Efficiency of Microfinance: Do Subsidies Matter?
Abstract:
Application of Data Envelopment Analysis (DEA) to the microfinance institutions is a recent
phenomenon. While no publications include the role of subsidies in their analysis of microfinance
efficiency, this paper measures the impact of various types of subsidies on the efficiency of
microfinance institutions (MFIs). Our results suggest that while subsidies globally decrease
efficiency of MFIs, subsidized borrowings improve the efficiency of MFIs as assessed using the
non‐parametric DEA. Particularly larger institutions are more efficient. Nevertheless, no effect of
the percentage of women, the clientele targeted (outreach) or a specific ownership structure is
found.
Key Words: Microfinance, Subsidies, Efficiency, Non‐parametric analysis
JEL Codes: G21, H2, H21, C14
1. Introduction
Efficiency of the financial institutions is crucial in competitive markets. While efficiency of
conventional financial institutions has often been studied, analyses of the efficiency of
microfinance institutions (MFIs) are less frequent due to the late emergence of this sector.
Nevertheless, many MFIs are still dependent on subsidies, what has brought the debate on their
efficiency under the spotlight. In traditional Banking Literature, the evaluation of financial
performance by using non parametric efficiency techniques i.e. Data Envelopment Analysis (DEA),
is a very common practice (Athanassopoulos (1997); Seiford and Zhu (1999); Camanho and Dyson
(2005)). However, its application to the microfinance institutions is more recent phenomenon.
Exceptions using DEA approach in microfinance are Gutiérrez‐Nieto et al. (2007) and (2009)
analyzing the relationship between social and financial efficiency and Biener and Eling (2010) who
focus on the performance of microinsurance programmes. None of three publications include the
role of subsidies in their analysis of microfinance efficiency.
However the efficiency analysis of MFIs based on conventional production and
intermediation model approach in non‐parametric efficiency analysis framework is hard to justify
because of their reliance on subsidies. The overall equation linking capital and labor inputs into
profits and social change still proves difficult to master without accommodating the subsidized
inputs (Cull et al. 2007). Therefore, measuring their efficiency demands the role of subsidies to be
accounted for, an area, largely neglected in the efficiency and productivity analysis of microfinance
institutions. Moreover, some donors and practitioners are concerned that excessive subsidization
will hamper the promise of sustainability of MFIs but also undercut both scale and efficiency
within the MFI, and possibly distort the market by favouring more inefficient institutions
(Armendariz and Morduch, 2010; Hudon and Traca, forthcoming) . To date only a few studies have
been done which have taken into account the role of subsidies into the assessment of financial
performance or efficiency of MFIs by employing parametric techniques (Cull et al., 2007 ; Hudon &
Traca, forthcoming; Hudon, forthcoming), let alone non‐parametric efficiency analysis. One
exception is Caudill et al. (2009) who use a transcendental logarithmic form to estimate efficiency
on a sample of institutions from Eastern Europe and Central Asia.
This paper fills this gap by incorporating the role of subsidy dependence in the DEA analysis
of MFIs. This quality financial information has been generated directly from the audit reports1 of
the 179 MFIs for two years (2005 and 2006). As a starting point, this essay calculates total
subsidies aggregating the revenue grants received by the MFIs and two opportunity costs: the
concessionary borrowings and the subsidized equity. With subsidy data at our disposal, this study
aims to resolve this key issue. Do these subsidies improve the performance of MFIs by enhancing
their efficiency? To that extent, this study aims to investigate in particular, some specific
hypothesis related to the efficiency of microfinance by employing with and without subsidy
analysis. Another important relationship to be estimated is between the efficiency of MFIs with
their financial sustainability and subsidization.We will also test if outreach, the ownership
structure, the size and the percentage of women of the MFIs have an impact on the global
efficiency.
Our results do not show some positive trends of subsidy dependence on the efficiency of
MFIs. The DEA efficiency scores show only a marginal positive impact of total subsidies on the
financial efficiency of MFIs when they are included as output. On the one hand, Subsidy
dependency clearly decreases efficiency of MFIs. Subsidizing credit lines would however have
better impact in terms of efficiency rather than revenue grants or subsidized equity. On the other
hand, particularly larger institutions are more efficient. No effect of the percentage of women, the
clientele targeted (outreach) or a specific ownership structure is found.
1 The audit reports have been taken from the Mix Market Website (hhp://www.mixmarket.org). The MIX MARKET is a global, web‐based microfinance information platform. It provides information to sector actors and the public at large on Microfinance Institutions (MFIs) worldwide, public and private funds that invest in microfinance, MFI networks, raters/external evaluators, advisory firms, and governmental and regulatory agencies
The paper is organized as follows. In the next section, to start off, we review the literature
on efficiency and public policy in microfinance. In Section 3, we describe the theoretical
background of non parametric efficiency analysis and the subsidy dependence index that we use in
the paper. We then present in Section 4 the database and some basic descriptive statistic. Section
5 provides efficiency results with and without total subsidies as output of the efficiency indicator.
Section 6 highlights the empirical evidence by employing the regression analysis. Finally, a
conclusion is given at the end.
2. Efficiency and Public Policy in Microfinance
Like the conventional financial institutions, the efficiency and productivity of MFIs has
generally been measured by conventional financial ratios or indicators such as staff productivity or
operating expense ratio (Balkenhol, 2007). An example is Hudon and Traca (2009) who use staff
productivity. Most new studies on cost efficiency use more sophisticated indicators of efficiency
such as data envelopment analysis (DEA) or stochastic frontier analysis (SFA) to calculate this
frontier. Gutiérrez‐Nieto et al. (2007) use DEA to a sample of 30 Latin American MFIs to test
twenty‐one specifications. They apply principal component analysis to explain efficiency scores by
means of four principal components. Their results show different rankings using DEA and more
conventional benchmarks and financial indicators.
Hermes et al., (forthcoming) estimate efficiency of 435 MFIs with stochastic frontier
analysis and find that outreach and efficiency of MFIs are indeed negatively correlated, what
suggests some trade‐off between these dimensions. Focusing on the country‐level financial
environment and using the same dataset and methodology, Hermes et al. (2009) show that
financial development and MFIs efficiency are positively correlated. Using data from Eastern
Europe and Central Asia, Caudill et al. (2009) estimate the cost function for MFIs using the
transcendental logarithmic form for all estimations. Their results show that larger MFIs offering
deposits operate more cost effectively over time.
Most microfinance programs have started with the support of donors since the emergence
of the sector during the 70s. After many years of operations, many of these MFIs still rely on
donors’ funds to finance their growth or even cover their operations. Hence a deeper
understanding of the true costs associated with subsidization of microfinance to the society, and
its impact on the efficiency of MFIs is needed. Balkenhol (2007) argues that efficiency should be
regarded as a key indicator by donors who not only focus on social and financial performances.
Although public policy has been largely debated in the microfinance sector, relatively few
studies have provided empirical evidences on the impact of subsidies. Morduch (1999) report the
difficulties to get reliable data on subsidies and all sort of donations. Subsidies traditionally come
from direct grants or donations, that are often reported but may also indirect with in‐kind asset or
training facilities. A large part of them are in fact granted through subsidized credit lines or soft
loans much cheaper than the market rates. These adjustments can make a big difference. For
instance, Morduch (1999) calculated that the sum of the direct and indirect subsidies to the
Grameen between 1985 and 1996 reached $144 million while the Grameen reported $ 1.5million.
As suggested by Hudon and Traca (forthcoming), theoretical arguments on the efficiency
effects of subsidies go in both directions. On the one hand, effects of soft‐budget and the
traditional moral hazard argument, related to the information problems of donors, suggest that
high level of subsidization would decrease the incentive to be efficient. An example of this is Bhutt
and Tang (2001) who argue that subsidies to microfinance NGOs end up funding inefficient and lax
management practices resulting in limited outreach and high loan default. On the other hand,
subsidies granted for infrastructure or human resources can help MFIs to increase their efficiency.
Many institutions that are now profitable and efficient have been subsidized in the past.
Caudill et al. (2009), Cull et al. (2009) and Hudon and Traca (forthcoming) incorporate
measures of subsidies in their analysis of efficiency of MFIs. Cull et al. (2009) analyze standard
efficiency indicators from the Microbanking Bulletin and suggest that subsidy does not necessarily
reduces the efficiency of MFIs but the nongovernmental microfinance organizations do tend to
have higher operating costs. Hudon (forthcoming) find that the level of subsidies granted per year
is related to the management quality but not the subsidies divided by the gross loan portfolio or
the total equity. Caudill et al. (2009) find that MFIs receiving lower subsidies operate more cost
effectively over time. Hudon and Traca (forthcoming) find subsidies have helped MFIs to be more
efficient until a threshold. Nevertheless, Caudill et al. (2009), Cull et al. (2009) and Hudon and
Traca (forthcoming) measure of efficiency is based on the stock of subsidies received in the past,
donated equity. In the next section, we will describe the Data Envelopment Analysis (DEA)
approach and the subsidy indicators.
3. Description of efficiency and subsidy indicators
3.1. Efficiency: DEA Model and Input Output Variable
For the efficiency analysis of the microfinance institutions, a two‐stage analysis has been carried
out. Data Envelopment Analysis (DEA) approach is used to estimate technical and pure efficiency
scores of the MFIs. The advantages of using the DEA technique to gauge efficiency are well
documented in the literature. DEA framework can handle multiple outputs and inputs. Thus, in the
context of MFIs efficiency analysis, it can incorporate both the outputs of outreach and
sustainability along with other inputs into a single framework. Neither has it required any price
information for the dual cost function nor parametric functional form for the production function.
Table 1 depicts the summary of inputs and outputs selected for this study. The main
objective of estimating a production function is to explain the quantity of output produced given
certain levels of inputs and other relevant factors that might explain the quantity of output
produced. In traditional financial literature two models i.e. Production Model and Intermediation
Model are popular depending upon what one thinks an institution do. The majority of the studies
in banking efficiency literature are based on the input oriented constant returns to scale CCR
model (Charnes et al, 1978). In the production model approach, financial institutions are treated
as firms that use physical input, employees and expend money in order to obtain deposits, grant
loans and collect revenues. We assume the output oriented Production model with variable
returns to scale is better suited to microfinance institutions rather than constant returns to scale
model. MFIs are indeed interested in increasing outreach i.e. lending loans to poor people which
commensurate with not only their social mission but also contributes towards sustainability as
well by collecting more revenues from lending. In addition to that they compete in an imperfect
economic environment as the markets for MFIs are not as well developed as the conventional
banking sector. And they always have restricted amount of money and human resource (Inputs) to
spend on unlike commercial banks which can generate money from shareholders. In the context of
output oriented model, this essay asks a specific question “By how much the output quantities are
proportionally expanded without altering the input quantities used?”
The selection of specifications with correct inputs and outputs in the context of MFIs is very
important. Based on the literature, we have selected a few inputs and outputs. This study uses LR‐
ACE2 as a general specification where gross loan portfolio and financial revenues are taken as an
output and assets, operating costs and number of staff as an input. In addition to that, we have
also used specifications L‐ACE and R‐ACE, where the former put emphasis on granting loan as main
objective of MFIs and latter signifies revenue collection as main objective of MFIs. The other
specifications used are basically the different combination of treating subsidies as an input and
output with the above general specifications.
2 The left part in all the specifications show outputs and the right part depict inputs.
Table 1
Inputs and Outputs in Efficiency Specifications
Variable Variable name Not. Definition Unit Papers using this input/ output
Input Total assets A Total of all net asset accounts ($)
Berger and Humphrey (1997),
Seiford and Zhu (1999) and Luo
(2003).
Input Operating Cost C
Expenses related to operations,
such as all personnel expenses,
rent and utilities, transportation,
office supplies, and depreciation
($)
Athanassopoulos (1997), Berger
and Humphrey (1997) and Pastor
(1999).
Input Number of
Staff E
The number of individuals who are
actively employed by the MFI. No.
Athanassopoulos (1997), Berger
and Humphrey (1997), Sherman
and Gold (1985), Seiford and Zhu
(1999) and Luo (2003
Output Gross loan
portfolio L
Outstanding principal balance of
all of the MFI’s outstanding loans ($)
Sherman and Gold, 1985;
Athanassopoulos, 1997; Berger
and Humphrey, 1997; Wheelock
and Wilson, (1999).
Output Financial
revenue R
Revenue generated from the gross
loan portfolio and from
investments plus other operating
revenue
($)
Pastor (1999) and Seiford and Zhu
(1999)
Output Financial
Revenue‐
Subsidy
Rs Financial revenues without Total
subsidies (R‐S) ($)
On top of the traditional variables used in DEA specification, we also include a new output variable
incorporating subsidies. To this end, we remove the total subsidies received by the MFI from its
financial revenues. This is done to adjust the financial revues of the MFI, part out its outputs, as if
they would not have received any subsidy.
Subsidies into DEA Framework
Many indicators of subsidies are used in the literature. An historical indicator is the subsidy
dependence index (SDI). The SDI is ratio of subsidy received by a MFI to revenue from loans to the
target group; it indicates whether a MFI could compensate society for the opportunity cost of
public funds used in a short time frame and still show a profit (Schreiner and Yaron, 2002). Other
indicators used in the literature include donated equity (Caudill et al., 2009), donated equity
divided by total equity (Hudon and Traca, forthcoming), donated equity divided by the age or
outstanding loan of the MFI (Hudon, forthcoming; Hudon and Traca, forthcoming) or subsidized
borrowing (Caudill et al., 2009). Similarly to Caudill et al. (2009), we use a mix of subsidy
indicators.
The Formula for our Subsidy Indicator (SUBLOAN) is:
( )[ ]LPS
LPSubBor
LPSubEq
LPRGcmAmERGSAdjustedTotalSUBLOAN
LPortfoliodingLoansPOuts=++=−×+×+==
tanubsidies
(1)
Where:
LP = Average annual outstanding loan portfolio of the MFI
RG = Revenue Grant
E = average annual equity;
m = Market Interest rate: Interest rate the MFI is assumed to pay for borrowed funds if access to
concessional borrowed funds (A) were eliminated.
A = Average annual outstanding concessionary‐borrowed funds (Average public debt)
c = Interest rate paid on concessionary borrowed funds (Public debt)
Hence, our measure of total subsidies (S) is composed of two parts. The first one are the direct
grants or money transfers (RG) coming from the income statement and which are also used in
Caudill’s (2009) direct grant. Donated equity, used by Hudon and Traca (forthcoming) and Hudon
(forthcoming) is the stock of grants received in the past.
The second part is related to opportunity costs. These are costs that MFIs do not have to
pay, either through subsidised equity (SubEq) or concessionary borrowings (SubBor). Similarly to
Morduch (1999) and Schreiner and Yaron (2002), subsidised equity is calculated by multiplying the
average equity in the year by the commercial market interest rate. Concessionary borrowings are
calculated by multiplying the total borrowings of the MFI by the difference between the market
interest rate and what the MFI actually pays. These three components are part of Yaron (1992)
subsidy dependence index (SDI) who also adds the total profit or loss of the MFI.
4. Database and descriptive statistics
Our data come from the Audit Reports of MFIs taken from Microfinance Information
eXange Inc website. Through this information exchange platform, individual MFI can provide
financial and outreach data and the Mix Market ranks these data for quality using a 5‐star system,
where 5 is the most complete data available, while 1 is the least complete data available (usually
the number of borrowers and some other outreach indicators but little financial information). Only
5‐star MFIs, which include MFIs reporting audited financial statements and a Rating on their online
profile, have been incorporated. 179 MFIs in 54 countries have been chosen based on the clarity
of their respective Audit Reports in general and subsidy figures in particular. The most important
variable to extract from the audit reports for subsidy calculations is the public debt/concessional
borrowing. Therefore MFIs have been selected in large part on the quality and clarity of public
debt figures in their respective audit reports.
Our sample for this essay consists of total 340 observations (170 for the each year 2005
and 2006) which is assumed to be representative of the whole microfinance sector. For example,
the 1,084 MFIs in the 19th MicroBanking Bulletin [MBB] (MicroBanking Bulletin, 2009) yield an
average Operational Sustainability of 111% compared to ours of 124.4%. The average nominal
yield is 31% in the MBB and 30.3% in our database and, finally, the average staff productivity is
103 in the MBB while it is 143 borrowers per staff in our database. As most samples using
databases from rating agencies or Mix Market, our sample is probably biased versus the largest
and better managed around the world. Given the well‐established concentration of microfinance
clients in the largest institutions (Honohan, 2004), the sample is however representative of the
universe of microfinance activity. 46% MFIs of our sample are registered as NGO, followed by the
NBFIs (29%). MFIs with “Bank” status constitute only 16% of total sample. Almost half of the MFIs
(48%) offer both group and individual lending services followed by MFIs which lend exclusively to
the individuals (32%). Geographically one‐third of MFIs locate in Latin America (33%) and almost
one‐fourth in Africa (23%). South Asian MFIs constitute only about 14% of the total MFIs in the
sample.
INSERT TABLE 2 (Appendix)
Table 2 presents a summary statistics of the variables used as an inputs and outputs in the
DEA framework along with other social and organizational variables used in the regression
framework. The statistics for total subsidy and its components (subsidized equity, subsidized
borrowings and revenue grants) are also given.
Table 3 presents the efficiency analysis of the MFIs bifurcated into various categories. The
results for the technical efficiency can been calculated using constant returns to scale efficiency
(crs), variable return to scale efficiency (vrs) and scale efficiency (Scale). However in the Table 3,
the average efficiency scores for specification LR‐ACE have been presented employing only much
realistic output oriented model with variable returns to scale. Which shows that on average CA
&EE and Latin American MFIs are the efficient ones while ME & NA and South Asians are the worst
ones relatively. MFIs with cooperative and NGO status are more efficient than others while MFIs
with individual and village lending methodology are on average more efficient than others.
Moreover MFIs with no saving designs; those which are not regulated and those which do not
provide other services are on average more efficient than their counterparts.
INSERT TABLE 3 (Appendix)
The correlation matrix in Table 4 reveals important relationships among the input‐output
variables used in the efficiency specifications; those used as categorical variables and other
organizational variables. The table stresses that variables are generally correlated between them,
except for subsidies. However, the correlation coefficients remain relatively low. They do not
exceed 0.8, the level at which collinearity problem appears (Kennedy, 2008).
INSERT TABLE 4 (Appendix)
5. Efficiency With and Without Subsidy Analysis
We first compare the average values of the technical efficiency results of with and without
subsidy specifications. The specifications entertained are LR‐ACE vs. LRs‐ACE (without subsidies)
and R‐ACE vs. Rs‐ACE (without subsidies). Since less output is generated without subsidies, the
efficiency will decrease. What is of interest is the magnitude of the decrease.
Comparing the general efficiency specification LR‐ACE (with subsidies) vs. LRs‐ACE (without
subsidies) in Table 5; we can see that efficiency decreases in the three cases when subsidies are
removed. Averages of CRSTE, VRSTE and SE have decreased from 0.582, 0.688, and 0.831 to 0.528,
0.634 and 0.806 respectively. The average efficiency scores can also be interpreted in another way.
For example, scores of 0.582 and 0.688 show that average output of MFIs can be increased by
41.8% and 31.2% with the same use of inputs assuming constant and variable returns to scale
respectively. Considering specification R‐ACE (where MFIs sole objective is to increase revenues),
the decrease in efficiency (from 0.480, 0.566 and 0.840 to 0.266, 0.288 and 0.753 for CRSTE, VRSTE
and SE respectively) is more resounding when subsidies have been subtracted from the revenues
in specification Rs‐ACE (without subsidy). For specification LR‐ACE, 13 and 20 MFIs have become
fully efficient with the injection of subsidies by employing CRSTE and VRSTE respectively. While 22
MFIs become fully scale efficient with the subsidies. On the other hand 11 and 16 MFIs previously
fully efficient become less efficient with subsidies for CRSTE and VRSTE respectively. Further, 10,
32 and 20 MFIs remain fully efficient irrespective of the subsidies for CRSTE, VRSTE and SE
respectively.
Table 5 also shows the results, in terms of the z‐value and significance levels, of the
Wilcoxon signed rank test of the differences between with and without‐subsidy efficiency scores.
The comparison between with and without‐subsidy efficiency scores for specifications LR‐ACE and
R‐ACE indicate that the efficiency of MFIs are significantly reduced (as reflect the positive sign)
without subsidies for all CRSTE, VRSTE and SE.
INSERT TABLE 5 (Appendix)
To sum up, the above efficiency analysis show that subsidies enable to increase efficiencies
significantly when incorporated into the non‐parametric DEA framework. Nevertheless, if we want
to analyze whether subsidies can make a significant difference towards efficiencies, we will
employ a parametric analysis and including some control variables. The following section will
investigate this issue in parametric framework using the Tobit regression analysis.
6. Tobit Regression Approach
6.1. Methodology
Tobit Regression analysis are carried out to test a series of hypotheses concerning the
relationship between financial efficiency and other indicators related to MFIs productivity,
organizational, outreach, sustainability and social impact amid subsidies. The model is censored if
one can at least observe the exogenous variables while it is truncated if the observations outside a
specified range are totally lost” (Amemiya, 1984:3). In this case, a Tobit censored regression model
is appropriate3 because it can accommodate the censored DEA efficiency score since the values of
the dependent variable lie between 0 and 1 with some values achieving the highest value of 1. This
study has taken the output oriented technical efficiency with variable returns to scale as
dependent variable for Tobit regression analysis using the panel data.
The Equation is as follows
+ C +LogWomen + LogAge + hLogOutreac + LogBor + OSS + SUB + SUB + = LogEff
iii7i6
i5i4i32i2i1i
εμβββββββα
(2)
Where iLogEff is the logarithm of the efficiency indicator; iSUB is the subsidy variables,
composed of SUBLOAN (total subsidies divided by loan portfolio), RG (revenue grants by loan
portfolio), SUBBOR (subsidised borrowings divided by loan portfolio) and SUBEQ (subsidised equity
divided by loan portfolio). Similarly to Hudon and Traca (forthcoming) we include the square of the
subsidy variable 2iSUB is the square term of the subsidy variables. iOSS is the operational self‐
sufficiency, iLogBor is the logarithm of the number of borrowers; hLogOutreac is the logarithm of
3 For literature see for example Chakraborty et al., 2001 ; McCarty and Yaisawarng, 1993; Gilen and lall, 1997 and Chilingerian, 1995 among others
our outreach indicator (loan size divided by GNI per capita in purchasing power parity);
LogAgei is the logarithm of the number of years since the MFI has been operating and
iLogWomen is the logarithm of the percentage of female borrowers, for MFIi.
Ci are the controls for Region, Status, Lending Methodology, Saving, Regulated and Other services
i.e. health, education etc in addition to providing financial services or not. Geographic region in
which the MFI operates are classified into 6 regions: Africa (A); East Asia and the Pacific (EA&P);
Eastern Europe and Central Asia (EE&CA); Middle East and North Africa (MENA); Latin America and
the Caribbean (LAC); South Asia (SA). Lending methodology is classified into 4 categories:
Individual (I); Individual & Solidarity/Group (IS); Group/Solidarity (S); Village banking (V). MFIs are
classified into 5 categories of ownership structure: Nongovernmental organizations (NGO); Bank
(B); Non‐banking financial intermediaries (NBFI); Rural Bank (RB); Cooperatives (Coop.).
The omitted variable categories are: for region, Africa; for status, Non Banking Financial
Institution (NBFI); for lending methodology, Group lending; and others are MFIs with no saving
feature, not regulated and no other services.
The base regression above describes the determinants of efficiency with particular
emphasis on the impact of subsidies on the efficiency. The SDI square term is included in the
regression to investigate the potentially marginal impact of subsidization on the efficiency as
suggested by Hudon and Traca (forthcoming). A random‐effect Tobit model has been employed to
analyze the panel data. It should be noted that Honore (1992) has developed a semi parametric
estimator for fixed‐effect Tobit models but there does not exist a sufficient statistic allowing the
fixed effects to be conditioned out of the likelihood. Therefore, unconditional fixed‐effects
estimates are biased (Stata 9 manual).
6.2. Results
Equations (1) ‐ (8) in Table 6 present the overall regression equations with base efficiency
specification LR‐ACE as dependent variable. The sample size consists of 165 MFIs out of 179 MFIs
for which the subsidy figures are available for both years. 14 MFIs have been dropped due to the
unavailability of women borrower information and negative values of subsidized borrowing
variable. Regression 1 includes the quadratic form of composite subsidy indicator (SUBLOAN)
along with all the control variables. The relationship between subsidy dependence and efficiency
does not hold as evident from the insignificant coefficient of the linear term of SUBLOAN.
However, the quadratic term is positive and significant which shows that after the threshold level,
more subsidization start contributing towards efficiency. Nevertheless, only one MFI, FACECAM in
our sample is above this threshold for year 2005. The linear coefficient of subsidy explains
variation in efficiency only once we drop the highly significant borrower variable in regression 2,
where the sign of the linear term becomes negative and significant.
INSERT TABLE 6 (Appendix)
The coefficients of the size (borrowers) in all the equations are also found to be significant
and positive even after dropping the age and OSS variables in Regression 3 & 4 (with quadratic and
with out quadratic term respectively) thus depicting that with increase in clients, MFIs tend to be
more efficient. This is in line with Caudill et al. (2009) who also find that especially large MFIs are
becoming more efficient over time. The age of the institution and the Operational Self Sufficiency
do not explain any variation in efficiency as evident from their insignificant coefficients in all the
regression. The total subsidy (SUBLOAN) in Regression 5 and its decomposition (Regression 6‐8)
into subsidized equity (SUBEQ), subsidized borrowing (SUBBOR) and revenue grants (RG) gives
contrasting but interesting results. Unlike Regression 1, when the quadratic form is not included in
Regression 5, the relationship between efficiency and subsidization becomes positive as evident
from the positive linear term of SUBLOAN though the impact is insignificant. In Regression 6 and 8,
subsidized equity (SUBEQ) and revenue grants (RG) does not explain any variation in the efficiency
as their coefficients are insignificant. However in Regression 7, subsidized borrowings (SUBBOR)
do explain significant variation in the efficiency of MFIs. Its significant positive coefficient depicts
that the subsidized borrowing contributes towards efficiency of MFIs. The regression equations
containing the quadratic terms for all the components of subsidy have also been investigated
(though not presented in the paper) but the impact of both linear and quadratic terms of subsidy
on the efficiency remains insignificant.
Our regression analysis show that lending to women does not explain any variation in
efficiency of microfinance as its positive coefficient is insignificant in all the regressions. Similarly,
no significant relationship has been found between the outreach and efficiency contrary to
Hermes et al., (forthcoming) who found that outreach and efficiency are negatively correlated.
Further, no impact of the ownership status of the MFI has been found on the efficiency. This
would mean that it is not because an MFI is registered as a cooperative, rural bank, NGO or bank
that it is more or less efficient. On the other hand, some evidence has been found that MFIs
offering other services in addition to the provision of credit (credit‐plus) are found to be efficient
only at 10% level in Regression 6 and 8 when subsidized equity and revenue grants explain
variation in efficiency of MFIs respectively. Which depicts that MFIs exhort subsidized funds in the
form of equity and revenue grants are more likely to be efficient in credit‐plus
activities. Interestingly, we have found evidence that MFIs which are regulated are more efficient
in most of the regressions. Thus depicts that MFIs (mostly banks and NBFI status) which are
subject to prudential regulation and supervision are more efficient unlike institutions relying on
other people’s money (e.g. donor‐supported NGOs) which are legally registered, but not regulated
or supervised. For other controls, MFIs lending methodology does not explain any variation in
efficiency. However, much in line with the reality, MFIs located in the Latin America along with
those in CA & EE region are found to be the efficient one unlike their counterparts in South Asia,
which are found to be inefficient.
7. Conclusion
At the outset of this paper, we endeavored to resolve few key issues. How to incorporate
the subsidies into the non parametric DEA framework to investigate the efficiency of microfinance
institutions? What factors are important in determining the efficiency of microfinance and how
much of these factors are driven by the subsidies in determining the efficiency of MFIs? In other
words, How efficiency relates to the various organizational and structural variables amid the
presence of subsidies? The way subsidy has been calculated in this paper, by summing the
subsidized equity, subsidized borrowing and revenue grants, allows us to compare the impact of
different components of subsidies on the efficiency of MFIs. By employing non‐parametric
efficiency framework, a comparison of efficiency scores with and without subsidies depict that the
average efficiency scores are improved when subsidies enter into the DEA framework. Specifically,
there exist numbers of MFIs which become 100% efficient with subsidies and conversely, there
exist MFIs which were previously 100% efficient but become less efficient once subsidies have
been accounted for. The issue of the impact of subsidization on the efficiency in the presence of
several organizational, social and structural variables has been addressed by employing random‐
effect Tobit model.
We find some evidence of positive association between subsidy dependence and efficiency
but it is only established for subsidized borrowing. Nevertheless, no significant effect is found for
total subsidies, subsidized equity and revenue grants. This suggests that subsidizing credit lines
would have better impact in terms of efficiency rather than grants or subsidized equity. Even with
the inclusion of quadratic term of subsidies, the relationship remains insignificant for all the
regressions up to the threshold level after which more subsidization contributes towards
efficiency. Only once we drop the size variable, the relationship become negative and significant
up to the threshold level. The size of the MFI is a major driver of efficiency, MFIs serving a large
number of clients are particularly more efficient that smaller ones. From social perspective,
lending to women borrowers and MFIs outreach do not explain any variations in efficiency.
However we do find evidence that those MFIs, which are regulated and provide credit‐plus
services, are more efficient.
This essay adds to the existing literature by taking on the issue of subsidies for the first
time in evaluating the efficiency of microfinance based on the data of 170 MFIs worldwide. From
policy perspective, valuable lessons can be drawn for the entire stakeholder in microfinance
Industry on the basis of this research work. For microfinance practitioners, it serves as a
performance evaluation guide to enhance the efficiency and in the course of that meeting the dual
objectives of outreach and sustainability. A with and without subsidy analysis based on the
efficiency scores of their respective MFIs can help them identify the efficiency‐enhancing role of
subsidies. In particular, the message is clear for those socially driven MFIs cater to poor and
women borrowers, to devise new income enhancing and enterprise development schemes which
can go a long way in enhancing their efficiency. From private investor’s perspective, it identifies
those MFIs which are successful in achieving maximum efficiency by a proper mix of inputs and
outputs. Even the social investors can benefit by analyzing mission‐driven MFIs in the sample
which are successful in increasing their outreach. For academia and researchers, this research
opens a new avenue of research by examining the role of subsidies on the efficiency of
microfinance in both non‐parametric and parametric framework.
Bibliography
Amemiya, T. (1984). Tobit models: a survey. Journal of Econometrics 24, 3‐61.
Armendáriz de Aghion, B and Jonathan Morduch (2004). Microfinance: Where do we Stand?, in
Charles Goodhart Edition. Financial Development and Economic Growth: Explaining the links
(London: Palgrave)
Athanassopoulos A, D. (1997). Service quality and operating efficiency synergies for management
control in the provision of financial services: Evidence from Greek bank branches. European
Journal of Operational Research Vol. 98, pp. 300‐313.
Balkenhol, B. (2007b). Microfinance, Efficiency Return and Public Policy. Microbanking Bulletin,
Issue 14, Spring.
Banker, R.D., R.F. Charnes, & W.W. Cooper (1984). Some Models for Estimating Technical and
Scale Inefficiencies in Data Envelopment Analysis. Management Science vol. 30, pp. 1078–1092.
Berger, A. N. and Humphrey, D. B. (1997). Efficiency of financial institutions: International survey
and directions for future research. Europeon Journal of Operational Research. Vol. 98, pp. 175‐212.
Camanho, A. S. and Dyson, R. G. (2005). Cost efficiency, production and value‐added models in the
analysis of bank branch performance. Journal of Operational Research Society, Vol. 56, pp. 483‐
494.
Chakraborty, K., B. Biswas, and W.C. Lewis. (2001). Measurement of technical efficiency in public
education: A stochastic and non‐stochastic production function approach. South. Econ, J., Vol. 67,
No. 4, pp. 889‐905.
Charnes, A., Cooper W. W. and Rhodes, E. (1978). Measuring the efficiency of decision making
units. European Journal of Operational Research Vol. 2, pp. 429‐444.
Chilingerian, J.A. (1995). Evaluating physician efficiency in hospitals: A multivariate analysis of best
practices. European Journal of Operational Research 80, 548‐574.
CGAP (2003). Microfinance Consensus Guidelines. Definitions of Selected Financial Terms. Ratios
and Adjustments for Microfinance. 3rd edn. Consultative Group to Assist the Poorest: Washington.
DC. USA.
Congo. Y. (2002). Performance of Microfinance Institutions in Burkina Faso. Discussion Paper No.
2002/01, WIDER, United Nations University, Helsinki, Finland.
Cull. R., Asli Demirguç‐Kunt & Morduch. J. (2007). Financial performance and outreach: a global
analysis of leading microbanks. The Economic Journal, Vol. 117, pp. F107‐F133
Fare, Rolf, S. Grosskopf and M. Norris, Z. Zhang (1994). Productivity Growth, Technical Progress,
and Efficiency Change in Industrialized Countries. American Economic Review (1994) 66‐83.
Farrel, M. (1957). The measurement of productive efficiency. Journal of royal statistical society,
Vol. 120, No. 3, pp. 253‐2811
Gillen, D. and Lall, A. (1997). Developing measures of airport productivity and performance: and
application of Data Envelopment Analysis. Transportation Research – E 33, 4, 261‐273.
Goodman. P. (2005). Microfinance Investment Funds. Key Features, ADA Publications, Luxemburg.
Gutierrez‐Nieto, B., Serrano Cinca C. and Mar Molinero C. (2009). Social Efficiency in Microfinance
Institutions. Journal of the Operational Research Society, Vol. 60 No. 19, pp. 104‐119
Gutiérrez‐Nieto B. and C. Serrano‐Cinca (2007). Factors Explaining the Rating of Microfinance
Institutions. Nonprofit and Voluntary Sector Quarterly, Vol. 36, pp. 439‐ 464.
Gutierrez‐Nieto, B., Serrano Cinca C. and Mar Molinero C. (2007). Microfinance Institutions and
Efficiency. OMEGA: International Journal of Management Science, Vol 35, No. 2, pp. 131‐142
Honoré, B.E. (1992). Trimmed LAD and least squares estimation of truncated and censored
regression models with fixed effects. Econometrica 60, 533–65.
Hermes, Niels & Lensink, Robert & Meesters, Aljar, (forthcoming). Outreach and Efficiency of
Microfinance Institutions. World Development, 39, 2011
Hudon, M. (Forthcoming) Management of microfinance institutions: Do subsidies matter?. Journal
of International Development.
Hudon, M. and Traca, D. (Forthcoming). On the efficiency effects of Subsidies in Microfinance : An
empirical Inquiry. World Development.
Hulme, David and Paul, Mosley (1996). Finance Against Poverty, Volume I and II, London:
Routledge, ISBN 0‐415‐09544‐1 and 0‐415‐12430‐1.
International Monetary Fund. International Financial Statistics, 2005 & 2006, Washington, D.C.
Inter‐American Development Bank (1994). Technical Guide for the Analysis of Microenterprise
Finance Institutions. Microenterprise Division. Washington. D.C.
Luo, X. (2003). Evaluating the profitability and marketability efficiency of large banks. An
application of data envelopment analysis. Journal of Business Research. No.56: 627‐635.
Mixmarket (2007). The Microfinance Information eXchange (MIX). Available at
http://www.mixmarket.org/en/what.is.mix.asp.
McCarty, T., A. and S, Yaisawarng (1993). Technical Efficiency in New Jersey School Districts in
H., Fried, C. A. K. Lovell and S. S. Schmidt, eds., The Measurement of Productive Efficiency:
Techniques and Applications, Oxford University Press, New York, pp. 271‐287.
Morduch, J. (1999a). The Microfinance Promise. Journal of Economic Literature, Vol. 37, pp. 1569‐
1614.
Morduch. J. (1999b). The role of subsidies in microfinance: evidence from the Grameen Bank.
Journal of Development Economics, Vol. 60, pp. 22‐248.
Pastor, J. M. (1999). Efficiency and Risk Management in Spanish Banking: a method to decompose
risk. Applied financial economics, Vol. 9, pp. 371‐384
Schreiner, M. (1997). A Framework For the Analysis of the Performance and Sustainability of
Subsidized Microfinance Organizations With Application to BancoSol of Bolivia and to the
Grameen Bank of Bangladesh. PhD dissertation. The Ohio State University
Schreiner, M., and Yaron, J. (1999). The subsidy Dependence Index and recent attempts to adjust
it. Savings and Development, Vol. 23, No. 4, pp. 375‐405.
Seiford, L. M. and Zhu, J. (1999). Profitability and marketability of the top 55 U.S. commercial
banks. Mngt Sci, Vol. 45, pp. 1270–1288.
Sharma, M. (2004). Subsidy Dependence and Financial sustainability in development banks‐ A case
study of Small Pacific Island Country. The University of South Pacific, Fiji Islands.
Sherman, H. D. and Gold, F. (1985). Bank branch operating efficiency: Evaluation with data
envelopment analysis. J Banking Finance, Vol. 9, pp. 297‐315.
Wheelock, David C & Wilson, Paul W, (1999). Technical Progress, Inefficiency, and Productivity
Change in U.S. Banking, 1984‐1993, Journal of Money, Credit and Banking, Blackwell Publishing,
vol. 31(2), pages 212‐34, May.
Yaron, J. (1992a). Successful Rural Finance Institutions. Discussion Paper No. 150, Washington.
D.C.: World Bank, ISBN 0‐8213‐2018‐1.
Table 2: Variable Description and Summary Statistics
Subsidized Funds Obs Definition Unit Mean Med. Min Max
Total Subsidy/Loan Portfolio The sum of revenue grants, subsidised equity
and concessionary (subsidised) borrowings $ 0.1423 0.0768 0.001 3.46
Subsidized Equity/ Loan
Portfolio 340
Equity grants are the sum of direct grants DG
and paid‐in capital PC. $ 0.0720 0.0489 0.0005 0.8758
Subsidized Borrowings/ Loan
Portfolio 340
Concessionary loans multiplied by the difference
between market interest rate and what MFIs
actually pay.
$ 0.0374 0.0187 ‐0.0424 1.9006
Revenue Grants (RG)/ Loan
Portfolio 340
Cash gifts except for the accounting choice to
record them as revenues rather than as direct
injection to equity.
$ 0.0329 0 0 1.2201
Inputs & Outputs Obs Definition Unit Mean Med. Min Max
Total asset (A) 340
Average of current year (t) and previous year (t‐
1) assets. It includes all asset accounts net of all
contra‐asset accounts, such as the loan‐loss
allowance and accumulated depreciation.
$(000s) 46000 16000 328 566100
Operational cost (C) 340
Expenses related to operations, such as all
personnel expenses, rent and utilities,
transportation, office supplies, and depreciation
$(000s) 4700 2100 96 77300
Staff (E) 340 The number of individuals who are actively
employed by the MFI. No. 620 221 9 24457
Total Subsidy (S) 340 The sum of revenue grants, subsidised equity
and concessionary (subsidised) borrowings $(000s) 2700 1100 6.583 112700
Outstanding Loan Portfolio
(LP) 340
The outstanding principal balance of all of an
MFI’s outstanding loans $(000s) 33000 11000 272 328700
Financial Revenues 340
Revenue generated from the gross loan portfolio
(R) and from investments plus other operating
revenue
$(000s) 9200 3500 71 150000
Organizational variables
Operational Self
Sufficiency(OSS) 340
Financial Revenue (Total)/ (Financial expense +
Loan loss provision expense + Operating
expense)
(%) 124.36 121.44 3.57 254.88
Loansize/GNIpc 340 Average loan size/ GNI per capita $ 0.9438 0.4225 0.0248 31.34
GNI per capita (current) 340 Gross national income divided by the population. $ 1404 1090 160 6070
MFI age 340 The years since MFI has started operations No. 13.92 12 3 51
Borrowers 340
The number of individuals who currently have an
outstanding loan balance with the MFI or are
responsible for repaying any portion of the Gross
No. 112366 26516 949 5163279
Loan Portfolio
Women borrowers 3304 Percentage of borrowers who are women % 64.12 60.6 8.6 102.1
Source: Own calculations
Table 3: Average Efficiency Scores (vrs)
Region Status Lending
methodology Regulated
Savings Providers
Other services
Africa 0.6638 Bank 0.6459 I 0.7048 Not 0.7046 Not 0.6809 Not 0.6852
South Asia 0.6475 Coop. 0.8406 I & S 0.6532 Yes 0.6539 Yes 0.6708 Yes 0.6573
Latin America 0.6944 NBFI 0.6124 S 0.6154
ME & NA 0.5620 NGO 0.7256 V 0.7664
CA & EE 0.7234 R. Bank 0.5298
EA & P 0.6483
Source: Own calculations Average scores are based on total of 340 observations.
4 10 observations have been reduced because five MFIs have no women borrowers information available.
Table 4: Correlation Matrix T. Sub SubEq SubBor RG OSS Age Women Outreach Borr. L. port Rev. Asset Cost Staff Region Status Lending Saving Regulat Other
T. Sub 1.000
SubEq 0.768** 1.000
SubBor 0.798** 0.483** 1.000
RG 0.724** 0.334** 0.310** 1.000
OSS ‐0.259** ‐0.140** ‐0.139* ‐0.31** 1.000
Age ‐0.122* ‐0.126* ‐0.054 ‐0.104 0.071 1.000
Women 0.137** 0.075 0.081 0.156** ‐0.093 0.136* 1.000
Outreach ‐0.034 0.010 ‐0.011 ‐0.073 0.066 ‐0.150** ‐0.297** 1.000
Borr. 0.004 ‐0.023 ‐0.037 0.068 0.279** 0.255** 0.137* ‐0.058 1.000
Input‐output Efficiency Variables
L. port ‐0.141* ‐0.186** ‐0.071 ‐0.079 0.201** 0.202** ‐0.138* 0.036 0.581** 1.000
Rev. ‐0.0809 ‐0.1301 ‐0.024 ‐0.041 0.136 ‐0.189** ‐0.095 0.008 0.539** 0.941** 1.000
Asset ‐0.126* ‐0.173** ‐0.071 ‐0.055 0.167** 0.222* ‐0.143** 0.050 0.588** 0.966** 0.915** 1.000
Cost ‐0.063 ‐0.130* 0.004 ‐0.032 0.031 0.164** ‐0.102 0.027 0.413** 0.905** 0.956** 0.898** 1.000
Staff 0.000 ‐0.038 ‐0.041 0.078 0.228** 0.260** 0.123* ‐0.004 0.972** 0.634** 0.602** 0.659** 0.487** 1.000
Categorical Variables
Region ‐0.090 ‐0.113* ‐0.070 ‐0.028 0.114* 0.124* 0.279** ‐0.176** 0.226** 0.140** 0.136 0.112* 0.091 0.225* 1.000
Status 0.123* 0.083 0.074 0.124* 0.095 0.267** 0.546** ‐0.326** 0.070 ‐0.341** ‐0.299* ‐0.371** ‐0.362** 0.011 0.171** 1.000
Lending 0.123* 0.125* 0.013 0.153** ‐0.166** 0.012 0.544** ‐0.228** ‐0.103 ‐0.293** ‐0.264* ‐0.297** ‐0.242** ‐0.120* 0.098 0.348** 1.000
Saving ‐0.102 ‐0.143** ‐0.016 ‐0.086 ‐0.077 0.162** ‐0.100 0.127* 0.111* 0.224** 0.207** 0.264** 0.235** 0.149* ‐0.167** ‐0.389** ‐0.098 1.000
Regulated ‐0.056 ‐0.047 0.072 ‐0.159** 0.044 ‐0.079 ‐0.258** 0.184** 0.005 0.161** 0.155** 0.192** 0.185** 0.036 ‐0.176** ‐0.370** ‐0.238** 0.336** 1.000
Other 0.099 0.063 0.090 0.071 0.019 0.103 0.389** ‐0.152** 0.168** ‐0.018 ‐0.005 ‐0.028 ‐0.028 0.156* 0.252** 0.345** 0.256** ‐0.167** ‐0.226** 1.000
Source: Authors own calculations. Total No. of observations are 340. *Significance level at 5%; **Significance level at 1%
Table 5: Efficiency Analysis (With and Without Subsidies)
LR‐ACE Wilcoxon
Signed ranked test
R‐ACE Wilcoxon
Signed ranked test
technical efficiency
with subsidies
without subsidies
with
subsidies without subsidies
constant (CRSTE)
0.582 0.528 8.254** p>0.000
0.480 0.266 13.697** p>0.000
variable (VRSTE)
0.688 0.634 7.193** p>0.000
0.566 0.288 13.905** p>0.000
scale efficiency (SE)
0.831 0.806 7.954** p>0.000
0.840 0.753 9.613** p>0.000
Source: Author’s own calculations. All values are average efficiencies of the total MFIs.
Table 6: Tobit Regression Analysis (Random Effect Model) Dependent variable: Efficiency (LR‐ACE)
z‐values in parentheses *significant at 10%; ** significant at 5%; *** significant at 1%. Source: Authors calculations based on data compiled from the audit reports of MFIs and from the Mix Market website.
(1) (2) (3) (4) (5) (6) (7) (8)
LogAge ‐0.070 (‐1.30)
0.014 (0.27)
‐0.065 (‐1.20)
‐0.067 (‐1.24)
‐0.070 (‐1.29)
‐0.065 (‐1.20)
LogOSS 0.051 (0.65)
0.099 (1.24)
0.053 (0.67)
0.035 (0.45)
0.054 (0.69)
0.046 (0.56)
LogBorrower 0.086
(4.02)*** 0.077
(3.96)*** 0.082
(4.22)*** 0.089
(4.20)*** 0.084
(3.85)*** 0.093
(4.38)*** 0.085
(4.08)*** Total Subsidy/loan (SUBLOAN) Subsidized Equity/loan(SUBEQ)
Subsidized borrowing/loan (SUBBOR)
Revenue Grants/loan (RG)
‐0.204 (‐1.08)
‐0.345 (‐1.82)*
‐0.204 (‐1.09)
0.089 (0.98)
0.099 (1.06)
‐0.070 (‐0.29)
0.382 (2.02)**
0.090 (0.42)
SUBLOAN‐sqr 0.123 (1.85)*
0.150 (2.21)**
0.120 (1.79)*
Loan Size/GNIpc ‐0.008 (‐0.26)
‐0.035 (‐1.18)
‐0.008 (‐0.29)
‐0.007 (‐0.23)
‐0.006 (‐0.20)
‐0.010 (‐0.35)
‐0.005 (‐0.18)
‐0.008 (‐0.28)
Women Borrower 0.016 (0.23)
0.025 (0.35)
0.008 (0.11)
0.012 (0.17)
0.020 (0.28)
0.020 (0.28)
0.017 (0.25)
0.020 (0.28)
Bank ‐0.069 (‐0.90)
‐0.031 (‐0.40)
‐0.065 (‐0.87)
‐0.070 (‐0.94)
‐0.073 (‐0.95)
‐0.069 (‐0.89)
‐0.073 (‐0.95)
‐0.069 (‐0.89)
Cooperatives 0.129 (1.11)
0.083 (0.70)
0.099 (0.86)
0.117 (1.02)
0.147 (1.26)
0.130 (1.12)
0.153 (1.33)
0.137 (1.18)
NGOs ‐0.027 (‐0.43)
‐0.048 (‐0.74)
‐0.038 (‐0.60)
‐0.038 (‐0.61)
‐0.029 (‐0.45)
‐0.021 (‐0.33)
‐0.037 (‐0.59)
‐0.023 (‐0.36)
Rural Bank 0.054 (0.33)
‐0.156 (‐0.96)
‐0.002 (‐0.01)
0.015 (0.09)
0.067 (0.40)
0.060 (0.35)
0.075 (0.45)
0.061 (0.37)
Individual 0.113 (1.13)
0.130 (1.39)
0.107 (1.16)
0.114 (1.24)
0.120 (1.30)
0.122 (1.32)
0.124 (1.35)
0.118 (1.27)
Individual & Group 0.016 (0.20)
0.048 (0.58)
0.010 (0.13)
0.019 (0.23)
0.024 (0.30)
0.029 (0.35)
0.022 (0.28)
0.025 (0.30)
Village Banking ‐0.099 (‐0.89)
‐0.063 (‐0.55)
‐0.092 (‐0.83)
‐0.108 (‐0.97)
‐0.115 (‐1.03)
‐0.108 (‐0.96)
‐0.098 (‐0.89)
‐0.115 (‐1.02)
C.Asia & E.Europe 0.178 (2.01)**
0.124 (1.39)
0.194 (2.22)**
0.199 (2.27)**
0.183 (2.07)**
0.172 (1.94)*
0.174 (1.97)**
0.179 (2.01)**
E. Asia & Pacific ‐0.016 (‐0.16)
‐0.040 (‐0.40)
‐0.026 (‐0.26)
‐0.026 (‐0.26)
‐0.016 (‐0.16)
‐0.029 (‐0.29)
‐0.014 (‐0.15)
‐0.023 (‐0.23)
Latin America 0.231
(3.36)*** 0.199
(2.86)*** 0.218
(3.26)*** 0.228
(3.40)*** 0.240
(3.48)*** 0.229
(3.33)*** 0.229
(3.36)*** 0.237
(3.38)***
M. East & N. Africa ‐0.135 (‐1.14)
‐0.136 (‐1.11)
‐0.104 (‐0.89)
‐0.109 (‐0.93)
‐0.139 (‐1.17)
‐0.150 (‐1.25)
‐0.129 (‐1.08)
‐0.149 (‐1.24)
S. Asia ‐0.153 (‐1.70)*
‐0.039 (‐0.45)
‐0.129 (‐1.46)
‐0.131 (‐1.47)
‐0.154 (‐1.70)*
‐0.163 (‐1.80)*
‐0.153 (‐1.71)*
‐0.159 (‐1.77)*
Other Services 0.080 (1.51)
0.082 (1.51)
0.072 (1.36)
0.076 (1.45)
0.084 (1.59)
0.089 (1.67)*
0.079 (1.50)
0.088 (1.66)*
Savings ‐0.077 (‐1.22)
‐0.050 (‐0.77)
‐0.078 (‐1.24)
‐0.068 (‐1.08)
‐0.066 (‐1.05)
‐0.067 (‐1.07)
‐0.073 (‐1.17)
‐0.066 (‐1.05)
Regulated 0.081 (1.49)
0.112 (2.03)**
0.087 (1.60)
0.093 (1.69)*
0.087 (1.60)
0.093 (1.69)*
0.072 (1.31)
0.094 (1.71)*
Constant ‐1.626
(‐3.26)*** ‐1.266
(‐2.52)***‐1.424
(‐4.06)***‐1.537
(‐4.43)***‐1.753
(‐3.53)***‐1.583
(‐3.23)*** ‐1.744
(‐3.62)*** ‐1.671
(‐3.35)***
No of Observations 330 330 330 330 330 330 330 330 No of Groups 165 165 165 165 165 165 165 165 Log Likelihood ‐154.42 ‐162.30 ‐155.45 ‐157.05 ‐156.12 ‐156.63 ‐154.64 ‐156.59 Wald Chi‐Square 73.25 54.45 70.72 66.87 69.12 67.88 72.72 67.99 Prob > Chi‐Square 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000