Promoting Entrepreneurship at the Base of the Social Pyramid via Pricing Systems: A case Study

• Author: R. Pino‐Mejías,J. Lara‐Rubio,A. Blanco‐Oliver
• Date: 01 January 2017
• Published date: 01 January 2017
• DOI: http://doi.org/10.1002/isaf.1400

RESEARCH ARTICLE

Promoting Entrepreneurship at the Base of the Social Pyramid

via Pricing Systems: A case Study

J. Lara‐Rubio

1

|A. Blanco‐Oliver

2

|R. Pino‐Mejías

3

1

Department of Financial Economics and

Accounting, Faculty of Economics and

Business, University of Granada, Granada,

Spain

2

Department of Financial Economics and

Operation Management, Faculty of Economics

and Business Sciences, University of Seville,

Seville, Spain

3

Department of Statistics and Operational

Research, Faculty of Mathematics, University

of Seville, Seville, Spain

Correspondence

A. Blanco‐Oliver, Department of Financial

Economics and Operation Management,

Faculty of Economics and Business Sciences,

University of Seville, Av. Ramon y Cajal, 1,

41018, Seville, Spain.

Email: aj_blanco@us.es

Summary

Historically, microfinance institutions (MFIs) have played a significant social role by helping

people at the base of the socio‐economic pyramid escape from social exclusion through the cre-

ation of microenterprises. However, international banks have recently started competing in the

microfinance sector. In this adverse environment, MFI management tools should be more innova-

tive and technologically advanced to increase efficiency, solvency and profitabilityand to compete

with commercial banks on equal terms. This study therefore strives to develop a credit‐risk man-

agement tool based on a multilayer perceptron (MLP) credit‐scoring model for a Peruvian MFI, and

to calculate the capital requirements and microcredit pricing on both internal ratings‐based (IRB)

and standardized approaches, analysing the impact of these models on the management of the

MFI. Our findings show that the implementation of an IRB approach with default probabilities

obtained from an MLP credit‐scoring model produces the best benefit by the MFIs in terms of

higher accuracy (reduction of misclassification costs by 13.78%), lower capital requirements

(in the range of 8.5–78%) and the best risk‐adjusted interest rates. Furthermore, with the

establishment of interest rates adjusted to the real risk of each client, MFIs are fairer and more

socially engaged by preventing economically viable low‐risk projects from becoming unviable

due to excessive interest rates. This leads to the creation of more small businesses by people

from the base of the socio‐economic pyramid and greater economic development and social

cohesion. The IRB model should therefore be implemented to improve MFI solvency,

profitability, efficiency, survival, management and social performance.

KEYWORDS

credit risk management, internal ratings‐based approach, microfinance institutions, pricing

1|INTRODUCTION

The high rates of interest charged by the microfinance institutions

(MFIs) are often much higher than normal bank rates (Rosenberg

et al., 2013)—the average effective interest rate of the MIX market

sample MFIs for the year 2009 was 25.38% (Roberts, 2013), oscillating

from 25% to 30% for the period 2004–2011 (Rosenberg et al., 2013).

These high interest rates greatly limit the profitability of new

microenterprises, thereby possibly rendering them unviable in some

cases—in the year 2007 Mexican MFI Compartamos disclosed in its ini-

tial public offering files that it was charging its customers an interest

rate of 90% on microcredits (Malkin, 2008). Another negative point

of the excessively high interest rates is that these could cause debt

traps for the poor (Taylor, 2011). This seriously damages the

reputation that the microfinance industry has as a paradigm to help

the poorest people (Sun and Im, 2015).

To reduce the interest rates is key for MFIs not to lose their social

essence while being able to compete with conventional banks that

have entered the microcredit market. To do so, MFIs should follow

the Basel II internal ratings‐based (IRB) recommendations and custom-

ize the interest rates that they charge their clients through a compre-

hensive measurement of the risks inherent to each borrower.

Nevertheless, in order to assign interest rates adjusted to the risk

of each borrower it is necessary to implement a pricing system under

the Basel II IRB (Basel Committee on Banking Supervision, 2006). This

also permits the determining of the capital requirements MFIs ask for.

Yet since the input variable in order to apply IRB is the probability

of default of each borrower, first of all a credit‐scoring model has to be

Received: 28 June 2015 Revised: 26 July 2016 Accepted: 20 September 2016

DOI 10.1002/isaf.1400

12 Copyright © 2016 John Wiley & Sons, Ltd.Intell Sys Acc Fin Mgmt. 2017;24:12–28. wileyonlinelibrary.com/journal/isaf

developed. Additionally, credit‐scoring systems offer other major

advantages to the financial intermediaries who adopt them: the cost

of credit analysis is reduced, cash flow is improved, faster credit

decisions are enabled, losses are reduced, a closer monitoring of

existing accounts is possible and prioritizing collections are allowed

(West, 2000). Consequently, according to Rhyne and Christen (1999),

credit scoring is one of the most important uses of technology that

may affect the management of MFIs. Therefore, pricing and credit‐

scoring models are two complementary tools based on the IRB

approach that provide MFIs with relevant management improvements

in terms of more efficiency and lower default losses (West, 2000),

more risk‐sensitive capital requirements (Repullo and Suarez, 2004)

and interest rates better adjusted to the risk of each borrower

(Ruthenberg and Landskroner, 2008).

In this framework, the main aim of this study is to analyse the

effects that both the Basel II IRB approach and credit‐scoring models

have on the financial management of MFIs in terms of the calculation

of risk‐adjusted interest rates and the capital requirements needed.

To do so, we use a multilayer perceptron (MLP) credit scoring to

calculate the probability of default and benchmark its performance

against the Basel II standardized approach and also with respect to

two alternative IRB models which determine the probability of default

using logistic regression (LR). Consequently, two research questions

are answered in this paper:

1. Will the effectiveness of the IRB approach outperform the

effectiveness of the standard approach?

2. Will the accuracy capacity of MLP models surpass the perfor-

mance of traditional LR models?

The remainder of our paper proceeds as follows. In Section 2 we

perform a literature review and show the hypotheses that will be

tested in this paper. In Section 3 we provide details of our data set

and undertake a detailed examination of the variables available to

predict default. In Section 4 the results of this study are shown and

there is a discussion of them. In particular, we pay special attention

to the improvement in the MFIs’management that the IRB approach

provides. Then we develop several credit‐scoring models specifically

designed for MFIs. Finally, in Section 5 we provide the main conclu-

sions and implications of this study.

2|EXTANT LITERATURE

2.1 |Credit‐Scoring Models in Microfinance

To date, risk management in microfinance is a field that has been stud-

ied very little, since the advances in risk management are often imple-

mented with substantial delays. The first credit‐scoring model applied

to microfinance was developed by Vigano (1993). He applied discrim-

inant analysis to a sample which contained 100 cases and 53 predictor

variables of a MFI from Burkina Faso. Its main drawback was the small

sample size: only 100 cases. Sharma and Zeller (1997) made a model

for an MFI in Bangladesh. They took a database with 868 borrowers

and, after applying a Tobit methodology based on maximum‐likelihood

estimation, five significant variables were obtained out of the 18 initial

variables. Reinke (1998) used a probit model for the development of a

credit scoring in a South African MFI. To do so, a sample size with

1641 cases was used, obtaining eight significant explanatory variables.

Zeller (1998) designed a credit‐scoring model using the Tobit method-

ology. He used a sample of 168 borrowers of MFI from Madagascar,

getting seven significant variables out of the 18 initial variables.

Schreiner’s (1999) model was implemented in Bancosol, an MFI from

Bolivia. Using a sample of 39,956 borrowers and applying a binary

LR, the nine original variables were incorporated into the final model.

Also in Bolivia, Vogelgesang (2003) showed two models for two differ-

ent MFIs. He implemented a random utility model to two data sets,

one of which contained 8002 cases and the other 5956 cases.

Kleimeier and Dinh (2007) formulated a scoring model for Vietnam’s

retail banking, also using LR. The sample contained 56,037 cases, 17

of the 22 original variables being significant. Finally, Rayo et al.

(2010) designed a model for a Peruvian MFI employing LR. Based on

a sample of 16,157 cases, leaving out 25% of them to validate the

model, they obtained 12 significant variables out of the 41 initially

considered.

Notwithstanding, the main disadvantage of the aforementioned

models proposed by the microfinance literature is that they all have

strong limitations when applied in the real world due to the strict

assumptions (linearity, normality and independence among predictor

variables) of the statistical techniques that have been used for their

development—for more details about the problems of the traditional

statistical methods, see Eisenbeis (1977) and Karels and Prakash

(1987). For this reason, in recent years nonparametric statistical

models such as the k‐nearest neighbour algorithm (Henley and Hand,

1996), support vector machines (Vapnik, 1998), decision tree models

(Davis et al., 1992) and neural network models (Patuwo et al., 1993;

Trinkle, 2005) have been successfully applied to credit‐scoring prob-

lems, even in the microfinance industry (Blanco et al., 2013). Among

these techniques, artificial neural networks (ANNs) are one of the most

powerful tools for pattern classification due to their nonlinear and non-

parametric adaptive‐learning properties. These are the main reasons

for their being used in this study (Trinkle and Baldwin, 2007; Angelini

et al., 2008).

An ANN is a nonparametric technique that emulates the neural

activity in the human brain by transforming inputs into desired outputs

using highly interconnected networks of relatively simple processing

elements, often termed neurons or nodes.

In this sense, Akkoç (2012), based on data from an international

bank operating in Turkey, showed that an ANN improves the results

obtained by parametric approaches (linear discriminant analysis, LR)

but is slightly surpassed by a three‐stage hybrid adaptive neuro fuzzy

inference system credit‐scoring model. Similarly, Cubiles‐de‐la‐Vega

et al. (2013) compared data mining techniques with traditional

methods. They found that an ANN obtains the best accuracy perfor-

mance, above linear discriminant analysis, quadratic discriminant

analysis, LR, support vector machines, classification and regression

trees, bagged classification tree, random forest, adaptive boosting,

binominal boosting and L

2

boosting. Sustersic et al. (2009), on

consumer credit data, also found that ANN credit‐scoring models

reached higher precision ability than LR. Despite this remarkable

J. LARA‐RUBIO ET AL.13