World Development Vol. 39, No. 6, pp.

938–948, 2011
Ó 2011 Elsevier Ltd. All rights reserved
0305-750X/$ - see front matter
www.elsevier.com/locate/worlddev
doi:10.1016/j.worlddev.2009.10.018

Outreach and Efficiency of Microfinance Institutions

NIELS HERMES
University of Groningen, The Netherlands

ROBERT LENSINK
University of Groningen, The Netherlands
Wageningen University and CREDIT, The Netherlands
University of Nottingham, UK

and
ALJAR MEESTERS *
University of Groningen, The Netherlands
Summary. — This paper uses stochastic frontier analysis to examine whether there is a trade-off between outreach to the poor and effi-
ciency of microfinance institutions (MFIs). We find convincing evidence that outreach is negatively related to efficiency of MFIs. More
specifically, we find that MFIs that have a lower average loan balance (a measure of the depth of outreach) are also less efficient. More-
over, we find evidence showing that MFIs that have more women borrowers as clients (again a measure of the depth of outreach) are less
efficient. These results remain robustly significant after having added a number of control variables.
Ó 2011 Elsevier Ltd. All rights reserved.

Key words — microfinance, outreach, sustainability

1. INTRODUCTION tract increased commercial funds, which may contribute to

supporting the outreach goal of MFIs. They may enlarge the
Microfinance institutions (MFIs) focus on providing credit amount of loans to the poor and/or ensure the provision of
to the poor who have no access to commercial banks, in order such loans for a longer period of time. Thus, the absolute
to reduce poverty and to help the poor with setting up their number of poor people that have access to MFIs may be in-
own income generating businesses. In the literature, this focus creased. Moreover, increased competition, technological
is generally described as outreach. Because providing credit to change, and financial market policies, which focus on
the poor in many cases is a very costly activity, focusing on strengthening market forces and improving the stability of
outreach may, at least potentially, conflict with the financial MFIs, may positively contribute to the efficiency of MFIs.
sustainability of MFIs. Therefore, Western donors and NGOs This, in turn, may help generating more financial resources
have provided financial support by offering MFIs loans with which the poor can be helped. Under these circumstances,
against below-market interest rates, helping them in lending outreach and financial sustainability and efficiency seem to be
to domestic small companies and poor agents. compatible objectives.
Recently, however, there seems to be a shift from subsidiz- Yet, focusing on financial sustainability and efficiency may
ing MFIs institutions to a focus on financial sustainability also go at the cost of lending to the poor. As lending money
and efficiency of these institutions. This goal stresses the to the poor—especially the very poor and/or the rural
importance of being able to cover the cost of lending money poor—can be very costly, the outreach and sustainability goal
out of the income generated from the outstanding loan portfo- may be conflicting. Especially in policy circles there is a hefty
lio and to reduce these costs as much as possible. Among other debate on the compatibility versus the trade-off between sus-
things, this increased focus on financial sustainability and effi- tainability and outreach. Whereas the so-called welfarist view
ciency is due to a number of developments the microfinance stresses the importance of outreach and the threat of focusing
business has been recently confronted with, such as the too much on sustainability, the institutionalist view claims
increasing competition among MFIs, the commercialization that MFIs should focus on sustainability.
of microfinance (i.e., the interest of commercial banks and While from a policy perspective it is very important to know
investors to finance MFIs), technological change that also whether the strife for financial sustainability and efficiency is
has become available for, and implemented in microfinance, compatible or conflicting with the outreach goal, there are sur-
and financial liberalization and regulation policies of the gov- prisingly few studies that have investigated this issue in a sys-
ernment (Rhyne & Otero, 2006). These developments have in- tematic and appropriate way. Most studies only provide
duced microfinance institutions to change their behavior, and anecdotal evidence and use small datasets and/or simple anal-
to broaden their services and activities. yses, except for Cull, Demirgücß-Kunt, and Morduch (2007).
The question that pops up is whether and to what extent
shifting the focus towards increased financial sustainability
and efficiency has implications for the outreach of MFIs. On
*
the one hand, the commercialization of microfinance may at- Final revision accepted: October 30, 2009
938
 OUTREACH AND EFFICIENCY OF MICROFINANCE INSTITUTIONS 939

Our study aims at going beyond the existing empirical anal- $2.3 billion ( CGAP, 2007). Yet, the increased interest from
yses in two important ways. First, we provide an indepth anal- commercial players may have also raised the need for MFIs
ysis of the potential compatibility or trade-off between to become financially sustainable and enhance their efficiency.
efficiency of MFIs and their outreach by using a large dataset, Moreover, two additional recent developments have helped
containing information for a large number of MFIs over a MFIs to improve their sustainability and efficiency. First, new
longer period of time than previous studies in this field. Sec- banking technology, such as charge cards, ATMs, the use of
ond, we use different measures of sustainability. In particular, cell phones, and the internet has begun to enter the microfi-
we use stochastic frontier analysis (SFA)—a technique that nance business, helping to reduce costs and improve the deliv-
has not been used extensively in the field of microfinance— ery of services (Kapoor, Ravi, & Morduch, 2007; Rhyne &
to measure the efficiency of individual MFIs. We then link Otero, 2006). Second, several developing countries have re-
the efficiency measures obtained from the SFA to measures cently liberalized financial markets, while at the same time
of outreach. For the analysis we use data for 435 MFIs, which installing regulations to help in improving the stability of the
we obtained from MixMarkete over the period 1997–2007. microfinance business. These changes of financial market pol-
The remainder of the paper is organized as follows. Section icies may also contribute to improving the sustainability and
2 discusses the literature on the relationship between financial efficiency of microfinance (Hartarska & Nadolnyak, 2007).
sustainability, efficiency, and outreach of MFIs. In Section 3 The above developments and the resulting emphasis on sus-
we set out the research methodology and explain the SFA in tainability and efficiency of MFIs may go at the cost of their
some detail. Section 4 continues with a description of the data- outreach, however. Reaching the poor and providing them
set, after which the estimation results are presented in Section with credit may be very costly. Making very small loans in-
5. In Section 6 we summarize the main findings and provide volves high transaction costs, in terms of screening, monitor-
the conclusions we derive from the analysis. ing, and administration costs, per loan. Several authors,
therefore, argue that the unit transaction costs for small loans
to the poor are high as compared to unit costs of larger loans
2. FINANCIAL SUSTAINABILITY, EFFICIENCY, (Conning, 1999; Hulme & Mosley, 1996; Lapenu & Zeller,
AND OUTREACH: A SHORT DISCUSSION OF 2002; Paxton & Cuevas, 2002). Thus, there may be a trade-
THE LITERATURE off between efficiency and outreach, implying that the shifting
focus towards increasing sustainability and efficiency reduces
Recently, MFIs have been confronted with a number of the scope for the more traditional aim of many MFIs, which
challenges that have affected their way of doing business. 1 is lending to the poor.
First, in several countries competition among MFIs has in- What is the evidence on this trade-off between efficiency and
creased rapidly (Rhyne & Otero, 2006). The consequences of outreach? In policy circles there has been a hefty debate on this
this increased competition for MFIs can be manifold, for issue between the welfarists, who propagate the dominance of
example, lower interest rates, lower costs, more efficiency, the outreach goal (Hashemi & Rosenberg, 2006; Montgomery
and the introduction of new financial services, such as saving & Weiss, 2005; Woller, 2002), and the institutionalists, who
accounts and insurance services. Bolivia is an example of a stress the importance of sustainability and efficiency (Christen,
country that has experienced increasing competition in the 2001; Isern & Porteous, 2005; Rhyne, 1998). Both camps pro-
microfinance industry since the late 1990s. Since then interest vide (in many cases mostly anecdotal) evidence to support
rates have gone down from 30% in 1998 to 21% in 2005. More- their view. Recently, however, representatives from both
over, Bolivian MFIs have become more efficient and they have camps seem to have moved towards the center, concluding
increased the range of financial services they offer to their cli- that, under certain conditions, sustainability and outreach
ents (Rhyne & Otero, 2006). may be compatible (Morduch, 2005).
Second, commercial banks have started to become inter- In the academic literature, however, we find surprisingly few
ested in providing microfinance, since in the past MFIs have rigorous testings of this issue. The most comprehensive study
shown that this can be a successful and profitable business. is from Cull et al. (2007). They examine the financial perfor-
K-REP in Kenya and the Commercial Bank of Zimbabwe mance (using measures of profitability) and outreach in a large
are two examples of commercial banks that have become in- comparative study, based on a new and extensive dataset of
volved in lending to the poor (referred to as “downscaling”) 124 MFIs in 49 countries. The study suggests that MFIs that
recently. 2 Moreover, in some countries the government has focus on providing loans to individuals perform better in
actively stimulated commercial banks to become involved in terms of profitability. Yet, the fraction of poor borrowers
microfinance. 3 Again, this may have put pressure on MFIs and female borrowers in the loan portfolio of these MFIs is
to reduce interest rates and costs and raise efficiency. lower than for MFIs that focus on lending to groups. It also
Third, commercial banks and investors, especially those suggests that individual-based MFIs, especially if they grow
from developed countries, have become increasingly interested larger, focus increasingly on wealthier clients, a phenomenon
in financing MFIs. Large banks such as Citigroup, Deutsche termed as “mission drift.” This mission drift does not occur
Bank, and HSBC, for example, have separate microfinance as strongly for the group-based MFIs. Thus, Cull et al.
divisions, supporting activities of MFIs. The interest of multi- (2007) do find evidence for a trade-off between efficiency and
national banks is due to the so-called “double bottom line” of outreach. In a more recent, largely descriptive paper, Cull,
financing and supporting MFIs: it allows banks and investors Demirgücß-Kunt, and Morduch (2009) provide further evi-
to show their corporate social responsibility, while at the same dence indicating that a trade-off between outreach and com-
time these investments provide attractive risk-return profiles mercialization may exist.
(Deutsche Bank Research, 2007). The first example of com- Cull et al. (2007) support the findings of several earlier stud-
mercial capitalization of MFIs was the creation of an invest- ies, which, however, used less rigorous techniques and/or
ment fund called Profund in 1995, which raised $23 million smaller datasets. Olivares-Polanco (2005) investigates the
to finance Latin American MFIs. In 2006 private investment determinants of outreach in terms of the loan size of MFIs,
funds, also known as microfinance investment vehicles using data for 28 MFIs in Latin America for the years
(MIVs), held portfolios of MFIs shares with a total value of 1999–2001. The analysis includes only one observation for
940 WORLD DEVELOPMENT

each MFI in the dataset. Using simple OLS, Olivares-Polan- analysis (DEA). Desrochers and Lamberte (2003) measure
co’s study confirms the existence of a trade-off between sus- efficiency for a sample of 50 co-operative rural banks, using
tainability and outreach. Makame and Murinde (2006) different methodologies, such as SFA and the distribution free
analyze the outreach versus sustainability trade-off using a bal- approach. They focus on aspects of corporate governance and
anced panel dataset for 33 MFIs in five East African countries show that more efficient rural banks are the ones that are
for the period 2000–05. Using different measures of the depth better able to control agency costs. Gutiérrez-Nieto,
(loan size) and breadth (number of borrowers) of outreach, Serrano-Cinca, and Mar Molinero (2007), who also use
they find strong evidence for a trade-off between outreach DEA, investigate the efficiency of 30 Latin American MFIs
and sustainability and efficiency. In contrast, Gonzalez and and show that differences in efficiency levels are associated
Rosenberg (2006), using data of 2,600 MFI in 2004, suggest with the location of the MFIs (i.e., in which country they
that there seems to be no conflict between financial sustain- are) as well as with their institutional type. Paxton (2007) uses
ability and outreach. Although their dataset looks impressive, SFA to measure the efficiency of 190 Mexican Popular Savings
the limitation of the analysis is that their data are largely self- and Credit Institutions and concludes that differences in effi-
reported and unadjusted (Cull et al., 2009). ciency are associated with differences in technology, average
Navajas, Conning, and Gonzalez-Vega (2003) do not di- loan size, rural outreach, and the age of the institution. Caud-
rectly analyze the existence of the trade-off, but their study ill, Gropper, and Hartarska (2009) use data from 137 MFIs in
may have implications for outreach versus sustainability. They 21 Eastern European and Central Asian countries. They use a
discuss the Bolivian microfinance market developments since mixture modeling approach to estimate cost functions, allow-
the mid-1990s and show that due to increased competition ing for heterogeneity of cost functions of MFIs. Based on their
MFIs changed their lending technologies and the borrowers analysis, they show that MFIs become more efficient over
on which they focus their activities. In particular, their discus- time, yet this is dependent on their size and whether they offer
sion suggests that the new competitor in the market (in Boli- deposits, as well as on the extent to which they receive subsi-
via, this was Caja Los Andes) offered loan contracts that dies.
attracted less poor and more productive borrowers. Conse- We draw two conclusions from the above review of studies
quently, the first mover in the microfinance market (Bancosol) on the efficiency of MFIs. First, none of the above studies
had to adjust its lending policies and, according to Navajas explicitly deals with the question of the trade-off between effi-
et al. (2003), it switched to loan contracts that prevented the ciency and outreach. Although some studies (e.g., Caudill
less poor, more productive borrowers to move to Caja Los et al., 2009; Paxton, 2007) take into account determinants that
Andes. Implicitly, this suggests that competition leads to less are related to outreach, they all focus more generally on the
access to credit for the poorest, that is, less outreach. In a re- determinants of efficiency of these institutions. Second, the
lated paper, McIntosh, De Janvry, and Sadoulet (2005) focus above studies use data from relatively small samples of MFIs
on the effects of increased competition in microfinance. In and restrict themselves to one or a few countries, or to coun-
their study, they empirically show that wealthier borrowers tries located in just one region. Thus, in our view the paper
are likely to benefit from increasing competition among micro- definitely makes an important contribution to the existing lit-
finance institutions, but that it leads to lower levels of welfare erature on the efficiency of MFIs by explicitly focusing on the
for the poorer borrowers. This seems to support the view that trade-off between outreach and efficiency, using information
outreach is hurt by the pressure of competition on the business for a large set of 435 MFIs for which we have over 1,300 insti-
of microfinance. tutional-year observations.
To conclude, the above review shows that there is only lim-
ited empirical evidence on the compatibility or trade-off be-
tween sustainability and outreach of MFIs. The few studies 3. METHODOLOGY
available suggest that there is a trade-off, yet they mostly use
small datasets and/or simple analyses, except for Cull et al. In our analysis we measure cost efficiency in terms of how
(2007). Our study aims at going beyond the existing empirical close the actual costs of the lending activities of an MFI are
analyses in two important ways. First, we use a substantially to what the costs of a best-practice MFI would have been in
larger dataset, containing information for a large number of case it produces identical output under the same conditions.
MFIs over a longer period of time than any of the previous Cost efficiency measures the reduction in cost that could have
studies in this field. Secondly, we use different measures of sus- been achieved if an MFI were both allocatively and technically
tainability. In particular, we look at the cost efficiency of efficient. As cost functions are not directly observable, ineffi-
microfinance institutions. In order to do this, we formulate a ciencies are measured in comparison with an efficient cost
cost function, apply the so-called stochastic frontier analysis frontier. Most studies on cost efficiency use data envelopment
(SFA) to determine a cost frontier, and determine which fac- analysis (DEA) or stochastic frontier analysis (SFA) to calcu-
tors may explain the distance from the best practice cost func- late this frontier. We use SFA, since it controls for measure-
tion (i.e., cost inefficiency). We focus on cost efficiency, and ment errors and other random effects. 4 More specifically, we
not on profit efficiency (which has also been used in the liter- use the SFA suggested by Battese and Coelli (1995), hence-
ature on efficiency of banks), because for MFIs the focus with forth the BC model. An advantage of the BC model as com-
respect to efficiency is not on being profitable as such. Their pared to the standard two-step SFA of Aigner, Lovell, and
ultimate goal is to reduce poverty and they focus more on Schmidt (1977) and Meeusen and van den Broeck (1977) is
how to be cost efficient (and thus financially sustainable) in that the BC model simultaneously estimates the cost frontier
attaining this ultimate goal. and the coefficients of the efficiency variables. Thus, the model
To be sure, we are not the first analyzing the efficiency of we apply is a one-step approach using SFA to investigate the
MFIs and its determinants. Actually, there are a few recent determinants of the inefficiency of individual MFIs vis-à-vis a
studies that have looked into this issue. Qayyum and Ahmad common cost frontier. Wang and Schmidt (2002) show that a
(2006) narrowly focus on measuring the efficiency of 19 MFIs two-step approach suffers from the assumption that the effi-
in three South Asian countries (i.e., without looking at the ciency term is independent and identically truncated, normally
underlying determinants) for which they use data envelopment distributed in the first step, while in the second step the
 OUTREACH AND EFFICIENCY OF MICROFINANCE INSTITUTIONS 941

efficiency terms are assumed to be normally distributed and in Eqn. (5) are taken in logs. We also include a year dummy
dependent on the explanatory variables. Therefore, this meth- (YEAR), which runs from 1 to 11, the square of the year dum-
od inherently renders biased coefficients. my, and its interactions with the input variables to account for
The general BC model specifies a stochastic cost frontier technology changes over time. 7
with the following properties: In order to control for the fact that different types of MFIs
may have different cost functions, we add a vector of dummies
ln C i;t ¼ Cðy i;t ; wi;t ; qi;t ; bÞ þ ui;t þ vi;t ; ð1Þ
for the type of MFI (MFITYPE). In particular, cost functions
where Ci,t is the total cost MFI i faces at time t and may differ between types of MFI due to differences in the levels
Cðy i;t ; wi;t ; bÞ is the cost frontier. In this cost frontier, y i;t repre- of subsidies these institutions receive from outside. The data
sents the logarithm of output of MFI i at time t, wi;t is a vector we use (discussed in more detail below) do not provide de-
of the logarithm of input prices of MFI i at time t, q are MFI tailed information about subsidies received, which stresses
specific control variables and b is a vector of all parameters to the need for adding controls for the MFI type. 8 In the estima-
be estimated. The term ui,t captures cost inefficiency and is tion outcomes discussed below we report the results for the
independent and identically distributed with a truncated nor- specific dummy variables we have created for the type of
mal distribution. 5 vi,t captures measurement errors and ran- MFI. In particular, we have dummy variables for banks
dom effects, for example, good and bad luck, and is (BANK), cooperatives (COOP), non-bank financial institu-
distributed as a standard normal variable. Both ui,t and vi,t tions (NONBANK) non-governmental organizations (NGO),
are time and MFI specific and can be represented as: rural banks (RURBANK), and other organizations (OTHER).
The dummy variable OTHER is left out of the empirical anal-
ui;t  N þ ðmi;t ; r2u Þ; ð2Þ ysis for reasons of singularity.
Finally, we add a number of additional control variables.
vi;t  iidN ð0; r2v Þ: ð3Þ First, we include the equity to total assets ratio (EQUITY)
as a measure of the differences in risk taking by MFIs as is
Next, we model the inefficiency of an MFI as: suggested by, among others, Berger and De Young (1997),
X
mi;t ¼ d0 þ dn zn;i;t : ð4Þ Dietsch and Lozano-Vivas (2000), Lozano-Vivas, Pastor,
n
and Hasan (2001) and Grigorian and Manole (2006). Second,
as some other studies do, we also include loan loss reserves di-
In Eqn. (4), z represents the vector of n variables that deter- vided by gross loans outstanding (LLR) to control for differ-
mine the inefficiency (m) of MFI i at time t. The deltas repre- ences in the risk taking strategies among MFIs (Fries &
sent the coefficients. Eqns. (1) and (4) are solved in one step by Taci, 2005; Lensink, Meesters, & Naaborg, 2008).
using maximum likelihood. As mentioned before, the central aim of the paper is to
For the specification of the cost function we use the model investigate the trade-off (or compatibility) between sustainabil-
developed by Sealey and Lindley (1977), who state that a bank ity and efficiency versus outreach of MFIs. In the framework
acts as an intermediate between funders and borrowers. In we use, mi,t is our measure of inefficiency of an MFI. To ana-
particular, we use total expenses per unit of labor and the lyze the relationship between efficiency and outreach we spec-
interest expenses per unit of deposits held as input prices, ify an empirical model, in which the inefficiency variable is the
whereas we use the gross loan portfolio of an MFI as our mea- dependent variable and in which we have a number of mea-
sure of output. The cost function has a translog specification, sures of outreach. Additionally, we include a number of con-
and can be specified as follows: trol variables that may also influence the inefficiency of MFIs.
ln ðTC i;t Þ ¼ b0 þ b1 ln ðSALARY i;t Þ þ b2 ln ðRi;t Þ þ b3 The general specification of the inefficiency equations we
estimate is as follows:
 ln ðGLP i;t Þ þ b4 ln ðSALARY i;t Þ2 þ b5
mi;t ¼ d0 þ d1 ALBi;t þ d2 WOMAN i;t þ di¼3:::6 LOANTYPEi;t
 ln ðRi;t Þ2 þ b6 ln ðGLP i;t Þ2 þ b7
þ d7 AGEi;t : ð6Þ
 ln ðSALARY i;t Þ ln ðRi;t Þ þ b8
 ln ðSALARY i;t Þ ln ðGLP i;t Þ þ b9 In this equation m stands for the first moment of the ineffi-
ciency distribution for MFI i at time t. The higher this mo-
 ln ðRi;t Þ ln ðGLP i;t Þ þ b10 YEARt þ b11 YEAR2t ment, the more likely it is that the MFI is inefficient. The
first two variables in this equation are generally accepted mea-
þ b12 YEARt ðSALARY i;t Þ þ b13 YEARðRi;t Þ
sures of outreach. They have also been used in other studies
X
17 (e.g., Ferro Luzzi and Weber, 2006; Makame & Murinde,
þ bj MFITYPEi;t þ b18 EQUITY i;t þ b19 LLRi;t 2006; Olivares-Polanco, 2005; Paxton, 2007). These variables
j¼14 are central to our analysis. ALB is the log of the average loan
þ ui;t þ vi;t : ð5Þ balance per borrower (in US dollars). Higher values of ALB
indicate less depth of outreach, since in this case the MFI is
In Eqn. (5) TC represents total costs an MFI faces, SAL- expected to provide fewer loans to poor borrowers. WOMAN
ARY represents the price of a unit of labor for one year, R denotes the percentage of female borrowers in the total loan
is the interest expenses per unit of deposits held, GLP is the portfolio of the MFI. Higher values for this measure indicate
gross loan portfolio, and MFITYPE refers to the type of more depth of outreach, since lending to women is associated
MFI. TC is measured as the total expenses of an MFI; SAL- with lending to poor borrowers.
ARY is measured as the total operating expenses per employee We acknowledge that our measures are perhaps rough
of an MFI; R is the MFI’s total financial expenses per dollar approximations of outreach. First, they cover only one aspect
of deposits; and GLP is the gross loan portfolio of the or dimension, that is, the depth of outreach. As discussed in
MFI. 6 The cost function specification takes into account the Schreiner (2002), outreach may have several dimensions, such
individual input and output variables, the square of these vari- as the value a microfinance loan has for the client (i.e., the
ables, as well as combinations of these variables. All variables worth of the loan), the cost of the loan to the client, the
942 WORLD DEVELOPMENT

breadth of outreach, length of outreach and the scope of out- Table 1A. Description of the panel (MFIs per year)
reach. Yet, as Schreiner (2002) concludes, many of these Year Number of MFIs for which we have data in a particular year
dimensions are difficult to measure. This is especially true in
the setting of the large sample of MFIs we use in our paper. 1997 6
1998 19
Second, Paxton (2003) correctly argues that loan size may be
1999 30
related to the term or type of the loan granted, and/or it
2000 42
may be related to the lending methodology of the MFI. Using
2001 60
average loan size as a measure of outreach means that MFIs
2002 123
targeting service and trading activities will be classified as hav-
2003 190
ing better outreach than MFIs focusing on manufacturing and 2004 243
agricultural activities, assuming that the latter types of activi- 2005 294
ties require larger loans on average. However, given the data 2006 298
restrictions we are confronted with and given the fact that 2007 13
we aim at using a large panel dataset, average loan size and
loans to women are simply the best measures we can come Total 1,318
up with. Moreover, as was mentioned above, these measures
have been used in several other studies as well.
Before developing our expectations with respect to the
trade-off between outreach to the poor and efficiency of an Table 1B. Description of the panel (number of year observations per MFIs)
MFI, we want to stress once again that the dependent variable Number of year observations available Number of MFIs
of Eqn. (6) measures the extent to which an MFI is considered
to be inefficient. This means that we expect that for the out- 1 104
reach variable WOMAN the coefficient will be significantly po- 2 106
sitive. For the outreach variable ALB the expected coefficient 3 85
will be significantly negative. 4 49
5 48
The remaining variables in Eqn. (6) are control variables.
6 19
LOANTYPE is a vector of dummy variables indicating which
7 8
type of loans an MFI mainly provides. The inefficiency of an
8 7
MFI may depend on the type of loans it mainly provides.
9 5
From the literature we know that some types of loans demand
10 4
more efforts than other ones. We have four different dummies, 11 0
indicating that the MFI mainly provides individual loans (IN-
DIV), group loans (GROUP), village loans (VILLAGE), and Total 435
individual, group, and village loans (ALLTYPE). The group
of MFIs not included in one of the four dummies consists of
those institutions that do not report their main lending type.
This group of MFIs is left out of the empirical analysis for rea- Table 2. Loan type and country region
sons of singularity. Region Loan type
AGE is a measure of the age of the MFI, that is, the number
of years since its establishment. Adding this variable allows for Individual Mixed Solidarity Village Total
the possibility to test the hypothesis that older, more experi- Africa 29 87 51 3 170
enced MFIs are more efficient. An alternative hypothesis, East Asia and 49 40 3 1 93
however, may be that older institutions have had to learn the Pacific
how to cope with microfinance practices by trial and error, Eastern Europe 7 13 3 0 23
whereas more recently established institutions may profit from and Central Asia
the knowledge with respect to microfinance practices that has Latin America 99 22 0 0 121
been built-up during the past few decades. In other words, new and the Caribbean
MFIs may leapfrog older institutions in terms of the efficiency South Asia 0 40 7 10 57
of their activities. If the first hypothesis holds, the coefficient
Total 184 202 64 14 464
for AGE is negative and significant, whereas in the second case
the coefficient will be positive. 9

of observations per MFI. For almost 50% of all MFIs in

4. DATA our dataset, we have only one or two observations. Moreover,
there is not a single MFI for which we have data for the entire
The data on MFIs are taken from MixMarkete, a global 1997–2007 period.
web-based microfinance information platform. After adjust- Table 2 provides information on which loan type is used
ments for missing data we have information for 435 MFIs most often by MFIs in different country regions. The informa-
over a period of 11 years (1997–2007). Our full sample consists tion is taken from the Mix Microbanking Bulletin, which
of 1,318 observations. Appendix Table A1 provides the corre- makes a distinction between individual lending, group, or sol-
lation matrix of the variables used in the analysis. Tables 1 de- idarity lending, village bank lending, and mixed lending
scribes the dataset in terms of the number of MFIs per year for (meaning that an MFI does not focus on using either one of
which we have information. As can be seen from Table 1A we the three types of lending). 11 The table shows that especially
have only 6 observations for 1997 and 13 for 2007. 10 In be- the Latin American MFIs in our sample mainly provide indi-
tween these two years, the number of observations increases vidual loans, whereas for African MFIs group lending is rela-
from 19 in 1998 to 298 in 2006. Table 1B shows the number tively more important.
 OUTREACH AND EFFICIENCY OF MICROFINANCE INSTITUTIONS 943

Table 3. Descriptive statistics of outreach measures per loan type presented in columns [1–6]. The results in columns [1–3] are
Individual Mixed Group Village Total based on a cost function excluding the year dummy variables.
In columns [4–6] the results for the three specifications are pre-
Average loan balance sented when we include the year dummy variables in the cost
per borrower
frontier. These six specifications are the ones on which our
Mean 1,132 567 115 85 715
analysis is primarily focused.
St. dev. 904 791 46 38 858
Next, we add different sets of the control variables, which we
Obs. 184 202 64 14 464
have discussed above, to the three specifications of columns
% loans below US$300 [4–6]. These extended specifications are presented in columns
Mean 54 67 94 95 71 [7] and [8]. Since our data have a panel structure, all estima-
St. dev. 21 31 12 7 29 tions have been carried out using pooled regressions.
Obs. 23 42 11 10 86 Panel A of table 4 refers to the estimation results of the cost
% woman borrowers
frontier. A positive coefficient implies an outward shift of the
Mean 43 64 65 97 58
cost function, and hence—ceteris-paribus —higher costs. The
St. dev. 24 26 24 5 27 estimation results for the cost function appear to be as ex-
Obs. 138 168 60 14 380 pected in most cases: the coefficients for SALARY and GLP
are always significant and positive. The coefficient for R is neg-
Average savings ative, which is not as expected. Yet, several of the interaction
balance per saver (US$) and quadratic terms are significant as well and some of them
Mean 1,892 2,332 37 22 1,751 are positive, which makes it difficult to directly observe the
St. dev. 10,415 26,409 60 32 18,412 marginal effect of R on total costs. We use the so-called delta
Obs. 172 183 62 14 431 method to calculate the average marginal effect of R on total
% clients in bottom costs. 13 The calculations show that the marginal effect is sig-
half of the population nificant and positive. 14 This leads us to conclude that our
Mean 12 13 0 50 17 specification of the cost frontier fits the theory reasonably well.
St. dev. 16 6 – 0.5 15 All dummy variables for the type of MFIs, as well as the risk
Obs. 5 21 1 4 31 taking variables EQUITY and LLR are statistically significant
in all specifications in Table 4, indicating that the type of MFI
and the risk taking strategy of an MFI indeed affect the cost
Table 3 provides descriptive statistics with respect to loan frontier.
types and different measures of outreach. The table suggests The coefficient for the year dummy variable, which is in-
that MFIs that focus on group lending and village banking cluded in columns [4–8], is always negative and statistically
are associated more with lending to the poor than MFIs focus- significant, indicating that total costs have reduced over time,
ing on individual lending. For instance, the average loan bal- which we interpret as being the result of technological
ance per borrower is much lower for MFIs focussing on group changes. This may be explained by a learning curve effect:
lending and village lending as compared to MFIs, which focus due to the strong growth of the microfinance business world-
on individual lending. In addition, the percentage of female wide, knowledge and technology has increased and has spilled-
borrowers is higher for MFIs focussing on group and village over, all contributing to making people more experienced in
lending as compared to those MFIs that mainly lend on an managing MFI activities and their costs.
individual basis. 12 Panel B of table 4 refers to the estimation of the inefficiency
equation. The results in columns [1–6] suggest that there is
strong evidence for a trade-off between outreach to the poor
5. ESTIMATION STRATEGY AND RESULTS and efficiency of MFIs. In all equations, the coefficient for
ALB is negative and highly significant. This suggests that
Table 4 provides the estimation results with respect to the MFIs with lower average loan balances (i.e., those that focus
relationship between outreach and efficiency. The procedure more on lending to the poor) are less efficient. In addition,
we have used to generate these results is carried out as follows. the results indicate that MFIs that focus more on female bor-
As was mentioned before, the approach we follow (i.e., the rowers are less efficient, since the coefficient for WOMAN is
SFA proposed by Battese and Coelli (1995)) simultaneously positive and significant, except for the specification presented
estimates the cost frontier and the inefficiency equation. With in column [5].
respect to the cost function we do the estimations using the The results for the outreach variables do not change when
specification of Eqn. (5) without (columns [1–3]) and with we include different sets of control variables. The coefficient
the year dummy and its interactions with the input variables for ALB remains to be negative and highly significant in all
(columns [4–8]). specifications presented in columns [7] and [8]. Also for WO-
As was mentioned above, our main focus is not on the spec- MAN the results remain the same given different specifica-
ification of the cost frontier, but on the specification of the tions. 15
inefficiency equation, and on the trade-off between outreach With respect to the results for the control variables, panel B
and efficiency in particular. With respect to the inefficiency of Table 4 shows the following. First, of the loan type dum-
equation (6) we follow a specific-to-general approach (Brooks, mies included in the analysis GROUP and ALLTYPE always
2002). This allows us to explicitly investigate the sensitivity of have a negative and significant coefficient (see columns [7] and
the results regarding the trade-off between outreach and effi- [8]), indicating that MFI focusing on group lending and/or on
ciency to different specifications of the inefficiency equation. combining group, individual, and village lending are more effi-
We start the analysis by including the two outreach variables cient. Especially the results with respect to GROUP are inter-
in the inefficiency equation separately. We then include both esting. Our results indicate that group lending practices are
outreach variables together in one regression. In panel B of generally less costly, supporting the view that this lending
Table 4, the results for these three different specifications are technique helps reducing information costs related to lending
944 WORLD DEVELOPMENT

Table 4. Results of the estimations

[1] [2] [3] [4] [5] [6] [7] [8]
Panel A – The cost frontier
SALARY 1.995*** 2.115*** 2.118*** 2.021*** 2.125*** 2.128*** 2.135*** 2.010***
(0.187) (0.227) (0.214) (0.186) (0.227) (0.213) (0.211) (0.213)
R 0.150** 0.102 0.249*** 0.130* 0.027 0.198** 0.184** 0.193**
(0.073) (0.096) (0.084) (0.076) (0.099) (0.086) (0.087) (0.086)
GLP 0.699*** 0.669*** 0.648*** 0.696*** 0.673*** 0.627*** 0.671*** 0.716***
(0.071) (0.089) (0.084) (0.072) (0.095) (0.084) (0.085) (0.086)
SALARY  R 0.018** 0.012 0.025*** 0.015* 0.007 0.024*** 0.021** 0.021**
(0.008) (0.010) (0.009) (0.008) (0.011) (0.009) (0.009) (0.009)
R  GLP 0.000 0.001 0.003 0.000 0.001 0.003 0.004 0.005
(0.004) (0.005) (0.005) (0.004) (0.005) (0.005) (0.005) (0.005)
SALARY  GLP 0.026*** 0.025*** 0.016** 0.026*** 0.024*** 0.015* 0.016** 0.01
(0.007) (0.008) (0.008) (0.007) (0.008) (0.008) (0.008) (0.008)
SALARY2 0.067*** 0.081*** 0.082*** 0.072*** 0.086*** 0.085*** 0.085*** 0.082***
(0.013) (0.015) (0.014) (0.013) (0.016) (0.015) (0.015) (0.014)
GLP2 0.012*** 0.013*** 0.012*** 0.012*** 0.013*** 0.012*** 0.011*** 0.007**
(0.002) (0.003) (0.003) (0.002) (0.003) (0.003) (0.003) (0.003)
R2 0.001 0.001 0.001 0.001 0.002 0.001 0.001 0.001
(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
BANK 0.457*** 0.300* 0.471*** 0.402*** 0.219 0.409*** 0.374** 0.383**
(0.149) (0.175) (0.157) (0.150) (0.174) (0.157) (0.158) (0.160)
COOP 1.003*** 0.897*** 1.004*** 0.942*** 0.791*** 0.922*** 0.912*** 0.950***
(0.148) (0.173) (0.155) (0.149) (0.173) (0.156) (0.155) (0.157)
NONBANK 0.706*** 0.511*** 0.730*** 0.650*** 0.415** 0.658*** 0.620*** 0.640***
(0.149) (0.173) (0.157) (0.150) (0.172) (0.156) (0.157) (0.159)
NGO 0.709*** 0.477** 0.751*** 0.650*** 0.342* 0.666*** 0.665*** 0.704***
(0.151) (0.186) (0.160) (0.153) (0.190) (0.160) (0.160) (0.162)
RURBANK 0.956*** 0.841*** 0.982*** 0.885*** 0.713*** 0.891*** 0.904*** 0.983***
(0.152) (0.182) (0.162) (0.153) (0.184) (0.163) (0.163) (0.166)
LLR 2.585*** 2.537*** 2.403*** 2.605*** 2.465*** 2.392*** 2.345*** 2.347***
(0.206) (0.277) (0.248) (0.206) (0.286) (0.247) (0.248) (0.251)
EQUITY 0.396*** 0.242*** 0.342*** 0.392*** 0.251*** 0.347*** 0.287*** 0.292***
(0.058) (0.064) (0.064) (0.058) (0.063) (0.063) (0.065) (0.065)
YEAR 0.190*** 0.204** 0.169** 0.156** 0.133*
(0.062) (0.096) (0.076) (0.075) (0.075)
YEAR2 0.007*** 0.008*** 0.008*** 0.007*** 0.007***
(0.002) (0.003) (0.003) (0.003) (0.003)
SALARY  YEAR 0.009 0.007 0.003 0.003 0.001
(0.006) (0.009) (0.007) (0.007) (0.007)
R  YEAR 0.002 0.003 0.005 0.005 0.005
(0.004) (0.004) (0.004) (0.004) (0.004)
Constant 8.411*** 8.033*** 8.782*** 7.808*** 7.169*** 8.093*** 8.487*** 8.209***
(0.850) (1.084) (1.026) (0.861) (1.113) (1.048) (1.051) (1.048)

Observations 1,304 1,061 1,061 1,304 1,061 1,061 1,061 1,061

Panel B – The inefficiency equation

ALB 0.219*** 0.187*** 0.212*** 0.181*** 0.187*** 0.190***
(0.019) (0.021) (0.019) (0.021) (0.021) (0.021)
WOMAN 0.402*** 0.148** 0.516 0.135** 0.140** 0.139**
(0.150) (0.064) (0.494) (0.062) (0.062) (0.063)
INDIVIDUAL 0.073 0.074
(0.123) (0.123)
GROUP 0.207*** 0.196***
(0.069) (0.069)
VILLAGE 0.041 0.000
(0.047) (0.050)
ALLTYPE 0.089* 0.078*
(0.046) (0.046)
AGE 0.005***
(0.002)

Constant 2.049*** 0.129 1.912*** 2.017*** 0.472 1.943*** 2.004*** 1.893***

(0.119) (0.490) (0.173) (0.120) (1.941) (0.210) (0.196) (0.190)
Observations 1,304 1,061 1,061 1,304 1,061 1,061 1,061 1,061
Standard errors in parentheses.
*
Significant at 10%.
**
Significant at 5%.
***
Significant at 1%.
 OUTREACH AND EFFICIENCY OF MICROFINANCE INSTITUTIONS 945

to the poor more than other lending techniques do. This result method that has been used only very recently in the literature
has also been reported in a recent paper on the determinants of on microfinance.
cost efficiency of MFIs (Caudill et al., 2009).
Third, the coefficient for the variable AGE is positive and
significant, indicating that older MFIs are less efficient (see 6. CONCLUSIONS
column [8]). This supports the view that more recently estab-
lished institutions profit from the knowledge with respect to This paper has used SFA to examine whether there is a
microfinance practices that has been built-up during the past trade-off between outreach to the poor and efficiency of MFIs.
few decades. Based on the existing knowledge base new MFIs Using a sample of more than 1,300 observations, we find con-
may leapfrog older institutions in terms of the efficiency of vincing evidence that outreach is negatively related to effi-
their activities. 16 ciency of MFIs. More specifically, we find that MFIs that
In order to investigate whether our results are influenced by have a lower average loan balance, which is a generally ac-
the fact that for several MFIs we only have a few observations, cepted measure of the depth of outreach, are also less efficient.
we also estimate the models using data of MFIs for which we Moreover, we find evidence showing that MFIs that have
have at least four, respectively, at least five year observations. more women borrowers as clients—again a measure of the
The results are robust for these subsamples of our total data- depth of outreach—are less efficient. These results remain ro-
set. The coefficients for ALB and WOMAN remain to be bustly significant after having added a number of control vari-
highly significant and negative and positive, respectively. 17 ables.
Summarizing the results in Table 4, we find support for In view of the current move to commercialization of the
the fact that there is a trade-off between outreach and effi- microfinance industry this appears to be bad news. Commer-
ciency of MFIs in our sample. These findings remain to be cialization may induce a stronger emphasis on efficiency.
significant even after controlling for a number of control Our study suggests that improving efficiency may only be
variables. Our results do seem to support findings of Cull achieved if MFIs focus less on the poor. It should be noted,
et al. (2007), the only other serious empirical analysis of however, that our results do not necessarily imply that a stron-
the relationship between sustainability/efficiency and out- ger focus on efficiency is bad for poverty reduction. As Zeller
reach, even though they look at profitability rather than effi- and Johannsen (2006) have pointed out, due to spill-over ef-
ciency. Cull et al. (2007) suggest that more profitable MFIs fects, MFIs that strive for efficiency, and score low on out-
focusing on extending individual loans have less poor and reach to the poor, may ultimately cause a higher poverty
female borrowers. Moreover, they show that as these MFIs reduction at the macro level than MFIs that score high on out-
grow larger they increasingly turn to wealthier clients, which reach indicators. This assumes that these efficient MFIs are
indicates “mission drift.” able to contribute to improving economic conditions at the lo-
We believe we have made an important contribution to the cal, regional, and country levels, and that these contributions
scarce academic work on the trade-off discussion. First of all, ultimately are higher than the contributions to poverty allevi-
we use a substantially larger dataset than Cull et al. (2007) ation made by MFI concentrating on outreach rather than
have in their analysis. Moreover, we use information of MFIs efficiency. To the best of our knowledge, until now no study
over a longer period of time than any of the previous studies in has empirically investigated the existence (let alone the size)
this field. Secondly, we use measures of sustainability that are of the effects of increased efficiency of MFIs at the regional
different from these other studies. In particular, we look at the or macro (country) level. Further research is needed to look
cost efficiency of microfinance institutions, applying SFA, a into this issue more carefully.

NOTES

1. This section is partly based on Rhyne and Otero (2006). 5. Thus, the total costs a MFI faces are never lower than the costs of the
frontier. For a graphical representation of the frontier and its dynamics,
2. Bell, Harper, and Mandivenga (2002) describe the process of down- see Berger, Hancock, and Humphrey (1993). The authors show how
scaling of these two banks. Isern, Ritchie, Crenn, Cook, and Brown (2003) inefficiency is determined by both technical and allocative inefficiency.
list 227 local commercial banks that are active in the market for
microfinance. Segrado (2005) describes the involvement of commercial 6. Data for TC, SALARY, R and GLP are not directly available from the
banks in microfinance in Egypt. dataset we have used for this study (MixMarkete; see Section 4 for a
description of this source). Instead, information in terms of ratios, such as
3. In Malaysia, Nepal and Thailand, for instance, the government has total costs to total assets are given only. This is why we have multiplied
initiated programs stimulating commercial banks to develop microfinance these ratios with total assets to obtain data for TC, SALARY, R, and
activities. In India the National Bank of Agriculture and Rural Devel- GLP. Thus, TC is measured as the total expenses to total assets ratio times
opment (NABARD) recently initiated a program to involve private banks total assets in US dollars. SALARY is the operating expenses to total
in microfinance. Seibel and Dave (2002) provide a discussion of the assets ratio times total assets in US dollars, divided by the total number of
commercial aspects of the NABARD program. employees. R is the financial expenses to total assets ratio divided by the
total deposits to total assets. GLP is the gross loan portfolio to total assets
4. Non-parametric techniques do not allow for measurement error and ratio times total assets in US dollars.
luck factors. These techniques attribute any deviation from the best-
practice MFI to technical inefficiency. For a more extensive review of the 7. Including an interaction term of the year dummy with the output
non-parametric and the parametric approach, see Matousek and Taci variable GLP appeared to be problematic for some of the models we
(2004). estimated, possibly due to problems of multicollinearity between this
946 WORLD DEVELOPMENT

interaction term and the other interaction terms that include the year WOMAN. The average savings balance per saver is sometimes used as a
dummy. measure of outreach. Yet, we have not used this measure in our analysis,
since not all MFIs are allowed to offer savings accounts due to legal
8. By adding dummies for different types of MFIs we assume that restrictions. Using this measure would therefore, at least potentially, result
subsidy levels are similar for the same type of MFIs. We admit this is only in biased outcomes.
an approximation of the differences of levels of subsidies.
13. See Oehlert (1992) for a specification of this method.
9. We also included size effects (measured as the number of active
borrowers to which the MFI provides loans) and regional effects (by 14. For completeness, we also calculated the marginal effects of SAL-
including regional dummies for African, and East Asian and Pacific ARY and GLP on total costs with the help of the delta method. The results
countries) in the original empirical specification of the inefficiency show also for these two variables the marginal effect is significant and
equation. Yet, in the empirical analysis these variables turned out not to positive.
be statistically significant, which is why we have excluded them from the
discussion in the main text of the paper. The results of these regressions 15. These results also hold for specifications of the model including size
are available on request from the authors. and region effects (i.e., control variables that were not significant; see
previous Footnote 9).
10. We retrieved the data from the MixMarket website in early 2008. By
that time, only few MFIs had finished their annual report, which explains 16. We also included the dummies for the type of MFI in the inefficiency
why we have a low number of observations for the year 2007. Information equation (the estimation results are available from the authors on request).
on MFIs for the 1990s and early 2000s is available only on a limited scale The results show that if we do this, these dummy variables are all
through the MixMarket website. insignificant. We also find this result if we include the dummies in the
inefficiency equation, but leave them out of the cost frontier estimations.
11. Note that for several MFIs in our dataset information on the loan These results clearly indicate that the type of MFI is primarily relevant for
type is not available, which also why the sum of the rows do not add up to the cost frontier, while there is no indication that it is a relevant issue for
the total amount reported in the last column of Table 2. explaining differences in inefficiency between different MFIs.

12. Unfortunately, for two of the five measures of outreach presented in 17. The estimation results of these robustness checks are available from
the table, we have insufficient information in our dataset, which is why we the authors on request.
only carry out the analysis with the outreach variables ALB and

REFERENCES

Aigner, D., Lovell, C. A. K., & Schmidt, P. (1977). Formulation and Dietsch, M., & Lozano-Vivas, A. (2000). How the environment determines
estimation of stochastic frontier production function models. Journal banking efficiency: A comparison between French and Spanish banks.
of Econometrics, 6(1), 21–37. Journal of Banking and Finance, 24(6), 985–1004.
Battese, G. E., & Coelli, T. J. (1995). A model for technical inefficiency Ferro Luzzi, G., & Weber, S. (2006). Measuring the performance of
effects in a stochastic frontier production function for panel data. microfinance institutions. Unpublished working paper, Geneva: Uni-
Empirical Economics, 20(2), 325–332. versity of Geneva.
Bell, R., Harper, A., & Mandivenga, D. (2002). Can commercial banks do Fries, S., & Taci, A. (2005). Cost efficiency of banks in transition:
microfinance: Lessons from the commercial bank of Zimbabwe and the Evidence from 289 banks in 15 post-communist countries. Journal of
co-operative of Kenya. Small Enterprise Development Journal, 13(4). Banking and Finance, 29(1), 55–81.
Berger, A. N., Hancock, D., & Humphrey, D. B. (1993). Bank efficiency Grigorian, D. A., & Manole, V. (2006). Determinants of commercial bank
derived from the profit function. Journal of Banking and Finance, 17(2– performance in transition: An application of Data Envelopment
3), 317–347. Analysis. Comparative Economic Studies, 48(3), 497–522.
Berger, A. N., & De Young, R. (1997). Problem loans and cost efficiency Gonzalez, A., & Rosenberg, R. (2006). The state of microfinance:
in commercial banks. Journal of Banking and Finance, 21(6), 849–870. Outreach, profitability and poverty. Presentation at the World Bank
Brooks, C. (2002). Introductory econometrics for finance. Cambridge: conference on access to finance.
Cambridge University Press. Gutiérrez-Nieto, B., Serrano-Cinca, C., & Mar Molinero, C. (2007).
Caudill, S. B., Gropper, D. M., & Hartarska, V. (2009). Which Microfinance institutions and efficiency. Omega, 35(2), 131–142.
microfinance institutions are becoming more cost-effective with time: Hartarska, V., & Nadolnyak, D. (2007). Do regulated microfinance
Evidence from a mixture model. Journal of Money, Credit and Banking, institutions achieve better sustainability and outreach: Cross-country
41(4), 651–672. evidence. Applied Economics, 39(10), 1207–1222.
AP, C. G. (2007). CGAP Brief. Washington, DC: CGAP (April). Hashemi, S., & Rosenberg, R. (2006). Graduating the poor into mircofi-
Christen, R. P. (2001). Commercialization and mission drift: The nance: Linking safety nets and financial services. Focus note No. 34,
transformation of microfinance in Latin America. Occasional paper Washington, DC: CGAP.
No. 5, Washington DC: CGAP. Hulme, D., & Mosley, P. (1006). Finance against poverty. Volumes 1 and 2.
Conning, J. (1999). Outreach, sustainability and leverage in monitored London: Routledge.
and peer-monitored lending. Journal of Development Economics, 60(1), Isern, J., Ritchie, A., Crenn, T., Cook, T., & Brown, M. (2003). Review of
51–77. commercial bank and other formal financial institution participation in
Cull, R., Demirgücß-Kunt, A., & Morduch, J. (2007). Financial perfor- microfinance. Washington DC: CGAP.
mance and outreach: A global analysis of lending microbanks. The Isern, J., & Porteous, D. (2005). Commercial banks and microfinance:
Economic Journal, 117(1), F107–F133. Evolving models of success. Focus Note No. 28, Washington, DC:
Cull, R., Demirgücß-Kunt, A., & Morduch, J. (2009). Microfinance meets CGAP.
the market. Journal of Economic Perspectives, 23(1), 167–192. Kapoor, M., Ravi, S., & Morduch, J. (2007). From microfinance to m-
Desrochers, M., & Lamberte, M. (2003). Efficiency and expense preference finance. Innovations: Technology, Governance. Globalization, 2(1–2),
in Philippines’ cooperative rural banks. CIRPEE working paper 03-21, 82–90.
Québec: Université Laval. Lapenu, C., & Zeller, M. (2002). Distribution, growth, and performance of
Research, Deutsche. Bank. (2007). Microfinance. An emerging investment the microfinance institutions in Africa, Asia and Latin America: A
opportunity. Frankfurt am Main: Deutsche Bank. recent inventory. Savings and Development, 26(1), 87–111.
 OUTREACH AND EFFICIENCY OF MICROFINANCE INSTITUTIONS 947

Lensink, R., Meesters, A., & Naaborg, I. (2008). Bank efficiency and Qayyum, A., & Ahmad, M. (2006). Efficiency and sustainability of
foreign ownership: Do good institutions matter?. Journal of Banking microfinance institutions in South Asia. Islamabad: Pakistan Institute of
and Finance, 32(5), 834–844. Development Economics (PIDE).
Lozano-Vivas, A., Pastor, J. T., & Hasan, I. (2001). European bank Rhyne, E. (1998). The yin and yang of microfinance: Reaching the poor
performance beyond country borders: What really matters?. European and financial sustainability. Microfinance Bulletin, 6–8.
Finance Review, 5(1–2), 141–165. Rhyne, E., & Otero, E. (2006). Microfinance through the next decade:
Makame, A. H., & Murinde, V. (2006). Empirical findings on cognitive Visioning the who, what where, when and how. Paper commissioned by
dissonance around microfinance outreach and sustainability. Unpub- the Global Microcredit Summit 2006. Boston, MA: ACCION Interna-
lished paper. University of Birmingham, Birmingham. tional.
Matousek, R., & Taci, A. (2004). Banking efficiency in transition Schreiner, M. (2002). Aspects of outreach: A framework for discussion of
economies: Empirical evidence from the Czech Republic. Economics the social benefits of microfinance. Journal of International Develop-
of Planning, 37(3–4), 225–244. ment, 14(5), 591–603.
McIntosh, C., De Janvry, A., & Sadoulet, E. (2005). How competition Sealey, C. W., Jr., & Lindley, J. T. (1977). Inputs, outputs, and a theory of
among microfinance institutions affects incumbent lenders. The Eco- production and cost at depository financial institutions. Journal of
nomic Journal, 115(506), 987–1004. Finance, 32(4), 1251–1266.
Meeusen, W., & van den Broeck, J. (1977). Efficiency estimation from Segrado, C. (2005). The involvement of commercial banks in microfinance:
Cobb–Douglas production functions with composed error. Interna- The Egyptian experience. Unpublished paper. University of Toronto,
tional Economic Review, 18(2), 435–444. Toronto.
MixMarkete. (2007). <www.mixmarket.org>. visited October. Seibel, H. D., & Dave, H. R. (2002). Commercial aspects of SHG banking
MIX Microbanking Bulletin, Trend Lines 2003–2005 MFI Benchmarks. in India. Paper presented at the seminar on SHG-bank lending
Montgomery, H., & Weiss, J. (2005). Great expectations: Microfinance and programme, New Delhi.
poverty reduction in Asia and Latin America. ADB Research Institute Wang, H. J., & Schmidt, P. (2002). One-step and two-step estimation of
paper series No. 63, Manila: ADB. the effects of exogenous variables on technical efficiency. Journal of
Morduch, J. (2005). Smart subsidies for sustainable microfinance. Finance Productivity Analysis, 18(2), 129–144.
for the Poor: ADB Quarterly Newsletter of the Focal Point for Woller, G. (2002). The promise and peril of microfinance commerciali-
Microfinance, 6, 1–7. zation. Small Enterprise Journal, 13(4), 12–21.
Navajas, S., Conning, J., & Gonzalez-Vega, C. (2003). Lending technol- Zeller, M., & Johannsen, J. (2006). Is there a difference in poverty
ogies, competition and consolidation in the market for microfinance in outreach by type of microfinance institution? The case of Peru and
Bolivia. Journal of International Development, 15(6), 747–770. Bangladesh. Paper presented at the global conference on Access to
Oehlert, G. W. (1992). A note on the delta method. The American Finance: Building Inclusive Financial Systems, organized as part of the
Statistician, 46(1), 27–29. annual conference series of The World Bank and the Brookings
Olivares-Polanco, F. (2005). Commercializing microfinance and deepening Institution in Washington, DC, May 30 and 31.
outreach: Empirical evidence from Latin America. Journal of Micro-
finance, 7(1), 47–69.
Paxton, J. (2003). A poverty outreach index and its application to
microfinance. Economics Bulletin, 9(2), 1–10.
Paxton, J. (2007). Technical efficiency in a semi-formal financial sector:
The case of Mexico. Oxford Bulletin of Economics and Statistics, 69(1), APPENDIX A
57–74.
Paxton, J., & Cuevas, C. (2002). Outreach and sustainability of member- See Table A1.
based rural financial intermediaries. In M. Zeller, & R. L. Meyer
(Eds.), The triangle of microfinance. Financial sustainability, outreach,
and impact. Baltimore and London: Johns Hopkins University Press.

Table A1. Correlation table

TC SALARY R GLP BANK COOP NONBANK NGO RURBANK LLR EQUITY
Panel A
TC 1.000
SALARY 0.610 1.000
R 0.084 0.113 1.000
GLP 0.964 0.548 0.082 1.000
BANK 0.530 0.351 0.013 0.500 1.000
COOP 0.329 0.005 0.310 0.282 0.309 1.000
NONBANK 0.082 0.004 0.125 0.063 0.268 0.386 1.000
NGO 0.140 0.279 0.222 0.172 0.198 0.286 0.248 1.000
RURBANK 0.127 0.115 0.013 0.094 0.158 0.228 0.197 0.146 1.000
LLR 0.098 0.029 0.048 0.019 0.014 0.066 0.067 0.008 0.001 1.000
EQ/TA 0.191 0.165 0.064 0.183 0.135 0.079 0.236 0.068 0.131 0.079 1.000
YEAR 0.060 0.085 0.041 0.099 0.049 0.051 0.082 0.002 0.122 0.066 0.096
INDIVIDUAL 0.204 0.192 0.033 0.204 0.096 0.080 0.004 0.107 0.128 0.021 0.139
GROUP 0.177 0.063 0.144 0.190 0.131 0.165 0.067 0.099 0.118 0.050 0.201
VILLAGE 0.138 0.222 0.098 0.117 0.105 0.152 0.344 0.038 0.078 0.009 0.259
ALLTYPE 0.045 0.089 0.132 0.059 0.048 0.070 0.094 0.058 0.036 0.055 0.051
AGE 0.314 0.067 0.002 0.321 0.009 0.074 0.113 0.013 0.274 0.035 0.158
ALB 0.418 0.651 0.154 0.459 0.373 0.149 0.164 0.402 0.005 0.030 0.299
WOMAN 0.134 0.265 0.235 0.174 0.114 0.206 0.059 0.465 0.021 0.023 0.076
948 WORLD DEVELOPMENT

Table A1—continued
YEAR INDIVIDUAL GROUP VILLAGE ALL-TYPE AGE ALB WOMAN
Panel B
TC
SALARY
R
GLP
BANK
COOP
NONBANK
NGO
RURBANK
LLR
EQ/TA
YEAR 1.000
INDIVIDUAL 0.133 1.000
GROUP 0.133 0.173 1.000
VILLAGE 0.006 0.092 0.097 1.000
ALLTYPE 0.041 0.042 0.045 0.024 1.000
AGE 0.085 0.304 0.054 0.126 0.043 1.000
ALB 0.133 0.193 0.150 0.248 0.138 0.076 1.000
WOMAN 0.001 0.205 0.091 0.059 0.166 0.075 0.462 1.000

Available online at www.sciencedirect.com