«Abstract In microcredit institutions, credit ocers play a prominent role in loan granting decisions. Indeed, they collect eld data, meet with the ...»
Credit Ocers and Loan Granting in
Micronance: Brazilian Evidence
This version: December 21, 2010
In microcredit institutions, credit ocers play a prominent role
in loan granting decisions. Indeed, they collect eld data, meet with
the applicants, and provide personal recommendations to the credit
committee that takes the nal decisions (loan approval/denial, and
loan size). This paper oers the rst precise quantication of the
degree of inuence of the credit ocers on the nal decision making.
Based on a detailed database from a Brazilian microcredit institution, we are able scrutinize the process that drives the determination of loan size within the institution. The partiality of the credit ocers is analyzed through the lens of gender bias, providing a measure of the extent of the agency problem at stake. The results show that there is indeed a gender gap in loan size generated by the MFI, and that this gap is almost exclusively attributable the credit ocers. In conclusion, this paper conrms that, despite monitoring, the credit ocers remain by far the dominant decision-makers in microloan granting.
Keywords: Microcredit, Gender, Credit Ocer, Loan Size, Allo- cation Process JEL codes: O16, D82, J33, L31.
UMR 201 - Développement et Sociétés (Paris I Sorbonne / IRD) and CERMi, Email:
firstname.lastname@example.org † Université Libre de Bruxelles (ULB), SBS-EM, Centre Emile Bernheim, and CERMi, Email: email@example.com 1 Electronic copy available at: http://ssrn.com/abstract=1729234 (...) the actions of loan ocers have substantial and sometimes unexpected and unintended consequences for the actual direction and outcome of many credit programs. (Dixon, Ritchie and Siwale, 2007, p. 8) 1 Introduction Asymmetric information is the main problem faced by the lending industry (Stiglitz and Weiss, 1981). To tackle this problem, bankers typically com- bine two strategies: credit scoring and relationship lending. On the one hand, credit scoring is the process by which lending institutions assess the creditworthiness of potential borrowers from their personal and/or business characteristics (Hand and Henley, 2007; Lewis, 1994). On the other hand, re- lationship lending is a time-consuming process by which credit ocers learn about their clients' creditworthiness (Berger and Udell, 1995; Boot, 2000) and oer them progressively increasing loans after timely repayments (Egli, 2004).
For reasons likely pertaining to low technology and relatively cheap human capital, credit scoring plays a less prominent role in the microcredit industry 1 than in mainstream banking. As a consequence, credit ocers benet from more leeway to allocate loans Armendáriz and Morduch (2010). This paper oers the rst quantication of the inuence of loan ocers on microcredit granting. This is made possible thanks to access to an exceptionally detailed database from Vivacred, a Brazilian Micronance Institution (MFI).
Microcredit ocers are hard to monitor, notably because of the decentralization of the microlending methodology (Fuentes, 1996; Warning and Sadoulet, 1998; Churchill, 1999; Aubert, de Janvry and Sadoulet, 2009; Dixon, Ritchie and Siwale, 2007). Moreover, the demand for microloans still excess by far the supply (de Janvry, McIntosh and Sadoulet, 2010). Consequently, the credit ocers could be tempted to follow their personal preferences, rather 2 the MFI's optimum, when selecting the borrowers.
On the other hand, most MFIs are socially-oriented, and often subsidized, institutions who need to stick to moral standards. Therefore, they need their sta to make decisions in line with their mission statement and sustainability 1 See Tra and Lensink (2007) for a comparative discussion on the lending practices of formal and informal credit markets.
2 The MFI's optimum needs not be understood as solely referring to prot maximization.
Other optimizations may also be considered here. For instance, Conning (1999); McIntosh and Wydick (2005); Ghosh and Van Tassel (2008); Armendáriz and Szafarz (2010) propose models built on social objective function.
2 Electronic copy available at: http://ssrn.com/abstract=1729234 concern. As a matter of facts, monitoring credit ocers is a major and dicult task for the MFIs' managers.
From the researcher's viewpoint, disentangling objective creditworthiness assessment from the subjective - and hence possibly discriminatory - judgments requires observing the decision process that takes place within the MFI. More precisely, it is necessary to: 1) establish how the credit ocers' recommendations are determined with respect to the applicant's characteristics, and
2) determine to which extent the credit committee critically examines these recommendations when making the nal decision. Moreover, the decisions at stake concern not only loan approval, but also the credit conditions.
The current paper addresses this issue through the lens of disparate treatment. Indeed, previous work (Agier and Szafarz, 2010a,b) has demonstrated that in Vivacred, a Brazilian MFI, women entrepreneurs receive smaller loans than their male counterparts, all other things being equal. Building on these ndings, we dissect here the decision mechanism that leads to such an outcome. Namely, we attribute to the credit ocers and the credit committee their respective responsibilities in the existing gender gap in loan size. The results show that, beside the 65% gender gap attributable to the applicants themselves (women ask for smaller loans), the female loan downsizing is mainly caused by the credit ocers (27%). Notably, instead of correcting the existing gender bias the credit committee tends to marginally reinforce it (7%).
Deriving those estimates is made possible by using partial least square (PLS) regression, and observing the complete three-step process starting with the application (requested amount), then followed by the credit ocer's recommendation, and ending with the credit committee's decision. Indeed, our database allows to trace any loan application that reaches the MFI. On top of these pieces of information, it includes all personal and business characteristics of the loan applicants. By taking into consideration all screening variables collected by the MFI, our results suer as few as possible from the missing variable problem that often plagues studies on creditworthiness assessment (Ross and Yinger, 2002).
The paper is organized as follows. Section 2 describes the database. Section 3 identies each MFI actor's responsibility in female loan downsizing. Section 4 concludes.
3 2 The Loan Granting Process in Vivacred Our unique database comes from Vivacred, a non-prot microcredit institution operating in Rio de Janeiro favelas, over the period 1997-2007 (11 years).
It includes all pieces of information gathered by the six branches of Vivacred.
The study is based on exhaustive data concerning 34,000 applications and 3 32,000 actual loans.
The credit ocers play an important role in clientele selection. Indeed, they are in charge of collecting all relevant information on the loan applicants, and making proposals to the credit committee based on their own evaluation. The full decision process in Vivacred is summarized by gure 1.
The granting process starts with the applicant entering a loan request (step 1). The application les are entrusted with credit ocers on a geographic basis in order to reduce the MFI operational costs. The designated credit ocer meets with the applicant and guarantor, if any, and collects the relevant data, and makes a recommendation to the credit committee (step 2).
This step is particularly demanding since the credit ocer needs to examine in details the applicant's business balance sheet and the household's budget.
Lastly, the full application le - including the credit ocer's recommendation 5
- is examined by the credit committee that has the nal word on the loan approval/denial and the loan size (step 3).
The credit ocer has a face-to-face contact with each applicant, while this is not the case for the credit committee. The credit ocers also spend more 3 The contracts with incomplete specications, the loans to Vivacred's employees, and the few group loans were removed.
4 The information on each applicant includes: private and professional location, birth date, birth state, marital status, gender, dependent(s), profession, bank references, partner's ID, current account, family consumption, family external income, full credit history (as a borrower, a borrower's partner, or a guarantor), the business characteristics and nancial statements, and all credit conditions (loan size, duration, full credit history) if a loan is granted (Agier and Szafarz, 2010b).
5 Actually, the so-called credit committee refers to a single person who is either the branch manager or a senior credit ocer, depending on the requested amount.
4 time on each individual le. For these reasons, it is likely that subjectivity aects the ocer's recommendation more than it aects the committee's decision.
Our dataset contains not only the actual loan contracts, but also all the applications presented to the committee, whether approved or denied. We are therefore able to trace the entire decision-making process that can be decomposed in three steps: 1) the amount requested by the loan applicant,
2) the credit ocer's recommendation, and 3) the nal loan size xed by the credit committee.
Table 1 presents the overall and gender-disaggregated descriptive statistics.
They concern the requested amount, the ocer's proposed amount, and the nal loan size, both in absolute terms and in proportion to the requested amount. In each case, a t-test for equal means between genders is performed.
a All nancial values are in monthly BRL (Real), the Brazilian currency. Over the period under consideration, the BRL uctuated between 0.270 and 0.588 USD.
b t-test for equal means between genders; *** p0.01, ** p0.05, * p0.1 Vivacred claims no special focus on female population. Its clientele is genderbalanced with about 50% women. Men and women face similar approval rates (94.5%), but women receive smaller loans, in absolute terms (BRL 891 versus BRL 1,137) as well as proportionately to the requested amount (78% versus 79.3%). This gender gap shows o at all step of the loan granting process. Indeed, women request smaller loans than men (BRL 1,237 against BRL 1,518). Then, the credit ocers perpetuate the gap gender in their recommendations (BRL 921 against BRL 1,168). Lastly, the nal decision made by the credit committee goes in the same direction (BRL 891 against 1,137).
Unconditional means thus indicate that the credit ocers and the credit committee do not compensate for the initial gender gap. Dierent objective factors could explain this phenomenon. For instance, men and women dier in within-household's situations and in business sizes.
5 In the next section, we will therefore draw regressions controlling for all characteristics pertaining to the borrowers, their businesses and the loan specications. In that way, we will solely keep the component of the gender gap unexplained by the objective characteristics. This will enable us to focus on the subjective appreciations made rst by the credit ocers, and subsequently by the credit committee. As our unique database allows to observe the three steps of the granting process, we are able to determine each actor's responsibility in the gender gap.
3 Who is responsible for the gender gap?
The typical agency problem embedded in microcredit granting is examined by Aubert, de Janvry and Sadoulet (2009). In the same line, Labie et al. (2010) show that credit ocers are more reluctant to serve disabled applicants than other sta members.
Agier and Szafarz (2010b) show that a gender gap in loan size is present beyond all the objective characteristics available to the MFI. Moreover, two thirds of the gender gap is attributed to the dierence in requests between male and female applicants. The remaining third originates from the MFI, underlining the discriminatory aspect of the gender gap. This paper further exploits the same database and disentangles the responsibilities of the credit ocers and the credit committee for this gender gap.
Unlike mortgage loan applications that are typically approved or denied as such, productive loans can be easily sized by the lender. Therefore, observing both the loan size and the requested amount allows to detect credit rationing.
In that way, we fully incorporate the request eect, and avoid distorting the impact of the gender dummy variable.
In the last step, the actual loan size resulting from the credit committee's decision is explained by the gender dummy, the controls, the residual requested amount (due to the client), and the residual proposed amount (due to the
This estimation procedure allows to evaluate the impact of the requested and proposed amounts independently from the control variables. The remaining gender gap, if any, is then attributable to the credit committee. Indeed,
7 its decision by considering the client's characteristics (including gender and requested amount), and the ocer's recommendation. Its decision can thus be gender-related either directly, or through the ocer's recommendation. In the former case, female loans may be lower either because female requests are lower (eR aF ), or because the committee is gender-biased (eF ). In the latter case, the gender gap arising at the committee level is a pure consequence of the ocer's proposal.
In turn, a gender gap emanating from the credit ocer can be attributable to either the applicant's request (eP dR aF ), or to gender-discrimination (eP dF ).
Consequently, the credit committee is potentially contaminated by any of those two sources of female loan downsizing through the ocer's proposal.
Figure 2 represents the four possible channels for a gender gap in loan size.