oped countries.

In developed countries, formal impact evaluation of business training is scarce. For

instance in France, business training is recognized as a salient component of French

micro�nance (Camdessus 2010). Nevertheless, extensive research on its impact is

absent and hardly dissociable from the microcredit itself (Balkenhol et al. 2013).

Moreover, in the EU most micro-entrepreneurs do not consider that they do need

BDS and prefer using informal sources of training o�ered by family, friends or media

(Lammermann et al. 2007). Lammermann et al. (2007) additionally highlight that

in Europe, one of the main challenges remains the increase of business training e�-

ciency.

We contribute to this literature by providing in Chapter 2 formal evidence on busi-

13


ness training e�ciency using data from a French MFI. Whereas, training appears to

have mitigated impact on the probability of default of the borrowers, we identify a

signi�cant positive impact on the survival time of the loans. Interestingly, McKenzie

and Woodru� (2014) in a review of existing literature on the impact evaluation of

business training programs report studies which �nd no signi�cant impact of busi-

ness training on business survival (Bruhn and Zia 2012; Calderon et al. 2012), a

negative impact for male entrepreneurs (Mano et al. 2008; Giné and Mansuri 2014)

and even a positive impact for female entrepreneurs (Valdivia 2012). Our �ndings,

somehow, exacerbate the debate on business training e�ciency. However, in contrast

to previous studies, we do not use RCTs or experiments to evaluate the impacts of

business training.11 Obviously, further research is needed to assess training e�ciency

in developed countries.

3. State intervention on the microcredit market

State intervention on the credit market has historically been used as a mean to transfer

resources to disadvantaged segments of population. However, many government programs

involved in �nancial services have failed and some of them continue to jeopardize the

micro�nance sector. This evidence suggests that not all the interventions are equally

e�ective. Some of the policies can even be counterproductive. For instance, Du�os and

Imboden (2004) argue that such interventions as interest rate ceilings, direct credit delivery,

subsidized lending programs are harmful for the development of the micro�nance sector. In

contrast, indirect interventions such as maintaining the macroeconomic stability, involving

private sector in poverty reduction initiatives, adjusting the regulatory framework to ease

11Dehejia and Wahba (1999) highlights the di�culties to assess treatment e�ects using non-experimental data.

14


competition and improving supervision have positive consequences.

Three out of four chapters of this thesis tackle state intervention on the microcredit market

through the lens of theory and empirics. This section allows us to elaborate the context

and importance of this hot topic in micro�nance. Before debating the how of intervention,

we discuss the why of the state intervention on the microcredit market.

(a) Why should state intervene on microcredit market?

Public interventions are generally used to correct a market failure. A market failure

arises when a competitive market fails to achieve e�cient allocation of resources.

One one hand, such failures occur due to informational asymmetries. On the other

hand, markets may perform ine�ciently due to the presence of externalities (Green-

wald and Stiglitz 1986). Microcredit sector may indeed exhibit important externali-

ties through its links with other markets.

Further, we revise these two sources of market ine�ciencies. For a more ample dis-

cussion of the rationales for state intervention on the rural markets see Besley (1994).

i. Imperfect information

Informational asymmetries are more acute on the microcredit market as com-

pared to classical credit market for at least two reasons. First, microcredit aims

at micro-enterprise start-up. Thus, there is neither previous relationship with

the banker nor veri�able credit history that could be used to reduce the infor-

mational gap (Bruhn-Leon et al. 2012). Second, micro-borrowers lack collateral

that is traditionally used to signal the quality of the project. More generally,

15


"the smaller the company, the bigger the information asymmetry" (Bruhn-Leon

et al. 2012).

Imperfect information on the credit market concerns either the type of the bor-

rower (adverse selection) or his action (moral hazard). One example of market

failure caused by the informational gap is credit rationing. It occurs either when

among identical applicants for a loan some receive credit and others do not or

when there is a group of individuals who do not receive credit whereas they

would receive a credit if loan supply was higher (Stiglitz and Weiss 1981). In

reality, credit rationing is hardly measurable due to unobserved credit demand

and supply (Aubier and Cherbonnier 2007). The reluctance characterizing clas-

sical credit market presents an opportunity for the micro�nance sector. Indeed,

microcredit can be viewed as a mean to decrease credit rationing problem as it

aims at �nancial inclusion of individuals who are rejected from classical �nan-

cial markets. Government intervention through loan guarantees can be used to

mitigate credit rationing issue (Aubier and Cherbonnier 2007; Gale 1990). In

the �rst chapter of this thesis we provide a theoretical model to illustrate this

mechanism. In addition, we include subsidies in the theoretical model of the

third chapter of this thesis.

Armendariz and Morduch (2010) argue that another possible solution to over-

come informational imperfection could be linking mainstream banks with local

markets, by employing well informed agents for example. However, this solutions

seems to be poorly suitable to developed economies where local markets are less

well informed compared to developing countries.

16


ii. Links with other markets

In addition to business start-up, micro�nance may lead to positive spillovers,

or indirect bene�ts. For example, a micro-entrepreneur gives a good example

to his kids, may hire other employees, and hence positively contributes to the

entire economy. In this context, Schreiner and Morduch (2001) illustrate a very

simple rationale for state intervention. It is possible that the costs of integrating

a micro�nance program exceed the bene�ts for a poor micro-borrower. How-

ever, the sum of all the bene�ts (direct and indirect) can exceed the costs of

micro�nance supply. In this case, there are clear grounds for subsidizing access

to micro�nance. However, it is di�cult to estimate the real size of these total

bene�ts (Armendariz and Morduch 2010).

Armendariz (2009) argues that micro�nance, impacts directly the income di-

mension of poverty, and indirectly education and health. Emran et al. (2011)

use the link between micro�nance and the labor market to explain some of the

existing puzzles on the micro�nance market. In addition, this study provides a

rationale for a smaller demand for microcredit in developed economies consist-

ing in a more developed labor market. Holvoet (2004) reports a positive link

between micro�nance and schooling or literacy. By the same token, Leatherman

et al. (2012) provide evidence on positive impacts of micro�nance programs and

clients' health.

More generally, Lensink and Scholtens (2004) �nd that �nancial development

has an indirect positive impact on economic growth by weakening the negative

e�ects of in�ation uncertainty.

In a nutshell, there exists a variety of links between micro�nance and other mar-

kets. In such a context, government intervention is welcome. Indeed, it can

17


reduce the costs of participation in a micro�nance program if the total spillovers

are worth it. In this thesis we do not explicitly model the link between mi-

cro�nance and other markets. These more general models are left for further

research.

(b) How does state intervene on the microcredit market?

To mitigate market failures or address potential externalities discussed above, gov-

ernments may choose to support micro�nance sector (for example through subsidiza-

tion), to regulate micro�nance activities or to intervene directly. Importantly, many

direct interventions have failed in the past and/or had extremely negative impacts

on economic development (Du�os and Imboden 2004). These types of interventions

are not under the scope of this thesis. Conversely, we contribute to the debate on the

e�ciency of subsidization and regulation of the micro�nance sector. In the following,

we illustrate some pieces of this debate and depict our main �ndings.

i. Regulation

The European micro�nance market is shaped by heterogeneous economic en-

vironments, political approaches toward socio-economic activity, and legal and

regulatory frameworks. Micro�nance activities in developed countries are reg-

ulated for at least two reasons. First, MFIs o�er �nancial services, and su-

pervision is necessary to protect clients against the consequences of asymmetric

information, market power and negative externalities (Freixas and Rochet 1997).

Second, MFIs typically bene�t from public subsidies and charitable donations.

18


Regulators are thus concerned with the possibility of money diversion. However,

while everyone agrees on the need for supervising subsidized MFIs, regulatory

design represents a real challenge. Adequate rules should bring e�cient control

without hampering MFIs in achieving their social mission, which mainly consists

in delivering credit to poor entrepreneurs.

Concerning the impact of the regulation on MFIs' performance, existing liter-

ature provides mixed evidence. Armendariz and Morduch (2010) content that

the existing regulations are poorly adapted to this young industry. By the same

token, Demirguc-Kunt et al. (2008) state that, regulatory prudential measures

aimed at �nancial stability can hamper the potential of banks to serve small

projects. Using data for 114 MFIs from 62 countries, Hartarska and Nadolnyak

(2007) �nd that regulations do not directly a�ect operational self-sustainability

and outreach. Cull et al. (2009) emphasize that complying with regulations is

costly to MFIs and may result in the exclusion of potential borrowers. The pros

and cons of loan ceilings are discussed in a CGAP report (CGAP and World

Bank 2012). The report states that ceilings constrain MFIs to focus on poor

clients but prevent holders of large projects from gaining access to �nance. Ceil-

ings also reduce cross-subsidization opportunities.

We address the impacts of loan-ceiling regulation in chapters three and four

of this thesis. In the third chapter we show that compliance with regulation

triggers mission drift. Using data from a French MFI before and after the in-

troduction of the ceiling, we �nd that the unregulated MFI preferred �nancing

smaller projects. In contrast, after ceiling enforcement, the MFI preferred �-

nancing larger projects. Chapter four provides additional evidence for mission

drift, but from a di�erent standpoint. In this empirical chapter we �nd that a

gender neutral MFI grants smaller loans to women applicants when it is ceil-

ing constrained. We explain that mission drift could arise due to co-�nancing

19


schemes with mainstream banks.

Further, we analyze another salient form of state intervention on the microcredit

market consisting in subsidization.

ii. Subsidization

The debate on the e�ciency of the subsidies in micro�nance is not new. The

relationship between microcredit and subsidies is historical. Grameen Bank, for

example, has constantly bene�ted from subsidies despite reporting pro�ts (Mor-

duch 1999b).12 In terms of the social bottom line, subsidized programs perform

better (than unsubsidized ones) in outreaching the poorest borrowers (Morduch

2000). Unsubsidized MFIs sacri�ce one dimension of their social performance

either by setting higher interest rates, targeting richer clients or decreasing the

share of female borrowers (D'Espallier et al. 2013).

However, there is no convincing evidence on the overall negative impact of the

subsidies on the micro�nance market. Armendariz et al. (2013) show that argu-

ments against subsidization of micro�nance are misleading, as subsidies' uncer-

tainty is associated with mission drift. Formal assessment of the role of subsidies

is missing as the academic literature on micro�nance subsidization remains rel-

atively scarce. This is mainly due to di�culties in obtaining high quality data.

One exception is Hudon and Traça (2011) who �nd that subsidies generally in-

crease the e�ciency of MFIs. This may be related to the concept of "smart

subsides" de�ned by Armendariz and Morduch (2010, pp. 333) as "carefully

designed interventions that seek to minimize distortions, mistargeting and inef-

12Cull et al. (2007) con�rm the existence of MFIs having achieved the "ultimate promise of micro�-nance" (i.e. self-sustainability and large outreach to the poor). However, according to this study suchMFIs are mainly exceptions.

20


�ciencies while maximizing social bene�ts". Mieno and Kai (2012) also advocate

the use of such subsidies. They �nd that subsidies received at the early stage

reduce the cost pressure for start-up MFIs and therefore allow them to achieve

economies of scale. Finally, Armendariz et al. (2013) argue that subsidization

is e�cient as long as there is no uncertainty regarding the timing or the amount

of subsidies.

We study the impact of micro�nance subsidization on the outreach in chapter

1. We additionally assume a fully subsidized MFI in the third chapter in line

with Bendig et al. (2012) who argue that subsidies remain crucial for MFIs in

Europe.

4. How to model MFIs?

(a) The double bottom line objective

The double bottom line objective is the hallmark of the micro�nance sector. It ren-

ders the modeling of this market signi�catively di�erent from the standard economic

theory with pro�t-maximizing �rms. Therefore, the social performance objective has

to be added to the classical models. For instance, McIntosh and Wydick (2005) model

a non pro�t MFI as an outreach maximizer (in terms of the number of �nanced bor-

rowers). The maximization of the objective takes place under a net budget-balancing

constraint, participation constraint, and non-negativity constraint. By the same to-

ken, Armendariz and Szafarz (2011) choose an objective function maximizing MFI's

21


outreach under a budget balancing constraint with full subsidization.

Three out of four chapters of this thesis include theoretical modeling. In the �rst

and third chapters we study the approval process. The social dimension is present

in our objective functions through increased breadth of outreach, in line with other

papers modeling socially minded MFIs. In the second chapter we do not model the

approval process. Here we use an objective function maximizing the expected pro�t

to the MFI, conditional on approval. Hence, this model is still suitable for a socially

minded MFI.

(b) The role of subsidization

State interventions that decrease the interest rate charged to microborrowers have

positive e�ects on the crowding-in of the safer borrowers.13 Therefore, this type of

subsidies mitigate credit rationing.

Both chapters one and three deal with subsidized MFIs. In chapter one we study

how state intervention through loan guarantee or business training subsidization can

increase the number of borrowers �nanced by an MFI. In chapter three we simply

assume that a socially minded MFI is fully subsidized and maximizes the number of

�nanced borrowers.

(c) Modeling non-�nancial services

As we have previously argued, business training is an important component of micro-

13According to Stiglitz and Weiss (1981) higher interest rates may result in riskier pools of borrowersas safer project holders do not a�ord borrowing at high interest rates. Consequently, total returns to thebank may decrease.

22


�nance in the developed countries. Two chapters of this thesis deal with this issue.

In the �rst chapter we study the link between business training provision and �nan-

cial inclusion. In the second chapter we study how business training is assigned to

microborrowers and give some insight into its impact on business survival.

Akin to Tirole (2006) and Barry and Bruno (2008), we model business training as

an action from the MFI which can increase the probability of project success. Barry

and Bruno (2008) is the only contribution we are aware of which provides a model

for mentoring in micro�nance in developed countries.

Finally, the provision of this non-�nancial service is costly. These costs cannot be

passed on to microborrowers due to the regulatory framework on interest rates caps

in most developed economies. Consequently, it is salient to account for these costs

in the theoretical modeling. As a result, in chapters 1 and 2 the costs of business

training enter directly into the expected pro�t of the MFI.

(d) A review of other particularities of micro�nance models

Micro�nance borrowers generally lack collateral. Hence, one of the challenges of

this sector is to �nd innovative ways to substitute traditional collateral. Though

this particular challenge is not explicitly addressed in this thesis we provide a brief

discussion of the existing mechanisms as they constitute an important feature of the

micro�nance. We don't integrate the features discussed bellow in our modeling as

they are e�ective mainly in developing countries. Concerning developed countries,

we conjecture that the lack of physical collateral is overcome by channels driven by

trust and reciprocity (Cornée and Szafarz 2013). However, the integration of these

channels in the modeling of micro�nance in developed countries is left for further

research.

23


Until recently the economic literature on micro�nance was mainly focused on the

group lending mechanism. This innovative way to reach poor individuals consists

in providing credit to groups where all members are jointly liable and exert peer

monitoring. Pioneering studies on the success behind group lending mechanism are

Stiglitz (1990, Varian (1990, Besley and Coate (1995) highlighting the role of peer

monitoring, mutual insurance and social collateral respectively. Lending can be either

simultaneous or sequential.14 Despite the popularity of the group lending, in this

thesis we rather model individual lending. Our motivation consists in the poor �t

of this methodology to a framework of richer economies (Armendariz and Morduch

2000; Townsend 2003).

Group lending is not the only ingredient of micro�nance success. In addition, some

pioneers of this type of lending (ex. Grameen Bank in Bangladesh and BancoSol in

Bolivia) are moving to individual lending contracts. Moreover, group lending does not

always dominate individual lending due to its costs (Giné et al. 2010). Concerning

other innovative mechanisms,Armendariz and Morduch (2000) illustrate how non-

re�nancing threats, direct monitoring and regular repayment schedules enforce the

incentives for loan repayment.

Regular repayment schedules are considered to be e�cient due to the inculcation

of the �scal discipline among borrowers, possibility for the MFI to early identify

potential problems in repayment and start dealing with them, or their substitutability

to imperfect saving vehicles. Jain and Mansuri (2003) suggest that regular high

frequency repayment schedules are used by MFIs to mitigate information asymmetries

and are only e�cient in environments where informal lending is well developed. They

additionally argue that a regular installment plan does better than a requirement for

co-�nancing. Informal lending from moneylenders is nonexistent in the developed

economies. Consequently, according to Jain and Mansuri (2003) regular installment

14For pros and cons of the sequential lending within a group of two borrowers see Aniket (2007).

24


contracts will not make a di�erence in this setting. This may explain why in Europe

co-�nancing requirements are possible. More precisely, in the third chapter of this

thesis we model a co-�nancing scheme between an MFI and a mainstream bank.

While in Jain and Mansuri (2003) setting the MFI relies on the monitoring from

an informal moneylender, in our setting the MFI is temped to free-ride on bank's

screening process, which comes before the microcredit approval.

(e) The summary of the PH.D. dissertation chapters

In chapter one we analyze how various forms of state intervention can impact mi-

cro�nance institutions' lending behavior. Using a simple model where entrepreneurs

receive individual uncollateralized loans, we show that, not surprisingly, state inter-

vention through the loan guarantee increases the number of entrepreneurs receiving a

loan. However, after modeling business development services provided by the micro-

�nance institution, we show that the loan guarantee can have a counterproductive

e�ect by reducing the number of entrepreneurs bene�ting from such services. We

therefore analyze an alternative policy: business development services subsidization.

We show that if business development services are e�cient enough and are targeted

toward less performing borrowers then - for �xed government expenditures - such

subsidies do better in terms of �nancial inclusion than the loan guarantee. Moreover,

we argue that - under similar conditions - business development services subsidization

alone does better in terms of �nancial inclusion than a mix of policies.

In chapter two we analyze how decisions of a micro�nance institution (MFI) on busi-

ness training provision can impact borrowers' behavior. In the theoretical model we

assume a situation where the MFI - through business training - and the borrower -

through e�ort - can act on the probability of borrower's project to succeed. We show

25


that, in contexts where under symmetric information the MFI optimally provides

more business training to the borrower with lower probability of success, superior

information (when the MFI has better information on probability of success than the

borrower) can lead the MFI not to provide business training (or to provide less busi-

ness training) to riskier borrowers. In this last case, because of a "looking-glass self"

e�ect, MFI's choice of business training impacts borrower's belief about his risk. We

then test this prediction using data from a French MFI. By means of empirical mod-

els, taking into account both the credit-granting and the training-granting processes,

we analyze how training programs are assigned to di�erent borrowers. Con�rming

our theoretical reasoning, we �nd a non-monotonic relationship between the MFI's

decision to provide business training and the risk of micro-borrowers. The probability

to receive business training appears to increase with risk for low-risk borrowers and

to decrease with risk for high-risk borrowers.

In chapter three we focus on loan ceilings imposed by regulators to subsidized mi-

cro�nance institutions (MFIs) in most developed countries. Micro-entrepreneurs in

need of above-ceiling loans are left with the co-�nancing option, which means secur-

ing the above-ceiling share of the loan with a regular bank, and getting a ceiling-high

loan from the MFI. Co-�nancing is attractive to MFIs because it allows them to

free-ride on the regular banks' screening process. Therefore, loan ceilings can have

the perverse e�ect of facilitating the co-�nancing of large projects at the expense

of micro-entrepreneurs who need below-ceiling loans only. This is the gist of our

theoretical model. We test the predictions of this model by exploiting the natural

experiment of a French MFI that became subject to the French EUR 10,000 loan ceil-

ing in April 2009. Di�erence-in-di�erences probit estimations con�rm that imposing

loan ceilings to MFIs can have unexpected and socially harmful consequences.

The fourth and last chapter compares the loans granted to male and female en-

trepreneurs by a French micro�nance institution (MFI). The sample period is split

26


in two: before and after the MFI implemented France's regulatory EUR 10,000 loan

ceiling. In the �rst period, the MFI does not co-�nance projects with mainstream

banks and loan size is gender-insensitive. In the second period, the MFI does co-

�nance above-ceiling projects with mainstream banks, and we observe a gender gap

in loan size. The results suggest that co-�nancing leads the originally gender-neutral

MFI to import disparate treatment from mainstream banks.

27


28

Chapter 1

State intervention and the microcredit market:

The role of business development services1

1.1 Introduction

In this chapter we examine the role of government intervention on the microcredit market. We

analyze the case of non-pro�t micro�nance institutions that provide individual loans and bene�t

from state intervention through direct subsidies or loan guarantees.2 This intervention seems

to be mainly due to the positive e�ect of microcredit on employment and poverty alleviation

through self-employment and entrepreneurship. Facilitating the access to microcredit is in turn

expected to bene�t the state by reducing other social expenses. Although, to our knowledge, it

has never been showed by rigorous evaluation methods, most of the actors of the sector expect

microcredit to create externalities on social expenses. In this chapter we analyze and compare

various forms of public subsidies.

Microcredit is generally de�ned as a small loan to individuals in poverty designed to encour-

1This chapter is based on a joint work with Renaud Bourlès, forthcoming in Small Business Economics,doi:10.1007/s11187-014-9578-0

2Individual lending, as well as state intervention, are classical in developed countries andare spreading in developing ones. For a discussion about government intervention in micro�-nance in Latin America see the article "Governments in micro�nance: threat or opportunity?" byP. Bate at http://www.iadb.org/en/news/webstories/2007-11-09/governments-in-micro�nance-threat-or-opportunity,4134.html, accessed 21 November 2013.

29

State intervention and the microcredit market

age entrepreneurship or access to employment. Microborrowers often lack collateral, rarely have

steady employment and their credit history can hardly be veri�ed. More generally, these in-

dividuals cannot meet the minimum requirements to access the traditional credit market, and

microcredit is often considered to be a solution to exclusion from the traditional banking system

and, consequently, to credit rationing.

According to the academic literature, state intervention via loan guarantees (as opposed to di-

rect subsidies) is considered to be the most e�cient measure in dealing with credit rationing. By

participating in this speci�c market, government impacts both the pure rationed borrowers (who

do not receive credit despite sharing the same characteristics with accepted borrowers and are

willing to pay a higher interest rate) and the redlined borrowers (who do not receive credit at any

interest rate because their projects do not generate a high enough return to the lender).3 In the

case of small-business lending, micro�nance institutions (MFIs hereafter) are strongly involved

in business development services. They entail devices o�ered in addition to loans that aim at

increasing the chances for the project to succeed. These devices mostly consist in training pro-

grams, for example, in accounting or management. In this chapter we shed light on the impact

of state intervention on business development services provision.

To do so, we base our work on Tirole's model of credit rationing in which borrowers are hetero-

geneous according to their project return and can enhance the probability of project success by

exerting a costly and unobservable e�ort (Tirole 2006). To adjust to the case of microcredit, we

model borrowers without any initial capital endowment and MFIs that lend without collateral

requirements.

In this basic setting, we �rst introduce state intervention through the loan guarantee (which

is common in microcredit mostly in developed countries4). We allow the state to pay back to

the lending institution a proportion of the capital lost if the entrepreneur's project fails. Not

3For detailed de�nitions of di�erent types of credit rationing see Ja�ee and Stiglitz (1990) pp. 847�849.4For example, the European Commission and the European Investment Bank started providing loan

guarantee for microcredits in the European Union by launching the European Progress Micro�nanceFacility in 2010.

30

1.1. Introduction

surprisingly, we �nd that whatever the size of the guarantee, such a policy increases the number

of entrepreneurs that receive a loan by widening the range of project returns optimally �nanced

by the bank.

The key contribution of our work consists in analyzing how the loan guarantee interacts with

business development services (or training more generally), another key feature of microcredit

targeting small businesses. To develop the analysis, we allow MFIs to invest in a device that

increases the probability of project success. In the absence of the state guarantee, business de-

velopment services crowd-in a number of the excluded borrowers if and only if the relative gain

generated by this measure is lower than its relative cost. However, when both business devel-

opment services and the state guarantee are modeled, the loan guarantee can have a "perverse"

e�ect, since it can reduce the incentive for the MFIs to provide business development services. In

particular, assuming that project returns are uniformly distributed among borrowers, we show

that the number of additional borrowers �nanced through business development services is larger

when state does not guarantee loans. The intuition behind this result is that, from the point

of view of the MFI, the loan guarantee decreases the expected return on business development

services.

Such a counterproductive e�ect leads us to model an alternative policy that would consist in

subsidizing business development services. To be able to compare between policies, we analyze

under what circumstances a government with a �xed budget would prefer to subsidize business

development services (BDS hereafter) rather than to guarantee loans. We show that subsidiz-

ing BDS brings better results (in terms of �nancial inclusion) than the loan guarantee provided

that the BDS are e�cient enough and target the projects with the lowest return. Moreover, we

�nd that by mixing policies (i.e. by providing both loan guarantee and BDS subsidization) the

state can get rid of the perverse e�ect. Nevertheless, the largest �nancial inclusion is achieved

when the entire state budget is allocated to BDS subsidization and when BDS target otherwise

excluded borrowers.

We now provide an overview of the existing literature with reference to this chapter. As we

31


have already mentioned, microcredit provides a solution to the borrowers who are excluded from

the traditional credit market. In the academic literature these individuals are denoted as either

"rationed" or "redlined" borrowers (see e.g. Stiglitz and Weiss 1981; Ja�ee and Stiglitz 1990).

Stiglitz and Weiss (1981) show in particular that, for a given interest rate, there exists a critical

value of return below which the bank does not �nance the project. One of the aims of our chapter

is to analyze how such a threshold evolves depending on the state intervention in the case of

microcredit (i.e. of uncollateralized loans). Note that we do not model explicitly "small" loans

which is unarguably an important characteristic of microcredit. Therefore, our model could be

understood as a model of social banking (micro�nance institutions being a particular example

of social banks). Nevertheless, by modeling other important aspects of micro�nance, such as the

lack of collateral requirements, the presence of the loan guarantees, and business development

services, we proceed in the following sections by applying our model to the micro�nance �eld.

Our chapter is not the �rst to study the e�ect of the loan guarantee. Craig et al. (2007) analyze

empirically the case of Small Business Administration, a program providing small �rm loan guar-

antees in the USA, and �nd a positive and signi�cant link between the level of SBA lending and

local economic growth. In a subsequent paper, Craig et al. (2008) �nd a positive link between

the average annual level of employment in the local market and SBA lending. These papers

present a rationale for government intervention in small �rm lending in general, but especially

in microcredit lending that directly promotes self-employment and small start-ups.

The importance of government intervention in credit rationing is also highlighted in the case of

France in a paper by Aubier and Cherbonnier (2007). They show evidence that credit rationing

was signi�cant during the 2001-2004 period for small and medium-sized enterprises. State in-

tervention, mostly through loan guarantee, is presented as a mean to reduce credit rationing.

Facilitating access to entrepreneurship then bene�ts the state by reducing other expenses (unem-

ployment bene�ts, etc.) according to Brabant et al. (2009), in a report for the French Ministry

for the Economy and Finance. From a theoretical point of view, Emran et al. (2011) analyze

how microcredit market interacts with labor market in a macroeconomic model. In the present

32

1.1. Introduction

chapter we disregard the interactions with other markets and focus on partial equilibrium on the

microcredit market. More precisely, we do not study the �nancial e�ciency of state intervention

and exogenously assume that the state's objective is to improve �nancial inclusion, that is to

crowd-in more entrepreneurs.

Regarding the comparison of various policies on the credit market, Gale (1990) analyzes the

e�ects of federal policies on credit allocation and economic e�ciency in a model with adverse

selection. He argues that the loan guarantee is more e�cient than pure direct lending programs

and pure interest subsidy as it operates through raising the return to the bank. Adding business

development services to the analysis and focusing on moral hazard rather than adverse selection,

we enrich this discussion regarding indirect subsidies. More precisely, we show that the loan

guarantee might be less e�cient than other indirect subsidies that can impact the (expected)

return to the bank. However, contrarily to Gale (1990) we do not analyze the e�ects in terms of

welfare.

The relationship between microcredit and subsidies is historical. Grameen Bank, for example,

has constantly bene�ted from subsidies despite reporting pro�ts (Morduch 1999b).5 Moreover,

subsidized programs perform better (than unsubsidized ones) in outreaching the poorest bor-

rowers (Morduch 2000). Unsubsidized MFIs sacri�ce one dimension of their social performance

either by setting higher interest rates, targeting richer clients or decreasing the share of female

borrowers (D'Espallier et al. 2013).

Still, the academic literature on micro�nance subsidization remains relatively scarce, mainly due

to di�culties in obtaining high quality data. One exception is Hudon and Traça (2011) who �nd

that subsidies generally increase the e�ciency of MFIs. This may be related to the concept of

"smart subsides" de�ned by Armendariz and Morduch (2010, pp. 333) as "carefully designed

interventions that seek to minimize distortions, mistargeting and ine�ciencies while maximizing

social bene�ts". Mieno and Kai (2012) also advocate the use of such subsidies. They �nd that

5Cull et al. (2007) con�rm the existence of MFIs having achieved the "ultimate promise of micro�-nance" (i.e. self-sustainability and large outreach to the poor). However, according to this study suchMFIs are mainly exceptions.

33


subsidies received at the early stage reduce the cost pressure for start-up MFIs and therefore

allow them to achieve economies of scale. Finally, Armendariz et al. (2013) argue that subsidiza-

tion is e�cient as long as there is no uncertainty regarding the timing or the amount of subsidies.

More generally, academic literature on microcredit design is mainly about developing countries

where group lending was � until recently � both the norm and the explanation for the success of

microcredit. Townsend (2003), however, questions this idea and argues that the choice between

individual and group lending is not simple. Particularly, group-lending prevalence depends on

the economy-wide average wealth: richer economies should experience less group lending. This

analysis might explain why individual lending is prevalent in developed countries. Still, the key

role of peer-lending in explaining high repayment rates in microcredit in developing countries has

been recently challenged by Giné and Karlan (2009) and individual lending now also spreads in

developing countries (for example in Grameen Bank in Bangladesh6 and in BancoSol in Bolivia).

The originality of our work lies mostly in the modeling of business development services (i.e.

training of the entrepreneurs by the MFI) that complements microcredit as a tool of �nancing

excluded individuals. Non-�nancial services provided by MFIs are termed "Micro�nance-Plus"

in Lensink and Mersland (2009). These kinds of programs are very popular in developed coun-

tries where they generally take the form of entrepreneurial training. In developing economies,

however, this "plus" services often take the form of social trainings, including health or educa-

tional services.

Several papers empirically assess the impact of these types of non-�nancial services. One exam-

ple is Karlan and Valdivia (2011) who study training programs in Peru using randomized control

trials and show that they have little e�ect for borrowers in this context. Another example is

Lensink et al. (2011b) who use data for MFIs in 61 countries. They show that MFIs providing

both �nance and business development services have similar performance as MFIs providing no

6For a discussion on the reasons of shifting from group lending to individual lending seethe article by M. Yunus "Grameen Bank II: Lessons Learnt Over Quarter of A Century" athttp://www.grameen.com/index.php?option=com_content&task=view&id=30&Itemid=0, accessed 22November 2013.

34

1.2. The model

"plus" services. However MFIs with social services do signi�cantly better in terms of outreach.7

The rest of the chapter is structured as follows. In the next section we lay out the basic model

and analyze the "laissez-faire" benchmark. We then introduce successively the state guarantee

(section 3) and business development services (section 4). After having shown that the state

guarantee can have a counterproductive e�ect on business development services, we analyze

the alternative policy of business development services subsidization in section 5. In section 6

we model a mix of policies where state both guarantees loans and subsidizes BDS. Section 7

concludes and presents some possible extensions and limitations of the model.

1.2 The model

Our modeling is based on the classical corporate �nance model (see e.g. Tirole 2006). It consists

of a continuum of risk neutral entrepreneurs,8 each endowed with a project that needs a �nancing

D (identical for all agents). Each project can either succeed and generate a return of ρD or fail

and give zero return (the invested capital is then lost). The return on investment (ρ) is assumed

to be heterogeneous among agents (and distributed on[ρ, ρ]). To increase the probability of

success, an entrepreneur must exert a costly e�ort (unobserved by the MFI). For simplicity, we

assume that there are only two possible levels of e�ort, high (the entrepreneur behaves) and low

(the entrepreneur misbehaves). The probability of success with high e�ort (p) is higher than the

probability of success with low e�ort (p): p > p. However, if an entrepreneur chooses to exert

a low e�ort, he receives a private bene�t, ψ. If the entrepreneur behaves he receives no private

bene�t.

The principal (an MFI, or more generally a social bank) then chooses the projects she invests

in (that is the borrowers she lends D to) and sets the interest rate (r). We assume that the

expected pro�t of the MFI has to be zero for each contract. This framework corresponds well to

7For detailed examples of MFIs providing (themselves or not) non-�nancial services see Dunford (2001).8Risk aversion of borrowers won't impact our results, as there are no �rst derivative e�ects.

35


situations where MFIs are not for pro�t or face important competition.9 However, this setting is

less appropriate for markets where competition is low and MFIs claim for-pro�t objectives (for

example Compartamos of Mexico, see Rosenberg 2007).

The moral hazard issue comes from the unobservability of entrepreneurs' e�ort by the MFI.10

For an entrepreneur to exert high e�ort, the interest rate has to be incentive compatible. The

zero expected pro�t condition together with the incentive compatibility constraint will therefore

give the minimum project return threshold to receive �nancing and the interest rate.

In contrast to Tirole (2006), we assume that entrepreneurs have no capital to invest in their

project. This di�erence allows our model to capture the speci�c feature of the microcredit mar-

ket where borrowers often lack collateral.

Moreover, in line with Tirole (2006), we assume that the projects are only viable when the en-

trepreneur behaves, meaning that (i) the Net Present Value (NPV) in this case is positive i.e.

pρ > 1 ∀ρ, or pρ > 1 and (ii) the NPV of the projects is negative if the borrower misbehaves,

meaning that pρ < 1− ψD ∀ρ or pρ < 1− ψ

D .11

Let us �rst solve the model under "laissez-faire", that is without state intervention. The en-

trepreneur receives the total return of the project net of the capital due to the bank. He receives

ρD− (1 + r)D if the project succeeds and zero if it fails. We assume that ρ > 1 + r. Therefore,

the entrepreneur will face the following incentive compatibility constraint:

p [ρD − (1 + r)D] ≥ p [ρD − (1 + r)D] + ψ (1.1)

9Cull et al. (2011) argue that micro�nance industry faces increasing competition and McIntosh andWydick (2005) show that competition among MFIs decreases their capacity to use cross-subsidization.

10We however assume here that the actual investment and the success of the project are veri�able. Inother words, we do not consider the case where delinquent borrowers cannot be compelled to reimbursetheir credit (see Anderson et al. 2009).

11We keep these assumptions on the viability of the project for the rest of this chapter. The presentationof the conditions on NPV changes when the loan guarantee and BDS are introduced. Because of theirlimited interest we do not present them for each model. Note that they do not alter our results.

36

1.2. The model

This amounts, for a given interest rate, to the minimum return for which the borrower exerts

high e�ort:12

ρmin =ψ

D∆p+ (1 + r) (1.2)

where ∆p = p− p.13

When borrowers exert high e�ort, the expected pro�t of the MFI writes (note that it is indepen-

dent of project return):

E (π) = p (1 + r)D −D (1.3)

and the zero pro�t condition gives the benchmark interest rate:

r =1− pp

(1.4)

Introducing the latter expression for the interest rate in (1.2), we �nd that the bank will invest

in all the projects generating a return higher or equal to the threshold ρmin:

ρmin =ψ

D∆p+

1

p(1.5)

Up to now, our modeling of microcredit was limited to a classic loan without collateral. However,

at least two other major aspects are key to microcredit: state guarantee and business development

services. Let us successively include these two patterns starting with the loan guarantee.

12Note that the minimal threshold for the project return is indeed always greater than 1 + r. This willalways be the case in the rest of the chapter.

13A stronger moral hazard issue consists in the incentive for the borrower not to leave with the cash.In our model, this constraint would correspond to p being high enough. Still, it seems that in real worldsuch an incentive is driven by future borrowing opportunities and sustainable �nancial inclusion (forexample through the inclusion in the mainstream banking sector by the creation of a credit history). Amore complete model would therefore include the value of future opportunities � from the viewpoint ofthe borrower � in case of success. This will not change our results. More precisely, in the case of a netpresent value of future borrowing opportunities V , independent of project's present return ρ, this wouldjust add a term −V/D to equation (1.2). This term being present in all the models presented hereafter,it does not impact our comparisons and conclusions.

37


1.3 The introduction of the loan guarantee

The loan guarantee is an essential tool for the expanding microcredit market.14 By reducing

the risk taken by the MFI, such a policy aims at crowding in a part of the initially excluded

borrowers.

In accordance with real world experience (see previous footnote), we assume that the state

guarantees a proportion γ < 1 of the outstanding loan if the project fails. As it only impacts

the consequence of project failure for the MFI, it changes neither borrowers' return nor their

incentive compatibility constraints. However, the zero expected pro�t condition then becomes:

E (π) = p (1 + rγ)D + (1− p) γD −D = 0 (1.6)

leading to an interest rate equal to:

rγ =1− pp

(1− γ) (1.7)

which is, not surprisingly, lower than the benchmark interest rate. We end up (using (1.2)) with

ργ =ψ

D∆p+

1− γ (1− p)p

(1.8)

where ργ represents the minimum return that a project should generate to be �nanced by the

MFI, in the presence of the state guarantee.

Therefore, as expected, the minimum project productivity threshold required for �nancing de-

creases due to the loan guarantee (ργ < ρmin). The intuition behind this result is simple: the

14For example, in France, several public organisms guarantee capital in case of loss: the "Fonds deCohésion Sociale" or Caritas (50% of the outstanding principal and unpaid interest) for consumer loans(that aim at �nancing goods that contribute to job seeking, such as cars, computers, business suits,etc...) and "France Active Garantie" (70% of the outstanding principal) for self-employment or smallbusiness loans. These guarantees are free from the MFI's point of view. More recently the EuropeanCommission and the European Investment Bank started providing up to 75% guarantee for microcreditsin the European Union through the European Progress Micro�nance Facility.

38

1.4. Modeling Business Development Services (BDS)

interest rate represents an "insurance" for the bank against high-risk agents. With the loan

guarantee, the government will bear a part of this costly "insurance". The MFI will provide

microcredits at a lower interest rate. Hence, a higher number of the entrepreneurs will optimally

exert high e�ort and a higher number of the projects will be �nanced. Thus, the loan guarantee

reduces credit rationing and can therefore allow the state to save on other social expenses, such as

unemployment bene�ts.15 It is important to emphasize that we do not investigate the �nancial

e�ciency of a loan guarantee program. The total gains from successful microcredit cannot indeed

be easily identi�ed as it may lead to lower unemployment bene�ts, better education for children

or better health for example. Obviously, there is a range of non-appropriable bene�ts ignored by

the single market approach that should be taken into account. However, the cost-bene�t analysis

is beyond the scope of this chapter.16

1.4 Modeling Business Development Services (BDS)

As we have already noted, business development services are another key feature of micro�nance.

MFIs often provide services that aim at increasing the probability of entrepreneurs' projects to

succeed (for example accounting or management trainings that help microborrowers to run their

business).

Akin to the "advisor model" presented by Tirole (2006), we model business development services

as an action provided by the MFI (at a �xed cost K per contract) that increases by an amount ε

the probability of entrepreneur's project to succeed. For the sake of simplicity, BDS are modeled

as uniformly increasing the probability of project success.17 If the MFI provides BDS, the

probability of success with high and low e�ort becomes respectively pε = p+ε and pε

= p+ε. The

15According to Brabant et al. (2009), it is cheaper � in the case of France � to subsidize microcreditthan to pay welfare bene�ts to microborrowers.

16Such an analysis would still be very di�cult to implement, as noted in Armendariz and Morduch(2010).

17In a broader model, BDS could be correlated to the level of e�ort put in the project, its intrinsicquality or the entrepreneur's ability.

39


independence of the increase in the probability of success as a result of BDS from borrower's e�ort

considerably simpli�es the model. It implies that entrepreneur's behavior does not depend on the

choice of MFI to provide BDS (it does not change ∆p). Therefore, the incentive compatibility

constraint in the presence of BDS remains the same (inequality (1.1)) and the minimum project

return compatible with e�ort is still de�ned by equation (1.2). However, relaxing this assumption

� and allowing for some complementarity between BDS and e�ort � will loosen the incentive

constraint of the borrowers. This will therefore tend to increase the range of borrowers �nanced

in the presence of business development services.

1.4.1 In the laissez-faire case

Let us assume for now that the lending institution bears the cost of the business development

services, K (independent form the project productivity ρ). In case it provides BDS, the expected

pro�t of the MFI is:

E (π) = (p+ ε) (1 + rε)D −D −K = 0 (1.9)

and the equilibrium interest rate charged to clients receiving BDS is:

rε =1− (p+ ε)

p+ ε+

K

(p+ ε)D(1.10)

Note that it is not straightforward to compare the equilibrium interest rate in the presence of

BDS and the benchmark interest rate r. Finally, using (1.2) and (1.10), we �nd the minimum

return required by the MFI when it engages in BDS:

ρε =ψ

D∆p+

1

p+ ε+

K

(p+ ε)D(1.11)

Lemma 1.4.1 The availability of business development services will increase the �nancial in-

40


clusion of borrowers (i.e. ρmin > ρε) if and only if

ε

p>K

D(1.12)

that is if and only if the relative gain in probability of success generated by business development

services exceeds its relative cost.18

. In other words, the condition in Lemma 1 states that BDS will be provided only in case where

the cost of the training is not too high to the MFI.19 If the latter condition is not satis�ed,

i.e. if ρmin < ρε, no BDS will be provided as they will not crowd-in any additional borrower.

On the contrary, if ρmin > ρε, all the entrepreneurs with projects generating a return belonging

to the interval [ρε, ρmin) will be �nanced and will receive BDS. Regarding the entrepreneurs

with project return higher than ρmin, the MFI is indi�erent between providing them BDS (and

charging them the interest rate rε) or not (and charging them the interest rate r). However, one

can imagine that the presence of capacity constraints (for training groups for example) insures

that only borrowers who need BDS in order to be �nanced (those with ρ < ρmin) will receive

these services.

1.4.2 In the presence of the state guarantee

Let us now study how business development services interact with the state guarantee. This is

a promising analysis as intuition suggests that state intervention might lower the incentive for

the MFI to provide such services.

18Alternatively we could assume that BDS increase project's return ρ by an amount τ , without im-pacting the probability of success. In this case the interest rate charged to the borrower is strictly highercompared to the benchmark case and the incentive compatibility constraint is di�erent from inequality(1.1). However, this alternative modeling does not alter our results which will still depend on the tradeo�between the relative gains and costs of BDS. For instance, the condition in Lemma 1 would write τp > K

D .19Assuming complementarity between BDS and e�ort, the condition in Lemma 1 becomes weaker

εp + ψ

D ·∆pε−∆p∆p∆pε

(p + ε) > KD , where ∆pε > ∆p represents the di�erence between the probabilities of

success with and without e�ort in the presence of BDS. Moreover, if ∆p < ∆pε, then ρε < ρmin wouldnot necessarily imply rε < r.

41


When the state guarantees a proportion γ of the loan, the zero pro�t condition of an MFI

providing BDS writes:

E (π) = (p+ ε) (1 + rγε)D + (1− (p+ ε)) γD −D −K = 0 (1.13)

implying:

rγε =1− (p+ ε)

p+ ε(1− γ) +

K

(p+ ε)D(1.14)

While it is easy to note that rγε < rε (the state guarantee decreases the interest rate), the

comparison of rγε with rγ is not trivial. Put di�erently, as in the previous section, depending on

their cost, business development services may increase the interest rate.

As previously, our simpli�cations ensure that borrowers' behavior is not impacted by BDS.

Therefore, using equation (1.2), we obtain the minimum return required by the bank in the

presence of both business development services and the state guarantee:

ργε =ψ

D∆p+

1− γ (1− (p+ ε))

p+ ε+

K

(p+ ε)D(1.15)

We therefore have ργε < ρε and obviously, in the presence of BDS, the state guarantee increases

the range of borrowers who are �nanced. However, it is not clear whether BDS are actually used

in the presence of the state guarantee (that is if ργε < ργ). Lemma 2 provides the condition

under which BDS crowd-in additional borrowers when loans are guaranteed.

Lemma 1.4.2 In the presence of the state guarantee, the provision of BDS by the MFI will

increase the �nancial inclusion (i.e. ργε < ργ) if and only if

ε

p>

K

(1− γ)D(1.16)

. Under condition (1.16), the MFI will provide BDS to borrowers with project returns between

ργε and ργ , and will be indi�erent between providing or not BDS to borrowers with ρ > ργ .

42


Still, in the presence of capacity constraints for training programs, it is consistent to assume

that BDS are only o�ered to borrowers with ρ < ργ , who would not be �nanced otherwise.

Comparing Lemma 1 and Lemma 2, it appears that condition (1.12) is weaker than condition

(1.16). This might indicate that BDS crowd-in less borrowers when the state guarantees loans.

This is clearly the case when KD < ε

p <K

(1−γ)D as then no borrowers are crowed-in in the presence

of the state guarantee through BDS, contrarily to what would happen without state intervention.

Whether this is also the case when BDS are used in the presence of loan guarantee (that is when

εp > K

(1−γ)D ) depends on the distribution of project returns. Business development services

then crowd-in borrowers with project returns in between ρε and ρmin in the absence of state

intervention; and in between ργε and ργ if the state guarantees loans.

In the simple case of a uniform distribution of project returns, less borrowers will be �nanced

through BDS in the presence of the state guarantee if ργ − ργε < ρmin − ρε. As we have

ργ − ργε =εD (1− γ)−Kpp (p+ ε)D

(1.17)

and

ρmin − ρε =εD −Kpp (p+ ε)D

. (1.18)

Proposition 1 holds.

Proposition 1.4.1 Under condition (1.16), if the distribution of the project returns is uniform,

the number of additional entrepreneurs �nanced through business development services is larger

without the state guarantee.

. The intuition behind this �nding relies on the return to the MFI from BDS in the presence of

the state guarantee which writes:

εD[(1 + r)− γ]

and is decreasing in γ. This negative relation explains why the MFI is less incited to provide

BDS if it bene�ts from the loan guarantee. As a result, less borrowers are crowded-in through

43


BDS under the loan guarantee compared to no state intervention. Proposition 1 can be related

to classical results in the insurance literature. In our context, BDS can indeed be understood

as a self-protection e�ort (an e�ort that decreases the probability of incurring a loss) exerted by

the MFI. The loan guarantee can be interpreted as a (free) insurance for the MFI (by decreasing

the size of the potential loss). Therefore, our result is related to the substitutability between

insurance and self-protection, found when the price of insurance is independent of the e�ort of

self-protection (Ehrlich and Becker 1972).

Proposition 1 might however not hold if the distribution of the project returns is not uniform. In

particular, if project returns are highly concentrated on the interval [ργ , ργε], then a smaller inter-

val would not necessarily result in a smaller number of �nanced projects. Moreover, Proposition

1, only holds under condition (1.16) implying that the cost of BDS should be low enough.

1.5 An alternative policy: business development ser-

vices subsidization

The possible perverse e�ect that the state guarantee can have on business development services

leads us to analyzing an alternative policy that consists of BDS subsidization. Such a policy en-

compasses both the direct subsidization of the cost of BDS paid by the MFI and the subsidization

of NGOs or associations that o�er BDS to microborrowers. Our approach can again be related

to papers in the insurance �eld that analyze subsidies or the public provision of preventive goods

(see e.g. Arnott and Stiglitz 1986; Lee 1992). However, all these papers mainly discuss the e�ect

of such policies on the price of insurance which is absent in our model (as we assume that the

loan guarantee is free from the viewpoint of the MFI).

The aim of this section is to identify when BDS subsidization will do better in terms of �nancial

inclusion compared to the loan guarantee coupled with unsubsidized BDS. To do so, we compute

the minimum project return required by the MFI when the government subsidizes business de-

44

1.5. An alternative policy: business development services subsidization

velopment services and compare it to the one with the state guarantee and unsubsidized BDS.

This comparison will then allow us to de�ne the most e�ective policy when the objective of

the government is to increase the number of the entrepreneurs �nanced under a �xed budget

constraint.

We assume that if the government chooses to subsidize BDS, it bears the total cost of the pro-

gram. Under subsidization, the MFI bene�ts from the increase in the probability of success of

the projects without paying the cost of BDS. The zero pro�t constraint for subsidized borrowers

writes:

E (π) = (p+ ε) (1 + r)D −D = 0 (1.19)

which gives as interest rate:

r =1− (p+ ε)

p+ ε(1.20)

Using (1.2), the minimum project productivity threshold is then:

ρ =ψ

D∆p+

1

p+ ε(1.21)

that we compare with the minimum project return threshold under the state guarantee and

unsubsidized BDS, (i.e. ργε).

Lemma 1.5.1 The necessary conditions for BDS subsidization to increase the �nancial inclusion

with respect to loan guarantee (i.e. ρ < ργ and ρ < ργε) are

ε

p+ ε> γ(1− p) (1.22)

and

K > γD(1− (p+ ε)) (1.23)

45


. The above Lemma states that BDS subsidization will crowd-in more borrowers than the loan

guarantee if (i) BDS are e�cient enough and (ii) the amount spent on BDS subsidization (for

the borrowers with lowest project returns) is greater than the expected amount spent on the

loan guarantee.

Lemma 3 provides conditions under which subsidizing BDS for borrowers with a return ρ < ρmin

will crowd-in more borrowers than guaranteeing a proportion γ of all the loans. An interesting

question consists in analyzing to what extent these conditions can be ful�lled when the state

budget is held �xed amongst the two policies. To answer this question, we assume that the

government faces a �xed budget equal to nsK, where ns is the number of borrowers bene�ting

from BDS subsidization. We infer the rate of guarantee γ corresponding to the �xed budget

condition.20 Letting n be the total number of �nanced borrowers, and nε be the number of

borrowers �nanced through unsubsidized BDS,21 the �xed budget condition writes:

[nε(1− (p+ ε)) + (n− nε)(1− p)] γD = nsK (1.24)

leading to

γ =nsK

[nε(1− (p+ ε)) + (n− nε)(1− p)]D. (1.25)

When ns = nε = n, i.e. when all the borrowers receive BDS and BDS are fully subsidized, we

have:

γ =K

D (1− (p+ ε))(1.26)

20For the sake of simplicity we assume that the guarantee rate adjusts such that the total expected ex-penditure in the case of the loan guarantee is equal to the total expenditure in the case of full subsidizationof BDS. An alternative strategy could be considering partial subsidization of BDS.

21nε is smaller or equal to n as we have seen in section 4.2 that the MFI is indi�erent between o�eringor not BDS to clients with project returns higher than ργ . The MFI might have an incentive not to o�erBDS to all the borrowers due to capacity constraints.

46

1.5. An alternative policy: business development services subsidization

implying ρ = ργε. Therefore, provided that BDS are e�cient enough in the presence of the loan

guarantee (i.e. inequality (1.22) holds22), when all the borrowers are o�ered BDS and BDS are

fully subsidized, BDS subsidization is exactly equivalent to loan guarantee in terms of �nancial

inclusion.

However, as γ is increasing in both ns and nε, as soon as either nε < n or ns < n, that is as

soon as BDS or subsidies are targeted toward borrowers with the lowest returns, we have:

γ <K

D (1− (p+ ε))(1.27)

implying that condition (1.23) holds. In this case two e�ects are at stake. First, if in the presence

of the loan guarantee the MFI chooses not to provide BDS to all the borrowers (nε < n), the

probability of failure for projects without BDS will be equal to 1 − p. The expected cost of

guaranteeing them would then amount to γD(1 − p) > γD(1 − (p + ε)) per project. Second,

if the state does not subsidize BDS for all the borrowers (ns < n), it will save K on n − ns

borrowers while it would still spend some money on them in the case of the loan guarantee.

These two e�ects corroborate that the loan guarantee is more expensive than BDS subsidization

when the total number of �nanced borrowers is held constant. We summarize our results in the

following proposition:

Proposition 1.5.1 If BDS are e�cient enough (that is if condition (1.22) is satis�ed) and are

targeted toward the borrowers with the lowest project returns (either directly by the MFI or through

subsidies), then the state can crowd-in more borrowers with the same budget by subsidizing BDS

rather than guaranteeing loans.

. The intuition behind this result is straightforward. It states that, by concentrating its e�ort

on the otherwise excluded borrowers, the state can increase �nancial inclusion. This is easier to

implement with BDS subsidization, as we have shown above that the MFI is indi�erent between

22Note that inequality (1.22) (implying that ρ < ργ) is equivalent to inequality (1.16) (implying thatργε < ργ) when ρ = ργε.

47


o�ering or not BDS to projects with the highest returns. Moreover, although it seems easy (for

the state) to make subsidies to BDS scarce, guaranteeing only some of the loans (that would have

the same e�ect) appears to be more di�cult. Such a strategy is also in line with reality, where

MFIs do not usually provide training to each client (for example due to capacity constraints)

but all the loans are guaranteed. Finally, it should be noted that this analysis only holds under

condition (1.22), that is if ε is high enough and γ is low enough. Therefore, if BDS technology is

not e�cient enough or if the level of guarantee is very high, BDS subsidization will not crowd-in

more borrowers compared to the loan guarantee.

In this section we have focused on a policy consisting of BDS subsidization and we have compared

it with the loan guarantee. More generally, the government might choose to mix policies by both

guaranteeing loans and subsidizing BDS. In the next section, we model a mix of these two policies

and show that it can eliminate the perverse e�ect of the loan guarantee discussed in section 4.2.

Finally, we analyze its impact on the �nancial inclusion.

1.6 Mixing policies

In this section we assume that the government both guarantees loans and subsidizes BDS. We

analyze if this mix can remove the perverse e�ect previously identi�ed and improve �nancial

inclusion. To do so, we study a case where the state guarantees a proportion γ′ < 1 of the

outstanding loan if the project fails and partly subsidies BDS, in the sense that it �nances a

part α ≤ 1 of its cost. In line with reality we assume that state keeps constant the guarantee

rate across all the borrowers. Following the analogy with (health) insurance literature developed

above, this study can be linked to papers analyzing the e�ciency of contracts covering both

disease prevention and treatment (Ellis and Manning 2007). Indeed, such a policy aims both at

decreasing the �nancial loss for the MFI in case of project failure and at increasing its incentive

to provide BDS. Ellis and Manning (2007) �nd that it is always desirable to o�er at least some

insurance coverage for prevention when individuals ignore its impact on prices.

48

1.6. Mixing policies

Let us �rst study to what extent the mix of policies can solve the perverse e�ect of the loan

guarantee. To do so, we compute the project return interval for borrowers �nanced through

BDS. Using equation (1.8), we derive that, without BDS, the MFI will �nance all projects with

a return higher or equal to:

ρ1 =ψ

∆pD+

1− (1− p)γ′

p(1.28)

If the state additionally �nances a proportion α of BDS cost, the zero-pro�t condition for bor-

rowers that bene�t from both policies writes:

E(π) = (p+ ε)(1 + r2)D + (1− (p+ ε))γD(1− α)−D − (1− α)K = 0 (1.29)

and the optimal interest rate for these borrowers is given by:

1 + r2 =1 + (K/D)(1− α)− (1− (p+ ε))γ′

p+ ε(1.30)

Therefore, under the mix of policies, the MFI �nances all the projects with returns higher than

ρ2 =ψ

∆pD+

1 + (K/D)(1− α)− (1− (p+ ε))γ′

p+ ε(1.31)

Note that ρ2 is decreasing in both α and γ′ suggesting that a higher α and γ′ imply larger

�nancial inclusion. The MFI will provide BDS under the mix of policies if ρ2 < ρ1.

Lemma 1.6.1 Business development services will increase �nancial inclusion when the state

guarantees loans and subsidizes a part of BDS (i.e. ρ2 < ρ1) if and only if:

ε

p>

(1− α)K

(1− γ′)D(1.32)

. As in previous sections, the condition in Lemma 4 states that the relative gain from training

has to be higher than its relative cost to the MFI. This condition will be easier to satisfy when α

is large and γ′ is small, meaning that subsidizing BDS is more e�cient when the loan guarantee

49


is low. Moreover, comparing (1.32) with (1.12), it appears that BDS are more likely to be used

in the presence of a mix of policies than in the "laissez-faire" case (that is condition (1.32) is

weaker than (1.12)) if and only if α > γ′.

When ρ2 < ρ1, that is under condition (1.32), BDS crowd-in borrowers with project return

between ρ2 and ρ1. Therefore, when project returns are uniformly distributed, the number of

borrowers receiving a loan as a result of BDS subsidization is given by:

ρ1 − ρ2 =ε(1− γ′)− p(K/D)(1− α)

p(p+ ε)(1.33)

Comparing this di�erence with ρmin − ρε (given in equation (1.18)), that is with the number of

borrowers �nanced through BDS in the "laissez-faire" case, we �nd the condition under which a

mix of policies eliminates the perverse e�ect of the loan guarantee.

Proposition 1.6.1 If the distribution of the project returns is uniform, the perverse e�ect high-

lighted in Proposition 1 will be eliminated by the mix of the loan guarantee and BDS subsidization,

if and only if

αK

γ′D>ε

p(1.34)

. Condition (1.34) shows that the perverse e�ect of the state guarantee on the number of

additional borrowers crowded-in through BDS can be avoided under the mix of policies. Again,

it will be easier to satisfy when α is large and γ′ is small. Moreover, using equation (1.32), it

appears that a necessary condition for BDS subsidization to actually crowd-in more borrowers

than in the "laissez-faire" case is α > γ′. This means that the part of BDS cost subsidized by

the state has to be larger than the part of the loan guarantee.

However, Proposition 3 does not show when the mix of policies will increase �nancial inclusion

compared to the loan guarantee or BDS subsidization alone. To study �nancial inclusion under

the mix of policies, let us compare ρ2 to ρ and ργε (we restrict the analysis to the case where

BDS are e�cient enough, that is where ρ2 < ρ1 and ργε < ργ).

50

1.6. Mixing policies

We �rst compare ρ2 with ργε, that is the project return threshold under the mix of policies (loan

guarantee at rate γ′ and BDS subsidization at rate α) with the project return threshold under

the loan guarantee (at a rate γ) and unsubsidized BDS. Using (1.15) and (1.31), we �nd that

ρ2 < ργε if and only if:

(1− (p+ ε))(γ − γ′)D < αK (1.35)

Inequality (1.35) states that the amount saved by decreasing the guarantee rate should be lower

than the cost of BDS subsidization (for a mix of policies to do better than loan guarantee). As in

the previous section when all the borrowers are o�ered BDS, under a �xed budget, both policies

are equivalent in terms of �nancial inclusion (that is αK = (1 − (p + ε))(γ − γ′)D). However,

when some borrowers are not o�ered BDS, following our previous reasoning, a mix of policies

will do better than the loan guarantee under a �xed budget. In this case, the government can

crowd-in the same number of borrowers at a lower cost, as it saves the guarantee rate di�erence

on borrowers that do not bene�t from BDS (then, for borrowers who bene�t from subsidies, we

have αK > (1− (p+ ε))(γ − γ′)D, i.e. condition (1.35) holds). We stress that this will only be

the case when BDS technology is e�cient enough, that is under condition (1.32).

Similarly, when comparing the mix of policies with BDS subsidization alone, we �nd using (1.21)

and (1.31) that ρ2 < ρ if and only if

(1− (p+ ε))γ′D > K(1− α) (1.36)

Akin to the previous discussion, under a �xed budget, we �nd that both policies are equivalent

(in terms of �nancial inclusion) when all the borrowers receive BDS. However, BDS subsidization

does better than the mix of policies when some borrowers are �nanced without receiving BDS.

Indeed, in this case (1.36) cannot be satis�ed as for borrowers �nanced without BDS (with ρ > ρ1)

the mix of policies is costly from the point of view of the state (due to the loan guarantee that

targets all the borrowers) contrarily to BDS subsidization. These results can be summarized in

51


the following proposition.

Proposition 1.6.2 When BDS are e�cient (that is under condition (1.22)) the �nancial inclu-

sion is maximized when the state subsidizes BDS (and does not guarantee loans) provided that

BDS subsidies are targeted enough.

. Of course, there are mechanisms that might undermine Proposition 1.6.2. First, BDS have to

be e�cient enough. Second, the MFI has to �nance projects that can be incentive compatible

without BDS (that is projects with returns higher than ρmin). This notably involves that such

projects are not �nanced by mainstream banks. This can be due to the presence of collateral

requirements or the presence of rationed borrowers (who do not receive credit despite sharing

the same characteristics as the accepted borrowers) highlighted by Stiglitz and Weiss (1981). If

the MFI has no such clients in its portfolio, a mix of policies might do better. This is due to the

loan guarantee that increases the range of project returns that are incentive compatible without

BDS, by reducing the interest rate. Moreover, the state might guarantee loans o�ered by MFIs

for exogenous reasons, such as possible correlations among project defaults that would jeopardize

MFIs.

Proposition 1.6.2, related to the substitutability between the loan guarantee and BDS subsidiza-

tion, seems to highly rely on the linearity of our model. Such a result is not typically obtained

in the case of risk-averse individuals in the insurance literature. In our context, borrowers' risk

aversion will not alter the �ndings as the loan guarantee does not change the consequences of

project failure for them. However, results might change if the MFI is not risk-neutral. This

question echoes the literature on risk aversion of nonpro�t organizations. Indeed both theoreti-

cal (Wedig 1994) and empirical (e.g. Preyra and Pink 2001 in the case of non pro�t hospitals)

papers indicate that nonpro�t organizations are risk-averse. Whether risk aversion applies to

MFIs remains an open question.

Another extension that might modify our results is the possibility of cross-subsidization by the

MFI. Cross-subsidization occurs when pro�ts made from lending to pro�table borrowers are be-

ing used to �nance non pro�table ones. Even though Armendariz and Szafarz (2011) identify

52

1.7. Concluding remarks

cross-subsidization as one of the components of MFI's social objectives, McIntosh and Wydick

(2005) show that the increasing competition faced by MFIs decreases their capacity to use cross-

subsidization. Moreover, in our context, the study of the cross-subsidization would require

assumptions on the distribution of ρ among the potential borrowers.

1.7 Concluding remarks

To conclude, we analyze in this chapter the impact of state intervention on �nancial inclusion

in the microcredit market where micro�nance institutions o�er individual loans and business

development services. We focus on the interaction between the loan guarantee and the choice

of the micro�nance institution to provide BDS. Our motivation relies on the intuition that

the loan guarantee might impact the MFI's involvement in business development services and

probably deteriorate their e�ciency in terms of �nancial inclusion. This intuition �nds its roots

in the substitutability between insurance and self-protection classically found in the insurance

literature. Indeed, the loan guarantee can be understood as an insurance against project failure

(for the MFI), whereas business development services act as a self-protection device by lowering

the probability of project failure.

By extending Tirole's (2006) model to the microcredit market with the loan guarantee and

business development services we prove that the state guarantee can be counterproductive in

terms of the number of entrepreneurs �nanced as a result of business development services (in

particular when the distribution of the project returns is uniform). This central �nding leads

us to study an alternative solution: business development services subsidization and then to

compare it to loan guarantee in terms of �nancial inclusion. We �nd that � for a �xed budget �

BDS subsidization can lead to higher �nancial inclusion than the loan guarantee, provided that

(i) BDS technology is e�cient enough and that (ii) BDS are targeted enough toward otherwise

excluded borrowers. Finally, we show that, even though it can eliminate the counterproductive

e�ect of the loan guarantee, a mix of the loan guarantee and BDS subsidization will not lead

53


to higher �nancial inclusion compared to BDS subsidization alone if BDS are targeted and are

e�cient enough.

One of the limitations of our model concerns the interactions of the microcredit market with the

missing markets. Indeed, state intervention in the credit market can have interesting implications

for the labor market for example (see Emran et al. 2011). In the present chapter we focus on the

"pure" impact of state intervention on the lending behavior of an MFI. The investigation of the

�nancial e�ciency of the public intervention is left for further research. This will in particular

be needed to explain why the state chooses to participate in the microcredit market. Moreover,

some of our results (mostly those on the mix of policies) seem to be driven by the linearity of

our model. It might be interesting to challenge them in the case of a risk-averse MFI or an MFI

using cross-subsidization.

54

Chapter 2

Informed Principal and the Microcredit

Market: Should Business Training be Targeted

towards the Least Able Borrowers?1

2.1 Introduction

Microcredit is a small-scale �nancial tool designed for individuals who are rejected from the

classical �nancial market.2 Micro-borrowers are often unemployed, lacking collateral and any

experience in starting a business. After having been widely used and studied in developing

countries, microcredit is now largely spreading to developed countries. While its main objective

- poverty alleviation - is common for both types of economies, its implementation has several

particularities in the developed countries. For example, individual lending is prevalent,3 loans

don't speci�cally target women and there is generally a high government implication through

1This chapter is based on a joint work with Renaud Bourlès, Dominique Henriet, and Xavier Joutard.2From this perspective, microcredit can be viewed as a solution to credit rationing occurring on the

classical credit market with imperfect information ine�ciencies. For details on credit rationing see Stiglitzand Weiss (1981).

3Group lending is a well known lending methodology in micro�nance which is particularly successfulin rural environments with tightly-knit networks (Postelnicu et al. 2013). Nevertheless, since recently indeveloping countries an increasing number of MFI's implement individual lending methodology (Armen-dariz and Morduch 2010).

55

Informed Principal and the Microcredit Market

guaranties or subsidies4 in industrialized economies. Another important characteristic of mi-

cro�nance in the developed countries is the presence of formal business training,5 also termed

Business Development Services (BDS). European Micro�nance Institutions (MFIs) have been

involved in BDS since their emergence (Lammermann et al. 2007).

Business training refers to di�erent kinds of additional non-�nancial support services that ac-

company loans. They may consist of de�ning and developing the business project (studying the

pro�tability, setting the commercial strategy and �nancing needs, administrative help), informa-

tion and help to obtain �nancing, trainings in accounting, management, marketing, law and the

follow-up of the project.

By means of a theoretical model and empirical evidences we study how an MFI assigns borrowers

to training programs. We use data on applicants of a French MFI. This MFI faced two types of

decisions. First, it approved or rejected applicants for a microcredit. Second, conditionally on

loan approval, it assigned some of the borrowers to a training program. We also have information

about unpaid installments and defaults of the clients, which we used to assess the riskiness of

the borrowers. Interestingly, data suggests that individuals assigned to training are not riskier

ex-post than individuals without training.This evidence re�ects two possible scenarios: either

training is targeted toward (ex-ante) high-risk individuals and is highly e�cient (as ex-post bor-

rowers with training are not riskier than borrowers without training) or training is not targeted

exclusively toward high-risk borrowers.

Existing literature is agnostic about the e�ciency of training in micro�nance (see for example

Karlan and Valdivia (2011, Lensink et al. (2011b) for developing countries, and Balkenhol et al.

(2013, Evans (2011, Edgcomb (2002) for developed countries). Overall, studies in both developed

4D'Espallier et al. (2013) argue that the lack of subsidies worsens social performances of MFI'sworldwide.

5Formal business training is provided through special arrangements, conditions or contracts between abusiness development agency/micro�nance institution and the entrepreneur (Lammermann et al. 2007).In developing countries training programs that accompany loans are less formal and often consist ofsocial development programs, such as information on health, civil responsibilities and rights, rules andregulations of the bank (McKernan 2002).

56

2.1. Introduction

and developing economies fail to corroborate the �rst scenario where business training is mainly

targeted toward the riskiest individuals and is highly e�cient. This gives credit to the second

scenario, where BDS do not necessarily target only high-risk borrowers.

Consequently, training assignment process appears to be somewhat puzzling and the aim of

this chapter is to tackle this puzzle. Generally, micro-entrepreneurs need �nancing to start-up

a business for the �rst time in their life. Therefore, they lack the necessary entrepreneurship

experience. It is plausible that the micro�nance institution is better informed than the borrower

about the potential success of the project due to its past experience. In this context, the contract

o�ered by the MFI (assignment to a training program or not) can reveal some information to

the borrower about his self and impact his actions. This mechanism, termed looking-glass self

(Cooley 1902), provides a rationale to the hypothesis where the MFI having superior information

might indeed not target business training exclusively toward high risk individuals.

In our theoretical model, the MFI will not provide training to some borrowers in order to shape

borrowers' behaviour. This is possible due to MFI's superior information about borrower's type.

In this context, we are looking for equilibria where assignment to a business training is not

necessarily bad news. To do so, we provide a theoretical model where both the MFI - through

business training - and the borrower - through e�ort - can impact the probability of success of

the project. Borrowers are heterogeneous on their type6 which enters the probability of success

of the project. The MFI, �rst, decides on loan granting and, second, on training assignment.

We �rst develop a discrete theoretical model with three types of agents. This simple model

allows us to illustrate the main intuitions underlining how MFI's decision on training assignment

varies with borrower's type. A more general continuous model is then presented. It allows us

to illustrate how a continuous level of business training varies with a continuous type of the

borrower. For both models we �rst present a framework under symmetric information where the

level of business training chosen by the MFI is decreasing with the type of the borrower. Second,

6By "type" we mean the intrinsic probability of the borrower to succeed. It can also be interpretedas the risk of the borrower.

57


we develop a model under reversed asymmetric information where the MFI has informational

advantage on the type of the borrower. This situation is unarguably plausible on the microcredit

marked where the experienced MFI faces an inexperienced borrower. Contrary to the �rst-best

solution, under reversed asymmetric information we show that a non-monotonic relation between

business training and the type of the borrower may arise. We more precisely show that there

exists a Perfect Bayesian Equilibrium where optimal level of business training is a non-monotonic

concave function of type (or risk): the optimal level of business training is �rst increasing with

the type of the borrower, and beyond a certain threshold it is decreasing.

Both models have advantages and drawbacks. The discrete model gains in simplicity and clarity

of the argument. It additionally allows us to compare the payo�s to the MFI under symmetric

and reversed asymmetric information and provide conditions under which the MFI is better-

o� when it has superior information. The continuous model has the inconvenience to be more

complex. Nevertheless, its advantage consists in results that are more directly testable in our

empirical exercise, where the type of the borrower and business training provision are modeled

as continuous variables. Therefore the presence of these two complementary models improves

the clarity of our argument and eases the transition to the empirical exercise.

In the empirical exercise we test for the existence of the looking-glass self mechanism using data

from a French MFI. We develop a credit scoring model using a bivariate probit model where we

control for individual unobserved heterogeneity. We proxy the type of the borrower discussed

in the theoretical model by his estimated intrinsic risk, which is modeled by means of a probit

regression. Using the simplest form of non-linearity, by introducing the estimated risk and its

square term in the business training probit regression, we �nd that the probability to be assigned

to a business training is �rst increasing with borrowers' type and beyond a certain threshold it is

decreasing. The mechanism behind the Perfect Bayesian Equilibrium outlined in the theoretical

model seems therefore to be veri�ed in real life.

This result, moreover, is robust to controlling for selection bias (Heckman 1979; Boyes et al.

1989) and to the use of the inverse of the survival time of the loan, instead of borrower's risk, as

58

2.1. Introduction

a measure of borrower's type (Roszbach 2004).

Our analysis contributes to at least three strands of literature concerning, respectively, the the-

oretical e�ect of reversed asymmetric information, the role of training programs in micro�nance

and the empirics of scoring models.

First our chapter is an interdisciplinary contribution focusing on sociological concepts such as

looking-glass self and self-con�dence. The looking-glass self e�ect occurs when social environ-

ment attempts to manipulate one's self-perception. This phenomenon has been widely studied

in sociological literature. "Looking-glass self" terminology is for the �rst time introduced by

Cooley (1902), who argues that people obtain a sense of who they are by observing how others

perceive or treat them. This concept has been further developed by Mead (1934) suggesting

that the sense of self is a by-product of social interactions. In economic literature this concept is

introduced latter by Benabou and Tirole (2003a) and Benabou and Tirole (2003b). Benabou and

Tirole (2003b) discuss three necessary ingredients for the looking-glass self e�ect to operate.96

First, the principal must have private information relevant to the agent's behavior. Second, the

principal must be incited to vary rewards or feedback according to her private information about

the agent. Finally, the agent must be aware about principal's superiority of information and

her motivations to impact agent's behavior. Benabou and Tirole (2003a) study many situations

where the principal might be better informed than the agent (for example at school, in the labor

market and family). Concerning help from the principal, authors show that it is always bad

news for the agent in equilibrium.7 Akin to Benabou and Tirole (2003a) we use a principal-agent

model where we incorporate the three ingredients discussed by Benabou and Tirole (2003b). In

contrast we will study situations where "help", through business training, is not always bad news

for the bene�ciary.

Principal's superiority of information assumption is a relatively novel approach in contrast to

7Other situations where help, more generally, can be detrimental to the agent are presented by Gilbertand Silvera (1996). Using di�erent experiments authors show that help can be used to undermine thebeliefs of the observers who might attribute a successful performance to help rather than to performer'sabilities.

59


the standard assumption where the agent is the only individual to observe his type. However,

it can only be applied in speci�c contexts. Ishida (2006), uses a principal-agent model to show

that promotions in the labor market can be used strategically in presence of the looking-glass

self e�ect. Villeneuve (2000) studies pooling and separating equilibria in the context where the

insurer evaluates risk better than its customers. Swank and Visser (2007) show how delegation

and increased attention from an informed employer can improve the motivation of an uninformed

employee. Nevertheless, authors highlight that their model only suits situations where the agent

is at the beginning of his career or performs tasks for the �rst time in his life, whereas the prin-

cipal has previous experience with similar tasks or agents. This framework suits remarkably well

the microcredit market, where micro-entrepreneurs borrow from an experienced MFI to start a

business for the �rst time. Consequently, the looking-glass self e�ect plays an important role in

the microcredit context. To our best knowledge, our contribution is the �rst one to introduce

reversed asymmetric information in micro�nance.

Importantly, the looking-glass self e�ect operates by impacting one's self con�dence. Self-

con�dence enhances motivation. This has important implications in economic contexts where

agents process information and make decisions (Benabou and Tirole 2002). Filippin and Paccagnella

(2012) argue that di�erences in self-con�dence may result in important education and income

gaps for equally talented individuals. In a laboratory experiment studying agents' search behav-

ior, Falk et al. (2006) �nd that individuals are generally uncertain about their abilities, that

they update their beliefs in response to the outcomes of their search and that subjects search too

much or too little as compared to a situation in which they are perfectly aware about their types.

Finally, self-con�dence can indeed be impacted by the outcomes on the microcredit market. As

underlined by Copisarow (2000), microcredits psychologically boost borrowers' self-con�dence

and self-esteem by giving them greater control over their lives and expanding their options.

Our chapter also contributes to the literature on the e�ect of training programs. Existing lit-

erature provides some evidence on the e�ciency of training programs on the job market. For

instance, Card et al. (2010) provide a meta-analysis of 97 studies focusing on the e�ectiveness of

60

2.1. Introduction

labor market programs, mainly in developed countries. Their analysis suggests that public sec-

tor employment programs are relatively ine�ective. In contrast, job search assistance has some

positive impacts, especially in the short run. Greenberg et al. (2003) synthesize �ndings of 31

evaluations on US government-funded training programs for the disadvantaged. They conclude

that women bene�t from large earning e�ects, whereas men and young individuals bene�t from

very small or insigni�cant e�ects. However, job market and micro�nance training seem to be

hardly comparable. Indeed, according to Drury et al. (1994), "training for entrepreneurship is

fundamentally di�erent from re-employment training. Its goal is not merely to provide business

skills, but to help develop a new and viable organization."

Concerning micro�nance trainings in developing economies, Karlan and Valdivia (2011) study

the e�ect of business training in FINCA-Peru using a randomized control trial. They �nd a

signi�cant impact of training on client retention in the MFI, business knowledge improvement

but little evidence on the pro�t or revenues increase. Lensink et al. (2011b) use data for MFIs in

61 emerging and developing countries and show that MFIs providing both �nance and business

development services have similar performance as MFIs providing no "plus" services. On the

other hand, McKernan (2002) �nds that noncredit aspects of micro�nance have positive e�ects

on borrowers' pro�ts. However, these noncredit aspects are composed of group cohesion, joint

liability and social development programs which are absent in developed countries. In devel-

oped countries, formal impact evaluation of BDS is scarce. For instance in France, business

training is recognized as a salient component of French micro�nance (Camdessus 2010). Never-

theless, extensive research on its impact is absent and hardly dissociable from the microcredit

itself (Balkenhol et al. 2013). Evans (2011) underlines some positive outcomes for BDS in the

framework of Women's Initiative for Self Employment in the United States (US). Nevertheless,

the author stresses that few programs are able to meet the demand as they face low budgets,

lack sta� and adequate products or services. Edgcomb (2002) provides �ve case studies for

institutions involved in business support services in the US and reports mitigated correlations

between training completion and successful entrepreneurship outcomes. Moreover, in the Euro-

61


pean Union (EU) most micro-entrepreneurs do not consider that they do need BDS and prefer

using informal sources of training o�ered by family, friends or media (Lammermann et al. 2007).

Lammermann et al. (2007) additionally highlight that in Europe, one of the main challenges

remains the increase of business training e�ciency. We contribute to this literature by providing

formal evidence on BDS e�ciency using data from a French MFI. Whereas, training appears to

have mitigated impact on the probability of default of the borrowers, we identify a signi�cant

positive impact on the survival time of the loans.

Our empirical strategy is based on a bivariate probit model where we jointly model two probit

equations: the �rst one modeling training assignment process and the second one modeling the

probability of default of the borrower. A comparable bivariate probit model has been developed

by Boyes et al. (1989) where the two probit equations concern the loan granting process and the

default of borrowers. To address selection bias issue in our setting (Heckman 1979), our chapter

pioneers the development of a trivariate probit model to test for the robustness of our baseline

bivariate probit model. Empirical literature, moreover, argues that despite default some loans

may still be pro�table to the bank which is concerned about the moment when a default occurs

rather whether a default occurs. Roszbach (2004) addresses this issue by providing a survival

time model. In line with this study, we use an alternative measure of risk in a bivariate mixed

model to check for the robustness of our �ndings. The originality of our chapter consists in the

development of formal empirical models, taking into account endogeneity issues, selection bias

or the survival time of the loans, to study how an MFI assigns di�erent borrowers to training

programs.

The remainder of the chapter is structured as follows. In section 2 we present the theoretical

model. We �rst present the discrete type model, followed by continuous type model. Data used

to test theoretical results is presented in section 3. In section 4 we present the econometric model

which we used to estimate the empirical results presented in section 5. We check for robustness

of our results in section 6. Section 7 concludes.

62

2.2. Theoretical model

2.2 Theoretical model

2.2.1 General framework

The agent, a borrower, has a project for which he needs �nancing. He has no collateral and

no personal investment. He needs to borrow from the bank the total amount of the project,

which we normalize to 1. The project will generate a return, ρ, in case of success and 0 in

case of failure. The principal, an MFI, demands a return of R = 1 + r in case of success with

R < ρ, where r is the �xed interest rate.8 It receives 0 in case of failure. The probability of

success (denoted p(θ, e, h)) depends on the type of the borrower θ, his e�ort e and the level of

business training from the MFI h.9 We assume the probability of success to be increasing in

these three terms. θ can also be interpreted as the intrinsic probability of success, depending on

borrower's and project's characteristics and excluding the e�ects of business training and e�ort

(i.e. p(θ, 0, 0) = θ). E�ort is costly for the borrower and business training is costly for the

MFI. The respective costs are denoted by ϕ(h) and ψθ(e) (i.e. we allow the cost of e�ort to be

type-dependent). The utility of the MFI is therefore given by

UP = p(θ, e, h)R− ϕ(h)

and the utility of the borrower is given by:

UA = p(θ, e, h)(ρ−R)− ψθ(e)

8A �xed interest rate is consistent with data where the MFIs �x the same interest rate for all theborrowers.

9In the context of micro�nance, business training may take di�erent forms. Generally, microborrow-ers follow various trainings in accounting or business management which are organized by the MFI orby its partners. From the approach of the literature on double-sided moral hazard (Casamatta 2003;De Bettignies and Brander 2007), business training may be interpreted as the e�ort provided be theMFI.

63


We follow the standard approach in banking modeling by assuming that the MFI is risk neutral.

To simplify the model we additionally assume that the borrower is risk neutral. However, releas-

ing this second assumption and modeling a concave utility function U(·) for the agent would not

change our result as long as U(0) = 0. We are going to analyze two information structures. In

the �rst one, the information is perfect and symmetric: both the borrower and the MFI observe

borrower's type θ. The MFI chooses h and the borrower chooses e simultaneously.

Figure 2.1: Timing of contracting under symmetric information

Our aim is to analyze if reversed asymmetric information can lead the MFI not to provide busi-

ness training to the lowest type, i.e. the riskiest, borrowers. We will focus on cases where,

under perfect information, the MFI provides a level of business training decreasing with type.

In that sense, absent the looking-glass self e�ect highlighted earlier, business training would be

considered as a bad news for borrowers (as it would re�ect a low probability of success).

In the second con�guration, we will assume reversed asymmetric information, that is a situation

where the borrower does not know his type, while the MFI does. In this case, the level of busi-

ness training chosen by the MFI (h) also conveys information about borrower's type, that might

in�uence borrower's behavior. In other terms, observing h, the borrower makes a belief about

his type that leads him to some level of e�ort.

We will show that, in contrast with the symmetric information case, a non-monotonic (concave)

relationship between business training and the type of the borrower, can emerge in some Perfect

Bayesian Equilibria (PBE). The purpose of the empirical section will then be to highlight this

peculiar feature of business training.

64


Figure 2.2: Timing of contracting under asymmetric information

2.2.2 A discrete model

Let us �rst give the basic intuitions of our model through a simple discrete model (e ∈ {0, 1},

h ∈ {0, 1}) with three types of borrowers: θ ∈ {θL, θM , θH}. Let ∀θ ψθ(0) = 0, ψ

θ(1) = ψ,

ϕ(0) = 0 and ϕ(1) = ϕ. Borrowers' types are de�ned by the return on e�ort and business

training. We make the following assumptions in the discrete model with symmetric information:

A1: The return of e�ort is increasing with type (that is p(θ, 1, h)− p(θ, 0, h) is increasing with

θ, ∀h).

A2: The return on business training is decreasing with type (that is p(θ, e, 1) − p(θ, e, 0) is

decreasing with θ, ∀e).

A3: The lowest-type borrowers do not have the incentive to provide e�ort (whatever the level

of business training):

p(θL, 1, h)− p(θL, 0, h) <ψ

ρ−R< p(θM , 1, h)− p(θM , 0, h) ∀h

A4: The MFI is not interested in training the highest-type borrowers:

p(θM , e, 1)− p(θM , e, 0) >ϕ

R> p(θH , e, 1)− p(θH , e, 0) ∀e

The third and forth assumptions are made to render our problem more pertinent.

65


Remark 2.2.1 This setting leads to a situation where, under symmetric information, the MFI

provides business training to the two lowest types: θL and θM and does not provide business

training to the highest type: θH . Borrowers of types θM and θH provide e�ort but the low type

θL does not.

. Under reversed asymmetric information, the borrower is not aware about his type. Only the

MFI observes the borrower's type. The action played by the MFI (assignment to a business

training or not) can therefore convey information to the borrower. The borrower forms beliefs

about his type after having observed the level of business training chosen by the MFI.

Our aim is to show that there exists a Perfect Bayesian Equilibrium in which the assignment to

a business training is a non-monotonic function of type, that is in which the MFI only trains

borrowers of type θM . In this case, a borrower observing that he is not trained, infers he is either

of low or high type. Let us note α the probability of the borrower to be θH when he receives

no business training from the MFI ((1 − α) is then probability of being θL). In other words, α

represents, in such an equilibrium, the belief of the borrower that he is of high type when he

observes that the MFI chooses not to train him. Moreover, in the considered equilibrium, when

the borrower observes that the MFI decided to train him he knows with certainty that he is of

type θM . This leads to the following proposition:

Proposition 2.2.1 Under reversed asymmetric information, there exists a PBE where the MFI

only helps θM -type borrowers and all borrowers exert e�ort, if and only if:

(1− α)(p(θL, 1, 0)− p(θL, 0, 0)) + α(p(θH , 1, 0)− p(θH , 0, 0)) ≥ ψ

ρ−R(2.1)

and

p(θL, 0, 1)− p(θL, 1, 0) <ϕ

R(2.2)

. Condition (2.1) insures that a borrower observing he is not helped optimally exerts e�ort,

whereas condition (2.2) states that the MFI prefers the situation where low-type borrowers exert

66


e�ort without being helped to the one of the equilibrium with symmetric information (where the

low-type borrower does not exert e�ort but receives help).

The intuition behind this result is rather simple. In the highlighted PBE, the MFI uses its

superior information to induce the low-type borrowers to exert e�ort. It does so by pooling them

with the highest type borrowers, those for whom e�ort is the most pro�table. This is done at

the expense of not providing them business training which is worth it under condition (2.2).

2.2.3 The continuous model

Let us now turn to a more complete model with a continuum of types, e�ort and business training:

θ ∈ [0, 1], e ∈ [0, 1] and h ∈ [0, 1]. In this case, we assume � for the sake of simplicity � that the

impact of business training or e�ort on the probability of success is decreasing with the type of

the borrower:

p′′1j < 0 for j = 2, 3

In the explicit example that we consider in this section, we more precisely assume

p (θ, e, h) = θ + (1− θ)1

2(e+ h) (2.3)

which is in line with our de�nition of θ as the intrinsic probability of success and with the litera-

ture on venture capital (Casamatta 2003), in which e�ort and business training are perfect sub-

stitutes considering their impact on the probability of success. Still, consistently with the discrete

model above, optimal e�ort will be increasing in type, due to a "discouragement" e�ect: a bor-

rower who "realizes" or is led to believe he has a small intrinsic probability of success, will be dis-

couraged to exert e�ort. This e�ect will materialize in our model through a type-speci�c cost of ef-

fort. More precisely, we need ψθ(e) to be such thatE (θ, h) = arg maxe

[p (θ, e, h) (ρ−R)− ψ (θ, e)]

is increasing in θ. One very simple way to model this property � that keeps the model tractable

� is to assume a linear e�ort with respect to the type of the borrower E (θ, h) = γθ. Therefore,

67


e�ort and borrower's type act as complements, as a higher type would induce higher e�ort. This

complementarity is in line with economic literature where the principal has a vested interest to

boost agent's self-esteem in order to increase his motivation (Benabou and Tirole 2002; Ishida

2006). Taking the probability of success as de�ned in (2.3) this can be obtained, for instance, by

assuming ψθ(e) = ρ−R4γ

1−θθ e2.

Regarding the cost of business training incurred by the MFI, a quadratic form ϕ(h) = ch2

2 (where

c is a positive constant) will be enough to insure that optimal level of business training is de-

creasing with type under symmetric information.

Indeed, under symmetric information, the program of the MFI is then given by:10

h∗ (θ) = arg max

h∈[0,1]

(θ + 1

2 (1− θ) (e+ h))R− 1

2ch2

s.t. e = γθ

We have the following remark:

Remark 2.2.2 Under symmetric information, the optimal level of business training provided by

the MFI is decreasing with the borrower's type:

h∗ (θ) =R

2c(1− θ)

. In other words, the MFI provides business training to those who need it the most. The optimal

level of business training is a decreasing a�ne function of the type.

Let us now turn to the case of reversed asymmetric information. Once again, our aim is to

show that (in a situation where the level of business training would be decreasing in type under

symmetric information), superior information can lead the MFI to train less the lowest-type bor-

rowers. In the case of a continuum of types, this would correspond to a non-monotonic (concave)

10Note that we do not model the approval process. Therefore, we do not make any assumption onMFI's position toward pro�ts. Consequently, our model is applicable both to non-pro�t NGOs and tofor-pro�t commercial institutions.

68


relationship between business training and type, i.e. an "exotic" pooling Perfect Bayesian Equi-

librium in which several non adjacent values of θ are associated with the same level of business

training. Let us show that such an equilibrium is possible.

Consider that business training is a concave non monotonic function of type h∗(θ) such that,

a level of business training h is associated with two possible types θ(h) and θ(h) (except at its

maximum). A borrower, observing a given level of business training infers some information

about his type. The inferred type of the borrower writes:

tθ(h) =θ(h)fθ (θ(h)) + θ(h)fθ

(θ(h)

)fθ (θ(h)) + fθ

(θ(h)

)where fθ (x) represents the belief of a borrower type θ on the distribution of types. This will

consist in a Perfect Bayesian Equilibrium if the optimal business training strategy when the

borrower's inferred type is tθ(·) is precisely h∗(θ), that is if h∗ is the solution of:

h∗ (θ) = arg max

h∈[0,1]

(θ + 1

2 (1− θ) (e+ h))R− 1

2ch2

s.t. e = γtθ(h)

The borrower updates his beliefs after having observed the level of business training chosen by

the MFI. This occurs due to the looking-glass self e�ect. The MFI is aware about borrower's

updating process. This mechanism is re�ected in the constraint on borrower's e�ort in the MFI's

maximization program. Here, borrower's e�ort depends on the updated type rather than on the

true type.

Proposition 2.2.2 There exists a PBE in which business training is a non-monotonic concave

function of type. It notably involves that borrowers' beliefs are decreasing in types: observing a

high level of business training is bad news for high-type borrowers but a good news for low-type

borrowers.

69


. To prove the existence of such an equilibrium, we analyze a particular concave business training

function, symmetric with respect to 12 : h∗ = σθ(1 − θ). In this case, the MFI o�ers the same

level of business training h to borrowers of type:

θ(h) =1

2−√

1

4− h

σand θ(h) =

1

2+

√1

4− h

σ

and (type-dependent) distributions of beliefs of the form:

fθ(x) = 1− σ

γ(θ − 1)

(x− 1

2

)

will lead to a Perfect Bayesian Equilibrium. Indeed, we then have the inferred type of the

borrower:

tθ(h) =1− 2 (θ − 1)

(14σγ −

hγ

)2

such that the optimal level of business training which is solution of

maxh∈[0,1]

(θ + 1

2 (1− θ) (e+ h))R− 1

2ch2

s.t. e = γtθ(h)

writes h(θ) = R2cθ (1− θ) which corresponds to the business training function h∗(θ) with σ = R

2c .

In the highlighted equilibrium, the beliefs of borrowers therefore correspond to distorted uniform

distributions: distorted upward for types θ < 1/2 and distorted downward for types θ > 1/2. In

this setting, for high-type borrowers business training is bad news, whereas it is good news for

low-type borrowers: for a given level of business training, low-type (resp. high-type) borrowers

put more weight on the higher (resp. lower) corresponding type. This allows the MFI to increase

the level of e�ort provided by lowest-type borrowers, by pooling them with a highest-type bor-

rowers.

70


The aim of the above argument was to show that there exist Perfect Bayesian Equilibria in which

business training is a non-monotonic function of type of the borrower. There may exist a lot of

other PBE in the considered setting and a lot of other settings in which such PBE exist.

Still, we have shown in this theoretical section that, because of superior information, a micro�-

nance institution might not want to train (or might want to train less) the lowest-type borrowers.

This comes from the fact that its business training decision conveys information about the ability

of the borrower. By pooling the lowest-type borrowers with highest-type borrowers, the MFI

uses the "looking-glass self" e�ect to induce them to exert more e�ort.

One might compare the PBE we highlight in this section to alternative strategies where the MFI

trains all the borrowers or randomly chooses the borrowers who will receive training. In these

limit cases the looking-glass self e�ect disappears. Accordingly, if we assume a simple case where

θ is uniformly distributed between 0 and 1 and all the borrowers have the same inference about

their type, the inferred type will be equal to 1/2. Consequently, the average e�ort without the

looking glass self e�ect becomes:

e =γ

2

This level of e�ort is strictly smaller than the one provided in presence of the looking-glass self

e�ect:

e = γtθ(h) =γ

2+ (1− θ)R

2c

[1

4− θ(1− θ)

]Notably, the di�erence in e�orts due to the looking-glass self is decreasing with borrowers' type.

In other words, low-type borrowers exhibit the highest increase in e�ort due to the looking-glass

self e�ect. The comparison of our PBE with the limit cases which exclude any type signaling

consists in comparing the tradeo� between e�ort and help. To outperform the highlighted PBE,

in the limit cases described above, higher level of business training should compensate for lower

level of e�ort and the cost of these strategies should not be too high for the MFI. Indeed, in

situations where the cost of business training provision is very high (which is likely to be the

case especially in the developed countries), the MFI will prefer to give incentives to borrowers

71


to exert higher e�ort rather than training them.

In the following section we provide the institutional context of the MFI providing data for our

study, which we present in section 4. In section 5 we test whether the equilibrium highlighted

by the theoretical model is indeed observed in reality. The two main hypotheses to be tested are

the following:

H1: The likelihood to receive help is increasing with type for low-type borrowers.

H2: The likelihood to receive help is decreasing with type for high-type borrowers.

In other words, we test whether business training is a non-monotonic concave function of bor-

rowers' type (measured by their risk or ex-ante intrinsic probability of default). The econometric

exercise will allow us to analyze whether the MFI internalizes the fact that its business training

decision impacts borrowers' behavior through the "looking-glass self" e�ect.

2.3 Institutional context of the MFI

CREASOL, the MFI providing data for our study, has been created in 2006, in the South of

France, under the status of a non-pro�t NGO at the initiative of a commercial bank in its

corporate social responsibility framework. This MFI generally targets individuals who have

di�culties in accessing �nancial services from mainstream banks. It targets mainly individuals

residing in the Provence-Alpes-Côte-d'Azur region. In line with its social mission statement,

most of MFI's clients are (long term) unemployed, have low education and income levels and

start a business for the �rst time in their lives. Most of them become self-employed in order to

exit unemployment status and/or to escape poverty. The MFI does not require any collateral or

guarantee from its clients. Therefore, the total pool of the applicants of this MFI is considered

"too risky" by most of the commercial banks.

In addition to microcredit services, this MFI is highly active in business training provision, in

cooperation with its partners. We have information about all the applicants who were accepted

72

2.3. Institutional context of the MFI

to receive a microcredit between May 2008 and May 2011.11 Conditional on acceptance, we

have information on business training provision. To our knowledge MFI's borrowers did not

bene�t from other trainings except those mentioned in the MFI's data set. The pool of MFI's

clients is almost evenly distributed between individuals with and without training (55% and

45% respectively). Hence, we have no evidence that the MFI targets primarily individuals who

may bene�t from the training. Unfortunately, we do not have data on business evolution (ex.

pro�ts, sales, etc.). This information is only available as a forecast during the application stage

in a business plan presented by the applicant. Hence, we don't study the link between business

training provision and business evolution. Nevertheless, our data set contains information on

borrowers' ex-post repayment behavior to MFI. We observe the number of unpaid installments

and the moment when they occurred. Consequently, our data set allows us to study the link

between business training provision and the likelihood of a business to succeed.

The timing of the relationship between the MFI and the borrower is the following. First the

borrower applies for a credit. Each individual can apply only once for a microcredit. This MFI

aims at �nancial inclusion of all the borrowers after the reception of the �rst microcredit.12 The

MFI decides whether to accept or reject borrower's application. The decision takes place on

several levels. First, the loan o�cer present the project during a credit committee. Then, the

credit committee takes a decision concerning loan granting. Second the MFI decides to provide

or not training to the selected applicants. Training is mandatory for the borrowers, they can

not refuse participation. Concerning the link between business training and the microcredit, we

do not have any evidence on clients' withdrawal from the MFI due to assignment to training.

Third, we observe for each client his/her microcredit reimbursement behavior. In other words,

we have information about the unpaid installments and the moment when they occurred.

11Our sample consists in the universe of the applicants.12As a consequence the MFI does not practice dynamic incentives through progressive lending.

73


2.4 Data

Our data from CREASOL allows us to model three di�erent (consecutive) processes:

1. credit-granting

2. assignment to business training

3. having three or more unpaid installments.

In Table 2.1 we present the descriptive statistics of our data along with the t-tests to compare

means between di�erent groups. We have individual information on 782 applicants received

between May 2008 and May 2011 for a business loan.13 These applicants represent the total

universe of information available to us. A large majority of these loans was dedicated to business

start-up, in contrast to business development. The average amount of the approved loans was

8, 900e, the average interest rate was 4.2%14 and the mean maturity was 52 months.

We will use this data in order to study how MFI's decision to train borrowers varies with their

intrinsic risk which is used to proxy the type of the borrower in the theoretical model. The risk

of the borrowers is estimated using individual, household and business characteristics presented

in Table 2.1.

13We do not study consumer loans, in contrast with Roszbach (2004).14The interest rate was �xed at 4% per year at the beginning of the period and reached 4.5% at

the end of the period of analysis. The interest rate is �xed and hence does not depend on borrower'scharacteristics.

74

2.4. Data

Table2.1:

Descriptive

Statistics

Applicants

Borrowers

Variables

TotalGranted

Rejected

t-test

BDS

NoBDS

t-test

Defaulting

Performing

t-test

Granted(%

)0.47

BusinessTraining(%

)0.26

0.55

0.49

0.57

-0.08

HighRisk(%

)0.10

0.22

0.19

0.25

-0.05

IndividualCharacteristics

Male(%

)0.62

0.61

0.63

-0.02

0.62

0.60

0.02

0.75

0.58

0.17***

Education(#

ofdiplomas)

1.84

1.89

1.80

0.09

1.89

1.89

0.00

1.46

2.01

-0.55***

Single(%

)0.55

0.53

0.57

-0.04

0.50

0.57

-0.07

0.63

0.51

0.13**

Unem

ployed

more

than12months(%

)0.39

0.33

0.44

-0.11***

0.37

0.28

0.08*

0.42

0.31

0.11*

Household

Characteristics

Household

income(kEUR)

1.33

1.49

1.19

0.30***

1.61

1.33

0.29**

1.11

1.59

-0.49***

Household

expenses(kEUR)

0.45

0.45

0.45

-0.01

0.47

0.42

0.06

0.47

0.44

0.03

BusinessCharacteristics

Low

personalinvestm

ent(%

)0.30

0.26

0.34

-0.08**

0.25

0.27

-0.02

0.38

0.23

0.15***

Assets(kEUR)

18.20

18.86

17.57

1.28

21.34

15.73

5.62**

12.19

20.74

-8.55***

Foodandaccommodationsector(%

)0.13

0.10

0.16

-0.06**

0.08

0.13

-0.04

0.09

0.11

-0.02

Grossmargin/Sales

0.75

0.74

0.77

-0.03*

0.74

0.74

0.00

0.71

0.75

-0.03

OtherCharacteristics

Other

dem

ands(%

)0.58

0.62

0.54

0.08**

0.82

0.38

0.44***

0.51

0.65

-0.15**

Loanofhonor(%

)0.48

0.47

0.48

-0.01

0.63

0.28

0.36***

0.43

0.48

-0.05

Sentbyamainstream

bank(%

)0.17

0.18

0.15

0.03

0.12

0.26

-0.14***

0.14

0.20

-0.06

NbofObservations

782

365

417

202

163

79

286

75


Descriptive statistics for the intrinsic risk of the borrowers modeled by a simple probit equation

are given in Table 2.2. We note that the average predicted intrinsic risk for the entire pool of

applicants is 0.24. Naturally, it decreases for the accepted applicants to 0.22. In Table 2.1 we

present the individual, household, business and other characteristics for applicants and e�ective

borrowers according to their status: accepted or rejected, with or without business training,

defaulting or performing.

Table 2.2: Descriptive statistics on the predicted intrinsic risk of the applicants

Mean Min Max SD

Accepted applicants 0.22 0 0.69 0.17

Rejected applicants 0.26 0 0.82 0.19

Total pool of applicants 0.24 0 0.82 0.18

Loan approval represents the �rst decision made by the MFI. Data on rejected applicants will

be used to correct for selection bias in a robustness check exercise. The availability of the in-

formation on rejected applicants presents an original characteristic of our data set. 47% of the

applicants have been granted a microcredit between May 2008 and May 2011.

In Table 2.1 we note that the proportion of long term unemployed applicants (more than 12

months) is signi�cantly greater among rejected projects. So is the case for applicants with low

personal investment15 (lower than 5%), projects in the food and accommodation sector, projects

having a larger ratio of gross margin to sales. This last �gure may appear surprising at �rst sight,

yet it suggests that the MFI privileges the amount of sales in the approval process, rather than

this ratio. Finally, accepted applicants come from households with higher incomes on average.

15Low personal investment is a dummy taking value 1 if the applicant's personal �nancial contributionto the project is lower than 5% of the project size. We take this cut-o� because this is the lowest cut-o�available in our data, after "No personal investment" and very few individuals provided no personalinvestment.

76

2.4. Data

A second decision concerns assignment of accepted applicants to a training program.16 55% of

the accepted borrowers were assigned to a training program. Trained and not trained borrowers

seem to hardly di�er with respect to their individual characteristics, according to Table 2.1.

We note that the proportion of long term unemployed individuals is larger for borrowers with

business training. Moreover, they have larger household incomes and their businesses exhibit

larger asset levels. Signi�cant di�erences arise for variables concerning other characteristics. In-

dividuals with business training are more likely to have other demands and bene�t from loans

of honor. This di�erentiation is in line with the context of microcredit where NGOs providing

training programs also provide loans of honor.17 The variable Other demands often include on-

going demands for a loan of honor. Hence, the link between the two variables and the likelihood

to be assigned to business training is direct. Interestingly, applicants sent by a mainstream bank

are less likely to be assigned to a training program. This is in line with our intuition. Borrowers

sent by a mainstream bank have either a co-�nancing credit from the bank (these are potentially

"high-type" clients) or have been rejected (these are potentially "low-type" clients). Our model

predicts that both high-type and low-type clients are the least likely to be trained.

In this chapter we are particularly interested in this second stage decision consisting in borrow-

ers' assignment to training programs. In our baseline model, we analyze how the MFI decides

to train a borrower conditional on loan granting. To do so, we will design a scoring model.

To build a scoring model, we de�ne as "defaulting" borrowers with 3 or more unpaid installment

in their credit history within the MFI. In other word this is data on ex-post defaults of MFI's

clients. 22% of all the accepted borrowers had 3 or more delayed payments in their credit history.

16Assignment to a training program can be interpreted as treatment and borrowers can be divided ina treated and control group respectively. From this perspective, our chapter can be assimilated to theliterature studying treatment e�ects. Nevertheless, treatment is obviously not assigned randomly in ourcase.

17A loan of honor is an interest free loan subsidized by the French government for individuals willingto start a business in order to get self-employed. The government delegates the disbursement of theseloans to NGOs which may also provide training programs.

77


We will denote these loans as "defaulting" in the following of the chapter.18 This de�nition is

close to the MFI's policy. In general, the MFI writes-o� all the loans having three or more consec-

utive delayed payments. Among the clients having received business training, 19% are defaulting

clients. This proportion amounts to 25% for clients without business training. However, this

di�erence is not signi�cant according to the t-test. Almost half of the defaulting loans have been

assigned to a training program, whereas among performing loans 57% was assigned to a training

program. Again the di�erence between the two groups is not signi�cant. To build our measure

of risk we will use individual, household and business characteristics. As Table 2.1 illustrates,

there are signi�cant di�erences between defaulting and performing clients. These di�erences are

usually (with some exceptions) of an opposite sign as compared to those for the approval deci-

sion. This presents an indicator of MFI's credit granting process quality. Defaulting clients are

more likely to be male, single, long-term unemployed, having lower education and income levels.

Concerning business characteristics, defaulting clients are signi�cantly more likely to have low

personal investment and lower level of assets. All these variables will be accounted for in the

design of the risk measure presented in the next section.

One potential problem of our data is that borrowers receive microcredits at di�erent moments.

Obviously, older clients are more likely to be defaulting compared to newly granted credits. To

deal with censored data issue, we will use, as a robustness check, an alternative measure of risk

consisting in the inverse of the expected survival time in a robustness check. Therefore, we

present in Table 2.3 descriptive statistics on the survival time of each microcredit.

We additionally have information on the amount of the loan of honor, amount of personal invest-

ment, the city or town of the applicants, juridical status of the business and forecasts for various

business variables (working capital requirements, net income, charges and revenues, etc.) from

the business plan. However, we do not use these variables either due to their low variability, high

correlations with variables included in the model or their low reliability.

18Note that the delayed payments need not to be consecutive or unpaid. However, most of delayedpayments in the database where consecutive.

78

2.5. Econometric model

Table 2.3: Descriptive statistics for survival time

Percentiles

Subsample Mean SD Min 5 10 25 50 75 90 95 Max

Ti, bad loans 340.1 237 0 61 92 184 274 457 668 822 1156

Ti, good loans 469.5 327.8 31 92 123 214 365 638 1003 1095 1279

In the next section we use this data to test the hypothesis presented in the theoretical model.

2.5 Econometric model

The purpose of this section is to study the relationship between the "type" of the borrowers and

their likelihood to be assigned to a training program by the MFI. As shown in the theoretical

model assignment to business training is not necessarily a monotonic function with respect to

borrower's type. The "type" of the borrower in the theoretical model, or in other words the

intrinsic probability of a business to succeed, corresponds in practice to the score given by the

MFI to each applicant during the approval process. Unfortunately, we do not have information

about this score. To bypass this limit we will use the information on the ex-post defaults of

the borrowers which is available from the MFI. The aim is to use ex-post information on credit

history to estimate the ex-ante score given by the MFI. To do so, we assume that the MFI has

a scoring strategy based on its previous experience. We attempt to infer MFI's scoring strategy

using the available information on ex-post defaults. To estimate the type of the borrower we will

use a probit equation to model the likelihood of a borrower to default.19 Among the explanatory

variables we will include individual, household, business characteristics. In addition, we will

control for the business cycles20 which obviously impact the likelihood to default. To do so, we

19DeYoung et al. (2008) show that credit scorings mitigate the information asymmetries associatedwith geographically distant small business borrowers.

20Source: Fiben, Banque de France.

79


will include quarterly growth rates of defaults (presenting a measure for economic health) and

start-ups (presenting a measure for competition) at the moment of business creation for each

micro-enterprise in our sample according to its sector of activity. Indeed, unfavorable economic

environment during the start-up phase can jeopardize the likelihood to survive of a business.

Data for business cycles exclusively cover French PACA Region, which corresponds to the region

where the MFI of our interest operates.

Akin to the theoretical model, in addition to individual, household, and business characteristics

which are the components of θ, ex-post default depends on business training provision (h) and

on borrower's e�ort (e). We will attempt to isolate these three e�ects (θ, h, e). To identify the

e�ect of business training we introduce in the default equation a dummy taking value one if a

borrower receives business training and zero otherwise. To isolate the e�ect of e�ort on the like-

lihood to default we will introduce a form of heteroscedasticity depending on business training

and on borrower's education level in the default equation using a bivariate probit model.

This approach allows us to estimate a variable Risk depending solely on individual, household,

and business characteristics to proxy the type of the borrower in the theoretical model. The

variable Risk, therefore, corresponds to 1− θ in our theoretical model. Nevertheless, the inter-

pretation of the PBE equilibrium outlined in the theoretical model is not altered by this inver-

sion: the non-monotonic concave relationship between business training and type θ corresponds

by symmetry to a non-monotonic concave relationship between Risk and business training, with

probability to receive business training increasing in risk for low-risk borrowers and decreasing in

risk for high-risk borrowers. Hence the hypotheses to test remain the same if we replace "type"

by "risk":

H1: The likelihood to receive help is increasing with risk for low-risk borrowers.

H2: The likelihood to receive help is decreasing with risk for high-risk borrowers.

To test this non-linear relationship between the likelihood to receive business training and bor-

rower's type, as suggested by the theoretical framework, we include the predicted Risk and

80


Risk2 variables in the Business training equation, which is also modeled using a probit equation.

Including Risk and Risk2 will allow us to test for the simplest form of non-monotonicity, that

is a quadratic relationship between the likelihood to receive business training and the predicted

risk of a client.

For the sake of comparison, we present a simple univariate model where we independently es-

timate two probit equations for business training and default, without heteroscedasticity. The

univariate model writes:

y∗1i = x′iβ1 + λ1Risk + λ2Risk

2 + ε01i y1i =

1 if y∗1i > 0 (Business training)

0 if y∗1i ≤ 0 (Otherwise)

(2.4)

y∗2i = w′iβ2+

2∑q=0

ωqdefaultqi+2∑q=0

ηqstartupqi+α1y1i+ε02i y2i =

1 if y∗2i > 0 (Default)


(2.5)

where Risk = Φ(w′iβ2) is the estimated intrinsic probability to default21 of the borrower and

Φ(·) is the normal cumulative distribution function and β2 is estimated using equation (2.5). λ1

and λ2 are the covariates of our main interest. If H1 is veri�ed, we expect λ1 to be signi�cantly

positive. If H2 is veri�ed, we expect λ2 to be signi�cantly negative.

x′i is a vector of variables Other demands, Loan of honor and Sent by a mainstream bank, w

′i is

a vector of various controls composed of individual, household, and business characteristics. q

measures the number of quarters after the credit granting occurred. For example, default0i is

the quarterly growth rate of business defaults in the sector of the enterprise i which occurred at

the moment the credit was granted to the enterprise i. Equivalently, startup2i is the quarterly

growth rate of new businesses created in the sector of the enterprise i which occurred two quar-

ters after the credit has been granted to the enterprise i. This measure of business cycles allows

21By intrinsic probability to default we mean the probability to default net of the e�ect of businesstraining and borrower's e�ort. However, in the univariate model, without heteroscedasticity we are onlyisolating the e�ect of business training in the default equation.

81


us to capture the economic environment at the moment the business has been started,22 one

quarter after the business has been started and two quarters after the business has been started.

y1i is a business training dummy, and ε01i and ε02i are independent error terms following a normal

distribution N (0, 1).

The results of the univariate model are presented in Table 2.4, columns (1) and (2) and com-

mented in the next section. We use this estimations from the univariate model as starting values

for the coe�cients of the control variables in the bivariate model.

We present the bivariate probit model (Boyes et al. 1989) which presents several advantages.

First, we jointly model two processes, business training decision and the probability to default

in presence of unobserved individual heterogeneity. Controlling for unobserved individual het-

erogeneity also allows us to take into account the fact that the MFI has soft information about

borrower's type (motivation, skills, personality, etc.), collected during the face to face meetings.

The MFI might use this information during business training provision process. Obviously, we

do not have access to this information. Hence, the bivariate model will allow us, at least partly,

to control for this informational asymmetry. By estimating simultaneously the two processes,

we can avoid to bias the estimator of the variance of the parameters estimates related to the

estimated expectation of Risk in the business training equation.

Second, it allows us to control for endogeneity between business training and default, as busi-

ness training can impact the default of the borrower. Isolating the e�ect of business training in

the default equation will allow us to better estimate the intrinsic probability to default of the

borrower, i.e. to better proxy borrower's type in the theoretical model.

Third, in order to capture the idea that two di�erent borrowers observing the same level of busi-

ness training have di�erent inference processes about their type assumed in the theoretical model,

we add heteroscedasticity in our model. The likelihood to default, estimated using realized data,

depends on borrowers' behavior, which in turn depends on the inferred type. The inferred type

22Generally, microcredit is used for business start-up, rather than for business development. Hence,the date of credit granting is used as a proxy for the date of business start-up.

82


depends on the information received from the MFI (business training or not) and the inference

process that can vary across individuals. In other words, the inference process introduces some

noise in borrower's behavior (or e�ort) and consequently some noise on his likelihood to default

which implies an increasing and non constant variance and can be naturally represented by a

scedastic function attached to the unobservable variable, vi. Hence, introducing heteroscedastic-

ity in the default equation represents a way to isolate the impact of e�ort on the probability to

default.

Controlling for the endogeneity between business training and default and introducing het-

eroscedasticity in the default process allow us to isolate in the default equation three di�erent

e�ects discussed in the theoretical model, i.e. business training, borrower's e�ort and borrower's

type e�ects.

The presence of the correlation between business training decision and risk is allowed by imposing

the following structure on the error terms:

ε1i = ρ1vi + ε01i

ε2i = ρ2ivi + ε02i

where the components ε01i, ε02i are independent idiosyncratic parts of the error terms and each

one is supposed to follow a normal distribution N (0, 1). The common latent factor vi entering

the compound terms ε1i, ε2i could be considered as an individual unobserved heterogeneity

factor. We assume that vi ∼ N (0, 1) and that this factor is independent of the idiosyncratic

terms. ρ2i ≡ ρ2exp(α2y1i + δEducationi), meaning that business training impacts indirectly

the probability of default through α2 (inference e�ect). We moreover assume that the inference

process depends on borrower's education level (or skills), through the coe�cient δ, which also

represents the indirect e�ect of education on probability of default.

The parameters ρ1 and ρ2 are free factor loadings which should be estimated. For identi�cation

reasons, we impose the constraint ρ2 = 1. Hence, the type of the borrower is proxied by Risk =

83


Φ(w′iβ2 + vi). We maximize the log of the likelihood function which is the sum of the individual

contributions to the likelihood (see Appendix A).

Obviously and akin to the theoretical model, this model is estimated only for granted loans,

as an individual can only be assigned to a training program if he has actually been granted

a microcredit. Hence, akin to the univariate case, we do not account for the selection bias in

the bivariate probit model, selection bias being left for the sake of a robustness exercise. The

bivariate probit model writes:

y∗1i = x′iβ1 + λ1Risk + λ2Risk

2 + ε1i (2.6)

and

y∗2i = w′iβ2 +

2∑q=0

ωqdefaultqi +2∑q=0

ηqstartupqi + α1y1i + ε2i (2.7)

with

Risk = Φ(w′iβ2 + vi)

The results of the estimation of the bivariate probit model with heteroscedasticity are presented

in Table 2.4, columns (3) and (4).

For the identi�cation of the model, it is important that the variables in the xi vector are di�erent

from the variables in wi vector. The three variables in the xi vector (loan of honor, other

demands and sent by a mainstream bank) have a direct link with the business training process,

as justi�ed in the Data section above. Hence, these three variables are perfect candidates to be

used exclusively in the Business training equation. Using business cycles in the default equation

also improves the identi�cation of our model. The growth rate of defaults and new created

businesses during the start-up phase cannot be used in the training equation as they occur after

training assignment took place. Hence, business cycles are perfect candidates to be used only in

the default equation to ensure the identi�cation of our model.

84

2.6. Econometric results

2.6 Econometric results

Our results are presented in Table 2.4.23 For benchmarking purposes, in column (1) and (2) we

present the results of univariate probit equations for business training and risk. They correspond

to the equations (2.4) and (2.5) estimated separately. In columns (3) and (4) we present the

estimations for the bivariate probit model.

The non-linear relationship between the likelihood to receive business training and the type of

the borrower is shaped by the coe�cients of Risk and Risk2. The Risk loading is positive in all

the speci�cations and the loading of the quadratic term is always negative, suggesting that the

MFI is more likely not to provide business training to borrowers with very low and very high

risk, as the probability to receive business training is �rst increasing with risk and, then, beyond

a certain threshold it decreases with risk. Nevertheless, this relationship becomes signi�cant (at

1% level) only in the bivariate probit model. We can compute using the estimators in column

(3) the risk threshold for which the probability to receive business training becomes decreasing

with risk. To do so we use the derivative:

∂Pr(y1i = 1|xi, Risk, vi)∂Risk

= (λ1 + 2λ2Risk)φ(·)

where φ(·) is a normal density which is always positive. Hence the sign of the previous derivative

is given by λ1 + 2λ2Risk. It will be positive for Risk smaller than 0.35 and negative otherwise.

We have estimated the Risk = Φ(w′iβ2 + vi) for each accepted client in our data. 77% of the

clients have an estimated risk smaller than 0.35 and 23% of the sample represent an estimated

risk higher than this threshold. The estimated probability to receive business training as a

function of the estimated risk of the borrowers is presented in Figure 2.3.

23In this chapter we are interested in the signs of the loadings and we will not study the sizes of marginale�ects. Hence all the results presented are the estimated coe�cients rather than marginal e�ects.

85


Table 2.4: Determinants of Business Training and Default Processes

Model Univariate probit Bivariate probit

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

Dependent variable: Business training Default Business training Default

Explanatory variables:

Risk 1.07 (1.4) 24.71*** (9.56)

Risk2 -1.20 (2.4) -35.61*** (13.34)

Other demands 1.07*** (0.17) 3.52*** (1.3)

Loan of honor 0.51*** (0.16) 1.46*** (0.54)

Sent by a mainstream bank -0.55*** (0.19) -1.76** (0.73)

ρ1 1.53** (0.63)

Business training (direct e�ect) -0.15 (0.18) -0.33 (0.23)

Male 0.62*** (0.2) 0.84*** (0.12)

Education (direct e�ect) -0.18** (0.07) -0.11* (0.04)

Single 0.06 (0.19) 0.33*** (0.12)

Unemployed at least 12 months 0.33* (0.19) -0.06 (0.1)

Household income (kEUR) -0.29** (0.12) -0.5*** (0.11)

Household expenses (kEUR) 0.66*** (0.23) 0.81*** (0.15)

Low personal investment 0.35* (0.19) 0.2* (0.11)

Assets -0.02*** (0.01) -0.04*** (0.01)

Food and accommodation sector 0.23 (0.31) 0.96*** (0.14)

Gross margin/Sales -1.03** (0.45) -0.64** (0.26)

Growth rate of defaults beginning of credit 0.009 (0.006) 0.009 (0.006)

Growth rate of defaults beginning of credit +1 0.008* (0.004) 0.01* (0.005)

Growth rate of defaults beginning of credit +2 0.007 (0.005) 0.004 (0.007)

Growth rate of start-ups beginning of credit 0.012*** (0.004) 0.015*** (0.004)

Growth rate of start-ups beginning of credit +1 0.006 (0.004) 0.005 (0.004)

Growth rate of start-ups beginning of credit +2 -0.001 (0.004) -0.002 (0.005)

Business training (indirect e�ect) 1.00** (0.47)

Education (indirect e�ect) -0.68* (0.36)

Intercept -0.75*** (0.22) -0.03 (0.48) -3.15*** (1.18) -0.27 (0.29)

-2 Log Likelihood 371 293 653

Observations 342 340 340

Standard errors in parentheses. ***p<0.01, **p<0.05,*p<0.1

86

2.6. Econometric results

0.4

0.5

0.6

0.7

0.8

0.9

1.0P

red

icte

d p

rob

ab

ilit

y t

o r

ece

ive

bu

sin

ess

tra

inin

g

0.0

0.1

0.2

0.3

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Pre

dic

ted

pro

ba

bil

ity

to

re

ceiv

e b

usi

ne

ss t

rain

ing

Predicted risk of the borrowers

Predicted training Poly. (Predicted training)

Figure 2.3: Bivariate Probit Model Estimations

The red line corresponds to a second degree �t curve for the predicted data. For the 23% of

clients above the 0.35 threshold, the likelihood to be assigned to a training program is decreasing

with Risk as the MFI is worried about the negative impact business training might have on their

inferred type.

Concerning other controls in the business training equation we observe a highly signi�cant posi-

tive relation between business training and other demands and loan of honor, which is in line with

the intuition: individuals receiving business training are more likely to apply for other demands

as they may be encouraged to do so in their training process, and they are more likely to receive

loans of honor as they are disbursed by the training institutions. Being sent by a mainstream

bank is negatively associated with the likelihood to receive business training. Individuals sent by

a mainstream bank are either rejected by the mainstream bank (and are likely characterized by

a very high risk) or are granted a co-�nancing credit by the mainstream bank (and are charac-

terized by a very low risk). In both of these situations we expect these individuals to be the least

likely to be assigned to a training program, due to potential motivation undermining (for high

87


risk individuals) or to the presence of a good performance which does not need business training

(for low risk individuals). ρ1 is signi�cant at 5%, suggesting that endogeneity bias is indeed an

issue and the bivariate model is better suited to deal with it, in contrast to the univariate model.

Hence, the bivariate model taking into account the individual unobserved heterogeneity is pre-

ferred to univariate equations. Overall, the univariate coe�cients for business training equation

are strongly underestimated compared to their bivariate equivalents.

Concerning the default equation, business training does not signi�cantly impact the likelihood

to default in univariate speci�cations and in the bivariate model, despite its negative sign (it

is signi�cant at 16% in the bivariate probit model). Male clients are signi�cantly more likely

to default compared to female clients. Higher education (measured in the number of achieved

diplomas) signi�cantly decreases the risk of the clients. Single individuals are signi�cantly more

likely to default in the bivariate probit model. Household income and expenses are respectively

important negative and positive determinants of the likelihood to default. Borrowers with low

personal investment seem to be riskier in contrast to businesses with larger assets. Businesses

in food and accommodation sector are riskier compared to other sectors of activity. Finally, the

gross margin to sales ratio is associated with lower business risk.

Interestingly, the growth rates of defaults and start-ups are positively correlated with risk. In-

deed, increasing growth rate of start-up is related to increasing competition in the sector, that

will negatively impact the performance of MFI's clients. Note that default rates are most impor-

tant one quarter after the beginning of the credit, whereas start-up rates are signi�cantly positive

at the moment of credit granting. Overall the signs of the loadings are in line with economical

intuition.

Concerning heteroscedasticity, the indirect e�ects of business training and education are signif-

icant at 5% and 10% levels respectively, with opposite signs. Higher level of business training

increases uncertainty about the risk of default. This seems to be in line with the intuition if

business training is, indeed, mainly reserved for "intermediary-risk" individuals. Finally, higher

level of education will signi�cantly decrease the variance of the unobserved individual hetero-

88

2.7. Robustness checks

geneity term, vi. In other words, there is more certainty about the risk of default for borrowers

with higher education.

In the next section we will perform several robustness checks to validate our main results for

the looking-glass self e�ect. We perform two robustness check consisting in controlling for the

selection bias and de�ning an alternative measure of risk.

2.7 Robustness checks

2.7.1 Correcting for selection bias

In this section we add to the previous bivariate models a third equation, namely the approval

decision which allows us to correct for the selection bias. Hence, we jointly model three di�erent

processes:

y∗0i = w′iβ0+

0∑q=−2

τqdefaultqi+

0∑q=−2

υqstartupqi+ε0i y0i =

1 if y∗0i > 0 (Granted)


(2.8)

y∗1i = x′iβ1+λ1Risk+λ2Risk

2+ε1i y1i =

1 if y∗1i > 0 (Business training)


(2.9)

where Risk = Φ(w′iβ2 + ρ2vi).

y∗2i = w′iβ2+

2∑q=0

ωqdefaultqi+

2∑q=0

ηqstartupqi+α1y1i+ε2i y2i =

1 if y∗2i > 0 (Default)


(2.10)

Note that we use the same explanatory variables w′i in the approval and risk equations as sug-

gested by Roszbach (2004). We moreover introduce business cycles in the approval equation

which may impact MFI's decision on loan granting. Hence, we introduce in the approval equa-

tion the growth rate of defaults and start-ups at the moment of approval, one quarter before

89


the approval, and two quarters before the approval. The presence of business cycles before the

moment when the approval takes place will improve the identi�cation of the trivariate model

as these variables are only present in the approval equation.24 In this model we allow for the

presence of correlation among both decisions (approval and business training) and risk equation

by imposing the following structure on the error terms:



ε2i = ρ2ivi + ε02i

As stressed in the previous section, the components ε00i, ε01i, ε

02i are independent idiosyncratic parts

of the error terms and each one is supposed to follow a normal distribution N (0, 1). The common

latent factor vi ∼ N (0, 1) is independent of the idiosyncratic terms. We introduce the same form

of heteroscedasticity as in the bivariate probit model, i.e. ρ2i ≡ ρ2exp(α2y1i+δEducationi). The

parameters ρ0, ρ1 and ρ2 are free factor loadings which should be estimated. For identi�cation

reasons, we impose the constraint ρ2 = 1. We maximize the log of the likelihood function which

is presented in the Appendix B. The results are presented in Table 2.5.

24We note that in the default equation business cycles are introduced at the moment of the beginningof the credit, whereas in the approval equation business cycles are introduced at the moment of approval,which does not necessarily coincide with the beginning of the credit. Hence, it is possible that there isno overlap between the business cycles variables in the approval and default equations.

90

Table 2.5: Determinants of Approval, Business Training and DefaultProcesses

Model Trivariate probit

(1) (2) (3)

Dependent variable: Approval Business training Default


Risk 27.04*** (10.47)

Risk2 -46.13** (20.93)

Other demands 3.17*** (1.2)

Loan of honor 1.44*** (0.48)

Sent by a mainstream bank -1.5*** (0.53)

ρ1 1.13** (0.46)

Business training (direct e�ect) 0.82*** (0.22)

Male -0.16 (0.13) 0.51*** (0.13)

Education (direct e�ect) -0.02 (0.05) -0.27*** (0.07)

Single -0.003 (0.14) 0.30** (0.14)

Unemployed at least 12 months -0.33** (0.13) 0.13 (0.12)

Household income (kEUR) 0.12* (0.07) -0.46*** (0.07)

Household expenses (kEUR) -0.24* (0.14) 1.17*** (0.13)

Low personal investment -0.31** (0.14) 0.27** (0.12)

Assets 0.003 (0.003) -0.02*** (0.01)

Food and accommodation sector -0.45** (0.19) 0.67*** (0.18)

Gross margin/Sales -0.57* (0.32) -0.28 (0.30)

ρ0 0.74*** (0.22)

Growth rate of defaults approval stage -2.93*** (0.79)

Growth rate of defaults approval stage -1 -2.32*** (0.72)

Growth rate of defaults approval stage -2 -0.44 (0.50)

Growth rate of start-ups approval stage -0.003 (0.02)

Growth rate of start-ups approval stage -1 -0.02 (0.01)

Growth rate of start-ups approval stage -2 -0.01 (0.01)

Growth rate of defaults beginning of credit -0.001 (0.008)

Growth rate of defaults beginning of credit +1 -0.006 (0.005)

Growth rate of defaults beginning of credit +2 0.002 (0.007)

Growth rate of start-ups beginning of credit 0.03*** (0.01)

Growth rate of start-ups beginning of credit +1 0.007 (0.005)


Business training (indirect e�ect) -14.17 (530)

Education (indirect e�ect) 0.32*** (0.09)

Intercept 0.77** (0.34) -3.67*** (1.42) -1.38*** (0.32)

-2 Log Likelihood 1518

Observations 662



This innovative model with three jointly estimated probit equations allows us to control for

the selection bias. Akin to Roszbach (2004) we use the same control variables for individual,

household and business characteristics in the approval and default equations. Controlling for

selection bias allows us to take into account the entire distribution of the variables, whereas in the

bivariate model we only use information on selected applicants. Selected applicants are expected

to have a lower risk compared to the rejected applicants. Hence, we expect the coe�cients of

Risk and Risk2 to be larger in absolute value after the correction of selection bias.

When we control for the selection bias, the non-monotonic relationship between the likelihood

to receive business training and the risk of the borrower remains signi�cant. We note that the

loadings for Risk and Risk2 are indeed larger in absolute value compared to the bivariate model.

The Risk threshold for which the likelihood to receive business training is reversed corresponds

to 0.29. 11% of our sample have an estimated risk higher than 0.29.

Coe�cients in the approval equation are expected to be of the opposite sign compared to the

risk equation, if the MFI's selection process is rational. Notably, some variables do not meet this

expectation. Male dummy, education level, single dummy, assets' level do not seem to impact

the approval decision despite a signi�cant impact in the default equation. In contrast, long term

unemployment and gross margin on sales are not signi�cant in the default equation. The most

striking result concerns the direct e�ect of business training. Using a trivariate model, we �nd

that business training increases the likelihood to default, whereas in the bivariate model this

coe�cient is not signi�cant. Interestingly, the indirect e�ect of business training is no longer

signi�cant and the indirect e�ect of the education becomes signi�cantly positive, suggesting that

default uncertainty increases for higher educated individuals.

Concerning business cycles we note that the growth rates of defaults are important in the approval

process, whereas a higher growth rate of start-ups at the beginning of the credit (i.e. higher

competition) positively impacts the likelihood to default.

Finally, both ρ1 and ρ0 are signi�cant suggesting that, indeed, there is correlation among the

error terms of the three processes which has to be taken into account.

92


2.7.2 An alternative measure of risk: the inverse of the survival

time

We extend the previous model by using additional information about the survival time of a

loan, Ti. In this model, the risk equation concerns the survival time before a default occurs

rather than only the event of a default. We calculate ti as follows. For defaulting loans ti is

the number of days between the date of loan granting and the date when default occurred. For

non-defaulting loans ti is the number of days between the date of loan granting and the date of

the data extraction. The survival time is then either perfectly observed (not censored) when a

default occurs y2i = 1, i.e. Ti = ti or is censored as the loan is still performing when y2i = 0,

i.e. Ti > ti. The bivariate mixed model will allow us to estimate the survival time for each loan.

To do so, we assume that the survival time follows the Weibull distribution which is the most

commonly used duration distribution in applied econometric work (Lancaster 1992).

Ti|vi, zi, y1i ∼Weibull(µi, σ) where µi ≡ exp(z′iβ2 + α1y1i + ρ2ivi)

where ρ2i ≡ exp(α2y1i + δEducationi). The expected survival time is given by:

E(Ti|zi, y1i, vi) = µ−1i Γ(1 +1

σ) (2.11)

where Γ(.) is the complete Gamma function (for more details see Lancaster (1992), Appendix

1) and σ is the Weibull scale parameter. Consequently, the risk to be in default is necessarily

related to the expectation of the survival time. Hence, a possible measure of this risk is naturally

given by:

E(Ti|zi, y1i, vi)−1 = µi[Γ(1 + 1

σ )]−1

For the process of business training decision, we replace the probability to be in default by the

alternative measure of risk, without the current decision y1i which should be obviously excluded

from the set of covariates of the measure of business training decision and without business

93


cycles, as they occur after the decision on business training took place.

y∗∗1i = x′iβ1 + λ1E(Ti|zi, vi)−1 + λ2E(Ti|zi, vi)−2 + ε1i (2.12)

where E(Ti|zi, vi) = exp(−z′iβ2 − vi)Γ(1 + 1σ ).

Concerning the identi�cation strategy, the expected survival time of a loan will be identi�ed

using the observed survival time, censored or not, attached to the granted loans. The identi�ca-

tion mechanism in this model, akin to the baseline model, is simply obtained by the non linear

function of the linear combination of the determinants of risk. We present the results of the

estimation for this model in Table 2.6.

The non-monotonic relationship between y2 and Risk is robust to the introduction of this al-

ternative risk measure.25 The threshold where business training becomes decreasing with risk

amounts to 1.78. After having estimated the risk, that is the inverse of the survival time, for

MFI's clients we �nd that only 2% of the sample has an expected risk higher than 1.78.

Importantly, in the mixed model the coe�cient for y2 becomes signi�cant in the reversed sur-

vival time equation. This �nding suggests that business training indeed is useful to increase the

chances to succeed of a business. The non signi�cance of this coe�cient in the bivariate probit

equation might be due to reduced variability in the risk variable which is a dummy. Neverthe-

less, the positive signi�cant impact in the trivariate model leaves the debate on the e�ciency of

business training highlighted in the existing literature open.

Finally Weibull parameter is signi�cant and positive, suggesting that risk is increasing with

time. The signs of other controls are in line with previous �nding, despite some signi�cance

di�erences. Long term unemployment becomes highly signi�cant. In contrast, assets and food

and accommodation sectors are no longer determinant in the reversed survival time equation.

25We multiply Risk variable by 100 to scale-down the estimated coe�cients and render them compa-rable to other loadings.

94


Table 2.6: Determinants of Business Training and the Inverse of the Survival Time

Model Bivariate Mixed

(1) (2)

Dependent variable: Business training Inverse of Survival Time


Risk 1.71*** (0.65)

Risk2 -0.48** (0.22)

Other demands 1.15*** (0.19)

Loan of honor 0.56*** (0.18)

Sent by a mainstream bank -0.53*** (0.2)

ρ1 0.23 (0.17)

Business training (direct e�ect) -0.86*** (0.23)

Male 0.67*** (0.22)

Education (direct e�ect) -0.49*** (0.1)

Single -0.16 (0.14)

Unemployed at least 12 months 0.53*** (0.14)

Household income (kEUR) -0.41*** (0.09)

Household expenses (kEUR) 0.8*** (0.18)

Low personal investment 0.48*** (0.14)

Assets -0.01 (0.01)

Food and accommodation sector 0.06 (0.21)

Gross margin/Sales -1.16*** (0.31)

Business training (indirect e�ect) -0.22 (0.14)

Education (indirect e�ect) 0.13*** (0.05)

Growth rate of defaults beginning of credit 0.006 (0.004)

Growth rate of defaults beginning of credit +1 0.01** (0.004)

Growth rate of defaults beginning of credit +2 0.01*** (0.003)

Growth rate of start-ups beginning of credit 0.005* (0.003)


Growth rate of start-ups beginning of credit +2 -0.01*** (0.003)

Weibull parameter 3.02*** (0.55)

Intercept -0.96*** (0.19) -5.57*** (0.36)

-2 Log Likelihood 1612

Observations 340


95


In this chapter we analyze how superior information can impact MFI's decisions concerning bor-

rowers' assignment to training programs. In the theoretical model we show that, in situations

where the relationship between business training and the type of the borrower is decreasing under

symmetric information, a non-monotonic relationship between business training and the type of

the borrower may occur under reversed asymmetric information, where the MFI has superior

information about borrower's type.

We test for the existence of this equilibrium using original data from a French MFI which in

addition to loan-granting assigned some of its clients to training programs. Using a bivariate

probit model to control for endogeneity between business training and risk of the borrower, we

have shown that a non-monotonic relationship between assignment to training and the risk is

indeed plausible. The MFI seems to take into account the "looking-glass self e�ect", that is the

fact that its choices impact borrowers' beliefs about their type, on the microcredit market.

Our chapter provides interesting evidence on how MFI's decisions might undermine borrowers'

motivation to exert e�ort. However, further research is required to better understand the condi-

tions on the shape of the beliefs and on how the MFI estimates the expected type of the borrower.

Further research is also needed concerning the assumptions we make in the theoretical model.

In this perspective it is worthy providing a link between the assumptions and data by testing

the hypothesis that business training has a greater impact on high-risk borrowers compared to

low-risk borrowers.

2.8 Appendix

A.Bivariate Probit Model: Likelihood Function

The individual contribution to the likelihood function given the common factor vi can be written

as follows:

96

2.8. Appendix

Li(θ|y1i, y2i, xi, wi, vi) = Φ

(x′iβ1 + λ1Φ(w

′iβ2 + vi) + λ2

[Φ(w

′iβ2 + vi)

]2+ ρ1vi

)y1i︸︷︷︸

P (y1i=1|vi,...)

·

[1− Φ

(x′iβ1 + λ1Φ(w

′iβ2 + vi) + λ2

[Φ(w

′iβ2 + vi)

]2+ ρ1vi

)](1−y1i)︸︷︷︸

P (y1i=0|vi,...)

·

[Φ(w

′iβ2 + α1y1i + vi)

]y2i︸︷︷︸P (y2i=1|vi,y1i,...)

·[1− Φ(w

′iβ2 + α1y1i + vi)

](1−y2i)︸︷︷︸P (y2i=0|vi,y1i,...)

Hence, in the �rst model with two simultaneous probit equations we �nally have to integrate Li

with respect to the density function of vi, by using the adaptive Gaussian quadrature integral

approximation, we maximize the log of the likelihood function.

l(θ|y1i, y2i, xi, wi)

=n∑i=1

ln

(∫Li(θ|y1i, y2i, xi, wi, vi)φ(vi)dvi

)

B.Trivariate Probit Model: Likelihood Function

The individual contribution to the likelihood function given the common factor vi can be written

97


as follows:

Li(θ|y0i, y1i, y2i, wi, xi, vi) = Φ(w′iβ0 + ρ0vi

)y0i︸︷︷︸P (y0i=1|vi,...)

·[1− Φ

(w′iβ0 + ρ0vi

)](1−y0i)︸︷︷︸P (y0i=0|vi,...)

·

Φ

(x′iβ1 + λ1Φ(w

′iβ2 + vi) + λ2

[Φ(w

′iβ2 + vi)

]2+ ρ1vi

)y0iy1i︸︷︷︸

P (y1i=1|vi,y0i=1,...)

·

[1− Φ

(x′iβ1 + λ1Φ(w

′iβ2 + vi) + λ2

[Φ(w

′iβ2 + vi)

]2+ ρ1vi

)]y0i(1−y1i)︸︷︷︸

P (y1i=0|vi,y0i=1,...)

·

[Φ(w

′iβ2 + α1y1i + vi)

]y0iy2i︸︷︷︸P (y2i=1|vi,y0i=1,y1i,...)

·[1− Φ(w

′iβ2 + α1y1i + vi)

]y0i(1−y2i)︸︷︷︸P (y2i=0|vi,y0i=1,y1i,...)

Hence, in the model with three simultaneous probit equations we �nally have to integrate Li

with respect to the density function of vi, by using the adaptive Gaussian quadrature integral

approximation, we maximize the log of the likelihood function.

l(θ|y0i, y1i, y2i, wi, xi)

=n∑i=1

ln

(∫Li(θ|y0i, y1i, y2i, wi, xi, vi)φ(vi)dvi

)

C.Bivariate Mixed Model: Likelihood Function

The individual contribution to the likelihood function conditional on vi using loan survival time

98

2.8. Appendix

can be written as follows:

Li(θ|y1i, y2i, ti, xi, wi, vi)

= Φ(x′iβ1 + λ1E(Ti)

−1 + λ2E(Ti)−2 + ρ1vi

)y1i︸︷︷︸P (y1i=1|vi,...)[

1− Φ(x′iβ1 + λ1E(Ti)

−1 + λ2E(Ti)−2 + ρ1vi

)](1−y1i)︸︷︷︸P (y1i=0|vi,...)[

σµσi tσ−1i exp {− (µiti)

σ}]y2i︸︷︷︸

f(ti|vi,y1i,...)

[exp {− (µiti)σ}](1−y2i)︸︷︷︸

P (Ti>ti|vi,y1i,...)

99


100

Chapter 3

Microcredit in Developed Countries:

Unexpected Consequences of Loan Ceilings1

3.1 Introduction

To favor �nancial services to poor entrepreneurs, US and European regulators have set upper

limits to the size of the loans that subsidized micro�nance institutions (MFIs) grant. By means

of a theoretical model and empirical evidence, this chapter shows that loan ceilings may paradox-

ically result in holders of large business projects being more likely to receive a microcredit. This

perverse e�ect is linked to the possibility of co-�nancing. More precisely, micro-entrepreneurs

holding large business projects requiring above-ceiling loans can secure the above-ceiling share

of the loans with a regular bank, and then apply for ceiling-high loans from the MFI. The co-

�nancing option is attractive to MFIs since it allows them to free-ride on the banks' screening

of applicants.

In this chapter we focus on loan ceilings imposed by regulators to subsidized micro�nance institu-

tions (MFIs) in most developed countries. Micro-entrepreneurs in need of above-ceiling loans are

left with the co-�nancing option, which means securing the above-ceiling share of the loan with

a regular bank, and getting a ceiling-high loan from the MFI. Co-�nancing is attractive to MFIs

1This chapter is based on a joint work with Ariane Szafarz.

101

Unexpected Consequences of Loan Ceilings

because it allows them to free-ride on the regular banks' screening process. Therefore, loan ceil-

ings can have the perverse e�ect of facilitating the co-�nancing of large projects at the expense of

micro-entrepreneurs who need below-ceiling loans only. This is the gist of our theoretical model.

We test the predictions of this model by exploiting the natural experiment of a French MFI that

became subject to the French EUR 10,000 loan ceiling in April 2009. Di�erence-in-di�erences

probit estimations con�rm that imposing loan ceilings to MFIs can have unexpected and socially

harmful consequences. The issue addressed in this chapter is speci�c to micro�nance in devel-

oped countries where MFIs remain niche institutions. This situation contrasts with the rapid

expansion of micro�nance in developing countries (Armendariz and Morduch 2010).2 MFIs in

developing countries typically supply standardized products � predominantly small loans � to a

large number of unbanked people. Due to existence of both banking coverage and social safety

nets, MFIs in developed countries target a limited number of micro-entrepreneurs disregarded by

commercial banks (Johnson 1998). These MFIs are meant to address a market failure and facil-

itate self-employment. According to Bendig et al. (2012), the �ve main objectives of European

micro�nance are: job creation, promotion of micro-enterprises, �nancial and social inclusion, and

empowerment of the speci�c target groups. In 2011, the MFIs active in the European Union

(EU) have granted more than 204,080 loans amounting EUR 1,047 million in total.3

In Europe, most MFIs bene�t from subsidies provided by local and/or national governments

(Bendig et al. 2012). Some are also �nanced by commercial banks in the framework of their

socially-responsible investment policy. Subsidies come in various forms, direct and indirect. Indi-

rect subsidies include: protection against default risk, tax incentives, loans at preferential rates,

and provision of business development services. As stressed by Hudon and Traça (2011), subsi-

dies are instrumental to MFIs, especially during their start-up phase. Subsidized micro�nance

can even reveal pro�table to public �nance (Evers et al. 2007; Brabant et al. 2009). Indeed,

2According to Lalwani and Kubzansky (2009), the sector serves 200 million customers worldwide.3The �gures largely underestimate the reality since they are based on responses from 108 MFIs among

the 376 contacted. Bendig et al. (2012) estimate that 500 to 700 MFIs are currently active in the EU,excluding credit unions and commercial banks.

102

3.1. Introduction

MFIs serve the poor and the unemployed, promote job creation, and so reduce the �nancial

burden of social welfare.

The literature on micro�nance in developed countries is scarce, likely because the sector is still

young and poorly delimited (European Commission 2012).4 The division between microcredit

and small business �nancing remains unclear. Depending on the provider, similar loans are clas-

si�ed as micro-loans, conventional loans, consumer loans, or SME loans.

In developing countries, small-business �nance and microcredit act as complements rather than

substitutes (Bauchet and Morduch 2013). In developed countries, the situation is trickier. The

division between businesses served by regular banks and MFIs is blurred, and some MFIs serve

clients who can borrow from banks. The reaction of the banking sector to the development of mi-

crocredit activities has been mixed. On the one hand, some banks have climbed the micro�nance

bandwagon by creating MFIs5 and/or by developing collaborations with MFIs. On the other

hand, the banking sector has been asking for better market delimitation and strict supervision

of micro�nance activities.

If o�ered the choice, most micro-entrepreneurs would prefer microcredit to a regular bank loan.

This is because socially-oriented and subsidized MFIs manage to screen their applicants less

severely than regular banks. MFIs also o�er attractive credit conditions, and some provide busi-

ness guidance. Therefore, many banks consider subsidized MFIs as a threat. At the request

of the banking sector, new rules have come into force (European Micro�nance Network 2012).6

The key features of existing regulatory frameworks concern access to data from credit bureaus,

interest-rate caps, access to �nancial markets, and loan ceilings.

4The European Commission has launched several initiatives to foster the development of the sector.For instance, the Joint European Resources for Micro to Medium Enterprises (JEREMIE) programallocates structural funds to European MFIs. The Joint Action to Support Micro�nance Institutions inEurope (JASMINE) program o�ers them technical assistance.

5For example, Fundació Un Sol Món in Spain was established by the savings bank Caixa Catalunyain 2000, and CSDL a non-bank MFI in France was established by a joint initiative by Crédit Municipalde Bordeaux and local authorities

6In Germany, only banks are allowed to grant credit, and MFIs act as simple intermediaries within atight collaboration with banks. For instance, the GLS bank delegates small-loan granting to local MFIs.

103


As far as loan ceilings are concerned, France has one of the most restrictive rules in the developed

world. The French Monetary and Financial Code (2007) stipulates that licensed MFIs are for-

bidden to grant loans above EUR 10,000. In contrast, the U.S Small Business Administration,

a federal agency promoting the creation and development of small businesses, has set a USD

50,000 cap to microcredit (Lieberman et al. 2012). The European Union recommends the use of

a EUR 25,000 ceiling (European Commission 2007). In practice, however, EU member states �x

their own ceilings. For instance, Hungary, Portugal, Slovakia, and the UK allow MFIs to grant

loans exceeding EUR 25,000 (European Commission 2007).7

The o�cial report on microcredit in France released by the General Inspection of Finances (Bra-

bant et al. 2009, p. 4) explains that the ceiling is imposed for�yet unspeci�ed�"cautionary

reasons" and that it is meant to keep mainstream banks �nance small businesses. In addi-

tion, the French authorities encourage bank-MFI cooperation. For instance, not-for-pro�t MFIs

such as ADIE (Association pour le Droit à l'Initiative Economique), CREASOL (Contraction

de Crédit Accompagnement Solidarité) and CSDL (Caisse Sociale de Dévelopement Local) have

received the public license for re�nancing their microcredit activity with bank loans (Valentin

et al. 2011).

The micro�nance literature provides mixed evidence on the impact of regulation on perfor-

mances of MFIs. Armendariz and Morduch (2010) content that the existing regulations are

poorly adapted to this young industry.8 Using data for 114 MFIs from 62 countries, Hartarska

and Nadolnyak (2007) �nd that regulations do not directly a�ect operational self-sustainability

and outreach. Cull et al. (2011) emphasize that complying with regulations is costly to MFIs

7In Europe, microcredit is provided by both banks and non-bank �nancial institutions. The EuropeanCommission (2007) provides a detailed overview of microcredit regulations in Europe. Microcredit pro-vision falls either under the harmonized banking-sector regulations, or under the far more heterogeneouslaws governing non-bank institutions. Only Romania and France have adopted rules speci�c to micro-credit, referred to as "special windows" (CGAP and World Bank 2012). A special window adopted inRomania in 2005 imposes a EUR 25,000 loan ceiling. In 2010, Italy has enforced two distinct ceilings: aEUR 25,000 one for business lending, and a EUR 10,000 one for social lending.

8For instance, Acclassato (2008) shows that legal interest rate caps are ine�cient in micro�nanceindustry.

104

3.1. Introduction

and may result in the exclusion of potential borrowers. The pros and cons of loan ceilings are

discussed in a CGAP report (CGAP and World Bank 2012). The report states that ceilings

constrain MFIs to focus on poor clients but prevent holders of large projects from gaining access

to �nance. Ceilings also reduce cross-subsidization opportunities.9 Our �ndings challenge these

statements.

The contribution of this chapter is twofold. First, it proposes a theoretical framework for dis-

cussing the consequences of loan ceilings, a key component in micro�nance regulation. Inspired

from Armendariz and Szafarz (2011), our model describes the loan allocation of a subsidized

socially-oriented MFI. It rests upon the realistic assumption that micro-entrepreneurs holding

large projects � i.e. projects requiring above-ceiling loans � have access to co-funding (Jain 1999).

Co-funding implies combining credit from a bank charging the market interest rate and from a

subsidized MFI charging a below-market rate. The bank's approval comes �rst.10 Hence, the

MFI has the opportunity to free ride on the bank's screening process. This opportunity drives

the MFI's preference for pre-screened applicants, i.e. holders of large projects, at the expense

of applicants holding smaller projects. In the micro�nance literature, such shift in clientele is

referred to as "mission drift".11 In sum, our model stresses that loan ceilings can have perverse

e�ects and trigger mission drift.

Second, we test the predictions of our theoretical model on real-life data. We exploit a natural

experiment, namely the conversion of a French unregulated NGO supplying microcredit into a

regulated MFI. This conversion occurred in April 2009 and implied immediate compliance with

the French EUR 10,000 ceiling. Interestingly, the unregulated institution did grant as many as

70% of above-ceiling loans. As expected, the conversion a�ected its loan allocation dramatically.

More precisely, di�erence-in-di�erences probit estimation shows that the change in status is as-

9MFIs use cross-subsidization when they partly cover the costs associated with serving the very poorby lending to wealthier, and hence more pro�table, clients.

10Typically, banks give conditional approvals.11Mission drift means that MFIs serve wealthier clients at the expense of poor ones (Ghosh and Van Tas-

sel 2008; Mersland and Strøm 2010; Armendariz and Szafarz 2011).

105


sociated with a rise in the MFI's approval rate. This result is consistent with the presence of

free-riding in the screening process. Moreover, the projects �nanced by the MFI became signif-

icantly larger after the enforcement of the ceiling. Overall, the empirical results seem to be in

line with the predictions of the theoretical model but they have to be interpreted with care due

to potential (self-)selection biases.

The rest of this chapter is structured as follows. Section 2 presents the theoretical model. Section

3 describes the context and the data. Section 4 outlines the empirical results. Robustness checks

are performed in Section 5. Section 6 concludes.

3.2 Theoretical Model

In this section, we build a simple one-period model inspired from Armendariz and Szafarz (2011).

The aim is to derive the impact of a loan ceiling on the loan allocation of a socially-oriented MFI.

The pool of applicants is composed of two groups of micro-entrepreneurs. The members of the

�rst group hold small projects and demand small loans. The members of the second group hold

large projects and demand relatively larger loans. The risk-neutral subsidized MFI maximizes

its outreach, i.e. its number of borrowers, under the budget constraint. Subsidies allow the MFI

to supply credit at below-market conditions. We assume cross-subsidization away by imposing

that both types of loans are costly to the MFI. We proceed in three steps. First, we present the

basic model without loan ceiling. Second, we add the ceiling to the picture and solve the model

again. Third, we compare the optimal loan allocations in the two situations. This comparison

will guide the empirical exercise in Section 4.

3.2.1 Loan Allocation without Ceiling

We consider a risk-neutral socially-oriented MFI supplying loans at below-market conditions to

micro-entrepreneurs. To ful�ll its social mission the MFI bene�ts from subsidy K. The MFI

106

3.2. Theoretical Model

maximizes its number of borrowers under the budget constraint. The pool of the applicants is

made of type-1 micro-entrepreneurs holding small projects and demanding loan size P1 and type-

2 entrepreneurs holding large projects and demanding loan size P2 with P1 < P2. For the MFI,

granting a loan to a type-1, respectively type-2, applicant entails a total cost of γ1, respectively

γ2. Costs are additive.12 The costs should be understood as the MFI's net cash-out �ows

associated with granting the loans. Actually, they aggregate cash-�ows of both signs. Positive

costs include the standard business costs associated with the loan granting activity. Negative

costs (or bene�ts) result from the expected returns on loan reimbursement. The net costs also

account for any other borrower-speci�c cash-�ows. For instance, if the MFI receives speci�c

subsidies for, say, serving�typically poorer�type-1 applicants, this extra budget is interpreted as

a negative component of γ1. To acknowledge that the MFI operates in a competitive environment

and o�ers below-market conditions, we assume that P1+γ1 > 0 and P2+γ2 > 0. This assumption

means that microcredit granting is costly and requires subsidization. In this way we also rule

out cross-subsidization opportunities, which are unrealistic in a competitive environment.

The program of the MFI writes:

max0≤n1,0≤n2

{n1 + n2} (3.1)

s.t. K = (P1 + γ1)n1 + (P2 + γ2)n2

In this simple linear set-up, the optimal loan allocation is a corner solution. To maximize

outreach, the MFI �nances a single type of projects. More precisely, the optimal numbers of

granted loans are:

n∗1 =

K

P1+γ1if P1 + γ1 ≤ P2 + γ2

0 if P1 + γ1 > P2 + γ2

(3.2)

12We rely on this simplifying assumption for tractability. More realistic representations of the costswould make the model more complex without a�ecting much its qualitative outcome.

107


and

n∗2 =

0 if P1 + γ1 ≤ P2 + γ2

KP2+γ2

if P1 + γ1 > P2 + γ2

(3.3)

In a ceiling-free environment, the MFI operates on its own. Its optimal loan allocation only

depends on loan sizes and costs. In particular, if the overall �nancial burden of serving type-1

applicants is lower than that of serving type-2 applicants (P1 + γ1 ≤ P2 + γ2), Eqs. (3) and

(4) show that the MFI will choose type-1 borrowers exclusively. This is the typical situation of

non-pro�t MFIs worldwide. When cross-subsidization is not a possibility, MFIs that do not drift

away from their mission o�er small loans only.

3.2.2 Loan Allocation with Ceiling

Let us now assume that the same MFI is constrained by loan ceiling S, where P1 < S < P2.

The MFI can still serve type-1 applicants in full. However, its type-2 applicants need external

�nancing to complement the maximal loan size, i.e. S, the MFI is allowed to supply. Since the

MFI proposes below-market conditions, the type-2 applicants are still interested in getting as

much credit as possible from the MFI. But before applying for microcredit, type-2 entrepreneurs

must secure a loan amounting P2 − S from a regular bank. We assume that the regular credit

market is competitive and entails credit-rationing (Stiglitz and Weiss 1981). Hence, regular

banks screen their credit applicants selectively.13

The MFI examines type-2 applications once a complementary loan has been secured. Inevitably,

the screening process implemented by the bank reduces the number of type-2 project holders who

manage to apply for microcredit. Let us denote N2 the number of type-2 applicants surviving the

bank's screening process. The others are rejected by the bank and disappear from the microcredit

market.14

13Modeling the bank's screening process would allow us to quantify further the MFI's loan allocation.14In the ceiling-free environment, we have implicitly assumed that the composition of the pool of

applicants is never binding for the MFI. Put di�erently, the MFI may �nd as many applicants of eachtype as its optimal loan allocation dictates. Lifting away this assumption would a�ect much the impact

108


Importantly, the surviving type-2 applicants to microcredit are pre-screened by the bank. In this

way, the loan ceiling reduces the informational asymmetry associated with type-2 applicants.

As a result, these applicants incur smaller costs than the unscreened type-2 applicants in the

ceiling-free environment. In other words, granting loans to type-2 applicants gives the MFI an

opportunity to free ride on the bank's screening process. Let us denote γ′2 < γ2 the cost associated

with type-2 pre-screened applicants. Ruling out cross-subsidization implies that S+γ′2 > 0. The

program of the MFI now writes:

max0≤n1,0≤n2≤N2

{n1 + n2} (3.4)

s.t. K = (P1 + γ1)n1 + (S + γ′2)n2

Again, the objective function is linear leading to corner solutions. However, the solution is

somewhat heavier to write down due to the restriction in type-2 applicants. The optimal numbers

of loans granted by the ceiling-constrained MFI are:

n∗∗1 =

K

P1+γ1if P1 + γ1 ≤ S + γ

′2

max

{0,

K−(S+γ′2)N2

P1+γ1

}if P1 + γ1 > S + γ

′2

(3.5)

and

n∗∗2 =

0 if P1 + γ1 ≤ S + γ

′2

min

{K

S+γ′2

, N2

}if P1 + γ1 > S + γ

′2

(3.6)

Eq. (3.5) and (3.6) reveal that the MFI's optimal allocation can include both types of borrowers.

If P1+γ1 ≤ S+γ′2, the MFI serves small project holders only. In contrast, when P1+γ1 > S+γ

′2,

the constraint on the number of pre-screened type-2 borrowers (n2 ≤ N2) can bite. In this case,

the MFI serves the N2 available large project holders, and it is left with no other choice but

supplying loans to type-1 applicants with the remaining budget. Alternatively, when constraint

of a loan ceiling. The only thing that really matters here is that the need for co-funding makes the MFItype-2 applicants strictly less numerous than otherwise.

109


n2 ≤ N2 does not bite, the MFI serves type-2 borrowers only. As a result, the MFI's optimal

loan allocation depends on the severity of the bank's screening process.

Evidently, this model makes sense only in an environment where commercial banks are interested

in co-�nancing micro-entrepreneurial projects with MFIs.15 In fact, banks may �nd co-�nancing

attractive since this is a way to share credit risks with the MFI. Given that many MFI applicants

hold start-up projects, the credit risks at stake may be high. Hence, imposing a loan ceiling to

the microcredit industry might ultimately ease �nancing large start-up projects.16 On the other

hand, banks that co-�nance projects with MFIs push the latter toward mission drift by inciting

them to disregard holders of small projects.

3.2.3 Comparison

Starting from the previous results, we now compare the optimal loan allocations without and

with a loan ceiling of level S. Since S < P2 and γ2 > γ′2, we have:

P2 + γ2 > S + γ′2 (3.7)

This inequality means that serving type-2 applicants is more a�ordable for the ceiling-constrained

MFI than for the ceiling-free one. Consequently, there are three possible cases (see Table 3.1). In

case I (P1+γ1 > P2+γ2 > S+γ′2), the ceiling-free MFI �nds type-1 projects costlier than type-2

ones. As a consequence, it only �nances type-2 projects. Introducing a loan ceiling cannot make

the situation worse for type-1 applicants, so that n∗1 ≤ n∗∗1 . Moreover, if the bank's screening is

severe enough to make constraint n2 ≤ N2 bite, then the inequality is strict. This would mean

15Otherwise N2 = 0 and type-2 applicants are unable to apply for microcredit. However, if banks arenot willing to fund type-2 projects partially, they should be even less keen to fund them in full. Type-2applicants would then �nd no way to �nance their projects. This market-failure situation is unlikely�butnot excluded�in a competitive credit market.

16Typically, entrepreneurs �nd it di�cult to gather su�cient funding for start-up project because oftheir informational opacity (Berger and Udell 1998). In this chapter, however, we disregard the indirectspillover e�ects of microcredit loan ceilings on bank lending and focus on the direct e�ect on MFIs.

110


that the ceiling-constrained MFI would serve some type-1 applicants even though the type-2 ones

have become less costly. This result is attributable to a bank-driven rationing e�ect. Case I is

probably the situation regulators have in mind when imposing loan ceilings. The increase in the

number of small projects �nanced appears as the consequence of imposing a tough prerequisite

to holders of large projects.

Table 3.1: Comparison of optimal loan allocations without and withceiling

CaseWithout ceiling With ceiling Outreach

n∗1 n∗2 n∗∗1 n∗∗2 Small projects Large projects

I. P1 + γ1 > P2 + γ2 > S + γ′2 0 K

P2+γ2max

{0,K−(S+γ

′2)N2

P1+γ1

}min

{K

S+γ′2

, N2

}n∗1 ≤ n∗∗1 n∗2 ≶ n∗∗2

II. P2 + γ2 ≥ P1 + γ1 > S + γ′2

KP1+γ1

0 max

{0,K−(S+γ

′2)N2

P1+γ1

}min

{K

S+γ′2

, N2

}n∗1 ≥ n∗∗1 n∗2 ≤ n∗∗2

III. P2 + γ2 > S + γ′2 ≥ P1 + γ1

KP1+γ1

0 KP1+γ1

0 n∗1 = n∗∗1 n∗2 = n∗∗2

Two polar subcases of case I stress the role of the spillover e�ect of the bank screening process

on the MFI optimal strategy. First, if the bank's screening has maximal severity (N2 = 0),

all type-2 applicants are rejected by the bank. As a result, the ceiling-constrained MFI serves

type-1 applicants only. This situation would correspond not only to a full segmentation of the

credit market, but also to a market failure. Indeed, type-2 applicants fail to obtain credit from

any source. The bank �nds them too risky while their demanded loan sizes are too high for the

MFI. Second, if the bank's screening is very soft(N2 ≥ K

S+γ′2

), the rationing e�ect disappears

and the MFI serves more type-2 applicants than in the ceiling-free situation (n∗2 < n∗∗2 ). This is

a consequence of the cost reduction associated with co-�nancing.

In case II (P2 + γ2 ≥ P1 + γ1 > S + γ′2), the ceiling-free MFI serves type-1 applicants only

while the ceiling-constrained MFI prefers type-2 applicants. The bank's screening reduces the

MFI's cost enough to reverse the MFI's preferences. Imposing a loan ceiling renders MFIs less

social in the sense of serving a smaller number of small project holders (n∗1 ≥ n∗∗1 ). However, the

111


severity of the bank's screening has a strong in�uence on the optimal allocation of the ceiling-

constrained MFI. If the screening is soft(N2 ≥ K

S+γ′2

), the MFI allocates its full budget to

type-2 projects. In contrast, when the screening is tough(N2 <

K

S+γ′2

), the surviving type-

2 projects are rationed. Once the MFI has exhausted this limited set of projects, it uses its

remaining budget to �nance type-1 projects. Either way, the outcome is detrimental to small

project holders. Case II illustrates the perverse e�ect of capping loan size. When case II applies,

instead of forcing the MFI to �nance small projects, the loan ceiling pushes the MFI toward

co-�nancing large projects with banks.

In case III (P2+γ2 > S+γ′2 ≥ P1+γ1), the loan ceiling has no impact on the MFI's loan allocation.

The MFI prefers type-1 projects to both screened and unscreened type-2 ones. Hence, the MFI

is insensitive to the bank's presence and serves type-1 applicants only.17

In sum, in case I the loan ceiling works in line with the regulator's intention. In case II, the

ceiling creates a perverse incentive to the MFI. In case III, the ceiling is useless. In practice,

predicting the precise reactions of MFIs to a given loan ceiling is complicated for several reasons.

First, same-jurisdiction MFIs exhibit substantial heterogeneity and attract di�erent groups of

applicants. Second, the impact of a loan ceiling crucially depends on the interaction of three

parameters: the level of the ceiling, the cost reduction associated with the bank's screening, and

the severity of this screening. Possibly, any reasonable loan ceiling will have the desired impact

on some MFIs but the perverse e�ect on others.

Still, the level of the ceiling matters. All other things equal, low ceilings reduce the prevalence

of case III, and make the MFI's optimal strategy more dependent on banks.18 For regulators,

identifying the ceiling that best �ts their objective is not an easy task. When imposed to MFIs

that spontaneously serve holders of small projects, ceilings may create mission drift. In contrast,

17This is not necessarily a market failure. Indeed, the bank could �nd type-2 projects attractive.Simply, capping the MFI's loan size is irrelevant to loan allocation. This is the typical situation wherethe regulation imposes a very high ceiling.

18In addition, the bank's screening severity could depend on the level of the loan ceiling. The lowerthe ceiling, the higher is the bank's exposure to credit risk.

112

3.3. Data and Context

ceilings may restore the social orientation of MFIs that are targeting more ambitious borrowers.

Our empirical results will demonstrate that the risk of mission drift is a real one.

3.3 Data and Context

In 2001, France adopted a regulatory "special window" devoted to non-bank �nancial institutions

providing credit for small start-ups, business developments, and buy-outs. The notable advan-

tage of complying with this regulation is the possibility of accessing funds on �nancial markets.

Funds accessibility reduces the MFIs' dependency on subsidies. In 2007, the special window has

been completed by The French Monetary and Financial Code (2007) allowing non-bank MFIs to

supply microcredit provided that they comply to the EUR 10,000 loan ceiling.19

This French EUR 10,000 ceiling is signi�cantly lower than the EUR 25,000 ceiling suggested by

the European Commission. The French regulators justify this choice by restricting the target

pool of borrowers of regulated MFIs to micro-businesses having no more than three employees.20

Interestingly, the regulators do not mention any other characteristics of the borrowers. The reg-

ulators' motivations contrast with the narrative of the micro�nance sector, which presents itself

as favoring self-employment for the unemployed and disadvantaged-group members. Currently,

France counts only three regulated non-bank MFIs: ADIE, CREASOL, and CSDL.21

To investigate the consequences of the French regulatory loan ceiling, we have hand collected

exhaustive data on the applicants and borrowers of CREASOL, an NGO that turned itself into a

regulated MFI in April 2009. Our database covers the 2008 - 2012 period, allowing us to observe

the loan allocation process under both statuses. We view the change in status as a natural ex-

19Some French commercial banks and cooperatives also supply microcredit. These institutions fallunder the Banking Law, and not the special window (European Commission 2007).

20The regulation also allows MFIs to grant consumer loans caped at EUR 3,000. However, there is nosuch loan in our sample.

21ADIE is the largest regulated MFI. In 2012, ADIE supplied 10,914 business microloans. Its year-endoutstanding amount was EUR 58,010,000 (Adie Annual Report 2012). In 2012, CREASOL supplied 648business microloans. Its year-end outstanding amount was EUR 3,526,000 (CREASOL Annual Report2012). CSDL has not released its 2012 �gures yet. In 2010, it supplied 285 business microloans for atotal amount of EUR 1,542,000. (CSDL Annual Report 2010).

113


periment representing a unique opportunity to scrutinize the impact of the French loan ceiling.

CREASOL was created in 2006 by a mainstream bank in the framework of its social respon-

sibility policy. Until April 2009, CREASOL operated under the unregulated NGO status. It

was subsidized by its parent bank and bene�ted from loan guarantees provided by the French

Government. The board of the legally independent NGO included a CEO, an executive director,

members of the parent bank, and independent members. As a non-regulated institution CREA-

SOL had no access to �nancing means other than subsidies.

In April 2009, CREASOL decided to become a regulated MFI in order to gain access to funds

at preferential rates. This transformation also resulted in a decrease of the dependence on its

parent bank. Since April 2009, the loan ceiling of EUR 10,000 is enforced. At the time, the

implementation of the new ceiling represented a real challenge for the managers and the cus-

tomers of CREASOL.22 Importantly, despite its statutory transformation the institution kept its

social mission unchanged. The target clientele is primarily composed of two types of borrowers:

unemployed individuals seeking self-employment, and start-ups lacking collateral.

Since its creation, CREASOL operates in line with the microcredit tradition and charges the

same interest rate to all its clients. The average loan duration is 51 months. Loans are repaid in

monthly installments. The annual interest rate is adjusted every two years to market conditions.

Over the sample period, it ranges between 4% and 5%, which is low considering the credit risk

associated with the �nancing of start-ups. In particular, the borrowers who managed to obtain

co-�nancing from CREASOL and a bank were charged a lower rate by CREASOL than by the

bank.

At the moment of application CREASOL observes the �nancial plan of the applicant. All the

sources of �nancing (consisting of personal investment, loans from a classical bank, loans of

honor, subsidies) are included in the �nancial plan. Hence, at the moment of the application to

the MFI the project holder has at least started his negotiation with a classical bank. During

22Interview with Daniel Boccardi and Christian Fara, the CEO and executive director of CREASOL,respectively. The interview was realized on the 27th of November 2013.

114


the credit committee, the MFI can either reject, approve or conditionally approve a microcredit

disbursement. In cases where the negotiation with a classical bank is not concluded the MFI

may approve a microcredit disbursement conditionally on bank approval. This is how the free-

riding on banks' screening process takes place. To test our assumption on free-riding during the

approval process, we perform a simple probit regression using data for the second period, i.e. the

period with ceiling, only. We �nd that having a bank loan positively impacts the probability of

being accepted by the MFI, all other things being equal.23 Concerning monitoring during the

credit relationship, to our knowledge, the MFI and the bank monitor clients separately. Hence

the free-riding only occurs at the approval stage. Moreover, there is no debt seniority in case of

default. This situation might appear puzzling compared to developing countries where MFIs are

usually better placed to screen the applicants in a context of local proximity or delegate screen-

ing to group peers in case of group lending. The micro�nance industry in developed countries is

younger and has not reached the scale of the developing countries yet. Hence, mainstream banks

might still be reluctant to entrust the screening of its clients to MFIs and rather rely on their

own classical credit scorings.24

Until 2010, CREASOL had six employees. In 2010, it experienced a signi�cant growth of its

lending activity, which resulted in the opening of two new branches and the hiring of four ad-

ditional employees. The loan granting process goes as follows. For each application, a loan

o�cer collects all the relevant information about the business, the �nancing structure, and the

applicant's individual and household characteristics. The loan o�cer makes a recommendation

to the credit committee, which has the �nal say. Typically, the decision boils down to approval

or denial of the demanded loan. Only in a small fraction of cases (7.6% in our sample) the

granted loan size is smaller than the demanded amount. Although the decision-making process

has two stages, we only recorded the �nal outcomes. Agier and Szafarz (2013b) show that the

loan o�cer's recommendation is mostly followed by the credit committee.

23Results available upon request.24DeYoung et al. (2008) show that credit scorings can mitigate the information asymmetries associated

with geographically distant small business borrowers.

115


Importantly, the MFI has hired new employees to create new branches that would use the same

lending principles as in the existing branches. Therefore, these new employees are not supposed

to be more business oriented compared to other employees. The CEO has not changed during

the observational period. A competitive explanation of the change in MFI's behaviour toward

loan applicants might be the desire to become more pro�table, as compliance with regulation

has opened access to loanable funds to the MFI.

Overall, our database includes exhaustive information on 1,097 credit applicants. The sample

period is split into two sub-periods, each one characterized by a speci�c regulatory status. During

the �rst sub-period (April 2008-March 2009), CREASOL was an unregulated NGO operating in

a ceiling-free environment. It treated 226 application �les. During the second sub-period (April

2009-June 2012), CREASOL was a regulated MFI constrained by the EUR 10,000 loan ceiling.

It received 871 loan applications.

Table 3.2 summarizes the characteristics of applicants and borrowers before and after the en-

forcement of the loan ceiling. These characteristics are split into three categories: �nancial,

business-related, and individual ones. The �nancial characteristics include the project size, the

demanded loan size, the actual loan size,25 and the existing sources of funds.

25Evidently the two average loan sizes are not measured from the same sample, which explains thediscrepancies observed despite the fact that CREASOL typically grants the demanded amounts.

116


Table 3.2: Descriptive Statistics: Characteristics of Applicants andBorrowersa

Applicants Borrowers

Without With Without With

Ceiling Ceiling t-test Ceiling Ceiling t-test

Financial Characteristics

Project size (kEUR) 30.22 30.62 0.40 26.55 35.08 8.53**

Demanded loan size (kEUR) 18.38 7.01 -11.36*** 16.62 7.05 -9.57***

Demanded loan size ≥ 10000 (%) 0.70 0.29 -0.41*** 0.65 0.31 -0.34***

Granted loan size 15.74 6.89 -8.84***

Having a bank loan (%) 0.03 0.27 0.25*** - 0.33 -

Bank loan (kEUR)b 46.53 40.87 -5.66 - 43.04 -

Having personal investment (%) 0.81 0.83 0.02 0.83 0.87 0.04

Personal investment (kEUR)b 6.85 7.08 0.2 6.00 8.21 2.21*

Having funds from other sources (%) 0.55 0.69 0.14*** 0.51 0.71 0.20***

Funds from other sources (kEUR)b 9.02 9.49 0.47 9.19 9.69 0.50

Business Characteristics

Start-up (%) 0.80 0.84 0.04 0.79 0.81 0.02

Services (%) 0.28 0.30 0.02 0.28 0.30 0.03

Trade (%) 0.22 0.31 0.08** 0.24 0.29 0.05

Accommodation and food service activities (%) 0.16 0.13 -0.04 0.13 0.11 -0.01

Construction (%) 0.09 0.11 0.02 0.11 0.11 0.01

Arts, entertainment and recreation (%) 0.07 0.04 -0.02 0.05 0.04 -0.01

Other sectors 0.18 0.12 -0.06** 0.19 0.14 -0.05

Individual Characteristics

Unemployed for more than six months (%) 0.55 0.59 0.04 0.49 0.55 0.06

Female applicant (%) 0.38 0.41 0.02 0.34 0.40 0.06

Single (%) 0.59 0.51 -0.08** 0.63 0.45 -0.18***

Education (nb. of achieved diplomas) 2.74 2.77 0.04 2.84 2.91 0.07

Average monthly household income (kEUR) 1.10 1.47 0.37*** 1.20 1.64 0.43***

Nb. of observations 226 871 100 521

aThe table gives mean values and t-test for equal means between the two sub-periods (without and with

loan ceiling).

bThe mean value is computed using only non zero data points.

Signi�cance: *** p<0.01, ** p<0.05, * p<0.1117


Interestingly, the applicants' average project size does not vary across periods, staying slightly

above EUR 30,000. This stability seems to indicate that there is a critical amount of cash needed

to start a micro-business in France. If so, our data would suggest that the EUR 25,000 ceiling

proposed by the European Commission is better adapted to the �eld than the French EUR 10,000

ceiling.

Despite the stability of the demanded loan sizes, we observe a signi�cant increase in the actual

loan size, which passed from EUR 26,550 to EUR 35,080. The increase is in line with the case

II outcome of our theoretical model where the bank's screening pushed the MFI toward funding

larger projects.

As a mechanical consequence of the loan ceiling, in the second period the demanded loan size

dropped dramatically (from EUR 18,380 to EUR 7,010). Likewise, the average loan size passed

from EUR 15,740 to EUR 6,890. In the ceiling-free context, 70% of the demanded loans surpass

EUR 10,000. This is additional evidence that the French ceiling is binding. It seems very low

compared to the needs of micro-entrepreneurs. In the second period, only 29% of the demanded

loans are equal to the ceiling value. This sharp drop in ceiling-high demands gives credence

to the assumption that the bank's screening process reduces the number of applicants holding

large projects. This explanation is also consistent with the fact that 27% of the second-period

applicants have previously secured a bank loan. Note that 54% of the applicants requesting an

amount higher or equal to EUR 10,000 held a bank loan.

Interestingly, in the ceiling-free situation, the few applicants with bank loans were all denied

microcredit.26 In the ceiling-constrained situation, the share of holders of a bank loan is higher

among borrowers (33%) than among applicants (27%). Holding a bank loan has moved from

being a liability to being an asset.

The ceiling seems to have no in�uence on the proportion of applicants/borrowers having a per-

sonal investment (around 82%).27 However, in the second period the size of the personal in-

26Precisely, there are six such applicants in our database. However, their average project size (EUR114,000) makes them be potential outliers.

27Financial support from family and friends is here considered as personal investment.

118


vestment rose signi�cantly among borrowers, and not among applicants. This suggests that the

regulated MFI prefers applicants with higher personal investments.

The proportion of applicants/borrowers having funds from other sources also increased sig-

ni�cantly after the enforcement of the ceiling. The loan cap seems to have incited micro-

entrepreneurs put e�ort in seeking additional funds rather than downsizing their projects.28

Fig.3.1 depicts the average �nancing plans submitted by CREASOL's applicants in the two peri-

ods. It shows that the share of project size demanded to CREASOL dropped from 67% to 41%.

In contrast, the share of the bank loan passed from 1% to 14%. Logically, the enforcement of

the loan ceiling coincided with applicants exhibiting a higher diversi�cation of funding sources,

but perhaps with the disappearance too of some applicants who would need above-ceiling loans

from CREASOL but failed to secure bank loans.

The business characteristics in Table 3.2 show that start-ups constitute the lion's share of CREA-

SOL's loan portfolio. Their proportion remains stable over time (around 82%). Likewise, there

is not substantial change in sector representation, except for the trade sector, which gained 9%

in the second period, but only among applicants. Among the individual characteristics, the two�

possibly interlinked�signi�cant changes concern an increase in the average household income and

a decrease in the proportion of single applicants. The regression analysis will control for all these

variables.

Fig. 3.2 features the relationship between project size and loan size.29 We represent the two

period-speci�c scattered plots and the corresponding regression lines. As expected, the rela-

tionship is positive. However, the scatter-plots exhibit strong di�erences. Under regulation

CREASOL �nanced larger projects than it did when unregulated.

28The main contributions to "other funds" come from subsidized loans (76%) or direct subsidies (19%)to micro-enterprises. For instance, Initiative France is a state-funded institution o�ering zero-interestloans to individuals willing to start, develop or buy-out a business. The 2012 average loan size releasedby Initiative France is EUR 8,340. (See http://www.initiative-france.fr/Creer/Pret-d-honneur). Nouvel

Accompagnement à la Création et la Reprise d'Entreprise (NACRE) is another public program supplyingto entrepreneurs business development services and zero-interest loans capped at EUR 10,000. Last, theunemployed seeking self-employment have access to grants from the national employment agency.

29Only actual loans are taken into account.

119


67%

41%

1%

14%

15%

18%

17%

26%

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Without ceiling With ceiling

CREASOL Bank loan Personal investment Funds from other sources

Figure 3.1: Applicants' project �nancing without and with ceiling

y = 0.537x + 2625.8

R² = 0.7587

y = 0.0291x + 6026.6

R² = 0.2155

0

5000

10000

15000

20000

25000

30000

35000

40000

0 50000 100000 150000 200000

Without ceiling With ceiling

Loan

Size

Project Size (in EUR)

Figure 3.2: Loan Size as a Function of Project Size

120


Moreover, Fig. 3.2 illustrates the impact of the loan ceiling. For the second sub-period, there

is indeed an accumulation of points hitting the upper limit for loan size of EUR 10,000.30 As

expected, due to the loan ceiling, when the project size increases, the share �nanced by CREA-

SOL mechanically decreases.

To test the implications of our model, we need to identify the empirical counterparts of the

theoretical notions of type-1 and type-2 applicants. In other words, we must �nd a size threshold

to categorize projects as "small" (type-1) or "large" (type-2). At �rst glance, the loan ceiling,

EUR 10,000, might seem an appealing candidate for this threshold. However, we are seeking a

size threshold whereas the ceiling is on loans. As descriptive statistics amply document, the loan

ceiling does not cap project size, because entrepreneurs do not seek full debt �nancing.31 For

this reason, determining a meaningful size threshold is not easy.32

To separate small and large projects in our sample, we combine two approaches. In line with

the theory, the �rst approach considers that type-1 projects are small enough to be �nanced in

full by a combination of personal investment, a loan from the MFI and the so-called "funds from

other sources," thus excluding bank loans. For each applicant, we compute the total amount

of money previously collected from all sources but bank loans. The average amount, computed

over the whole sample, is added to the loan ceiling to obtain a �rst proxy for the size threshold

equal to EUR 22,048. The second approach uses data from the ceiling-constrained period only, it

considers as type-1 projects those for which fewer than 50% of the applicants hold a bank loan.

Table 3.3 places the second proxy for the size threshold between EUR 25,000 and EUR 30,000.

Equipped with these two complementary approaches, we have decided to use the middle-of-the-

road EUR 25,000 threshold in the baseline regressions, while keeping a large spectrum of other

30Actually, before the change CREASOL had imposed to itself a maximal loan size of EUR 40,000.However, this limitation was hardly binding.

31To deal with moral hazard and adverse selection issues, �nancial institutions favor entrepreneurialprojects with already secured partial funding coming from personal money or funds provided by friendsand family (Manigart and Struyf 1997; Berger and Udell 1998).

32Undeniably, choosing this threshold is somewhat endogenous, and robustness checks will be requiredto examine the sensitivity of our results with respect to the threshold.

121


possibilities (between EUR 10,000 and EUR 40,000) for the purpose of robustness checks.

Table 3.3: Descriptive Statistics: Bank Loan and Approval Rate, with Ceiling only

Project Size Without bank loan With bank loan % applicants % borrowers

Range (EUR) Applicants Borrowers Applicants Borrowers with bank loan with bank loan

0-10,000 2 1 170 94 1% 1%

10,000-15,000 8 4 178 93 4% 4%

15,000-20,000 7 5 111 53 6% 9%

20,000-25,000 18 8 80 50 18% 14%

25,000-30,000 14 12 50 33 22% 27%

30,000-40,000 39 28 28 18 58% 61%

40,000-60,000 42 29 11 7 79% 81%

60,000-80,000 40 28 3 3 93% 90%

80,000-291,400 68 55 2 0 97% 100%

Total 238 170 633 351 27% 33%

The aim of our empirical exercise is to test the prediction of case II in the theoretical model

according to which a loan ceiling can trigger mission drift. More precisely, we will study how

the MFI's approval process changed after the enforcement of the loan ceiling. Did the MFI favor

loans to holders of large projects (i.e. type-2 applicants)? Did the MFI prefer holders of bank

loans? To o�er a �rst hint, Table 3.4 reports approval rates over the two periods, broken down

by project size. The approval rates over the two periods are compared by means of two-sided

t-tests.

122


Table 3.4: Descriptive Statistics: Project Sizes and Approval Rates without and withCeiling

Project Size Without ceiling With ceiling

Range (EUR) Applicants Borrowers Approval rate Applicants Borrowers Approval rate

0-10,000 31 14 45% 172 95 55%

10,000-15,000 31 14 45% 186 97 52%

15,000-20,000 32 21 65% 118 58 49%*

20,000-25,000 23 10 43% 98 58 59%

type-1 projects 117 59 50% 574 308 54%

25,000-30,000 20 10 50% 64 45 70%*

30,000-40,000 26 9 35% 67 46 69%***

40,000-60,000 44 16 36% 53 36 68%***

60,000-80,000 12 5 42% 43 31 72%**

80,000-291,400 7 1 14% 70 55 79%***

type-2 projects 109 41 38% 297 213 72%***

Total 226 100 44% 871 521 60%***

Stars report the signi�cance levels of the t-tests for equal approval rates without and with ceiling.

*** p<0.01, ** p<0.05, * p<0.1

The overall approval rate is signi�cantly higher after the enforcement of the loan ceiling (60%

against 44%). However, the signi�cance of the increase is not uniform across project sizes.33

We observe signi�cant di�erences (at the 5% level at least) for projects surpassing EUR 30,000,

while a 10% signi�cance level is obtained for the EUR 25,000-EUR 30,000 class. For smaller

projects, the approval rates do not signi�cantly change in the presence of the ceiling.34 This

suggests that the enforcement of the ceiling was followed by more favorable treatment for large

loans, which is consistent with case II in the theoretical model. In the next section, we use probit

di�erence-in-di�erences (di�-in-di�) estimation to further investigate how the loan ceiling has

a�ected the approval process.

33The only class of projects for which the approval rate decreased includes those sized between EUR15,000 and EUR 20,000. The di�erence (49% against 65%) is signi�cant at the 10% level only.

34The di�erence in signi�cance levels is not linked to statistical precision since the numbers of obser-vations are higher in small-size project classes.

123


3.4 Regression Analysis

Our theoretical model predicts that the impact of micro�nance loan ceilings depends on the

severity of the credit rationing exerted by regular banks. When rationing is strong (case I in

Table 3.1), the ceiling-constrained MFI is more social than its ceiling-free counterpart, insofar as

it �nances smaller projects. In contrast, when rationing is mild (case II), the ceiling-constrained

MFI �nances larger projects than its ceiling-free counterpart. This is because free-riding reduces

the MFI's screening costs of larger projects. Last, when the ceiling is high enough to be non-

binding, the MFI is insensitive to credit rationing by regular banks (case III).

Regarding CREASOL, the MFI under study, the descriptive statistics show that the ceiling is

strongly binding, so that case III is excluded. When CREASOL was unregulated, more than 50%

of its loans were sized above the ceiling. Likewise, case I seems unlikely since the enforcement

of the ceiling coincided with an increase in the proportion of large projects �nanced. To test

the model predictions in case II while controlling for applicants' characteristics, we now turn to

regression analysis.

Regulatory changes can be viewed as natural experiments, the consequences of which can be

explored econometrically by means of di�-in-di� estimation.35 The control group consists of

small projects, which are not a�ected by the ceiling (type-1 individuals in the theoretical model).

The treated group consists of large projects which had to secure external �nancing under ceiling

enforcement (type-2 individuals in the theoretical model). We, moreover, conjecture that type-

2 individuals who did not access external �nancing have abandoned their project rather than

scaled it down. This assumption allows us to ensure that the control group was not impacted by

the introduction of the ceiling, so that the di�erence between the two groups is not biased. We

additionally make the assumption that the pool of MFI's applicants would not have changed in

35Previous papers have studied in this way the impact of regulatory shifts on �rm �nancing. Using non-linear di�-in-di� estimation Kamar et al. (2009) show that the 2002 Sarbanes-Oxley Act decreased smallbusiness �nancing on capital markets. Using a linear probability model (OLS and �xed-e�ect model),Quinn (2014) �nds that the 2001 Moroccan corporate law harmed access to bank �nancing and made itharder for �rms to reach limited-company status.

124

3.4. Regression Analysis

the absence of loan ceiling.

Here, we aim to test the two predictions obtained in case II, provided in the last column of Table

3.1. These theoretical results concern outreach, i.e. the number of loans granted by the MFI.36

The �rst prediction (n∗1 ≥ n∗∗1 ) states that the total number of type-1 projects granted by the

MFI is not larger in the ceiling period than in the ceiling-free period. The second prediction

(n∗2 ≤ n∗∗2 ) reverses the inequality for type-2 projects. Together, the two inequalities describe the

ceiling-driven mission drift in case II. They are not, however, equivalent since the total number

of projects �nanced by the MFI is period speci�c, so that: (n∗1 + n∗2) 6= (n∗∗1 + n∗∗2 ).

To build the empirical counterparts of the two theoretical predictions, we use probabilities of

approval. The �rst testable hypothesis, H1 in Table 3.5, states that the approval probability

of type-1 projects is not larger in the ceiling period than in ceiling-free period. Similarly, H2

in Table 3.5 claims that the approval probability of type-2 projects is not smaller in the ceiling

period than in the ceiling-free period.

Table 3.5: Hypotheses to be Tested

Hypothesis Theoretical Empirical

prediction test

H1: The approval probability of type-1 projects is not larger n∗1 ≥ n∗∗1 δ ≤ 0

in the ceiling period than in the ceiling-free period.

H2: The approval probability of type-2 projects is not smaller n∗2 ≤ n∗∗2 δ + γ ≥ 0

in the ceiling period than in the ceiling-free period.

To implement the tests in Table 3.5, we estimate a di�-in-di� probit model explaining loan

approval as a function of the project type, the period, and control variables. Along with the

36Each prediction compares the success of type-i (i = 1, 2) applicants across the two periods. In themodel, all type-i projects have the same size, which makes the results easier to outline.

125


results in Section 3, we use the EUR 25,000 threshold to delimit type-1 (small) and type-2 (large)

projects. In the regressions, project type is captured by the dummy variable Large Project,

which takes value 1 for projects above EUR 25,000, and 0 otherwise. The presence of the loan

ceiling is represented by the dummy variable, Ceiling, which takes value 1 in April 2009 and

after, and 0 in March 2009 and before.

The estimated model is written:

E[Approval|Large Project, Ceiling,X] =

Φ(βLarge Project+ δCeiling + γCeiling ∗ Large Project+ θX) (3.8)

whereX is a vector of control variables including the constant term; β, δ, γ, and vector θ represent

the parameters to be estimated; and Φ(·) is the normal cumulative distribution function.

Coe�cients β and δ capture the e�ects of project size and period, respectively (Puhani 2012).

Our �rst hypothesis to be tested,H1, is about the impact of the ceiling on the approval probability

of type-1 projects. From Eq. (3.8), we have:

∆E[Approval|Large Project = 0, Ceiling,X]

∆Ceiling= Φ(δ + θX)− Φ(θX) (3.9)

As Φ(·) is strictly monotonic, δ and [Φ(δ + θX)− Φ(θX)] have the same sign. Hence, H1 can

be reformulated as: δ ≤ 0.

Similarly, we have:

∆E[Approval|Large Project = 1, Ceiling,X]

∆Ceiling= Φ(δ + γ + θX)− Φ(θX) (3.10)

The sum δ + γ captures the e�ect of the loan ceiling on type-2 projects, so that H2 can be

rephrased as δ + γ ≥ 0. The parameter inequalities associated with H1 and H2 are provided in

the last column of Table 3.5.

126


The theoretical predictions we seek to test compare the approval probabilities of �xed-type

projects before and after the ceiling is enforced. In addition, di�-in-di� estimation provides

fruitful comparisons between the treatments of same-period type-1 and type-2 projects. Indeed,

Eq. (3.8) implies that:

∆E[Approval|Large Project, Ceiling = 0, X]

∆Large Project= Φ(β + θX)− Φ(θX) (3.11)

and:

∆E[Approval|Large Project, Ceiling = 1, X]

∆Large Project= Φ(β + δ + γ + θX)− Φ(δ + θX) (3.12)

For instance, a negative β (resp. a positive β + γ) indicates that the �rst-period (resp. second-

period) loan approval of type-2 projects is tougher (resp. looser) than for type-1 ones.

Table 3.6 presents the estimation results. In Panel A, columns (1) to (4) report the estimates

obtained for four speci�cations of Eq. (3.8) corresponding to the progressive inclusion of control

variables. Panel B summarizes the results regarding our coe�cients of interest. The �rst two

lines in Panel B are designed to test hypotheses H1 and H2. The next two lines compare approval

probabilities of the two types of projects granted in the same-period.

127


Table 3.6: Probability of Approval

Panel A: Coe�cient estimates (standard errors in parentheses)

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

Ceiling (δ) 0.05 (0.13) 0.04 (0.13) 0.04 (0.13) -0.01 (0.14)

Large Project (β) -0.40** (0.17) -0.45***(0.17) -0.52***(0.18) -0.56***(0.19)

Ceiling*Large Project (γ) 0.89***(0.19) 0.89*** (0.19) 0.97*** (0.20) 0.93*** (0.21)

Having personal investment 0.26** (0.11) 0.29** (0.12) 0.27** (0.12)

Having funds from other sources -0.02 (0.09) -0.02 (0.09) 0.03 (0.09)

Start-up -0.38***(0.12) -0.27** (0.13)

Services -0.21 (0.14) -0.19 (0.14)

Trade -0.38***(0.14) -0.34** (0.14)

Food and accommodation -0.61***(0.16) -0.48***(0.17)

Construction -0.12 (0.17) -0.05 (0.17)

Arts and entertainment -0.43** (0.21) -0.32 (0.23)

Unemployed for more than 6 months -0.23***(0.09)

Female 0.01 (0.09)

Single -0.08 (0.09)

Education (nb. of achieved diplomas) 0.06** (0.03)

Household income 0.09** (0.04)

Constant 0.04 (0.11) -0.14 (0.14) 0.48** (0.20) 0.30 (0.23)

Nb. of observations 1,097 1,097 1,056 1,016

Panel B: Di�-in-di� estimates (p-values in parentheses)

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

δ 0.05 (0.68) 0.04 (0.75) 0.04 (0.75) -0.01 (0.97)

δ + γ 0.94***(0.00) 0.93*** (0.00) 1.01*** (0.00) 0.92*** (0.00)

β -0.40** (0.02) -0.45***(0.01) -0.52***(0.00) -0.56***(0.00)

β + γ 0.48***(0.00) 0.44*** (0.00) 0.46*** (0.00) 0.37*** (0.00)

This table reports the results of estimating a probit model in which the dependent variable is being granted a loan by the

MFI. Panel A reports coe�cient estimates and, in parentheses, standard errors. Panel B reports di�-in-di� estimates, and

in parentheses, p-values based on Wald tests. Large Project is an indicator for projects larger than 25,000. Ceiling is the

indicator for the period after the introduction of the loan ceiling (April 2009). Signi�cance: *** p<0.01, ** p<0.05, *

p<0.1.

The �rst speci�cation (column (1)) excludes any control variable. The second speci�cation (col-

umn (2)) controls for the sources of funding (personal investment and funds from other sources).

128


To avoid multicollinearity, we exclude variables related to bank loans.37 The third speci�cation

(column (3)) also includes sector characteristics. The last one (column (4)) adds borrowers'

individual characteristics.

Overall, the results are remarkably consistent across speci�cations.38 The coe�cient of the

Large Project dummy is signi�cantly negative. The coe�cient of the Ceiling dummy is in-

signi�cant, but the loading of the interaction term Ceiling ∗ Large Project, is signi�cantly

positive.

Panel B provides evidence on the relevance of case II in our model. Since δ is not signi�cantly

di�erent from zero, one cannot reject the hypothesis that type-1 projects have similar approval

rates in both periods. However, (δ+ γ) being signi�cantly positive suggests that type-2 projects

are treated more favorably by the ceiling-constrained MFI than by the ceiling-free one. In the-

oretical terms, our results support the following statements: n∗1 = n∗∗1 and n∗2 < n∗∗2 . These

expressions are special cases of the two inequalities characterizing case II in our model, namely:

n∗1 ≥ n∗∗1 and n∗2 ≤ n∗∗2 .

The last two lines in Table 3.6 deliver two additional insights. First, β is signi�cantly negative,

which indicates that the ceiling-free MFI favors small projects to large ones, all else equal. This

result con�rms that the MFI ful�lls its social mission faithfully, at least in the ceiling-free period.

Second, (β + γ) is signi�cantly positive. Thus, the bias toward small projects is reverted after

the enforcement of the loan ceiling. In the second period, the MFI grants loans to large projects

more easily than to their small counterparts. Although we fail to observe any change in the

approval of small projects, these �gures con�rm that large projects experience a major upward

shift in their approval rate. Regardless of the underlying mechanism, our results con�rm that

introducing a loan ceiling makes the MFI prefer to �nance large projects. This is the typical

37The dummy variable "Having a bank loan" and the size of the bank loan are both highly correlatedwith project size. These statistically inconvenient features corroborate our argument on the importanceof co-�nancing.

38Adding controls marginally increase both the signi�cance of the Large Project dummy and its (ab-solute) value. However, we interpret signs only, since the non-linearity of the model makes amplitudesirrelevant.

129


stigma of mission drift in micro�nance (Armendariz and Szafarz 2011).

The evidence of mission drift is however mitigated by the fact that, in absolute terms, people

with small projects do not endure harsher loan approval after the enforcement of the ceiling than

before it (n∗1 = n∗∗1 ). The harm exists only relative to applicants with large projects, who enjoy

signi�cantly better approval conditions once the loan ceiling comes into force. Thus, relatively

speaking, small projects are disadvantaged in the second period. With loan approval becoming

easier for large projects, but not for small ones, the share of small projects �nanced by the MFI

mechanically decreases. Put di�erently, the introduction of the ceiling has bene�ted to the hold-

ers of large projects only. This can be interpreted as a mild form of mission drift.

Contrasting with standard applications of di�-in-di� estimation, our econometric design is meant

to explore the impact of the regime change on an institutional variable, namely the MFI's ap-

proval probability, and not on the treatment of individuals, here the MFI's borrowers. This

design makes our estimates immune to the violation of the so-called "stable unit treatment value

assumption" (SUTVA), which states that the treatment of an individual has no spillover e�ect

on that of other individuals (Wooldridge 2002, pp. 629).39

In line with the prediction of case II in our theoretical model, the di�-in-di� regressions demon-

strate that the French loan ceiling pushed CREASOL to deviate from its social mission, which

consists in serving poor entrepreneurs disregarded by mainstream banks. This outcome sharply

contrasts with the regulators' expected scenario featured in case I of the model. The results,

however, are contingent on the level of the French ceiling, which is particularly low by developed-

country standards.

Generally speaking, determining a loan ceiling that prevents mission drift is di�cult, if not im-

possible, for several reasons. While high ceilings are useless (case III), the di�erence between

39Unarguably, our study is incompatible with SUTVA. Imposing a loan ceiling a�ects the whole poolof applicants, as our descriptive statistics con�rm. For type-2 applicants, the impact is direct since theyneed to �nd new sources of funds. Changes in the pool of type-1 applicants may result from downscaling.Self-selection can push otherwise ambitious applicants to spontaneously down-scale their projects in orderto become admissible by the MFI. Alternatively, project downscaling can follow on from loan denial bybanks.

130

3.5. Robustness Checks

case I (desired case) and case II (mission drift) is subtle. Case II can emerge when two condi-

tions are met. First, the ceiling needs to be low enough to be binding: the ceiling-constrained

MFI is unable to serve applicants who would be welcome otherwise. The lower the ceiling, the

higher is the probability for this situation to occur. The second condition relates to the presence

of mainstream banks willing to co-�nance projects with the MFI. If no such bank exists or if

the existing ones ultimately reject all the applicants, mission drift is impossible. In contrast,

when banks are willing to co-�nance projects, the MFI has the opportunity to free-ride on their

screening process. Free-riding makes larger projects less costly to handle and monitor. This

e�ect is mitigated by the severity of the bank screening process. Applications denied by banks

cannot reach the MFI anymore. As a result, the MFI can be rationed in large projects and keep

serving holders of below-ceiling projects, though in a reduced proportion. Our empirical exercise

shows that the combination of the two conditions for the emergence of mission drift is realistic.

Loan ceilings make perfect sense to counteract on mission drift in developing countries where

the credit market is highly segmented. In developed countries, things are di�erent. Regulators

of the micro�nance industry should take the risk of loan-ceiling-driven mission drift seriously.

3.5 Robustness Checks

In this section, we check the robustness of our regression results along two dimensions. First,

a key value set in the analysis is the threshold used to separate large projects from small ones.

Although the EUR 25,000 threshold was carefully justi�ed in Section 3, this value remains

somewhat arbitrary from the theoretical standpoint. We therefore test whether our empirical

results resist changes in this threshold. Section 5.1 runs the di�-in-di� regressions with six

di�erent values for the threshold. The second check concerns the length of the observation

period that followed enforcement of the loan ceiling. Possibly, the changes in loan approval

detected in the baseline regressions are, at least partly, due to events that occurred in this period

and had nothing to do with micro�nance regulation. To test this possibility, we reduce the time

131


span after the introduction of the ceiling, and rerun the regressions.

3.5.1 Speci�cation of Project Size

Table 3.7 presents the di�-in-di� estimates for speci�cation (4) in Table 3.6 (Panel B) using

six di�erent size thresholds, including the one used in the baseline regression, which serves as

a reference. Speci�cally, we consider the following cut-o�s to de�ne the Large Project dummy

variable: EUR 10,000, EUR 15,000, EUR 20,000, EUR 25,000 (reference), EUR 30,000, and EUR

40,000.

The �rst two lines of Table 3.7 suggest that the theoretical predictions of case II are con�rmed

for all the size thresholds used here. In the last two lines, the coe�cients lose signi�cance for the

thresholds of EUR 10,000 and EUR 15,000. This suggests that applicants with projects below

EUR 15,000 are not credit constrained with below-ceiling loans from the MFI.40 These results

not only con�rm the robustness of our previous results; they also con�rm the �ndings in Section

3 that a size threshold smaller than or equal to EUR 15,000 makes little sense.

Our database covers the May 2008-June 2012 period, and the loan ceiling was enforced in April

Table 3.7: Probability of Approval: Di�erent Speci�cations for Project Size

Threshold for Large Project dummy 10,000 15,000 20,000 25,000 30,000 40,000

δ 0.09 0.10 -0.07 -0.01 0.10 0.19(0.73) (0.60) (0.65) (0.97) (0.45) (0.11)

δ + γ 0.44*** 0.52*** 0.80*** 0.92*** 0.95*** 1.02***(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)

β -0.27 -0.30 -0.56*** -0.56*** -0.53*** -0.51**(0.33) (0.16) (0.00) (0.00) (0.01) (0.02)

β + γ 0.08 0.12 0.30*** 0.37*** 0.33*** 0.31**(0.49) (0.24) (0.00) (0.00) (0.01) (0.02)

Nb. of observations 1,016 1,016 1,016 1,016 1,016 1,016

This table reports di�-in-di� estimates for the probit model estimated in Table 3.6, speci�cation (4), with several cut-o�s

for Large Project dummy. In parentheses we present the p-values based on Wald tests. Ceiling is the indicator for the

period after the introduction of the loan ceiling (April 2009). Signi�cance: *** p<0.01, ** p<0.05, * p<0.1.

2009. The baseline regressions exploit the full database in order to gain on precision. As a result

40In our sample, only 3% of the applicants with projects below EUR 15,000 have a bank loan.

132


we are dealing with a one-year �rst period and an over-three-year second period. However, using

a relatively long second period increases the probability that the explained variable is in�uenced

by events not linked to the regulatory change under scrutiny.

To check whether this issue a�ects our empirical �ndings, we run the regressions with a reduced

time-span stopping in June 2010. The reduced sample is made up of 226 observations for the

�rst period and 268 observations for the second. Table 3.8 provides the descriptive statistics.

It suggests that enforcement of the loan ceiling resulted in the short run in a sharp drop in

the size of �nanced projects, from above EUR 30,000 to below EUR 20,000. The project size

stabilized later, apparently after more than one year, once the applicants started realizing that

co-�nancing was a feasible option. Likewise, the proportion of start-ups increased signi�cantly

in the short run. This is not surprising, as development projects typically require more funding

than start-ups. Interestingly, the proportion of long-term unemployed individuals increased in

the short run.

Table 3.9 features the estimation results for the reduced period. Apart from some lower signi�-

cance levels attributable to the smaller sample size, the �gures appear to be remarkably close to

those of the baseline regressions.

Overall, the robustness checks suggest that the mission-drift outcome resists changes in both

the size threshold and the period delimitation. In this way, the checks reinforce the empirical

validity of case II in our theoretical model.

133


Table 3.8: Descriptive Statistics: Reduced Observation Perioda

Applicants Borrowers

Without With Without With

Ceiling Ceiling t-test Ceiling Ceiling t-test

Financial Characteristics

Project size (kEUR) 30,22 19,62 -10,60*** 26,55 21,26 -5,29**

Demanded loan size (kEUR) 18,38 6,53 -11,85*** 16,62 6,11 -10,51***

Demanded loan size >=10000 (%) 0,70 0,21 -0,49*** 0,65 0,19 -0,46***

Granted loan size 15,74 5,96 -9,78***

Having a bank loan (%) 0,03 0,11 0,08*** - 0,11 -

Bank loan (kEUR)b 46,53 27,08 -19,45* - 30,07 -

Having personal investment (%) 0,81 0,79 -0,01 0,83 0,88 0,05

Personal investment (kEUR)b 6,85 4,80 -2,06** 6,00 5,37 -0,63

Having funds from other sources (%) 0,55 0,71 0,16*** 0,51 0,78 0,27***

Funds from other sources (kEUR)b 9,02 8,83 -0,19 9,19 8,99 -0,20

Business Characteristics

Start-up (%) 0,80 0,91 0,11*** 0,79 0,92 0,13***

Services (%) 0,28 0,31 0,03 0,28 0,35 0,07

Trade (%) 0,22 0,25 0,03 0,24 0,25 0,01

Accommodation and food service activities (%) 0,16 0,12 -0,04 0,13 0,06 -0,07*

Construction (%) 0,09 0,14 0,05 0,11 0,13 0,02

Arts, entertainment and recreation (%) 0,07 0,03 -0,04** 0,05 0,02 -0,03

Other sectors 0,18 0,15 -0,03 0,19 0,18 -0,01

Individual Characteristics

Unemployed for more than 6 months (%) 0,55 0,66 0,11** 0,49 0,65 0,16**

Female applicant (%) 0,38 0,34 -0,04 0,34 0,34 0,00

Single (%) 0,59 0,56 -0,03 0,63 0,52 -0,11*

Education 2,74 2,48 -0,26* 2,84 2,69 -0,15

Average monthly household income (kEUR) 1,10 1,26 0,16* 1,20 1,48 0,28*


aThe table gives mean values and t-test for equal means between the two sub-periods

bThe mean value is computed on the non-zero data points only.

Signi�cance: *** p<0.01, ** p<0.05, * p<0.1

134


Table 3.9: Probability of Approval: Reduced Period

Panel A: Coe�cient estimates (standard errors in parentheses)

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

Ceiling (δ) -0.17 (0.14) -0.19 (0.15) -0.16 (0.16) -0.17 (0.16)

Large Project (β) -0.40** (0.17) -0.50***(0.17) -0.48** (0.19) -0.52***(0.20)

Ceiling*Large Project (γ) 0.87***(0.25) 0.88*** (0.26) 1.02*** (0.27) 1.03*** (0.29)

Having personal investment 0.45*** (0.16) 0.37** (0.18) 0.37** (0.19)

Having funds from other sources 0.02 (0.13) 0.00 (0.14) 0.02 (0.14)

Start-up -0.29 (0.19) -0.17 (0.21)

Services -0.11 (0.19) -0.07 (0.19)

Trade -0.20 (0.19) -0.06 (0.20)

Food and accommodation -0.77***(0.23) -0.69***(0.24)

Construction -0.13 (0.23) -0.10 (0.24)

Arts and entertainment -0.45 (0.32) -0.42 (0.34)

Unemployed for more than 6 months -0.31** (0.14)

Female -0.03 (0.14)

Single 0.06 (0.14)

Education (nb. of achieved diplomas) 0.07* (0.04)

Household income 0.11* (0.06)

Constant 0.04 (0.11) -0.29* (0.16) 0.27 (0.26) -0.03 (0.33)


Panel B: Di�-in-di� estimates (p-values in parentheses)

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

δ -0.17 (0.25) -0.19 (0.19) -0.16 (0.32) -0.17 (0.30)

δ + γ 0.70***(0.00) 0.69*** (0.00) 0.86*** (0.00) 0.86*** (0.00)

β -0.40** (0.02) -0.50***(0.00) -0.48** (0.01) -0.52***(0.01)

β + γ 0.46** (0.02) 0.38* (0.05) 0.54** (0.01) 0.51** (0.03)

This table reports the results of estimating a probit model in which the dependent variable is being granted a microcredit.

The time span covers the period corresponding to one year before and one year after the enforcement of the loan ceiling.

Panel A reports coe�cient estimates and, in parentheses, standard errors. Panel B reports di�-in-di� estimates, and in

parentheses, the signi�cance (p-value) of these estimates based on Wald tests. Large Project is an indicator for projects

larger than 25,000. Ceiling is the indicator for the period after the introduction of the ceiling (April 2009). Signi�cance:

*** p<0.01, ** p<0.05, * p<0.1.

135


3.6 Conclusion

This chapter addresses the impact of loan ceilings on the microcredit market, both theoretically

and empirically. Our theoretical model applies to an MFI operating in a competitive credit mar-

ket, which is the case in most developed economies. The MFI is subsidized and o�ers loans at

below-market conditions. The presence of mainstream banks willing to co-�nance projects with

the MFI is a distinctive feature of our model, which also makes it speci�c to developed economies.

Relying on these assumptions, we show that imposing a low loan ceiling can trigger mission drift.

The MFI can be tempted to opt for a cost-reducing strategy including co-�nancing above-ceiling

projects with banks. This strategy is at the expense of holders of small projects, who need

below-ceiling loans. As a consequence, the ceiling-constrained MFI might end up granting larger

loans and attracting wealthier clients, a phenomenon called "mission drift" in the micro�nance

literature.

The second part of this chapter tests the prediction of our model by exploiting a natural exper-

iment. We bene�ted from detailed information on the applicants of a French MFI before and

after the enforcement of the French EUR 10,000 loan ceiling. The descriptive statistics shows

that in the ceiling-constrained MFI initiated co-�nancing large projects with banks. At the same

time, our di�-in-di� probit regressions con�rm that loan approval became signi�cantly easier for

holders of large projects. Mission drift is thus a real threat associated with the enforcement of a

loan ceiling. Therefore, regulators should pay attention to this possible outcome when imposing

loan ceilings to the micro�nance industry.

Binding loan ceilings encourage bank-MFI co-�nancing schemes and di�use the bene�ts of sub-

sidization across a pool of borrowers that goes beyond the typical target pool of MFIs. Actually,

co-�nancing schemes have both advantages and drawbacks. The very existence of subsidized

MFIs represents an opportunity for mainstream banks to �nd partners to share risks with. This

is especially relevant when it comes to �nance start-ups that would otherwise be denied from

access to the credit market. However, co-�nancing makes MFIs dependent on the screening

136

3.6. Conclusion

processes of mainstream banks. Eventually, banks can signi�cantly modify the pool of project

holders who end up applying for microcredit. In addition, this can perversely a�ect the credit

allocation of MFIs. In line with our theoretical model, our empirical analysis provides evidence

of this perverse e�ect. Incidentally, it con�rms that average loan size is a poor indicator of

mission ful�llment (Dunford 2002; Armendariz and Szafarz 2011). The size of a single leg of a

two-leg funding arrangement makes little sense. In our empirical study, the average loan size

of the MFI mechanically decreased after the enforcement of the loan ceiling, although the MFI

started (co-)�nancing larger projects.

The prevalence of bank-MFI co-�nancing schemes might also harm the disadvantaged segments of

the population who are typically targeted by MFIs. Notably, these segments include unemployed

persons, women, and migrants who seek �nancial empowerment through self-employment. When

�nding a paid job is di�cult for reasons pertaining to lack of diploma and/or discrimination in

the job market, self-employment remains one of the few possibilities left for escaping poverty.

Further research is needed to investigate whether�regulated or not�MFIs in developed countries

are able to e�ciently address the key issue of poverty alleviation.

Our theoretical model su�ers from several limitations. First, it assumes that the MFI maximizes

outreach, i.e. its number of borrowers. While this assumption is frequently used, the literature

has not yet met a consensus on the way to formalize the objective function of MFIs. In fact, there

are reasons to believe that the objective of MFIs is complex and institution-speci�c (Molenaar

2009; Hudon and Sandberg 2013). Second, we use a one-period model and consider two project

sizes only. These simpli�cations help deriving a three-case comparative analysis contrasting the

situations of the MFI with and without a loan ceiling. More sophisticated speci�cations could

deliver a more nuanced picture. In particular, the empirical analysis has shown that the personal

investment of the applicants matters.

Our database is remarkably detailed but still limited to a single institution, CREASOL. In ad-

dition, the change of status of this institution was not randomly assigned. We cannot rule out

that the decision of the managers to opt for the MFI status was, at least partly, driven by their

137


intention to serve holders of larger projects and become more pro�table. This however would

contradict the public statements according to which the status change was motivated by gaining

access to funds at preferential rates while keeping the social mission unaltered. In any case,

the loan ceiling made it possible to deviate�intentionally or not�from serving holders of small

projects.

This chapter emphasizes that regulations imposing loan ceilings on microcredit activities can

have unexpected and perverse consequences. From that perspective, working with a single in-

stitution is su�cient to make our point. Admittedly, the French ceiling is very low. It is even

the lowest loan ceiling found in developed countries. Addressing the reasons for this French

peculiarity goes beyond the scope of this chapter. But, whatever the reason, we hope that our

conclusions will raise concern among regulators of microcredit in developed countries.

Our main message to regulators is the following. Due to the pervasive diversity in the micro�-

nance industry, it is very di�cult, if not impossible, to �nd an optimal loan ceiling that would

be low enough to make a di�erence, but at the same time high enough to avoid mission drift. In

view of this problem, other regulation designs could be envisaged. Our case study emphasizes

that project size matters more than loan size when it comes to de�ning social lending. Therefore,

a regulatory route could be to impose ceilings to project size rather than loan size. Such a rule

could, however, be easily circumvented by arti�cially splitting large projects into smaller ones.

Alternatively, regulators could try delimiting the target pool of borrowers of subsidized institu-

tions. For instance, women and discriminated-against minorities could be targeted more speci�-

cally. In this way, micro�nance in developed countries would meet its original principle of serving

poor and disadvantaged populations. This is of utmost importance since the micro�nance sector

in developed countries is still very young. Regulations have a key role to play in shaping its

future.

138

Chapter 4

Loan Ceilings and Women's Access to Credit

in France 1

4.1 Introduction

Female entrepreneurship is advocated as a driving force in economic development. Even so,

access to credit is still a challenging barrier to women entrepreneurs. Two types of gender bias

in lending are documented in the literature. The �rst stems from harsher credit approval (Orser

et al. 2000; Cavalluzzo et al. 2002; Fay and Williams 1993). The second relates to credit con-

ditions, including collateral requirements and loan size.2 In France, women account for 47% of

the workforce but only 30% of entrepreneurs (Brana 2013). Hence, there are grounds to suspect

that women �nd it harder than men to set up a business.

We contribute to this stream of the literature by scrutinizing the loan allocation of a French

micro�nance institution (MFI). We have data on the applicants and on the borrowers from a

French MFI between April 2008 and June 2012. The MFI was an unregulated NGO before April

2009. In April 2009 the MFI becomes regulated and complies with EUR 10,000 loan ceiling. In

1This chapter is based on a joint work with Ariane Szafarz.2Riding and Swift (1990, Coleman (2000, Bellucci et al. (2010) �nd that collateral requirements are

gender-related in Canada, the UK, and Italy, respectively. Alesina et al. (2013) and Agier and Szafarz(2013a) show that female micro-entrepreneurs receive smaller loans than male ones, in Italy and Brazil,respectively.

139

Loan Ceilings and Women's Access to Credit in France

this paper we study how the allocation of microcredits has been impacted by the introduction

of such a ceiling.

In developed countries, gender discrimination has been detected in various economic activities.

The evidence is overwhelming that women are penalized on the job market (Altonji and Blank

1999; Blau and Kahn 2000), con�rming that competition is insu�cient to deter discriminatory

practices. Evidence is also found in other markets, such as those for cars (Ayres and Siegel-

man 1995) and housing (Page 1995). Gender discrimination can thus potentially interact with

economic decision-making in any area. Unfortunately, data are often insu�cient to assess this

situation. Because of US legal requirements,3 race and gender discrimination has been scruti-

nized in mortgage lending (Munnell et al. 1996; Han 2004) and in the small business credit

industry (Blanch�ower et al. 2003; Cavalluzzo and Wolken 2005). In mortgage lending, black

applicants face the worst denial rate (Schafer and Ladd 1981) while female applicants experience

disparate treatment (Ladd 1998). Stereotypes thus seem to have survived the enforcement of the

US Equal Credit Opportunity Act.

In Europe, discrimination in lending is di�cult to test directly because banks are not required

to release individual data. To get around this issue, we used an indirect identi�cation strategy,

which consists in observing the impact of mainstream banks' loan approval process on applicants

to an MFI. This is made possible by the French regulatory context. In France, licensed MFIs,

i.e. those allowed to �nance their activity through borrowing, are subject to a strict EUR 10,000

loan ceiling.4 However, a signi�cant percentage of entrepreneurs targeted by French MFIs5 have

business projects that require above-ceiling loans. To apply for microcredit, these entrepreneurs

have to secure co-�nancing from a mainstream bank beforehand. Accordingly, the gender and

3 The US lending industry is subject to anti-discrimination regulations including: the Fair HousingAct of 1968, the Equal Credit Opportunity Act (ECOA) of 1974, and the Home Mortgage DisclosureAct (HMDA) of 1975, which was amended in 1989 to make it mandatory for lenders to report race andethnicity of their loan applicants.

4This ceiling is signi�cantly lower than the EUR 25,000 threshold recommended by the EuropeanCommission.

5Most of them are unemployed people aiming at self-employment.

140

4.1. Introduction

project characteristics of microcredit applicants with above-ceiling projects partly reveal how

banks treat female applicants.

In France, as elsewhere, commercial banks are the main providers of small-business �nance

(Berger and Udell 1998; Cornée and Szafarz 2013), while MFIs are new players in the �eld.

By imposing a low loan ceiling on MFIs, the French regulator sought to preserve the banks'

prerogative to provide small businesses with loans above EUR 10,000 (Brabant et al. 2009).

In practice, however, the French regulation has led to projects being co-�nanced by banks and

MFIs. This outcome can be viewed as a somewhat unexpected byproduct6 of the particularly

low loan ceiling enforced by the French government (see previous chapter).7 It can be rational-

ized by the fact that co-�nancing is pro�table to credit providers through information sharing

(Bennardo et al. 2014) and natural complementarities (Fall 2010). In India, the ICICI Bank

has entered into partnership arrangements with 30 MFIs (Ananth 2005). According to these

arrangements, loan contracts are directly signed by the bank and the borrowers, and the MFI

acts as guarantor against defaults. This arrangement reduces the MFI's cost of capital while

preserving its incentives to monitor borrowers.

Under French regulations, business co-�nancing can be attractive to all parties involved, i.e. the

bank, the borrower, and the MFI, for several reasons. First, co-�nancing gives banks access to

new market segments while limiting their risk. Banks are typically reluctant to �nance credit-

history-free start-ups. Second, for borrowers lacking credit history co-�nancing may be the only

way to launch relatively large business projects at reasonable cost.8 Third, co-�nancing allows

ceiling-constrained MFIs to attract entrepreneurs with above-ceiling projects. To some extent it

6MFIs and banks have di�erent statuses. MFIs are subsidized institutions maximizing social perfor-mance within a budget constraint, while banks are driven by pro�t maximization (Aubert et al. 2009).However, Armendariz and Szafarz (2011) provide evidence that the social mission varies across MFIs.

7In the United States, the loan ceiling for microcredit is USD 50,000. The European Union (EU)recommends the use of a EUR 25,000 ceiling, but member states remain free to set their own rules. Somecountries (Romania, Italy) have adopted the EU recommendation, while others, like Hungary, Portugal,Slovakia, and the UK, allow MFIs to grant loans exceeding EUR 25,000. France is the only EU memberto impose a ceiling lower than the EU recommendation.

8In our data set, we observe that in three years out of four (2009, 2010, and 2011, but not 2012) theinterest rates charged by the banks are signi�cantly higher than that charged by the MFI.

141


also o�ers MFIs an opportunity to free-ride on banks' screening processes.9

At the same time, the feasibility of co-�nancing means that MFIs' pools of applicants become at

least partly shaped by banks. We use this feature as an identi�cation strategy for scrutinizing the

banks' attitude toward female loan applicants. More precisely, we exploit a natural experiment,

since we observe the full loan-granting process of an MFI before and after the loan ceiling is

introduced, i.e. before and after the emergence of co-�nancing with banks. Changes in the pool

of the applicants give us insights into the banks' loan granting process, while the pro�les of the

co-�nancing recipients tells us how the MFI reacts to this process.

Co-�nancing is still understudied. While the literature cites evidence of co-�nancing schemes link-

ing formal and informal institutions in developing countries (Jain 1999; Andersen and Malchow-

Møller 2006; Degryse et al. 2013), co-�nancing between banks and MFIs in developed countries

has not been reported so far. This might indicate that the French situation is fairly exceptional.

Alternatively, one could argue that the micro�nance industry in developed countries is still in its

infancy and has not yet fully exploited market opportunities. In 2010, developed countries ac-

counted for only 2.6% of the micro�nance clients reported in the Microcredit Summit Campaign

Report 2012.10

Based on partial-least-square estimations, our results con�rm that the introduction of the loan

ceiling has dramatically changed loan allocation in the MFI we have studied. Speci�cally, the

institution moved from a favorable allocation to women to a gender-neutral one. Before the

ceiling was introduced, the MFI granted loans without bank co-�nancing, and we �nd that the

MFI selected women with larger requested amounts, corresponding to more ambitious projects.

This �nding suggests that the MFI attempted to correct for gender discrepancies existing in

the demand side, as women tend to request signi�cantly smaller loans. After the ceiling was

brought in, the institution started co-�nancing above-ceiling projects with mainstream banks.

9In our data set, 71% of the applicants with a secured bank loan ended up with a co-�nancingarrangement.

10http://www.microcreditsummit.org/uploads/resource/document/web_socr-2012_english_62819.pdf

142

4.2. Data and Descriptive Statistics

Under ceiling enforcement the MFI no longer selected women with larger requested amounts.

Our �ndings suggest that co-�nancing has led an originally positively oriented MFI to give up

some control of its loan allocation decision to mainstream banks.

The remainder of this paper is structured as follows. Section 2 provides an overview of the data.

The econometric model in Section 3 pinpoints the relationship between gender, requested amount

and loan size. In section 4 we discuss the results. In section 5 we provide a robustness check

using a Heckman selection model to scrutinize MFI's approval process. Section 6 concludes.

4.2 Data and Descriptive Statistics

Individual data were hand-collected on the applicants and borrowers of a French MFI set up

in 2006. The database covers the period from 2008 to 2012 and includes detailed information

on 1,098 credit applicants. The MFI's pool of applicants is made up of unemployed people

seeking self-employment, and start-ups lacking collateral and credit history. Until April 2009,

the institution operated under the unregulated NGO status. As such, it was required to �nance

its activity through subsidies only, which restricted its growth. From then on, the NGO changed

its status to a regulated MFI in order to gain access to funds at preferential rates. Since then,

the MFI has been subject to the EUR 10,000 loan ceiling. Although the change of status enabled

the MFI to grow signi�cantly,11 the institution has preserved its social purpose.

Since it was founded, the MFI has used the typical individual microcredit lending methodology,

charging all borrowers the same interest rate. Over the sample period, the interest rate changed

slightly due to market conditions, but remained between 4% and 5% p.a., which is remarkably

low given the risks involved in start-up �nancing. Loans are to be repaid in monthly installments.

The average loan duration is 51 months. Loan applications are examined by a loan o�cer, while

the credit committee has the �nal say on loan approval. Typically, the decision is binary: the

11In 2010, the MFI opened two new branches and its sta� passed from six to ten employees.

143


credit committee either approves12 or denies a loan of the size requested by the applicant.13

Loan size is rarely questioned by the loan o�cers because the MFI delegates this task to third

parties. Speci�cally, NGOs are in charge of helping applicants de�ne their �nancial needs and

o�ering them business development services. The MFI has however no commitment toward these

independent-and typically subsidized-NGOs.

The full sample period is split in two. The �rst period (April 2008-April 2009) corresponds to

the status of an unregulated NGO. The second (May 2009-June 2012) is longer and begins with

the enforcement of the loan ceiling. During the �rst period, the MFI received 227 applications

and granted 100 loans. During the second, it received 871 applications and granted 519 loans.

In practice, the main di�erence between the two periods is the emergence of co-�nancing. Figs.

4.1 and 4.2 illustrate the change that took place in May 2009. In the �rst period (Fig. 4.1),

the MFI �nanced projects up to EUR 40,00014 without bank intervention. In the second period

(Fig. 4.2), the EUR 10,000 loan ceiling was introduced. The holders of below- and above-ceiling

projects follow distinct paths. For those below the ceiling, there is no change. They retain the

right to apply to the MFI directly. In contrast, holders of above-ceiling projects are required

to secure a partial bank loan before applying to the MFI. Their best interests dictate that they

apply to a bank for the portion of their desired loan exceeding EUR 10,000. Doing so has

two advantages. First, it maximizes their chances of obtaining a loan from the bank. Second, it

minimizes the �nancial burden of their debt since the MFI charges an interest rate that is typically

lower than that of mainstream banks. Projects denied by banks may be either abandoned or

downsized to an amount that does not require bank �nancing. However, downsizing can strongly

compromise the investment project. Therefore, we conjecture that projects requiring loans well

above EUR 10,000 that are denied by banks are mostly abandoned. In any case, we do not

12If a co-�nancing loan with a mainstream bank is under negotiation at the moment of microcreditapproval, the MFI can make a conditional approval which comes into force after bank's validation.

13In only 7.6% of our sample is the granted loan size smaller than the demanded one. This way ofdoing departs from the lending methodology adopted by many MFIs, which use loan size as a decisionvariable (Agier and Szafarz 2013b).

14This threshold was hardly binding.

144


observe the outcomes of the bank application process; we observe only the occurrence of bank

loans among MFI applicants as well as the size and interest rates of these loans.

Figure 4.1: Loan Allocation Process in the First Period

Figure 4.2: Loan Allocation Process in the Second Period

Table 4.1 displays descriptive statistics concerning the MFI's applicants, disaggregated by period

and gender, together with t-tests for equal means between men and women.15 To focus on

the demand side of the market, Table 4.1 reports statistics for applicants rather than actual

borrowers. However, the �gures for borrowers are displayed in Table 4.2. The proportion of

female applicants remains similar over the two periods: 38% in the �rst, 41% in the second.

For the �rst period, the t-tests in Table 4.1 do not detect any signi�cant di�erences in �nancial

characteristics between male and female applicants, suggesting that in the �rst period men and

women face the same �nancial constraints. However, in the second period, women apply to

15The sample size is smaller for the �rst period, which may result in larger standard deviations andless rejections of H0

145


Table 4.1: Descriptive Statistics on Applicants

First period: No ceiling Second period: Ceiling

Male Female t-test Male Female t-test

Loan approval rate 0.47 0.39 0.08 0.6 0.59 0Financial Characteristics

Requested amount (EURk) 18.51 18.1 0.41 7.14 6.8 0.34*Granted loan size (EURk) 15.08 17.01 -1.93 7.12 6.55 0.57**Project size (EURk) 30.64 29.17 1.47 32.4 27.59 4.81*Has bank loan (%) 0.03 0.01 0.02 0.27 0.28 -0.01Bank loan (EURk)a 40 92.2 -52.2 45.94 33.05 12.89**Has personal investment (%) 0.84 0.77 0.07 0.83 0.83 0Personal investment (EURk)a 6.79 5.24 1.55 6.62 5.54 1.08

Business CharacteristicsStart-up (%) 0.84 0.74 0.10* 0.86 0.83 0.03Food and accommodation (%) 0.1 0.27 -0.17*** 0.13 0.13 0Trade (%) 0.19 0.28 -0.08 0.26 0.37 -0.10***Services (%) 0.3 0.24 0.06 0.25 0.36 -0.11***Construction (%) 0.15 0 0.15*** 0.18 0.01 0.16***Arts and entertainment (%) 0.06 0.06 0 0.04 0.04 0Other sectors (%) 0.19 0.15 0.04 0.14 0.09 0.05**

Individual CharacteristicsUnemployed for at least 6 months (%) 0.57 0.52 0.05 0.59 0.6 -0.01Single (%) 0.53 0.69 -0.17** 0.49 0.54 -0.05Age (in years) 40.12 36.26 3.86*** 39.04 39.03 0.01Dependent children 0.86 1.04 -0.18 0.8 1 -0.20***Education (# quali�cations) 2.58 2.95 -0.38* 2.57 3.07 -0.51***Household income (EURk) 1.11 1.1 0 1.4 1.57 -0.17**Observations 140 87 518 353Computed only for non-zero points.

146


Table 4.2: Descriptive Statistics on Borrowers

First period: No ceiling Second period: Ceiling

Male Female t-test Male Female t-test

Financial CharacteristicsRequested amount (EURk) 16.27 17.3 -1.03 7.25 6.73 0.52**Granted loan size (EURk) 15.08 17.01 -1.93 7.12 6.55 0.57**Project size (EURk) 26.58 26.47 0.11 36.77 32.1 4.67Has bank loan (%) 0 0 0 0.32 0.33 0Bank loan (EURk) 0 0 0 45.96 37.72 8.25Has personal investment (%) 0.86 0.76 0.1 0.87 0.85 0.02Personal investment (EURk)a 54.44 48.73 5.71 7.56 6.59 0.98

Business CharacteristicsStart-up (%) 0.82 0.74 0.08 0.82 0.8 0.02Food and accommodation (%) 0.07 0.24 -0.17** 0.12 0.1 0.02Trade (%) 0.18 0.35 -0.17* 0.23 0.37 -0.14***Services (%) 0.3 0.24 0.06 0.25 0.39 -0.14***Construction (%) 0.17 0 0.17** 0.19 0.01 0.18***Arts and entertainment (%) 0.08 0 0.08* 0.05 0.04 0.01Other sectors (%) 0.2 0.18 0.02 0.17 0.1 0.07**

Individual CharacteristicsUnemployed for at least 6 months (%) 0.5 0.47 0.03 0.55 0.56 -0.02Single (%) 0.59 0.71 -0.11 0.46 0.45 0Age (in years) 40.92 35.74 5.18** 38.35 38.4 -0.05Dependent children 0.8 1 -0.2 0.78 0.98 -0.20**Education (# quali�cations) 2.71 3.09 -0.38 2.66 3.29 -0.63***Household income (EURk) 1.14 1.33 -0.19 1.54 1.78 -0.24**Observations 66 34 309 210Computed only for non-zero points.

147


the MFI for smaller loans than men and actually receive even smaller ones. The emergence of

co-�nancing is visible in Table 4.1 and Table 4.2. In the �rst period, the few applicants holding

a bank loan (2.7% of the sample) were all rejected by the MFI. In the second period 27% of

applicants and 33% of borrowers had previously secured a bank loan. Unarguably, co-�nancing

went from being a liability for the MFI in the �rst period to being an asset in the second.

Interestingly, we detect no gender gap in the likelihood of obtaining a bank loan. In contrast,

there is a huge gender gap in the size of the bank-supplied loans. Women also tend to undertake

smaller projects than men, probably because the loans they manage to obtain from banks are

on average 28% lower than those extended to men (EUR 45,940 against EUR 33,050). The

average gap (EUR 12,890) represents 47% of the average project size of female applicants (EUR

27,590). Table 4.2 shows that the sizes of the loans granted by the MFI to women exhibit

similar features. This gender-speci�c credit rationing is in line with previous evidence by Agier

and Szafarz (2013a), who detect a "glass-ceiling e�ect" at a Brazilian MFI, meaning that loan

approval is not discriminatory, but women with ambitious projects tend to receive smaller loans

than men. Presumably, the fact that gender-related disparate treatment in microcredit a�ects

credit conditions rather than loan approval is linked to the micro�nance tradition of serving

female borrowers (Armendariz and Morduch 2010).

Table 4.1 exhibits large gender disparities in business activities. Strikingly, the share of female

projects in the food and accommodation sector dropped in the second period. This decline could

be related to the economic crisis that made the sector less attractive.

Overall, the descriptive statistics point to the necessity of controlling for business sector in the

regression analysis.

Table 4.1 also highlights the change in applicants' individual characteristics. In the �rst period,

female applicants are younger than men and more often single. These signi�cant di�erences

disappear in the second period. Concurrently, other di�erences emerge. The second-period

female applicants have more dependent children than their male counterparts; they also exhibit

higher education levels and belong to wealthier households. The introduction of the loan ceiling

148

4.3. Regression Analysis: Methods

seems to have squeezed out young, single, and poorer female applicants.

Gender aside, Table 4.1 corroborates the �nding that the MFI's pool of applicants changed

dramatically in the second period. Practically non-existent in the �rst period, the occurrence of

bank loan holders among applicants jumped to 27%. Meanwhile, the loan approval rate increased.

This increase was probably driven by free-riding on bank screening, since the MFI partly relies

on the bank's approval decision. Therefore, MFIs decision process becomes partly controlled by

the mainstream bank. Importantly, the average size of the projects submitted to the MFI seems

insensitive to the introduction of the ceiling, remaining around EUR 30,000. This might indicate

that there is a critical size for entrepreneurial projects in France.

4.3 Regression Analysis: Methods

Our aim is to estimate the period-speci�c impacts of the applicants' gender on both requested

amount and loan size. We proceeded as follows. We run OLS regressions for the requested

amount and the loan size. We include a female dummy (F ) taking value one when the project

holder is a women, a period dummy (C) taking value one in period with ceiling, their interac-

tion term (F ∗ C), project size (PS, in euro), bank loan (BL, in euro, which is equal to zero if

there is no bank loan) and other controls (in X vector) including the applicants' business char-

acteristics (start-up, business sector) and individual characteristics (marital status, age, number

of dependent children, education level and household income). This methodology allows us to

mimic a di�erence-in-di�erences model which is not directly applicable for our research question.

We do not have appropriate control and treatment groups. Therefore, we cannot attribute the

observed di�erences, if any, exclusively to the treatment. Indeed, men are also impacted by the

introduction of the ceiling and it is unrealistic to assume that the EUR 10,000 threshold was not

a constraint for them.

The coe�cients of our main interest correspond to the female dummy and the sum of the female

and the interaction term loadings.

149


A parallel analysis of the requested amount and of the loan size gives us an insight on MFI's

selection process. If the MFI is gender-neutral we expect the coe�cients of female dummy and

of the interaction term to be of the same sign in the requested amount and loan size estimations.

Conversely, if we �nd a di�erence in coe�cients' signs (or signi�cance) across estimations we will

be able to identify a gender bias.

An alternative way to address the MFI's selection process consists in using Heckman's selection

model (Heckman 1979). We will present this model for the sake of a robustness exercise. It will

moreover allow us to analyze the approval decision of the MFI.

The OLS model for the requested amount and the loan size is given by:

yij = α0j + αFjFi + αCjCi + αFCjFi ∗ Ci + αPSjPSi + αBLjBLi + α′xjXi + εij (4.1)

Where i is the individual indicator, j takes value 1 if the dependent variable is the requested

amount and value 2 if the dependent variable is the loan size. Eq. (4.1) is estimated by using

the full sample of applicants when the dependent variable corresponds to the requested amount.

Conversely, it is estimated on the smaller sample consisting of actual borrowers only if the

dependent variable is the loan size. The MFI typically grants the loan size requested by the

successful applicants. The only decision in the hands of the MFI consists in loan approval

or denial. The subsequent sizes of approved loans are straightforward outcomes of the loan

allocation process. Hence, the outcome of the MFI's decision will the re�ected in the di�erence

in the results for the requested amount and loan size estimations.

The estimation of the Eq. (4.1) may be a�ected by multicollinearity issues, however. In both

periods, the project size depends on the applicant's characteristics, possibly including gender. In

addition, project holders with a bank loan have necessarily passed the bank's screening process.

Their loan size thus depends on both their project size and characteristics. To address this

double source of potential multicollinearity, we used the partial least squares (PLS) estimation

strategy suggested by (Agier and Szafarz 2013a). At the MFI level, this approach allowed us to

150


disentangle the impacts of demand-side and supply-side factors on both the requested amount

and the loan size.

Therefore, we performed a double PLS (2PLS) estimation. The �rst step of our methodological

strategy involved regressing project size on both gender, period, interaction term, and control

variables in the following way (henceforth, we drop the subscript i to ease the presentation of

the equations):

PS = γ0 + γFF + γCC + γFCF ∗ C + γ′xX +RPS (4.2)

where RPS is the residual project size net of the in�uence of applicant's characteristics and

period e�ect.

The second step consists in regressing the bank loan on gender, period dummy, interaction term,

residual project size, and the controls. Taking into account Eq. (4.2), we obtained:

BL = δ0+(δF+δPSγF )F+(δC+δPSγC)C+(δFC+δPSγFC)F ∗ C+δPSRPS+(δ′x+δPSγ′x)X+RBL

(4.3)

whereRBL is the residual bank loan size net of the in�uence of all the other explanatory variables,

including gender and project size. There are signi�cant correlations between the project size and

the bank loan and the other control variables as depicted in Tables 4.3 and 4.4. These tables

suggest that the 2PLS strategy is indeed necessary to correct for potential multicollinearity

among the variables.

151


Table4.3:

First

Correlation

Matrice

Requested

Fem

ale

Project

Bank

Personal

Havingpersonal

Start-up

Services

Trade

amount

size

loan

investm

ent

investm

ent

Requestedamount

1Fem

ale

-0.04

1Project

size

0.30

-0.06

1Bankloan

0.02

-0.04

0.90

1Personalinvestm

ent

0.17

-0.06

0.70

0.52

1Havingpersonalinvestm

ent

0.04

-0.01

0.21

0.12

0.26

1Start-up

-0.17

-0.06

-0.20

-0.20

-0.08

0.22

1Services

-0.07

0.09

-0.16

-0.12

-0.12

-0.02

0.07

1Trade

-0.08

0.11

0.09

0.12

0.07

0.00

-0.09

-0.41

1Foodandaccommodation

0.17

0.05

0.18

0.13

0.15

0.09

-0.12

-0.25

-0.25

Construction

-0.06

-0.25

-0.14

-0.12

-0.11

-0.06

0.11

-0.22

-0.22

Artsandentertainment

0.01

-0.01

0.02

0.02

0.02

0.02

-0.01

-0.14

-0.14

Unem

ployed≥6months

-0.08

0.00

-0.15

-0.18

-0.09

0.12

0.21

0.00

0.00

Single

0.02

0.07

-0.12

-0.13

-0.08

-0.02

0.07

0.04

-0.06

Age

-0.06

-0.04

-0.03

-0.03

0.06

-0.06

0.02

0.04

0.04

Dependentchildren

-0.03

0.09

-0.04

-0.02

-0.05

0.00

-0.01

-0.05

0.05

Education

0.06

0.15

0.11

0.06

0.13

0.05

-0.01

0.07

0.01

Household

income

-0.03

0.06

0.17

0.15

0.18

0.09

-0.04

0.05

0.04

Signi�cantcorrelations(10%

andless)in

bold.

152


Table4.4:

SecondCorrelation

Matrice

Foodand

Construction

Artsand

Unem

ployed

Single

Age

Dependent

Education

Household

accommodation

entertainment≥6months

children

income

Foodandaccommodation

1Construction

-0.14

1Artsandentertainment

-0.09

-0.08

1Unem

ployed≥6months

-0.02

0.00

-0.01

1Single

0.03

-0.04

0.03

0.10

1Age

-0.01

-0.05

-0.03

0.09

-0.05

1Dependentchildren

0.00

0.04

-0.06

0.00

-0.42

0.08

1Education

-0.11

-0.14

0.04

-0.03

-0.01

-0.05

-0.13

1Household

income

-0.06

-0.06

-0.02

-0.13

-0.43

0.08

0.23

0.21

1Signi�cantcorrelations(10%

andless)in

bold.

153


The estimation of Eq. (4.3) in itself reveals information on the screening process at banks.

The results should be interpreted with care, however, since they may be a�ected by a selection

bias. Indeed, we observe only the bank loans of successful applicants who subsequently applied

to the MFI.

The last step of our strategy consists in a double PLS (2PLS) estimation for the requested

amount and the loan size. We performed it by replacing PS by RPS (from Eq. (4.2)) and BL

by RBL (from Eq. (4.3)) in Eq. (4.1) for both the requested amount and the loan size. Our

main equation of interest became:

yj = θ0j + θFjF + θCjC + θFCjF ∗ C + θPSjRPS + αBLjRBL+ θ′xjX + εj (4.4)

where the θ loadings contain the links between the respective control variables and the project

size or the bank loan. For instance:

θFj = αFj + αPSjγF + αBLjδF + αBLjδPSγF

Using the 2PLS methodology, our coe�cients in Eq. (4.4) incorporate four di�erent e�ects.

Using the example of the female dummy, θFj is composed of:

1. a direct e�ect of being a female (αFj)

2. an indirect e�ect of being a female through the project size (αPSjγF )

3. an indirect e�ect of being a female through the bank loan (αBLjδF )

4. an indirect e�ect of being a female through the e�ect of the project size on the bank loan

(αBLjδPSγF ).

The other loadings follow the same structure due to the 2PLS speci�cation. This strategy

allows us to correct for potential multicollinearity issues between the project size and the control

154

4.4. Regression Results

variables and between the bank loan and the control variables (including project size). The next

section discusses the estimation results.

4.4 Regression Results

In the �rst period (April 2008-April 2009), the MFI has the legal status of an unregulated NGO.

It is not subject to the EUR 10,000 loan ceiling, and there is no co-�nancing with banks. Con-

versely in the second period (May 2009-June 2012) the MFI is regulated and subject to the EUR

10,000 ceiling. This period is characterized by the emergence of co-�nancing by the MFI and

mainstream banks (33% of MFI's borrowers had a co-�nancing from a mainstream bank in the

second period versus 0% in the �rst period).

In Table 4.5 we present the estimations of the Eqs. (4.2) and (4.3), in columns (1) and (3)

respectively. In column (2) we present a benchmark OLS estimation for the bank loan to better

understand the utility of the PLS method. Column (1) reports the loadings of the OLS regression

for the project size. We detect a signi�cant impact of gender on project size in both periods.

In the �rst period this e�ect corresponds to female dummy loading. In the second period it

corresponds to the sum of the female dummy and the interaction term loadings. Women seem

indeed to undertake smaller projects than men. However, the e�ect is more signi�cant in the

second period. In economic terms, the gender gap in project size is considerable since it amounts

to EUR 8,389 in the second period. The characteristics that exhibit a signi�cantly positive in-

�uence on project size are: education, household income and food and accommodation sector.

A signi�cant negative in�uence is observed for two sectors: services and construction. Finally,

start-ups and long-term unemployed individuals undertake smaller projects.

Columns (2) and (3) in Table 4.5 explain the size of the bank loan. Note that these equations

are estimated on the whole set of applicants. For those applicants without a bank loan this

variable is set to 0. In the simple OLS estimation the impact of being a women is not signi�cant.

However this results has to be interpreted with care as we only observe MFI's applicants who

155


Table 4.5: Regression results for Project Size and Bank Loan

Explained variable Project Size Bank Loan(1) (2) (3)

Estimation method OLS OLS PLS

Female -9.078* 1,243 -4,772***(5.051) (1,527) (1,524)

Ceiling -1.141 10,991*** 10,235***(3.432) (1,036) (1,036)

Female*Ceiling 0.689 -858.9 -402.4(5.510) (1,663) (1,663)

Female+Female*Ceiling -8.389*** 384.2 -5,174***(2.490) (755.8) (751.4)

Project Size 662.6***(9.695)

Residual Project Size 662.6***(9.695)

Star-up -12.98*** -882.5 -9,486***(2.935) (894.6) (885.7)

Services -9.752*** 2,533** -3,929***(3.500) (1,060) (1,056)

Trade 3.229 2,249** 4,389***(3.521) (1,063) (1,063)

Food and accommodation 17.31*** -534.0 10,935***(4.167) (1,269) (1,257)

Construction -16.03*** 1,387 -9,235***(4.350) (1,322) (1,313)

Arts and entertainment -0.839 3,445** 2,889*(5.655) (1,707) (1,707)

Unemployed for at least 6 months -6.187*** -2,951*** -7,050***(2.198) (666.1) (663.4)

Single -4.664* -886.4 -3,977***(2.521) (762.2) (760.8)

Age -0.0432 -13.63 -42.23(0.107) (32.23) (32.23)

Dependent children -2.658** 323.3 -1,438***(1.093) (330.9) (329.9)

Education (nb. of achieved diplomas) 1.754** -573.3*** 588.9***(0.707) (213.9) (213.2)

Household income 4.444*** -967.8*** 1,977***(1.076) (327.7) (324.9)

Constant 46.41*** -15,609*** 15,143***(6.975) (2,152) (2,105)

Observations 985 985 985R-squared 0.167 0.856 0.856Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1

156


have managed to secure a bank loan. Importantly, this dummy becomes signi�cant in column

(3) which is a simple PLS estimation. The loadings in column (3) are corrected for the multi-

collinearity issues discussed in the previous section. This result suggests that among applicants

with a bank loan women receive signi�cantly smaller bank loans than men with similar charac-

teristics, mainly because they undertake smaller projects. In economic terms, the e�ect is large

reaching EUR 5,174 in the second period.16 The period e�ect is strongly signi�cant and roughly

constant between columns (2) and (3). This �nding illustrates the necessity of bank co-�nancing

in the second period.

Moreover, the estimations show that the correlation between project size and bank loan size is

both high and signi�cant. This correlation explains the strong match (R-squared is equal to

86%). Globally, the project-size equation and bank-loan PLS equation deliver a consistent pic-

ture. Women-owned businesses and start-ups exhibit smaller project sizes and get smaller loans

from banks while both project size and bank loans are the largest for the food and accommo-

dation sector, for higher educated individuals and for richer households. Still, a few interesting

di�erences between columns (2) and (3) are worth mentioning. In particular, higher education

and household income are associated with smaller bank loans in column (2). This is in line with

the fact that banks grant credits at harsher conditions than MFIs or other subsidized institu-

tions. Consequently, project holders with larger opportunities (ex. with higher education, or

better �nancial characteristics) might be reluctant to borrow from banks.

These �rst and second step results do not give an optimistic insight concerning women's situation

in terms of project size and access to �nance. In table 4.6 we depict the loan allocation process

by the MFI of our interest. The estimation methodology follows the lines set out in Section 3.

It enables us to trace the impacts of gender and other characteristics throughout the MFI's loan

allocation process. The coe�cients in the �rst line in Table 4.6 illustrate the e�ect of being a

women on the requested amount (columns (1) to (3)) and on the loan size (columns (4) to (6))

16The coe�cient of the female dummy in the �rst period has to be interpreted with caution, as only 6applicants had a bank loan in the �rst period. Interestingly, none of them has been granted a microcredit.

157


Table 4.6: Regression results for the Requested Amount and Loan Size

Explained variable Requested Amount Loan Size(1) (2) (3) (4) (5) (6)

Estimation Method OLS PLS 2PLS OLS PLS 2PLS

Female -805.0 -2,941*** -1,624** 2,036** 110.4 1,287(674.5) (676.6) (673.2) (895.5) (899.4) (894.4)

Ceiling -8,788*** -9,056*** -11,881*** -6,476*** -6,718*** -9,242***(483.3) (479.9) (457.4) (596.3) (593.3) (577.7)

Female*Ceiling 811.6 973.7 1,085 -2,147** -2,001** -1,901**(734.5) (734.4) (734.4) (954.7) (954.7) (954.8)

Female+Female*Ceiling 6.62 -1,967*** -539.1 -110.5 -1890.3*** -614.0(333.8) (339.9) (331.8) (386.8) (393.7) (384.5)

Project Size 235.3*** 212.2***(10.33) (12.61)

Residual Project Size 235.3*** 52.41*** 212.2*** 48.72***(10.33) (4.281) (12.61) (4.656)

Bank Loan -0.276*** -0.276*** -0.247*** -0.247***(0.0142) (0.0142) (0.0171) (0.0171)

Residual Bank Loan -0.276*** -0.247***(0.0142) (0.0171)

Star-up -2,158*** -5,213*** -2,595*** -1,708*** -4,463*** -2,123***(395.3) (413.7) (391.1) (446.8) (486.3) (441.5)

Services 327.8 -1,967*** -882.4* -15.09 -2,084*** -1,115**(469.6) (469.7) (466.4) (526.1) (527.0) (522.8)

Trade -390.3 369.5 -842.0* -451.5 233.6 -848.9(470.6) (473.4) (469.3) (532.9) (535.3) (531.0)

Food and accommodation 1,477*** 5,550*** 2,532*** -60.15 3,612*** 914.9(560.3) (576.6) (555.3) (661.2) (677.9) (650.9)

Construction 408.7 -3,363*** -814.3 226.8 -3,174*** -896.4(584.1) (594.4) (579.8) (656.0) (670.1) (651.2)

Arts and entertainment -510.7 -708.2 -1,506** -803.8 -981.8 -1,694*(755.3) (754.8) (753.7) (888.0) (887.8) (888.6)

Unemployed for at least 6 months -835.8*** -2,292*** -345.5 -619.2* -1,932*** -192.9(297.2) (309.6) (293.0) (343.0) (357.4) (339.4)

Single -206.0 -1,304*** -205.7 -440.8 -1,430*** -449.4(336.8) (340.7) (336.0) (399.5) (406.5) (397.9)

Age -37.20*** -47.35*** -35.70** -32.14* -41.30** -30.88*(14.24) (14.25) (14.23) (17.01) (17.03) (17.01)

Dependent children -41.85 -667.3*** -270.4* -256.3 -820.3*** -465.6**(146.2) (147.1) (145.7) (181.2) (182.5) (180.7)

Education (nb. of achieved diplomas) -16.04 396.7*** 234.1** 91.35 463.4*** 318.2***(94.81) (94.54) (94.17) (113.4) (113.0) (112.6)

Household income -182.6 863.3*** 317.5** -208.3 734.6*** 246.9(145.4) (146.2) (143.5) (164.6) (164.6) (161.7)

Constant 15,678*** 26,599*** 22,419*** 13,047*** 22,893*** 19,159***(976.0) (954.1) (929.5) (1,152) (1,175) (1,128)

Nb. of observations 985 985 985 580 580 580R-squared 0.648 0.648 0.648 0.582 0.582 0.582Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1

158


in the period without ceiling. Conversely, the e�ect of being a women in the period with ceiling

is captured by sum of the female and the interaction term loadings in line 4. This table describes

both the demand from the MFI, through the requested amount, and its supply, through the loan

size.

Column (1) and (4) present simple OLS estimations following Eq. (4.1) for the requested amount

and loan size respectively. Remarkably, gender is not signi�cant for the requested amount neither

in the �rst nor in the second periods. This result is important since we have found that women

undertake smaller loans than men, meaning that women �nance a larger part of their projects

using a microcredit. Interestingly, the MFI in the �rst period selects women with larger projects.

This is in line with a naturally positive orientation of the MFI towards women. However, this

e�ect disappears in the period with ceiling, where the pool of the MFI is partly shaped by the

screening of the mainstream banks. Hence, the MFI has less control over its decision on loan

allocation, especially if it is tempted to free-ride on bank's screening process. Obviously, the pe-

riod dummy is strongly signi�cant in all the columns due to the introduction of the EUR 10,000

ceiling. As expected, project size has a signi�cantly positive impact on the requested amount

and on the loan size. Interestingly, a smaller bank loan is associated with a larger request and

microcredit from the MFI. This result suggests that the MFI indeed plays a role of expanding

the access to �nance for those who have di�culties to access traditional �nance. Conversely,

this might mean that the applicants prefer to apply for larger loans from the MFI due to more

favorable conditions. Gender aside, start-ups, long term unemployed and older project holders

request and receive smaller loans from the MFI.

The results are dramatically di�erent for simple PLS estimations in columns (2) and (5). The

loadings in these columns account for the link between the project size and the control variables.

In these speci�cations we only introduce the residual project size and keep the variable bank

loan unchanged. Women request signi�cantly smaller loans from the MFI in both periods. This

occurs because they undertake smaller projects. The MFI mitigates this gender bias by choosing

women with larger requested amounts in the �rst period. However, the MFI no longer does so

159


in the second period. We observe a signi�cant di�erence in the loan size for female borrowers

under ceiling enforcement. The size of the loading is relatively large (EUR 1,890) as compared

to the microcredit ceiling. We give the following interpretation for this result. Originally, the

MFI is positively oriented toward women. This is in line with its social bottom line. Therefore,

it attempts to correct for gender biases existing in the demand side of female project holders by

encouraging the �nancing of women with larger loan requests (according to the results in the

�rst line of Table 4.6). In the �rst period the MFI has the entire control upon the pool of its

applicants, who face relatively weak �nancial constraints. This is no longer the case under loan

ceiling. Mainstream banks at least partly shape the pool of the applicants through co-�nancing

schemes, whereas a socially oriented MFI may be tempted to free-ride on bank's screening pro-

cess. This interpretation is particularly realistic in the context where the introduction of the

ceiling has triggered co-�nancing arrangements. A competing scenario might be the willingness

of the MFI to become more �nancially sustainable, as compliance with regulation has opened

access to the MFI to loanable sources.

Gender aside, individuals having dependent children, being single, being in services or construc-

tion sectors request and receive smaller loans from the MFI because they undertake smaller

projects. Conversely, higher education and household incomes are associated with larger re-

quested and received amounts in PLS estimations.

Finally, in columns (3) and (6) we present the estimations of the Eq. (4.4) for the requested

amount and the loan size. Female applicants still request smaller amounts from the MFI without

ceiling (due to very few applicants with a bank loan in the �rst period). However, they align

their demands with men in the second period. Female applicants need to borrow more from the

MFI in the second period. We suppose this is due to the impediments faced by women in the

process of accessing classical �nance. Concerning loan supply, the MFI corrects the downward

bias present in the requested amounts by women. Indeed, the female dummy in the loan size

equation is not signi�cant. By the same token, we �nd no gender impact in the second period.

Therefore, the originally women-oriented MFI becomes gender-neutral. We interpret this �nd-

160

4.5. Robustness check: Heckman selection model

ing as a result of co-�nancing schemes with banks. The underlying e�ect might have occurred

through two di�erent channels. On one hand, banks have partly shaped the pool of the appli-

cants of the MFI. Indeed, holders of large projects have to secure a bank loan before applying

for microcredit. On the other hand, the socially oriented MFI might be tempted to free-ride o

bank's screening process.

Overall the results show that the situation of women worsened in comparison with the �rst-

period.

4.5 Robustness check: Heckman selection model

The objective of this paper is to analyze the loan allocation process of an MFI before and after a

change in regulation. We have data both on the applicants and on the borrowers of the MFI. An

alternative way to address this research question is to use a Heckman selection model (Heckman

1979). Indeed, the MFI �rst selects its applicants and further grants the microloans. The loan

size is only observed for accepted individuals.

Besides providing an alternative approach, this method has the advantage of giving a direct

insight on MFI's approval process. The existing literature provides evidence on harsher credit

approval for women (Orser et al. 2000; Cavalluzzo et al. 2002; Fay and Williams 1993). In this

section we will be able to test this previously found results using our data.

The Heckman model writes as follows:

LoanSize = θ0 + θFF + θCC + θFCF ∗ C + θPSRPS + αBLRBL+ θ′xX + ε

Approval = 1[λ0 + λFF + λCC + λFCF ∗ C + λPSRPS + λBLRBL+ λ′xX + λ′zZ + u > 0

]where 1 is an indicator function taking value 1 when the linear combination in the brackets is

positive. The Loan Size is only observed when Approval = 1. In Heckman selection model the

following assumptions are made: u ∼ N (0, 1) and E(ε|u) 6= 0. The Approval process is modeled

161


using a probit estimation. The variables in Z vector a�ect only the approval process. We use

business cycles consisting in the growth rate of defaults and of the start-ups in the sector of

activity of the enterprise i at the moment of credit granting, one quarter before credit granting,

and two quarters before credit granting. We moreover include in Z a dummy taking value 1 if

the applicant has a bank loan. This strategy will allow us to identify whether the MFI has a

preference toward �nancing applicants with a bank loan.

The results of this model are presented in Table 4.7. Concerning business cycles we only report

the signi�cant variables in the table.

In columns (1) and (2) we report the results without the inclusion of the residual project size

and residual bank loan. In columns (3) and (4) we present estimations using simple PLS, i.e.

with residual project size as an explanatory variable instead of the project size itself. In columns

(5) and (6) we present results for 2PLS estimations, i.e. where both the residual project size and

residual bank loan are included.

Concerning the probability of approval, gender is not signi�cant, suggesting that there is no

gender discrimination in MFI's approval process. Among the other explanatory variables, ser-

vices, trade, food and accommodation sectors, long-term unemployment, number of dependent

children, and the variables on business cycles stand out with a signi�cantly negative impact on

the probability of loan approval. Moreover, household income and having a bank loan increase

the probability of approval. This last �nding gives credit to our assumption that the MFI accepts

more easily individuals with bank loans due to free-riding opportunities.

Our previous results are remarkably stable after controlling for the selection bias. Nevertheless,

one �nding is particularly worth mentioning. In column (6) we observe that in the �rst period

the MFI is gender neutral. However, in the second period the sum of the female and interaction

term loadings is signi�cantly negative. This result suggests that there is indeed some bias in

MFI's loan allocation occurring in the second period which persists after correcting for multi-

collinearity issues using a 2PLS strategy.

Finally, the inverse of the Mills ratio is not signi�cant. Overall, the Heckman model con�rms

162

4.6. Conclusion

our main �ndings outlined in Section 4.4.

4.6 Conclusion

Gender-neutral regulations can result in gender-sensitive outcomes (Johnson and Nino-Zarazua

2011). Cull et al. (2011) show that pro�t-oriented MFIs respond to supervision by serving fewer

women in order to maintain pro�t rates. Our paper con�rms that apparently benign micro-

credit regulations, such as a loan ceiling, can a�ect access to credit for women entrepreneurs.

Speci�cally, we provide evidence that female micro-borrowers are harmed by the loan ceiling

imposed on licensed MFIs by the French regulator. We also o�er a possible rationale for the

mechanism underpinning this unexpected outcome. A low loan ceiling leads to the development

of co-�nancing schemes between MFIs and mainstream banks. In turn, the MFIs are a�ected

by whatever biases in loan granting the banks are prone to. Our empirical �ndings suggest that

an MFI which is originally positively oriented toward women, becomes at best gender-neutral.

However, we �nd evidence that ceiling enforcement can lead to gender biases in terms of lower

loans allocated to female borrowers.

Women are known to start businesses with smaller external �nance than men (Coleman 2000).

However, the evidence on gender discrimination in lending remains controversial. According to

Carter et al. (2007), many of the di�erences in the bank loans granted to male and female

entrepreneurs are attributable to structural dissimilarities. Therefore, any econometric analysis

that �nds gender discrimination is suspected of having missed relevant variables. In this paper,

we get around this issue by using an indirect identi�cation technique and by taking advantage

of a natural experiment. We detect gender biases in banks' loan allocation by observing their

impact on the applicants of an MFI which proved to be gender-neutral before co-�nancing with

mainstream banks was introduced.

The main limitation of our approach is the impossibility of estimating-and subsequently correct-

ing for-a self-selection bias that might have appeared in the pool of microcredit applicants after

163


Table 4.7: Robustness check: Heckman Selection Model

Explained variable Approval Loan Size Approval Loan Size Approval Loan Size(1) (2) (3) (4) (5) (6)

Estimation method Probit OLS Probit/PLS PLS Probit/2PLS 2PLS

Female -0.0937 2,082** -0.105 160.3 -0.104 1,341(0.200) (889.8) (0.201) (894.4) (0.200) (890.4)

Ceiling 0.211 -6,593*** 0.209 -6,835*** 0.208 -9,368***(0.150) (631.0) (0.149) (627.9) (0.145) (619.5)

Female*Ceiling 0.157 -2,227** 0.158 -2,081** 0.158 -1,982**(0.219) (956.5) (0.219) (956.6) (0.219) (956.6)

Female+Female*Ceiling 0.063 -145.0 0.053 -1,920*** 0.053 -640.1*(0.101) ( 388.9) (0.103) (394.4) (0.100) (384.1)

Project Size 0.00123 211.7***(0.00315) (12.52)

Residual Project Size 0.00123 211.7*** 0.00117 47.68***(0.00315) (12.52) (0.00169) (5.028)

Bank Loan -8.92e-08 -0.247*** -8.92e-08 -0.247***(4.58e-06) (0.0170) (4.58e-06) (0.0170)

Residual Bank Loan -8.92e-08 -0.247***(4.58e-06) (0.0170)

Star-up -0.195 -1,610*** -0.211* -4,358*** -0.210* -2,011***(0.122) (480.6) (0.127) (521.2) (0.122) (487.2)

Services -0.291** 90.20 -0.303** -1,974*** -0.303** -1,002*(0.146) (558.6) (0.147) (562.7) (0.146) (561.2)

Trade -0.441*** -288.9 -0.437*** 394.6 -0.437*** -691.7(0.149) (611.8) (0.150) (612.4) (0.149) (605.2)

Food and accommodation -0.584*** 164.5 -0.562*** 3,828*** -0.563*** 1,122(0.174) (781.8) (0.179) (787.5) (0.176) (755.3)

Construction -0.154 283.1 -0.174 -3,110*** -0.173 -824.5(0.181) (658.9) (0.185) (675.2) (0.181) (659.7)

Arts and entertainment -0.382 -675.0 -0.383 -852.7 -0.384 -1,568*(0.234) (913.2) (0.233) (913.3) (0.233) (912.7)

Unemployed for at least 6 months -0.179** -549.4 -0.186** -1,859*** -0.186** -114.1(0.0897) (365.0) (0.0936) (380.4) (0.0894) (368.4)

Single -0.142 -365.4 -0.148 -1,353*** -0.148 -368.4(0.102) (420.8) (0.103) (428.8) (0.102) (423.1)

Age -0.00375 -30.16* -0.00381 -39.30** -0.00380 -28.85*(0.00428) (17.26) (0.00428) (17.28) (0.00428) (17.29)

Dependent children -0.0870** -215.5 -0.0902** -778.1*** -0.0901** -422.2**(0.0439) (195.5) (0.0442) (197.7) (0.0438) (197.0)

Education (nb. of achieved diplomas) 0.0369 75.15 0.0391 446.4*** 0.0391 300.7***(0.0284) (116.5) (0.0283) (116.5) (0.0282) (116.4)

Household income 0.132*** -267.2 0.137*** 673.6*** 0.137*** 184.2(0.0453) (197.9) (0.0459) (200.2) (0.0453) (199.8)

Having a bank loan 0.269* 0.269* 0.269*(0.146) (0.146) (0.146)

Default growth rate in t -0.578** -0.578** -0.578**(0.229) (0.229) (0.229)

Default growth rate in t-2 -0.826*** -0.826*** -0.826***(0.242) (0.242) (0.242)

Constant 0.605** 13,439*** 0.662** 23,263*** 0.661** 19,515***(0.303) (1,363) (0.294) (1,360) (0.290) (1,307)

Inverse of Mills ratio -809.6 -809.6 -809.6(1,540) (1,540) (1,540)

Observations 985 985 985 985 985 985Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1

164

4.6. Conclusion

the loan ceiling came into force. In addition, we cannot fully exclude that internal or external

factors neglected in this study interfere with the change of the MFI's attitude toward female

borrowers. Among the external factors, one can think of the �nancial crisis, which overlaps

the �rst period of our sample. Among the internal factors, we can mention the growth of the

institution, which could have a�ected its governance. Although neither of these factors alone

could have generated the observed gender-sensitive change of attitude, it may well be that they

a�ected the loan granting strategy of the MFI.

Further research could build on our innovative methodology and investigate whether biases also

exist in bank lending against other discriminated-against segments of the population. Access to

MFIs' databases could make it easier to check whether banks exert disparate treatment based on

race and ethnicity. The literature �nds that non-white applicants can indeed be discriminated

against in lending (Storey 2004; Blanchard et al. 2008; Blanch�ower et al. 2003).

The success of the worldwide microcredit industry is at least partly attributable to its focus

on poor female entrepreneurs who desperately need funds to launch their businesses (Garikipati

2008; Guérin 2011). It is therefore important to avoid introducing regulations that can counter-

act the women's empowerment e�orts made by MFIs (Hudon and Sandberg 2013). If the aim

of the French regulator is to segment the credit market, then co-�nancing arrangements should

be prohibited. However, ruling out co-�nancing while maintaining a very low loan ceiling could

compromise the sustainability of the micro�nance industry, especially if subsidies dry up (Hudon

and Traça 2011). In any case, forcing MFIs to accept a loan ceiling that is too low to meet the

needs of micro-businesses is counterproductive. Evidently, the EUR 25,000 ceiling suggested by

the EU makes more sense than the EUR 10,000 cap introduced in France.

165


166

General Discussion

This PhD thesis studies micro�nance in developed countries. In contrast to developing coun-

tries where micro�nance has been widely investigated, our research question is relatively under-

researched and poorly documented. Therefore, this thesis represents a step toward a better

understanding of the opportunities and the challenges faced by this industry in the industrial-

ized world where micro�nance mainly consists of two devices, microcredit and business training.

In the �rst two chapters we tackle business training services which complement microcredit pro-

vision. Conversely, in chapters 3 and 4 we investigate the implications of microcredit regulation

on the social bottom line of MFIs. In what follows we will brie�y discuss our main �ndings and

will give an insight into further research on micro�nance in richer economies.

State intervention on the micro�nance market is not a recent phenomenon. In chapter 1, we

theoretically analyze the impact of state intervention on �nancial inclusion in the microcredit

market where micro�nance institutions o�er individual loans and business training services. We

focus on the interaction between the loan guarantee and the choice of the micro�nance institution

to provide business training. This research question has been motivated by the intuition that the

loan guarantee might impact the MFI's incentives to provide business training and probably de-

crease its e�ciency in terms of outreach. This intuition stems from the substitutability between

insurance and self-protection documented in the insurance market. Indeed, the loan guarantee

acts as an insurance against project failure for the MFI, whereas business development services

act as a self-protection device as they increase the probability of project success.

We have extended Tirole's (2006) �xed-investment model to the microcredit market by adding

167

General Discussion

the loan guarantee and business training. We show that the state guarantee can have counter-

productive e�ects in terms of borrowers �nanced through business training (in particular when

the distribution of the project returns is uniform). This central �nding leads us to study an

alternative solution: business training subsidization and then to compare it to loan guarantee in

terms of �nancial inclusion. We �nd that � for a �xed budget � business training subsidization

can lead to higher �nancial inclusion than the loan guarantee, provided that training technology

is e�cient enough and that training is targeted enough toward excluded borrowers. Finally, we

show that, even though it can eliminate the counterproductive e�ect of the loan guarantee, a

mix of the loan guarantee and business training subsidization will not lead to higher �nancial

inclusion compared to business training subsidization alone if training is targeted and is e�cient

enough.

In chapter 1 we have shown that business training can improve access to �nance through a di-

rect positive impact on the probability of success of a project. However, business training can

moreover impact the probability of success indirectly, through borrowers' behavior. This indirect

e�ect arises when the borrowers believe that the MFI has superior information about their type.

This situation has been applied to labor market (Ishida 2006), schooling and family (Benabou

and Tirole 2003a) and it suits particularly well the microcredit market. Indeed, on one hand,

micro-borrowers are inexperienced and start a business for the �rst time. On the other hand,

the MFI has superior information about micro-entrepreneurship through its past experience.

In the second chapter we account for this indirect impact of business training on borrowers be-

havior, i.e. on their incentive to exert e�ort through the so called looking glass self e�ect. In

particular, we analyze how superior information can impact MFI's decisions concerning borrow-

ers' assignment to training programs using both theoretical and empirical modeling. We show

that information plays an important role on the microcredit market. In such a context the MFI

might manipulate borrowers' self con�dence through the looking glass self e�ect. In the theo-

retical model of the second chapter we show that, in situations where the relationship between

business training and the type of the borrower is decreasing under symmetric information, a

168

General Discussion

non-monotonic relationship between business training and the type of the borrower may occur

under reversed asymmetric information, i.e. situations where the MFI has superior information

about borrower's type.

We test for the existence of this equilibrium using original hand-collected data from CREASOL, a

French MFI which in addition to loan-granting assigned some of its clients to training programs.

Using a bivariate probit model to control for endogeneity between business training and risk of

the borrower, we have shown that a non-monotonic relationship between assignment to training

and the risk is indeed plausible. The MFI seems to take into account the "looking-glass self

e�ect". In other words, the MFI takes into account the fact that its choices impact borrowers'

beliefs on the microcredit market. Overall, the econometric analysis validates the equilibrium

presented in the theoretical model.

The second chapter provides interesting evidence on how MFI's decisions about business train-

ing might undermine borrowers' motivation to exert e�ort. In the next two chapters we will no

longer focus on business training provision. In turn we will investigate the consequences of a

di�erent decision of CREASOL consisting in compliance with new regulation. We show that this

decision has also salient implications on microborrowers. But these consequences are no longer

based on informational asymmetries as in chapter 2. They rather undermine access to �nance

for some groups of applicants.

More precisely, in the last two chapters we focus on loan ceiling regulation which is common in

developed economies. For instance, the European Commission de�nes microcredit as a loan of

up to EUR 25,000 (European Commission 2009). However, in France microcredit associations

are constrained by a EUR 10,000 ceiling. CREASOL has complied with this ceiling in April

2009. In chapters 3 and 4 we study the social consequences of microcredit ceiling enforcement.

The third chapter addresses the impact of loan ceilings on the microcredit market, both theo-

retically and empirically. Our theoretical model applies to CREASOL, which operates along a

competitive classical credit market, which is the case in most developed economies. The MFI

grants microcredit services at below-market conditions. The presence of classical banks willing to

169

General Discussion

co-�nance projects with the MFI is a particular feature of our model, which also makes it speci�c

to developed countries. Relying on these assumptions, we show that imposing a low loan ceiling

can trigger mission drift. The MFI can be tempted to adopt a cost-reducing strategy consisting

in co-�nancing arrangements with banks for large projects. This strategy is at the expense of

holders of small projects, who need below-ceiling loans. As a consequence, the ceiling-constrained

MFI might shift toward granting larger loans and attracting wealthier clients. This phenomenon

is termed "mission drift" in the micro�nance literature.

The second part of the third chapter tests for the plausibility of the mission-drift scenario by

exploiting a natural experiment. We bene�ted from detailed information on the applicants of

CREASOL before and after the enforcement of the EUR 10,000 loan ceiling. Our descriptive

statistics show that co-�nancing of large projects with banks emerged after ceiling enforcement.

At the same time, our di�-in-di� probit regressions con�rm that loan approval became signif-

icantly higher for holders of large projects. Mission drift is therefore a real threat in contexts

where low loan ceilings are enforced and classical �nance is well developed. Therefore, regulators

should pay attention to this possible outcome when imposing loan ceilings to the micro�nance

industry.

In chapter 3 we have focused on the project size as a measure of the depth of outreach. In

chapter 4 we go a step further in analyzing perverse e�ects of loan ceilings. In this chapter we

use a di�erent measure of social performance consisting in the size of microcredits granted to

women. Particularly, we focus on the outcomes in terms of loan size across male and female

micro-entrepreneurs arising after loan ceiling enforcement.

Gender-neutral regulations sometimes trigger gender-sensitive outcomes (Johnson and Nino-

Zarazua 2011). Cull et al. (2011) show that pro�t-oriented MFIs respond to regulatory en-

forcement by serving fewer women in order to maintain pro�t rates. In chapter 4 we con�rm

that apparently benign microcredit regulations, such as a loan ceiling, can signi�cantly a�ect

access to credit for women entrepreneurs. Speci�cally, we provide evidence that female micro-

borrowers are worse-o� after ceiling enforcement by the French regulator. We also o�er a possible

170

General Discussion

rationale for the mechanism underlying this unexpected outcome. A low loan ceiling leads to

the development of co-�nancing schemes between MFIs and mainstream banks. In turn, MFIs

are a�ected by potential biases existing on the classical �nancial market. Our empirical �nd-

ings suggest that an MFI which is originally positively oriented toward women, becomes at best

gender-neutral. Furthermore, we �nd evidence that ceiling enforcement could possibly lead to

lower loans allocated to female borrowers.

Following our results in chapters 3 and 4, our main recommendations to regulators would be the

following. Signi�cant heterogeneity in the micro�nance industry makes it is extremely di�cult

to de�ne optimal loan ceiling that would be low enough to make a di�erence, but at the same

time high enough to avoid mission drift. Facing this issue, other regulation designs could be

developed. In chapter 3 we emphasize that project size is more informative as compared to the

loan size in de�ning social lending. Therefore, a potential regulation could consist in imposing

ceilings to project size rather than loan size. This alternative could, however, be easily overcome

by arti�cially splitting large projects into smaller ones.

In addition, regulators could attempt de�ning the target pool of borrowers of subsidized MFIs.

For instance, women and discriminated-against minorities could be targeted in priority. In this

way, micro�nance in developed countries would meet its original principle of serving poor and

disadvantaged segments of the population.

Importantly, the success of the worldwide microcredit industry is at least partly explained by

its focus on poor female entrepreneurs who desperately need funds to start their businesses

(Garikipati 2008; Guérin 2011). It is therefore important to avoid enforcing regulations that can

go against women's empowerment e�orts made by MFIs (Hudon and Sandberg 2013). If the aim

of the French regulator is to segment the credit market, then co-�nancing arrangements should

be prohibited. However, ruling out co-�nancing while maintaining a very low loan ceiling could

compromise the sustainability of the micro�nance industry, especially if subsidies dry up (Hudon

and Traça 2011). In any case, forcing MFIs to accept a loan ceiling that is too low to meet the

needs of micro-businesses is counterproductive. Evidently, the EUR 25,000 ceiling suggested by

171

General Discussion

the EU makes more sense than the EUR 10,000 cap introduced in France.

Finally, micro�nance in developed countries is still very young. Nascent regulation initiatives in

some of the European countries have a key role to play in shaping its future. We hope that our

conclusions from chapters 3 and 4 will raise concern among regulators of microcredit in developed

countries.

Further, we discuss some of the limits of this research. First, our database is remarkably de-

tailed but still limited to a single institution, CREASOL. Particularly, studying the e�ects of

the regulatory change is cumbersome in this context as the change of status of this institution

was not randomly assigned. Second, this thesis gives interesting insights concerning microcredit

in developed economies using a single-market approach. It would be interesting to study a more

general model including missing markets (see Emran et al. 2011). As we have discussed in the

introduction of this thesis micro�nance has signi�cant spillovers on the labour, schooling, health

markets and on the economical development more generally. Accounting for these links will allow

us to analyze a more general equilibrium and to elaborate consistent policy implications. This

analysis will be particularly useful to explain what are the gains for the state from participating

in the microcredit market.

Concerning further research on micro�nance in the developed countries, �rst, I would like to

focus on borrowers' incentives to reimburse their loans. Indeed, absent the mechanisms in force

under group lending methodology, dynamic incentives, compulsory savings, or direct monitoring

the reimbursement process remains somewhat puzzling in developed countries.

Notably, the purpose of this project is to study, using a theoretical model inspired from Benabou

and Tirole (2003a) and (Herold 2010), the link between trust and the type of the contracts o�ered

by the MFIs in developed countries and worldwide. The motivation for this research consists

in the fact that trust and reciprocity play an important role in micro�nance where individuals

lacking traditional forms of collateral are denied access to �nancial services. Hence, social capital

(or trust) is considered to be a substitute for formal collateral. For instance, Cornée and Szafarz

(2013) provide evidence for reciprocity shaping the incentives of credit reimbursement in social

172

General Discussion

banking. The predictions of the theoretical model will be tested using an empirical strategy with

reference to Algan and Cahuc (2010) and Aghion et al. (2010) who use comparable measures for

trust.

To analyze the link between the generalized (how much one trusts people in general) and in-

terpersonal (how much one trusts people from his network) levels of trust and the type of the

contracts o�ered by the MFIs we intend to use various data sets. First, comparable trust in-

dicators will be provided by the World Values Survey including surveys conducted from 1981

to 2008 in 87 di�erent countries. Second, information about di�erent types of contracts o�ered

by the MFIs is available from the Mix Market data set including information about more than

2000 MFIs in 119 countries. Besides the type of the contracts, there is information on di�erent

practices of the MFIs which can be used to proxy their monitoring intensity.

There are several assumptions that can be tested in this research project. The �rst assumption

relies on the fact that in countries with higher levels of trust, MFIs are more likely to o�er

incomplete contracts, i.e. contracts without (alternative forms of) collateral requirements. If

this assumption is true we would expect a negative correlation between the generalized level of

trust and the presence of group lending methodology and compulsory savings requirements. This

result would be in line with La Porta et al. (1997) who argue that in economies with higher

levels of trust, contracts are shorter and cover only broad contingencies.

The second assumption consists in the negative relationship between the intensity of monitoring

and trust as advocated by Bjørnskov (2009). The author argues that more trustful employees

need to be less monitored. If this assumption is true I would expect a negative correlation be-

tween the intensity of monitoring of the MFIs' borrowers and the level of generalized trust.

Finally, a third assumption is in line with Aghion et al. (2010) who argue that trust shapes

the level of institutions which in turn shape the level of trust. Similarly it would be interesting

to test whether higher levels of trust are associated with less enforced contracts which in turn

generate higher levels of trust.

In a nutshell, in developing countries micro�nance boasts impressively high repayment rates. To

173

General Discussion

explain this phenomenon researchers have provided several rationales, ranging from the mech-

anisms in force under group lending to dynamic incentives. However, micro�nance in the de-

veloped countries does not follow this type of lending methodologies. Tackling the puzzle on

repayment incentives of micro-borrowers in developed countries is one of the priorities on my

research agenda.

A second salient further research question concerns entrepreneurial motivation. Du�o (2010)

argues that the main challenge of the micro�nance sector is to di�erentiate between those indi-

viduals having a sense of entrepreneurial initiative and risk-taking (opportunity entrepreneurs)

and those who are constrained to become self-employed (necessity entrepreneurs). This question

is especially pertinent in the context of developed countries in presence of safety nets, which

weaken the incentives for self-employment.

According to the existing literature there are signi�cant di�erences between the two types of

entrepreneurs. For instance, Bhola et al. (2006) �nd that opportunity entrepreneurs are gen-

erally better educated and more risk tolerant. Moreover, they have higher opportunity costs of

self-employment, a larger professional network, a higher level of self-con�dence, an internal locus

of control, and a lighter perception of obstacles.

The purpose of this research project is to address both theoretically and empirically the per-

formance of micro-entrepreneurs with respect to their type. Using data from CREASOL, we

will measure performance by default status and the survival time. To identify the type of the

borrowers (necessity or opportunity) we have surveyed CREASOL's clients. Our data, allows

us to test for the validity of the above listed di�erences between opportunity and necessity en-

trepreneurs among CREASOL's clients. Out of 574 granted loans between April 2008 and April

2012, 294 micro-entrepreneurs participated in the survey. 49% of them have declared being neces-

sity entrepreneurs. 46% of the participants have declared being opportunity entrepreneurs and

5% identify themselves with both types. This survey allows us to investigate whether necessity

entrepreneurs are less likely to succeed in their business compared to opportunity entrepreneurs

according to some of the existing studies (Andersson and Wadensjo 2007; Acs and Varga 2005;

174

General Discussion

Hart 2003; Bates 1990) or whether there is no signi�cant di�erence between the two of them

(Block and Sandner 2009).

Following the literature on credit scoring models (Roszbach 2004), beyond default status, it is in-

teresting to study when the default occurred. Our intuition suggests that necessity entrepreneurs

are more likely to default at the early stage of their business. This can be explained by the fact

that these borrowers have a preference for an employee status as compared to the self-employed

status. Thus, if o�ered a job proposal at early stage of the project they could be more likely to

accept it.

Further, through the lens of a theoretical model, we will explain which of the di�erentiating

characteristics of the two types of borrowers have a greater impact the probability of success of

the project. Both types of borrowers are inexperienced. Arguably, opportunity entrepreneurs

receive a private bene�t from running a business. Nevertheless, they seem to be more risk-

tolerant. Hence, these borrowers might cause important losses to the MFI compared to necessity

entrepreneurs who are generally considered risk-averse in the existing literature. Consequently,

opportunity and necessity features do not correspond to the binary "good" and "bad" classi�ca-

tion in the standard principal-agent modeling.

Finally we will undertake the development of a separating self-selective contract for the necessity

and opportunity borrowers. The contract design might rely on dynamic incentives, di�erent

repayment schedules, loan size, interest rates, loan duration, co-�nancing with classical banks,

collateral, training programs, renegotiation, etc.

To conclude, this thesis expands existing literature on micro�nance to developed countries. In

this research we have tackled the functioning of business training and microcredit in presence of

subsidies, informational asymmetries and nascent regulation. Despite this step toward a better

understanding of the industry, many issues remain unsolved and need further investigation. For

instance, there is a �ourishing literature on the impact evaluation of micro�nance in developing

countries. Such studies would be especially helpful for the developed countries where safety nets

are well developed and governments are involved in supporting micro�nance. Therefore, the

175

Bibliography

question whether MFIs in developed countries are able to e�ciently address poverty alleviation

through social and �nancial inclusion is key for further research.

176

References

Acclassato, D. H. (2008). Les plafonnements de taux d'intérêt en micro�nance servent-ils

réellement les pauvres et petits opérateurs économiques? Mondes en développement (1),

93�109.

Acs, Z. and A. Varga (2005). Entrepreneurship, agglomeration and technological change. Small

Business Economics 24, 323�334.

Adie Annual Report (2012). http://www.csdl.asso.fr/wp-

content/uploads/downloads/2012/09/

rapport-dactivite-veroulle-en-ecriture.pdf (accessed May 16, 2013).

Aghion, P., Y. Algan, P. Cahuc, and A. Shleifer (2010). Regulation and distrust. The Quarterly

Journal of Economics 125 (3), 1015�1049.

Agier, I. and A. Szafarz (2013a). Micro�nance and gender: Is there a glass ceiling on loan

size? World Development 42, 165�181.

Agier, I. and A. Szafarz (2013b). Subjectivity in credit allocation to micro-entrepreneurs:

Evidence from Brazil. Small business economics 41 (1), 263�275.

Alesina, A. F., F. Lotti, and P. E. Mistrulli (2013). Do women pay more for credit? evidence

from italy. Journal of the European Economic Association 11 (s1), 45�66.

Algan, Y. and P. Cahuc (2010). Inherited trust and growth. The American Economic Re-

view 100, 2060�2092.

Altonji, J. G. and R. M. Blank (1999). Race and gender in the labor market. In O. Ashenfelter

and D. Card (Eds.), Handbook of Labor Economics: Volume 3C, pp. 3143�3259. Elsevier.

Ananth, B. (2005). Financing micro�nance � the ICICI Bank partnership model. Small En-

terprise Development 16 (1), 57�65.

Andersen, T. B. and N. Malchow-Møller (2006). Strategic interaction in undeveloped credit

markets. Journal of Development Economics 80 (2), 275�298.

177

References

Anderson, S., J.-M. Baland, and K. O. Moene (2009). Enforcement in informal saving groups.

Journal of Development Economics 90 (1), 14�23.

Andersson, J. P. and E. Wadensjo (2007). Do the unemployed become successful entrepreneurs?

A comparison between unemployed, inactive and wage-earners. International Journal of

Manpower 28, 604�626.

Aniket, K. (2007). Sequential group lending with moral hazard. Edinburgh School of Economics

Discussion Paper No. 136.

Armendariz, B. (2009). Micro�nance for self-employment activities in the European urban

areas: Contrasting Crédal in Belgium and Adie in France. Centre Emile Bernheim Working

Paper No. 09/041.

Armendariz, B., B. D'Espallier, M. Hudon, and A. Szafarz (2013). Subsidy uncertainty

and micro�nance mission drift. Centre Emile Bernheim Working Paper, available at:

http://ssrn.com/abstract=1731265.

Armendariz, B. and J. Morduch (2000). Micro�nance beyond group lending. Economics of

transition 8 (2), 401�420.

Armendariz, B. and J. Morduch (2010). The economics of micro�nance. MIT press.

Armendariz, B. and A. Szafarz (2011). On mission drift in micro�nance institutions. In B. Ar-

mendariz and M. Labie (Eds.), The Handbook of Micro�nance, pp. 341�366. Elsevier.

Arnott, R. and J. E. Stiglitz (1986). Moral hazard and optimal commodity taxation. Journal

of Public Economics 29 (1), 1�24.

Aubert, C., A. de Janvry, and E. Sadoulet (2009). Designing credit agent incentives to pre-

vent mission drift in pro-poor micro�nance institutions. Journal of Development Eco-

nomics 90 (1), 153�162.

Aubier, M. and F. Cherbonnier (2007). L'accès des entreprises au crédit bancaire. Economie

& prévision 1 (177), 121�128.

Ayres, I. and P. Siegelman (1995). Race and gender discrimination in bargaining for a new

car. The American Economic Review 85, 304�321.

Balkenhol, B., C. Guézennec, F. Lainé, and L. Nouaille-Degorce (2013). Le microcrédit pro-

fessionnel en France: quels e�ets sur l'emploi? Conseil d'Analyse Stratégique, Bureau

International du Travail, Finance Solidaire, No. 2013-07.

Barry, A. and O. Bruno (2008). Microcredit supply in developed countries: the role of men-

toring. GREDEG, WP No. 2008 - 04.

178

References

Bates, T. (1990). Entrepreneur human capital inputs and small business longevity. Review of

Economics and Statistics 72, 551�559.

Bauchet, J. and J. Morduch (2013). Is micro too small? microcredit vs. sme �nance. World

Development 43, 288�297.

Bellucci, A., A. Borisov, and A. Zazzaro (2010). Does gender matter in bank � �rm rela-

tionships? Evidence from small business lending. Journal of Banking & Finance 34 (12),

2968�2984.

Benabou, R. and J. Tirole (2002). Self-con�dence and personal motivation. Quarterly journal

of economics 117 (3), 871�915.

Benabou, R. and J. Tirole (2003a). Intrinsic and extrinsic motivation. The Review of Economic

Studies 70 (3), 489�520.

Benabou, R. and J. Tirole (2003b). Self-knowledge and self-regulation: An economic approach.

The psychology of economic decisions 1, 137�167.

Bendig, M., M. Unterberg, and B. Sarpong (2012). Overview of the microcredit sector in

Europe 2010-2011. European Micro�nance Network.

Bennardo, A., M. Pagano, and S. Piccolo (2014). Multiple bank lending, creditor rights, and

information sharing. Review of Finance.

Berge, L. I. O., K. Bjorvatn, and B. Tungodden (2011). Human and �nancial capital for

microenterprise development: Evidence from a �eld and lab experiment. CMI Working

Paper.

Berger, A. N. and G. F. Udell (1998). The economics of small business �nance: The roles

of private equity and debt markets in the �nancial growth cycle. Journal of Banking &

Finance 22 (6), 613�673.

Besley, T. (1994). How do market failures justify interventions in rural credit markets? The

World Bank Research Observer 9 (1), 27�47.

Besley, T. and S. Coate (1995). Group lending, repayment incentives, and social collateral.


Bhola, R., I. Verheul, R. Thurik, and I. Grilo (2006). Explaining engagement levels of oppor-

tunity and necessity entrepreneurs. EIM Business and Policy Research H200610.

Bjørnskov, C. (2009). Social trust and the growth of schooling. Economics of education re-

view 28 (2), 249�257.

179

References

Bjorvatn, K. and B. Tungodden (2010). Teaching business in Tanzania: evaluating participa-

tion and performance. Journal of the European Economic Association 8 (2-3), 561�570.

Blanchard, L., B. Zhao, and J. Yinger (2008). Do lenders discriminate against minority and

woman entrepreneurs? Journal of Urban Economics 63 (2), 467�497.

Blanch�ower, D. G., P. B. Levine, and D. J. Zimmerman (2003). Discrimination in the small-

business credit market. Review of Economics and Statistics 85 (4), 930�943.

Blau, F. D. and L. M. Kahn (2000). Gender di�erences in pay. Journal of Economic Perspec-

tives 14 (4), 75�99.

Block, J. and P. Sandner (2009). Necessity and opportunity entrepreneurs and their duration

in self-employment: evidence from german micro-data. Journal of Industry, Competition

and Trade 9, 117�137.

Boyes, W. J., D. L. Ho�man, and S. A. Low (1989). An econometric analysis of the bank

credit scoring problem. Journal of Econometrics 40 (1), 3�14.

Brabant, M., P. Dugos, and F. Massou (2009). Rapport sur le microcrédit. Inspection Générale

des Finances.

Brana, S. (2013). Microcredit: an answer to the gender problem in funding? Small Business

Economics 40 (1), 87�100.

Bruck, C. (2006). Millions for millions. The New Yorker, October issue.

Bruhn, M. and B. Zia (2012). Stimulating managerial capital in emerging markets: the impact

of business and �nancial literacy for young entrepreneurs. Mimeo. World Bank, Washington

DC.

Bruhn-Leon, B., P.-E. Eriksson, and H. Kraemer-Eis (2012). Progress for micro�nance in

Europe. EIF Working Paper 2012/13.

Calderon, G., J. M. Cunha, and G. De Giorgi (2012). Business literacy and development:

Evidence from a randomized controlled trial in rural Mexico. Mimeo. Stanford University.

Stanford, CA.

Camdessus, M. (2010). Rapport annuel de l'observatoire de la micro�nance. Banque de France.

Card, D., J. Kluve, and A. Weber (2010). Active labour market policy evaluations: A meta-

analysis. The Economic Journal 120 (548), F452�F477.

Carter, S., E. Shaw, W. Lam, and F. Wilson (2007). Gender, entrepreneurship, and bank

lending: The criteria and processes used by bank loan o�cers in assessing applications.

Entrepreneurship Theory and Practice 31 (3), 427�444.

180

References

Casamatta, C. (2003). Financing and advising: optimal �nancial contracts with venture cap-

italists. The Journal of Finance 58 (5), 2059�2086.

Caudill, S. B., D. M. Gropper, and V. Hartarska (2009). Which micro�nance institutions

are becoming more cost e�ective with time? Evidence from a mixture model. Journal of

Money, Credit and Banking 41 (4), 651�672.

Cavalluzzo, K. and J. Wolken (2005). Small business loan turndowns, personal wealth, and

discrimination. The Journal of Business 78 (6), 2153�2178.

Cavalluzzo, K. S., L. C. Cavalluzzo, and J. D. Wolken (2002). Competition, small business

�nancing, and discrimination: Evidence from a new survey. The Journal of Business 75 (4),

641�679.

CGAP and World Bank (2012). A guide to regulation and supervision of micro�nance. Wash-

ington, D.C.: CGAP and World Bank, October.

Christen, R. P. (2001). Commercialization and mission drift. CGAP Occasional Paper No.5.

Coleman, S. (2000). Access to capital and terms of credit: A comparison of men-and women-

owned small businesses. Journal of Small Business Management 38 (3), 48�48.

Conlin, M. (1999). Peer group micro-lending programs in Canada and the United States.


Cooley, C. H. (1902). Human nature and the social order. Transaction Books.

Copisarow, R. (2000). The application of microcredit technology to the UK: Key commercial

and policy issues. Journal of Micro�nance 2 (1), 13�42.

Cornée, S. and A. Szafarz (2013). Vive la di�érence: Social banks and reciprocity in the credit

market. Journal of Business Ethics, 1�20.

Craig, B. R., W. E. Jackson, and J. B. Thomson (2007). Small �rm �nance, credit rationing,

and the impact of SBA-guaranteed lending on local economic growth. Journal of Small

Business Management 45 (1), 116�132.

Craig, B. R., W. E. Jackson III, and J. B. Thomson (2008). Credit market failure interven-

tion: do government sponsored small business credit programs enrich poorer areas? Small

Business Economics 30 (4), 345�360.

CREASOL Annual Report (2012). available at http://www.crea-

sol.fr/upload/documents/creasol_rapportact2012.pdf (accessed May 16, 2013).

CSDL Annual Report (2010). available at http://www.crea-

sol.fr/upload/documents/creasol_rapportact2012.pdf (accessed May 16, 2013).

181

References

Cull, R., A. Demirguc-Kunt, and J. Morduch (2007). Financial performance and outreach: A

global analysis of lending microbanks. The Economic Journal 117, F107�F133.

Cull, R., A. Demirguc-Kunt, and J. Morduch (2009). Micro�nance meets the market. Journal

of Economic Perspectives 23(1), 167�192.

Cull, R., A. Demirgüç-Kunt, and J. Morduch (2011). Does regulatory supervision curtail

micro�nance pro�tability and outreach? World Development 39 (6), 949�965.

Cull, R., A. Demirguc-Kunt, and J. Morduch (2011). Micro�nance trade-o�s: Regulation,

competition and �nancing. In B. Armendariz and M. Labie (Eds.), The Handbook of Mi-

cro�nance, pp. 141�157. Elsevier.

De Bettignies, J.-E. and J. A. Brander (2007). Financing entrepreneurship: Bank �nance

versus venture capital. Journal of Business Venturing 22 (6), 808�832.

De Mel, S., D. McKenzie, and C. Woodru� (2008). Returns to capital in microenterprises:

evidence from a �eld experiment. The Quarterly Journal of Economics 123 (4), 1329�1372.

De Mel, S., D. McKenzie, and C. Woodru� (2014). Business training and female enterprise

start-up, growth, and dynamics: experimental evidence from sri lanka. Journal of Devel-

opment Economics 106, 199�210.

Degryse, H., L. Lu, and S. Ongena (2013). Informal or formal �nancing? Or

both? First evidence on the co-funding of Chinese �rms. Available at SSRN:

http://ssrn.com/abstract=2023751 or http://dx.doi.org/10.2139/ssrn.2023751.

Dehejia, R. H. and S. Wahba (1999). Causal e�ects in nonexperimental studies: Reevalu-

ating the evaluation of training programs. Journal of the American statistical Associa-

tion 94 (448), 1053�1062.

Demirguc-Kunt, A., T. Beck, and P. Honohan (2008). Finance for all?: Policies and Pitfalls

in Expanding Access. Washington DC: The World Bank.

D'Espallier, B., M. Hudon, and A. Szafarz (2013). Unsubsidized micro�nance institutions.

Economics letters 120 (2), 174�176.

DeYoung, R., D. Glennon, and P. Nigro (2008). Borrower� lender distance, credit scoring, and

loan performance: Evidence from informational-opaque small business borrowers. Journal

of Financial Intermediation 17 (1), 113�143.

Drury, D., S. Walsh, and M. Strong (1994). Evaluation of the edwaa job creation demonstra-

tion. Research and Evaluation Report Series 94-G, Washington, D.C.: U.S. Department

of Labor.

182

References

Du�o, E. (2010). Microcrédit, miracle ou désastre ? Le Monde.

Du�os, E. and K. Imboden (2004). The role of governments in micro�nance. CGAP Donor

Brief No. 19, June.

Dunford, C. (2001). Building better lives: Sustainable integration of micro�nance and edu-

cation in child survival, reproductive health, and HIV/AIDS prevention for the poorest

entrepreneurs. Journal of Micro�nance/ESR Review 3 (2), 1�25.

Dunford, C. (2002). What's wrong with loan size? Freedom from Hunger, Davis, CA. Found on

Alternative Finance web site: http://www. alternative-�nance. org. uk/cgi-bin/summary.

pl.

Edgcomb, E. L. (2002). What makes for e�ective micro-enterprise training? Journal of Mi-

cro�nance 4 (1), 99�114.

Ehrlich, I. and G. S. Becker (1972). Market insurance, self-insurance, and self-protection. The

Journal of Political Economy 4 (80), 623�648.

Ellis, R. P. and W. G. Manning (2007). Optimal health insurance for prevention and treatment.

Journal of Health Economics 26 (6), 1128�1150.

Emran, M. S., A. Morshed, and J. E. Stiglitz (2011). Micro�nance and missing markets. MPRA

Paper No. 41451.

European Central Bank (2012). Monthly bulletin. march 2012.

European Commission (2007). Expert group report, regulation of microcredit in Europe.

European Commission (2009). Accompanying document to the Decision of the European Par-

liament and of the Council establishing a European Micro�nance Facility for Employment

and Social Inclusion (Progress Micro�nance Facility), ex ante evaluation; SEC(2009) 907.

Commission Sta� Working Paper, Brussels, 02.07.2009.

European Commission (2012). Report from the commission to the european parliament and

the council (on the application of directive 2006/48/ec to microcredit) brussels, 18.12.2012.

European Commission (2013). SMEs' access to �nance survey.

European Micro�nance Network (2012). A collection of case studies on the legal and regulatory

framework for micro�nance provision. EMN Electronic Bulletin No. 1, December 2012.

Evans, C. E. (2011). The importance of business development services for micro�nance clients

in industrialized countries. Global Microcredit Summit 2011, Commissioned Workshop

Paper.

183

References

Evers, J., S. Lahn, and M. Jung (2007). Status of micro�nance in western Europe, an academic

review. EMN Issue Paper.

Falk, A., D. Hu�man, and U. Sunde (2006). Self-con�dence and search. Technical report, IZA

Discussion Papers.

Fall, F. S. (2010). La complémentarité banque/micro�nance dans le contexte Subsaharien :

Analyse à travers l'optique de G.B. Richardson (1972). Revue Techniques Financières et

Développement 101, 165�176.

Fay, M. and L. Williams (1993). Gender bias and the availability of business loans. Journal of

Business Venturing 8 (4), 363�376.

Filippin, A. and M. Paccagnella (2012). Family background, self-con�dence and economic

outcomes. Economics of Education Review 31 (5), 824�834.

Freixas, X. and J.-C. Rochet (1997). Microeconomics of banking, Second Edition. Cambridge,

MA: MIT Press.

Gale, W. G. (1990). Federal lending and the market for credit. Journal of Public Eco-

nomics 42 (2), 177�193.

Garikipati, S. (2008). The impact of lending to women on household vulnerability and women's

empowerment: evidence from India. World Development 36 (12), 2620�2642.

Ghosh, S. and E. Van Tassel (2008). A model of mission drift in micro�nance institutions.

Department of Economics, Florida Atlantic University, December.

Gilbert, D. T. and D. Silvera (1996). Overhelping. Journal of personality and social psychol-

ogy 70 (4), 678�690.

Giné, X., P. Jakiela, D. Karlan, and J. Morduch (2010). Micro�nance games. American Eco-

nomic Journal: Applied Economics 2 (3), 60�95.

Giné, X. and D. Karlan (2009). Group versus individual liability: Long term evidence prom

philippine microcredit lending groups. Working Papers 970, Economic Growth Center,

Yale University.

Giné, X. and G. Mansuri (2014). Money or ideas? a �eld experiment on constraintsto en-

trepreneurship in rural pakistan. A Field Experiment on Constraintsto Entrepreneurship

in Rural Pakistan (June 1, 2014). World Bank Policy Research Working Paper.

Gomez, R. and E. Santor (2003). Do peer group members outperform individual borrowers?:

A test of peer group lending using Canadian micro-credit data. Bank of Canada.

184

References

Greenberg, D. H., C. Michalopoulos, and P. K. Robins (2003). Meta-analysis of government-

sponsored training programs, a. Industrial and Labor Relations Review 57 (1), 31�53.

Greenwald, B. C. and J. E. Stiglitz (1986). Externalities in economies with imperfect infor-

mation and incomplete markets. The Quarterly Journal of Economics 101 (2), 229�264.

Guérin, I. (2011). The gender of �nance and lessons for micro�nance. In B. Armendariz and

M. Labie (Eds.), The Handbook of Micro�nance, pp. 589�612. Elsevier.

Han, S. (2004). Discrimination in lending: Theory and evidence. The Journal of Real Estate

Finance and Economics 29 (1), 5�46.

Hannam, M. and P. Cheng (2012). Financing sustainable micro-lending in London. Strategic

Change 21 (7-8), 369�378.

Hart, D. M. (2003). The Emergence of Entrepreneurship Policy. Cambridge University Press.

Hartarska, V. and D. Nadolnyak (2007). Do regulated micro�nance institutions achieve better

sustainability and outreach? Cross-country evidence. Applied Economics 39 (10), 1207�

1222.

Heckman, J. J. (1979). Sample selection bias as a speci�cation error. Econometrica 47 (1),

153�161.

Hermes, N. and R. Lensink (2007). The empirics of micro�nance: what do we know? The

Economic Journal 117 (517), F1�F10.

Hermes, N., R. Lensink, and A. Meesters (2011). Outreach and e�ciency of micro�nance

institutions. World Development 39 (6), 938�948.

Herold, F. (2010). Contractual incompleteness as a signal of trust. Games and Economic

Behavior 68 (1), 180�191.

Holvoet, N. (2004). Impact of micro�nance programs on children's education: Do the gender of

the borrower and the delivery model matter? Journal of Micro�nance/ESR Review 6 (2),

27�50.

Hudon, M. and J. Sandberg (2013). The ethical crisis in micro�nance. Business Ethics Quar-

terly 23 (4), 561�589.

Hudon, M. and D. Traça (2011). On the e�ciency e�ects of subsidies in micro�nance: an

empirical inquiry. World development 39 (6), 966�973.

Ishida, J. (2006). Optimal promotion policies with the looking-glass e�ect. Journal of Labor

Economics 24 (4), 857�877.

185

References

Ja�ee, D. and J. E. Stiglitz (1990). Credit rationing. In B. M. Friedman and F. H. Hahn

(Eds.), The Handbook of Monetary Economics, Vol. II, pp. 838�888. London-Singapore:

World Scienti�c Publishing.

Jain, S. (1999). Symbiosis vs. crowding-out: the interaction of formal and informal credit

markets in developing countries. Journal of Development Economics 59 (2), 419�444.

Jain, S. and G. Mansuri (2003). A little at a time: the use of regularly scheduled repayments

in micro�nance programs. Journal of Development Economics 72 (1), 253�279.

Johnson, S. (1998). Micro�nance north and south: contrasting current debates. Journal of

International Development 10 (6), 799�810.

Johnson, S. and M. Nino-Zarazua (2011). Financial access and exclusion in Kenya and Uganda.

The Journal of Development Studies 47 (3), 475�496.

Kamar, E., P. Karaca-Mandic, and E. Talley (2009). Going-private decisions and the Sarbanes-

Oxley Act of 2002: A cross-country analysis. Journal of Law, Economics, and Organiza-

tion 25 (1), 107�133.

Karlan, D. and M. Valdivia (2011). Teaching entrepreneurship: impact of business training on

micro�nance clients and institutions. Review of Economics and Statistics 93 (2), 510�527.

Kraemer-Eis, H., F. Lang, and S. Gvetadze (2013). European small business �nance outlook.

EIF Working Paper 2013/18.

La Porta, R., F. Lopez-De-Silane, A. Shleifer, and R. W. Vishny (1997). Trust in large orga-

nizations. American Economic Review 87, 333�338.

Ladd, H. F. (1998). Evidence on discrimination in mortgage lending. The Journal of Economic

Perspectives, 41�62.

Lalwani, N. and M. Kubzansky (2009). Stretching the fabric of MFI networks. Monitor Inclu-

sive Markets, December.

Lammermann, S., E. Zamorano, and P. Guichandut (2007). Micro�nance and business devel-

opment services in Europe. a guide on good practices. European Micro�nance Network,

April.

Lancaster, T. (1992). The econometric analysis of transition data. Cambridge University Press.

Leatherman, S., M. Metcalfe, K. Geissler, and C. Dunford (2012). Integrating micro�nance

and health strategies: examining the evidence to inform policy and practice. Health Policy

and Planning 27 (2), 85�101.

186

References

Lee, K. (1992). Moral hazard, insurance and public loss prevention. Journal of Risk and

Insurance 59 (2), 275�283.

Lensink, R. and R. Mersland (2009). Micro�nance plus. Working Paper, Kristiansand, Norway,

University of Agder.

Lensink, R., R. Mersland, and V. T. H. Nhung (2011a). Should micro�nance institutions spe-

cialize in �nancial services. In Second International Research Conference on Micro�nance,

Groningen, The Netherlands.

Lensink, R., R. Mersland, and V. T. H. Nhung (2011b). Should micro�nance institutions

specialize in �nancial services? Second International Research Conference on Micro�nance,

Groningen, The Netherlands, June 2011.

Lensink, R. and B. Scholtens (2004). Financial development and the transmission of monetary

shocks. University of Groningen.

Lieberman, I., J. Mudd, and P. Goodeve (2012). US micro�nance at the crossroads. avail-

able at http://www.theomegafoundation.ca/documents/Report - US Micro�nance at the

Crossroads.pdf.

Manigart, S. and C. Struyf (1997). Financing high technology startups in belgium: An explo-

rative study. Small Business Economics 9 (2), 125�135.

Mano, Y., A. Iddrisu, Y. Yoshino, and T. Sonobe (2008). How can micro and small enter-

prises in sub-saharan africa become more productive? the impacts of experimental basic

managerial training. World Development 40 (3), 458�468.

McIntosh, C. and B. Wydick (2005). Competition and micro�nance. Journal of Development

Economics 78 (2), 271�298.

McKenzie, D. and C. Woodru� (2014). What are we learning from business training and

entrepreneurship evaluations around the developing world? The World Bank Research

Observer 29 (1), 48�82.

McKernan, S.-M. (2002). The impact of microcredit programs on self-employment pro�ts: Do

noncredit program aspects matter? Review of Economics and Statistics 84 (1), 93�115.

Mead, G. H. (1934). Mind, self, and society. Chicago: University of Chicago Press.

Mersland, R. and R. Ø. Strøm (2010). Micro�nance mission drift? World Development 38 (1),

28�36.

Mieno, F. and H. Kai (2012). Do subsidies enhance or erode the cost e�ciency of micro�nance?

evidence from m� worldwide micro data. PRIMCED Discussion Paper Series 26.

187

References

Molenaar, K. (2009). Micro�nance, its concepts and development, lessons to draw for europe.

European Micro�nance Network.

Morduch, J. (1999a). The micro�nance promise. Journal of economic literature 37 (4), 1569�

1614.

Morduch, J. (1999b). The role of subsidies in micro�nance: evidence from the Grameen Bank.

Journal of development economics 60 (1), 229�248.

Morduch, J. (2000). The micro�nance schism. World Development 28(4), 617�629.

Munnell, A. H., G. M. Tootell, L. E. Browne, and J. McEneaney (1996). Mortgage lending in

Boston: Interpreting HMDA data. The American Economic Review 86, 25�53.

Orser, B., S. Hogarth-Scott, and A. Riding (2000). Performance, �rm size, and management

problem solving. Journal of Small Business Management 38, 42�58.

Page, M. (1995). Racial and ethnic discrimination in urban housing markets: Evidence from

a recent audit study. Journal of Urban Economics 38 (2), 183�206.

Postelnicu, L., N. Hermes, and A. Szafarz (2013). De�ning Social Collateral in Micro�nance

Group Lending. Working Papers CEB 13-050, ULB � Universite Libre de Bruxelles.

Preyra, C. and G. Pink (2001). Balancing incentives in the compensation contracts of nonpro�t

hospital CEOs. Journal of Health Economics 20 (4), 509�525.

Puhani, P. A. (2012). The treatment e�ect, the cross di�erence, and the interaction term in

nonlinear di�erence-in-di�erences models. Economics Letters 115 (1), 85�87.

Quayes, S. (2012). Depth of outreach and �nancial sustainability of micro�nance institutions.

Applied Economics 44 (26), 3421�3433.

Quinn, S. (2014). The value of corporate governance for bank �nance in an emerging economy:

Evidence from a natural experiment. Journal of Law, Economics, and Organization 30 (1),

1�38.

Reed, L. R. (2013). Vulnerability. The state of the Microcredit Summit Campaign Report.

Riding, A. L. and C. S. Swift (1990). Women business owners and terms of credit: Some

empirical �ndings of the Canadian experience. Journal of Business Venturing 5 (5), 327�

340.

Rosenberg, R. (2007). CGAP re�ections on the Compartamos Initial Public O�ering. Focus

note.

Roszbach, K. (2004). Bank lending policy, credit scoring, and the survival of loans. Review of

Economics and Statistics 86 (4), 946�958.

188

References

Schafer, R. and H. F. Ladd (1981). Discrimination in mortgage lending. MIT Press Cambridge,

MA.

Schreiner, M. (1999). Aspects of outreach: A framework for the discussion of the social bene-

�ts of micro�nance. Center for Social Development Working Paper No. 99-3, Washington

University in St. Louis.

Schreiner, M. and J. Morduch (2001). Replicating micro�nance in the United States: oppor-

tunities and challenges. Replicating Micro�nance in the United States. Washington, DC:

Fannie Mae Foundation.

Stiglitz, J. E. (1990). Peer monitoring and credit markets.World Bank Economic Review 4 (3),

351�366.

Stiglitz, J. E. and A. Weiss (1981). Credit rationing in markets with imperfect information.

The American Economic Review 42 (2), 393�410.

Storey, D. J. (2004). Racial and gender discrimination in the micro �rms credit market?:

Evidence from Trinidad and Tobago. Small Business Economics 23 (5), 401�422.

Swank, O. H. and B. Visser (2007). Motivating through delegating tasks or giving attention.

Journal of Law, Economics, and Organization 23 (3), 731�742.

The French Monetary and Financial Code (2007). art R518-61.

Tirole, J. (2006). The theory of corporate �nance. Princeton University Press.

Townsend, R. M. (2003). Microcredit and mechanism design. Journal of the European Eco-

nomic Association 1 (2-3), 468�477.

Valdivia, M. (2012). Training or technical assistance for female entrepreneurship? evidence

from a �eld experiment in Peru. Mimeo. GRADE.

Valentin, P., T. Mosquera-Yon, and C. Masson (2011). Le microcredit. Rapport du groupe de

travail du Cnis, No. 125, Septembre.

Van Velzen, R. (2014). The impact of business training on business outcomes: a random-

ized control trial on female micro�nance clients in Vietnam. Master thesis, University of

Groningen.

Varian, H. (1990). Monitoring agents with other agents. Journal of Institutional and Theoret-

ical Economics 146, 153�174.

Villeneuve, B. (2000). The consequences for a monopolistic insurance �rm of evaluating risk

better than customers: the adverse selection hypothesis reversed. The Geneva Papers on

Risk and Insurance Theory 25 (1), 65�79.

189

References

Wedig, G. J. (1994). Risk, leverage, donations and dividends-in-kind: A theory of nonpro�t

�nancial behavior. International Review of Economics & Finance 3 (3), 257�278.

Wooldridge, J. M. (2002). Econometric analysis of cross section and panel data. Cambridge:

MIT Press,.

190

theses.frAIX-MARSEILLE UNIVERSITÉ aFculté des Sciences Economiques et de Gestion Ecole Doctorale...

Documents

Transcript of theses.frAIX-MARSEILLE UNIVERSITÉ aFculté des Sciences Economiques et de Gestion Ecole Doctorale...