World Development: Syedah Ahmad, Robert Lensink, Annika Mueller

World Development 136 (2020) 105130
Contents lists available at ScienceDirect
World Development
journal homepage: www.elsevier.com/locate/worlddev
The double bottom line of microfinance: A global comparison between

conventional and Islamic microfinance
Syedah Ahmad a, Robert Lensink a,b,⇑, Annika Mueller a
a
Faculty of Economics and Business, University of Groningen, Groningen, The Netherlands
b
Development Economics Group, Wageningen University, The Netherlands
a r t i c l e i n f o a b s t r a c t
Article history: Conventional microfinance institutions (MFIs) can promote financial inclusion, but they also prompt eth-
Accepted 31 July 2020 ical concerns regarding the social consequences of commercialization and high interest rates. Islamic
MFIs, which adhere to Sharia’s prohibition of riba (usually interpreted as a ban on interest), present an
alternative. Differences between conventional and Islamic MFIs in terms of outreach and financial sus-
JEL classification: tainability remain underexplored; no comprehensive data set details Islamic MFIs either. With new data,
G21 collected with a global survey, the authors construct a unique panel of 543 conventional and 101 Islamic
L21
MFIs, operating in Islamic and non-Islamic countries. These data suggest that the market for Islamic
L31
Z12
microfinance is more important than previously recognized, has grown in recent years, and is likely to
continue growing in every region of the world. Statistical comparisons, using various estimation tech-
Keywords: niques, regarding the outreach and financial performance of Islamic and conventional MFIs also reveal
Islamic microfinance that the breadth and depth of Islamic MFIs exceed those of conventional MFIs, though conventional
Social entrepreneurship MFIs achieve stronger financial performance. This latter result is not robust though.
Ethical finance Ó 2020 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://
Outreach
creativecommons.org/licenses/by/4.0/).
Financial performance
1. Introduction tices adopted by Islamic MFIs, such as interest-free forms of finan-

cial access, also might allow for greater outreach but require more
Microfinance institutions (MFIs) generally strive to generate resources to manage, relative to conventional microfinance (Visser,
positive social impacts while simultaneously delivering sound 2013). To the best of our knowledge, rigorous comparisons of con-
financial performance to achieve a ‘‘double bottom line.” By the ventional and Islamic MFIs, in terms of outreach and financial per-
end of 2017, 981 MFIs had submitted performance reports to the formance measures, and thus their ability to achieve the double
Microfinance Information Exchange (MIX Market), which revealed bottom line, are lacking.
an estimated US$114 billion in loan volume and 139 million cus- The question of whether conventional and Islamic MFIs perform
tomers (Valette & Fassin, 2018). That is, this sector clearly has differently is particularly relevant, considering the increasing
expanded to comprise a vast variety of organizations, which appear interest in MFIs that offer products and services compliant with
heterogeneous in their approaches to achieving this dual mandate. Islamic financial principles (Abedifar, Molyneux, & Tarazi, 2013).1
In response, a debate has cropped up, regarding which types of A large proportion of the world’s poor (700 million in 2013; World
MFIs may be most successful in realizing both objectives (e.g., Bank, 2013) live in Muslim-majority nations, sparking interest in
Armendáriz & Morduch, 2010; Banerjee, Karlan, & Zinman, 2015; Islamic microfinance as a financial outreach tool. Even MFIs that pre-
Morduch, 2016). Some studies propose an influence of religion, viously offered only conventional microfinance products have
such as when Mersland, DÉspallier, and Supphellen (2013) argue started offering Islamic versions, marketing them as effective tools
that Christian-based MFIs earn lower profits but also incur lower to facilitate and encourage small businesses (Ahmed, 2002). Yet Isla-
funding costs than conventional MFIs. The specific financing prac- mic MFIs differ markedly from conventional MFIs on several key
1
The global financial crisis of 2007–2008 raised interest in Islamic finance in
⇑ Corresponding author at: Faculty of Economics and Business, University of general and Islamic microfinance in particular, because banks operating according to
Groningen, Groningen, The Netherlands. Islamic principles exhibited greater resilience to the crisis than their conventional
E-mail addresses: s.s.ahmad@rug.nl (S. Ahmad), b.w.lensink@rug.nl (R. Lensink), counterparts. Global Islamic finance assets in 2017 accounted for more than US$2.4
a.m.mueller@rug.nl (A. Mueller). trillion (Mohamed, Goni, & Hasan, 2018).
https://doi.org/10.1016/j.worlddev.2020.105130
0305-750X/Ó 2020 The Authors. Published by Elsevier Ltd.
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
2 S. Ahmad et al. / World Development 136 (2020) 105130
dimensions, such as their sources of financing, investment and pro- 2. Literature review and hypotheses development
duct portfolios, and management.2 For example, high interest rates
in the conventional MFI sector, in addition to being criticized as 2.1. Features of Islamic microfinance
unethical (Hudon & Sandberg, 2013), conflict with Islamic
prohibitions on microfinance products that involve paying or receiv- Islamic microfinance offers an alternative to conventional
ing riba.3 Conventional, for-profit MFIs charge significantly higher microfinance for meeting the financial needs of the poor and finan-
interest rates when markets are less competitive (Baquero, cially excluded (Karim, Tarazi, & Reille, 2008). The two microfi-
Hamadi, & Heinen, 2018), and several conventional MFIs have been nance forms differ considerably from an operational perspective
accused of acting like loan sharks, not only charging extremely high (Ahmed, 2002). Even if some fundamental similarities apply to
interest rates but also using aggressive collection methods the financial instruments or techniques, the products and services
(Boatright, 2014). provided by Islamic MFIs are free of particular elements
To investigate the possible trade-offs that Islamic MFIs confront (Obaidullah, 2008), because their business activities must adhere
in pursuing a double bottom line, we therefore construct a novel to halal (permissible) principles. For example, both conventional
data set that reflects a comprehensive, global mapping of Islamic and Islamic MFIs use equity and debt-based financing, but they
microfinance service providers. Using an online survey that we operationalize the instruments differently. Weill (2020) proposes
sent to all MFIs reporting to MIX Market, we identify MFIs cur- a summary of four main principles of Islamic (micro-)finance:
rently providing microfinance products in line with Islamic princi-
ples, as well as those that plan to provide such products in the (1) Interest is forbidden.
future. This classification is novel, in that it relies on direct survey (2) Lenders are rewarded through profit sharing, though the
questions about product offerings. We then align our findings with most popular Islamic microfinance products do not reflect
databases provided by the microfinance network for Arab coun- conventional profit-and-loss sharing principles, as we dis-
tries, Sanabel (2012) and the Islamic Banking Database (2014), cuss subsequently.
which establishes an MFI classification that is more comprehensive (3) The MFIs cannot finance activities considered sinful by
than previous approaches, in terms of the regions covered and Islam, such as maysir (gambling) (Chong & Liu, 2009), alco-
number of Islamic MFIs included. We thus create a detailed, consis- hol, or borrowing and lending to conventional MFIs that
tent map of the supply and demand sides of the market for Islamic charge interest.
microfinance products, according to global distribution trends. In (4) Contract terms should be entirely clear and eliminate any
total, we identify 644 MFIs by type and specify 101 of them, based contractual uncertainty, due to the prohibition of gharar
in 33 countries that can be classified as Islamic MFI providers. (uncertainty).
These comprehensive data suggest that the market for Islamic
microfinance is more important than is generally acknowledged, The Sharia-compliant financial products that Islamic MFIs offer
and its recent growth appears likely to persist, in every geograph- can be broadly categorized into three types: (1) equity financing
ical region. instruments, such as mudaraba and musharaka; (2) credit or debt
Using this newly constructed data set, we also undertake a com- financing instruments, including ijara, istisna, murabaha, qard e
parison of the performance of conventional MFIs and Islamic MFIs, hasan, and sala’m; and (3) other types of microfinancing, such as
according to the dual objectives of social benefits and financial per- asset-building products, typically in the form of saving accounts
formance. Fan, John, Liu, and Tamanni (2019) compare Islamic and (e.g., wadiah), investment deposits, or mutual insurance schemes
conventional MFIs too, using a sample of 300–600 observations, (e.g., micro-takaful). We address the first two categories in more
depending on the outcome variable, containing 316 MFIs. For this detail next but exclude the third category as this category is not
statistical analysis, we expand the sample to approximately 5000 relevant for the analysis in this paper.
observations, including 644 MFIs.4 The analyses suggest that Isla-
mic MFIs outperform conventional MFIs in terms of outreach, but
conventional MFIs might perform better financially. This latter result 2.1.1. Equity-like instruments
is not robust though, which might reflect the endogeneity problems Equity financing relies on profit-and-loss sharing (PLS) arrange-
that affect our results, despite our best efforts to reduce possible ments, rather than interest-based contracts, between an Islamic
sample selection problems by using cross-sectional, panel, and MFI and its clients, to conform with Islamic principles (Khan &
instrumental variable regression techniques. Mirakhor, 1992). For example, under a mudaraba or trustee financ-
In Section 2, we outline the main characteristics of and products ing contract, the MFI is the investor (financier), and the MFI’s client
offered by Islamic MFIs, along with a review of literature pertain- manages the enterprise. If the business generates profits, the fund-
ing to the social and financial performance of MFIs. We also ing parties split the gains according to some predetermined rule
develop some testable hypotheses for our quantitative analyses. (Visser, 2013). Thus, the profit shares are predetermined, but the
Section 3 presents the variables for our empirical analysis, provides profits are unknown in advance (Weill, 2020). If the business
some motivational statistics for our main analysis, and then details incurs a loss, it is borne exclusively by the Islamic MFI, but the
the empirical methodology. After we outline the results, according entrepreneur (i.e., client) receives no compensation. Under an
to our newly constructed data set and regression analyses, in Sec- equity partnership musharaka contract, both the MFI and the client
tion 4, we conclude in Section 5. instead contribute capital and share profits according to a prede-
fined rule, and they also jointly manage the business. Profits are
2
negotiated freely; losses are covered according to the capital con-
Differences in financial sustainability also might reflect distinct sources of
tributions of the MFI and the entrepreneur. Mudaraba is thus closer
funding, such as funding by Islamic charities and donations or other borrowers’
contributions, which are more prevalent in Islamic MFIs. to a limited partnership, whereas musharaka is similar to a busi-
3
Many studies use interest and riba interchangeably, but they are not exactly the ness model involving equity stakes with controlling rights.
same (Ugi, 2018). It is more accurate to state that Islam prohibits riba, not interest. Yet Although a strict application of PLS principles reduces the risk of
the ban on riba is widely interpreted as a ban on interest by fiqh scholars who
insolvency for an Islamic MFI, the shareholders’ risk also might
specialize in Islamic jurisprudence.
4
Other comparative studies of Islamic and conventional MFIs offer conflicting
transfer to depositors, in form of more volatile returns (Visser,
results, based on a more restricted set of Islamic MFIs, such as Widiarto and 2013). In practice, Islamic banks often stabilize the profit distribu-
Emrouznejad (2015) and Abdelkader and Salem (2013). tions to depositors, but Islamic (and conventional) MFIs cannot
S. Ahmad et al. / World Development 136 (2020) 105130 3
accept deposits (or savings) to begin with, and they often are not 2.2. Social and financial performance of MFIs
subject to regulations by a Central Bank or other monetary author-
ity. Moreover, PLS contracts may increase information asymmetry Microfinance institutions, both conventional and Islamic, have
problems, relative to debt-based interest rate contracts (Weill, social and financial objectives. On the one hand, MFIs aim to
2020). In particular, adverse selection problems probably grow reduce poverty by providing financial services to poor households
more acute with PLS contracts. Borrowers with low profit expecta- that have been excluded from the formal financial system. On the
tions prefer a PLS contract over an interest-based one, because the other hand, the MFIs themselves aim to achieve financial self-
amount they must share with the lender (i.e., a share of their small sufficiency, without the need for subsidies (Tulchin, 2003). Achiev-
profits) under a PLS contract probably is lower than a fixed interest ing both objectives simultaneously is referred to as attaining the
payment. But borrowers earning higher profits likely seek an microfinance promise (Morduch, 1999) or double bottom line
interest-based contract so they can pay a fixed payment, lower (Armendáriz & Labie, 2011). In practice, it remains difficult to reach
than the share of their profits. Moreover, interest-based contracts both objectives, which even may be subject to a trade-off. Accord-
incentivize borrowers to work harder to earn high returns, because ing to financial systems (Robinson, 2001) or self-sustainability
they still pay the same, fixed amount, regardless of their profits. In (Schreiner, 2002) approach, social performance and financial sus-
contrast, a PLS contract requires a higher payment if they earn tainability can go hand-in-hand, because reaching more customers
more, so borrowers might lack strong enough incentives to put should create economies of scale. If financially sustainable MFIs
forth stringent effort. Such a scenario is likely to lead to moral haz- attract more funds, it could increase their ability to serve more
ard problems. poor people. A poverty lending (Robinson, 2001; Schreiner, 2002)
perspective instead implies the necessary trade-off between social
performance and financial sustainability, because providing finan-
2.1.2. Debt-based instruments cial services to the poor is expensive and can persist only by MFIs
Debt-based instruments instead do not align with PLS princi- that receive subsidies (i.e., not financially self-sustainable). The
ples (Shahinpoor, 2009), so they may evoke higher risk for the bor- high costs of lending primarily stem from the transaction costs,
rower, because repayment does not depend on the borrower’s in that poor people often live in remote areas, and high fixed costs,
profits (Weill, 2020). For example, ijara is a lease purchase; the even for small loan amounts. Theoretically, it is not clear whether
MFI allows a client to use an asset it possesses, for a certain price social and financial objectives trade off or are compatible; empiri-
and period.5 With each payment, the lessee moves closer to a pur- cal studies appear necessary to address this debate.
chase and transfer of ownership of the leased asset. Unlike tradi- Different contributions from empirical microfinance literature
tional leasing though, all risks are borne by the MFI, including any also inform this discussion. In particular, some studies examine
impairment or damage to the leased asset caused by factors outside the social and economic impacts of microfinance on end-users;
the client’s control, such as weather events. These terms are stated in recent studies using randomized controlled trials tend to offer
advance. strong criticisms. For example, Banerjee et al. (2015) conclude,
Istisna is an exchange contract that defers payment and delivery from a study across eight countries, that microcredit fails to induce
of the product; the MFI might produce goods itself or buy them from transformative effects or raise households out of poverty. Dahal
a third party, then deliver them to the end customer (i.e., client). and Fiala (2020) instead argue that prior studies are severely
This contract refers to goods that have yet to be produced, such as underpowered, such that it is impossible to establish whether
a building or road. The end customer might pay when the contract and how microcredit affects welfare. With their non-
is signed or at subsequent stages in the manufacturing process. experimental study in Sierra Leone, Garcia, Lensink, and Voors
Instead, murabaha is a goods-financing contract, such that the MFI (2020) provide evidence that microcredit offered through group
acquires a requested product and resells it to the client for its cost lending systems helps people release their internal psychological
plus a markup to cover any service costs. It facilitates the purchase constraints and develop aspirational hope, which may provide a
and resale of commodities in rural areas in particular (Wilson, foundation for increased welfare in the future. Across these con-
2007). Murabaha can evoke high costs, often higher than conven- trasting views though, a general consensus indicates that microfi-
tional financial products, because the Islamic MFI must physically nance is not a panacea. In particular, the social impact of
handle the goods, ensure they are properly stored, and insure them microcredit appears lower than early predictions suggested and
(Visser, 2013). It also involves two sales transactions and thus probably cannot lift large segments of poor populations out of pov-
potentially two tax payments, though this issue should not be a con- erty, though that pessimistic conclusion may refer mainly to
cern in relation to value-added taxes (VAT). The VAT or other ad val- microcredit, not necessarily the wider range of microfinance activ-
orem tax on conventional sales that require interest payments by ities, such as microsavings and microinsurance. Furthermore, var-
definition will be lower than the tax imposed on a murabaha sale ious groups of vulnerable people might respond to microfinance
though, in which a markup gets added to the sales price.6 activities in distinct ways, as implied by recent survey research
Qard e hasan is an interest-free loan. The borrower repays the by Hansen, Huis, and Lensink (2020), Hermes and Lensink (in
principal, with no return, reflecting the Islamic precept that Muslims press), and Lensink and Bulte (2019).
should help those in need, such as by supporting rural households or The disappointing results of microcredit also have induced
giving prospective entrepreneurs a chance to start their business another stream of research that seeks tactics for improving its
(Abdul Rahman, 2007; Obaidullah, 2008; Wilson, 2007). A small impact (Lensink & Bulte, 2019). One option is to rethink the pro-
fee of approximately 0.5 percent may be charged to cover expenses. duct design; whereas traditional microcredit involves short-term,
Finally, sala’m is a forward sale, mostly used for agricultural group loans with rigid contract terms, more flexible repayment
financing. The quality, quantity, time, and price of the goods to terms, such as might be achieved through longer grace periods
be purchased must be fully specified, leaving no ambiguity (Field, Pande, Papp, & Rigol, 2013), could increase the impact of
(Dhumale & Sapcanin, 1999; Obaidullah, 2008). The goods microcredit. Another approach expands on microfinance, to go
included in these contracts cannot be gold, silver, or currencies. beyond providing credit and also offer varied financial and non-
financial services, such as gender-based and business training.
5
Here, we refer to a lease to buy (ijara wa iqtina, also known as ijara muntahia bi
Such offerings are broadly referred to as microfinance-plus
tamleek), not an operational lease. (Garcia & Lensink, 2019). Bulte, Lensink, and Vu (2017), with an
6
We thank an anonymous referee for noting this point. experimental study in Vietnam, suggest that combining credit with
business training may enhance the impacts. However, both more mance indicators. Nor does outreach reflect the overall impact of
flexible contract terms and microfinance-plus activities also have microfinance, which refers to the ultimate influence that financial
financial consequences and may demand additional subsidies, with services have on people’s welfare. Greater outreach, implying that
detrimental impacts on the financial self-sufficiency objectives. the MFI has provided financial services to more people or propor-
The pursuit of this second, financial objective seemingly has tionally more to the poorest people, might lead to positive impacts
prompted increasing commercialization of conventional microfi- but does not do so inevitably. Even further, we cannot confirm that
nance, which critics allege has created mission drift, increased con- average loan size, a commonly used measure of breadth, has any
sideration of wealthier clients, the exclusion of poor and female relationship with poverty levels, because MFIs might cross-
borrowers (Cull, Demirgüç-kunt, & Morduch, 2007), and a shift subsidize expensive, small loans with profitable, larger loans,
toward more individual lending. Notably, de Quidt, Fetzer, and which produces a higher average loan size (Armendáriz &
Ghatak (2018a) provide theoretical and empirical evidence that Szafarz, 2011). Alternatively, increased average loan sizes may
commercialization in conventional microfinance leads to greater stem from demand-side factors related to the type of clients
competition and shifts away from non-profit group lending and (Morduch, 2000) or reflect a progressive lending system in which
toward for-profit individual lending. Commercialization induces credit limits increase over time, conditional on repayments of pre-
greater competition too, which arguably should increase the funds vious loans. Overall then, outreach indicators provide, at best, only
available, such that it could have positive effects. With a simulation an imperfect indication of MFIs’ social performance.
exercise, de Quidt, Fetzer, and Ghatak (2018b) predict that the neg- Accordingly, empirical supply-side studies offer mixed results.
ative effects of monopolistic for-profit lending, due to commercial- Some studies suggest a negative relationship between outreach
ization, are almost entirely compensated for by the results of and financial sustainability (Cull et al., 2007; Hartarska, Shen, &
increased competition. Thus, the ultimate influence of commercial- Mersland, 2013; Hermes et al., 2011; Louis & Baesens, 2013);
ization is ambiguous, and in practice, the MFI’s internal organiza- others find no trade-off (Adhikary & Papachristou, 2014; Louis,
tion appears important in determining these effects. Churchill Seret, & Baesens, 2013). A few studies propose that the likelihood
(2019) provides empirical evidence that non-profit MFIs are more of a trade-off is contingent on factors such as the representation
socially driven than for-profit ones, which instead may be more of stakeholders on boards of MFIs (Hartarska, 2005), gender diver-
concerned about financial returns, perhaps at the cost of their sity in the board (Hartarska, Nadolnyak, & Mersland, 2014), or loan
social objective (Hermes & Lensink, in press). methodology (Tchakoute-Tchuigoua, 2012). Churchill (2019)
Another closely related stream of literature explicitly focuses on instead points to the type of outreach as an important contingency,
the supply side. That is, rather than investigating the impact of such that a trade-off may arise between financial sustainability and
microfinance on end-users by gathering individual-level data, depth, but complementarity marks the link between financial sus-
these studies address the MFIs themselves, with data at the MFI tainability and breadth. Churchill also shows that for-profit MFIs
level. Hermes and Hudon (2018) provide a systematic review of outperform non-profit MFIs in terms of financial sustainability
such studies, but for our purposes, we note an interesting point and breadth of outreach, but not on the depth of outreach, which
these studies raise, pertaining to the trade-off between social out- reiterates the likely importance of organizational structure. Finally,
reach and financial sustainability. From a supply-side perspective, with a comprehensive meta-analysis of relevant literature,
financial sustainability implies that MFIs’ activities do not result in Reichert (2018) simply concludes that there is no definitive answer
losses over time, such that they no longer need subsidies but still to the question of whether MFIs can achieve both goals
can continue to provide microcredit (Balkenhol, 2007; Quayes, simultaneously.
2012). Such financial sustainability can be measured with standard
financial ratios such as the return on equity (ROE) or the return on 2.3. Hypotheses development
assets (ROA), though some researchers use the operational self-
sufficiency of MFIs, which reflects whether they can cover their Using insights from our review of prior literature, which relates
costs with revenues, or their financial self-sufficiency, which indi- mainly to conventional MFIs, we seek to establish some predictions
cates whether they can operate without ongoing subsidies, soft regarding how conventional and Islamic MFIs might compare with
loans, or grants. Finally, studies that use data envelopment or regard to the trade-off. We do not aim to derive new theory to
stochastic frontier analyses tend to measure (financial) efficiency explain the differences between the two types of MFIs; rather,
(e.g., Caudill, Gropper, & Hartarska, 2009; Hermes, Lensink, & we try to establish testable hypotheses.
Meesters, 2011; Servin, Lensink, & van den Berg, 2012; for a With regard to social performance, we deliberately restrict our
meta-analysis, see Fall, Akim, & Wassongma, 2018). predictions to differences in outreach, rather than social impact;
From a supply-side perspective, outreach is the representation our research methodology (in line with prior supply-side studies)
of the social value created by MFIs, which can be measured by cannot identify impact, which would require data from end-
two dimensions: depth and breadth (Navajas, Schreiner, Meyer, users. Thus, the comparison focuses on the amount (breadth) and
Gonzalez-Vega, & Rodriguez-Meza, 2000; Schreiner, 2002). The type (depth) of borrowers served by the two types of institutions.
breadth of outreach reflects how many people the MFI serves, Both conventional and Islamic MFIs aim to provide financial access
reflecting its coverage in terms of the number of clients served. to borrowers neglected by mainstream financial institutions (Cull
The depth of outreach instead indicates whether an MFI serves et al., 2007; Kleynjans & Hudon, 2016; Strøm, DÉspallier, &
the poorest segments of the population (Brau & Woller, 2004; Mersland, 2014). Among conventional MFIs, non-profit versions
Schreiner, 2002), generally measured by either the average loan still exist, though several developments make them less promi-
size scaled by the gross domestic product (GDP) per capita of the nent, as detailed by de Quidt et al. (2018a), including a broad shift
focal country or by the ratio of female to the total number of bor- from non-profit to for-profit MFIs, increasing competition among
rowers. Unlike financial performance, which is relatively easy to the growing numbers of MFIs in each country, and the enhanced
quantify using available, validated finance and accounting mea- importance of individual lending at the expense of joint liability
sures, debate continues about how best to measure social perfor- group lending. According to Churchill (2019) findings, these three
mance. DÉspallier and Goedecke (2020) offer a survey discussion trends seem likely to lead to greater breadth of outreach but
and detail some disadvantages of standard outreach indicators; decreased depth among conventional MFIs. For example, rising
in particular, outreach cannot be identical to social performance interest rates and heightened penalties for non-repayment
and at best is a subdimension of a broad range of social perfor- (Dehejia, Montgomery, & Morduch, 2012) imposed by for-profit
MFIs likely lead to involuntary exclusions of the poorest borrowers whether the MFI offered financial products in line with Islamic prin-
from conventional MFIs. In contrast, Islamic MFIs explicitly inte- ciples, as well as the types of conventional and Islamic financial
grate a religious foundation for their operating principles, which products offered. We combined the results of our survey with infor-
means they are barred from offering products that involve interest mation gathered from two, more limited databases, namely, the
(Abedifar et al., 2013) and also try to avoid high commissions or Sanabel (2012) and the Islamic Banking Database (2014).8 For con-
fees (Beck, De Jonghe, & Schepens, 2013). Borrowers of Islamic ventional MFIs, we also asked about plans to include Sharia-
MFIs thus may be less likely to confront high borrowing costs or in- compliant products in future portfolios.
voluntary exclusion. Moreover, borrowers who strictly adhere to We combine our global survey data (collected in 2016) with
Islamic principles are de facto unable to borrow from conventional existing (financial and outreach-related) information for the years
MFIs, which do not adhere to Sharia. In this sense, conventional 1999–2016, obtained from MIX Market. For some variables, we
MFIs may encounter more voluntary financial exclusion, which also turned to other, existing data sources. Specifically, we obtain
may affect the breadth and depth of their outreach. The poorest country-level information about the percentage of the population
members of the population often adhere strictly to Sharia, due to that is Muslim from Kettani (2010).9 For information about the offi-
social and religious norms (El-Gamal, El-Komi, Karlan, & Osman, cial state religion, we rely on a detailed report by Barro and McCleary
2014), so we predict that voluntary exclusion may primarily affect (2005), updated with other publicly available sources.10 We obtain
the depth of outreach, though this assertion is not clear ex ante. GDP data from the World Bank (http://data.worldbank.org). The final
More relevant for the breadth of outreach, Islamic MFIs are not sample of MFIs that completed our survey consists of 644 MFIs in 86
allowed to borrow from or lend to conventional MFIs, which may countries for the period 1999–2016, around 11% of which are Islamic
induce liquidity problems (Weill, 2020). However, such limitations MFIs.
on breadth also could be compensated for by expanded access to
other funding sources, such as Islamic charities and donations or
contributions from other borrowers. Therefore, we hypothesize 3.2. Model specification
that Islamic MFIs perform, on average, better on outreach than
conventional MFIs, especially in terms of depth, whereas the out- One of the main challenges of comparing outreach and financial
comes for breadth are less clear, due to liquidity concerns. performance by Islamic versus conventional MFIs relates to endo-
Turning to financial sustainability, we offer several reasons Isla- geneity biases. That is, Islamic and conventional MFIs are not ran-
mic MFIs may be likely to underperform financially relative to con- domly distributed over different countries, and some (probably the
ventional MFIs. First, their operational and administrative costs most poor) Islamic borrowers are religiously prohibited from bor-
likely are higher (El-Zoghbi & Tarazi, 2013). Notably, Islamic MFIs rowing from conventional MFIs. We thus cannot randomly assign
frequently offer non-PLS murabaha contracts, tied to some asset potential borrowers, who might borrow from both types of MFIs,
(e.g., property, plant, equipment). The need to transfer such an to either Islamic or conventional MFIs, and there is no clear-cut
asset demands substantial operational costs, far more than manag- solution to this endogeneity issue.
ing a cash distribution. Murabaha also implies two sales transac- To address our central question, we apply both cross-sectional
tions instead of one, along with a higher sales price that and panel regressions, adopting several empirical approaches in
integrates a markup (and thus higher taxes). Furthermore, it an attempt to mitigate endogeneity concerns.11 First, noting that
requires MFIs to store and insure the asset, with further increased the main independent variable of interest (whether an MFI is Islamic,
costs. Another popular product is qard e hasan (El-Zoghbi & Tarazi, ISMFI) is time-invariant, we start with two cross-sectional
2013), which is easier to administer than murabaha but still is not approaches, using the between-subjects estimator and a Fama-
priced to cover all administrative and default costs. Second, even MacBeth regression.12 These regressions (as well as those detailed
the PLS contracts probably lead to more adverse selection and subsequently) include approximately 10 additional independent
moral hazard problems than interest-based contracts. Thus, for variables that allow us to control for selection based on observable
the (relatively small) group of Islamic MFIs that adopt PLS MFI characteristics (which we describe later in this section) to
schemes, profits still may be lower than those earned by conven- assuage concerns about omitted variable bias. Second, we also use
tional MFIs. The relatively high costs of providing Sharia- a random-effects model and exploit the panel dimensions of our
compliant products may explain why the development of Islamic data; we expect the estimates to be more efficient than in our
microfinance has lagged; Islamic microfinance products still serve cross-sectional approach. Time dummies control for year effects.
less than 1 percent of borrowers (El-Zoghbi & Tarazi, 2013). Noting Third, to address the endogeneity of the Islamic MFI variable, we
the generally higher operational costs and lower pricing of Islamic use an instrumental variables (IV) approach with two alternative
financial products, relative to conventional microfinance, we
hypothesize that Islamic MFIs are less profitable than conventional
MFIs.
8
In our data set of 101 Islamic MFIs, 89 were identified by our survey question;
Sanabel (2012) and the Islamic Banking Database (2014) helped us identify 12
additional Islamic MFIs that report to MIX Market.
3. Data, model specification, and variables 9
We calculate the Muslim population for each period as
n
PopulationFuture ¼ PopulationPresent xð1 þ iÞ , where i is the growth rate and n is the
3.1. Sample construction number of years, which is 10 years for our study. For details, see Appendix B.1.
10
We cross-checked the information with data from the Organisation from Islamic
Corporation. Accessed from http://insct.syr.edu/wp-content/uploads/2014/08/OIC_
To identify MFIs that offer Sharia-compliant products, we con- Member_States.pdf.
ducted a web survey in 2016 (for more details, see Appendix A), 11
In contrast, Fan et al. (2019) only provide standard ordinary least square
using invitation emails sent to key staff members of the 2,544 MFIs regressions for comparing conventional and Islamic MFIs
12
that report to MIX Market.7 The survey contained questions about Some MFIs that claim to be Islamic offer both Islamic and non-Islamic
microfinance products. We follow standard practice and classify MFIs as Islamic if
they offer both Islamic and non-Islamic microfinance products (e.g., Widiarto &
7
The MIX Market database provides information about a global set of registered Emrouznejad, 2015). Their financial statements do not report the data for Islamic and
MFIs (www.mixmarket.com). Being listed indicates the MFI’s willingness to comply non-Islamic microfinance products separately, so it is not possible for us to
with the data standards set by MIX Market, simply by the act of reporting data, yet differentiate the contribution and role of Islamic versus non-Islamic products offered
these data suffer from the well-known self-reporting biases. by these MFIs.
instruments, though the choice of instruments (described subse- Balance per Borrower/GNI per Capita” are in Table C.1 in Appen-
quently) is limited by data availability. Fourth, we use an inverse dix C.13
probability weighting (IPWIV) estimator to control for non- Finally, the sample includes only those MFIs that responded to
response bias and non-random selection into the survey. The use our survey, so we use an inverse probability weighting estimator to
of different estimation techniques may enhance confidence in the control for potential sample selection caused by non-responses
robustness of the results, though none of our identification strategies (Seaman & White, 2013). Briefly, this procedure works as follows:
can fully resolve endogeneity issues, so our results should be inter- We use a logit model to predict the probability that an MFI
preted as correlational, not causal. responds to the survey, according to a set of independent variables
In all our empirical models, we regress the measures of out- (year dummies, dummy variables for the legal status of the MFI,
reach and financial performance by an MFI i in country j (at time the variables previously included in the Z vector; see Table C.2.1
t, when we include the panel dimension), which we denote y on in Appendix C). The inverse values of these predicted probabilities
a time-invariant dummy variable that equals 1 if the MFI in ques- then serve as weights in the subsequent regressions (of the deter-
tion is Islamic, and 0 otherwise (ISMFIij ), in addition to a series of minants of outreach and financial performance), such that observa-
control variables, denoted by Z, that (may) vary across MFIs, coun- tions with characteristics similar to those MFIs that did not
try, and time. The precise definitions of the dependent variables respond to our survey take higher weights. We report results for
(outreach and financial performance) are in Table 4 and explained which we combine the inverse probability weighting estimator
in greater detail in Section 4.3.1. with the IV approach.14 (We also conducted these estimates without
Because our main independent variable ISMFIij is time- instruments, and those results are available on request.)
invariant, we start with a cross-sectional regression, and our first
estimate uses a between-subject estimator, such that we run an 4. Empirical results
ordinary least squares (OLS) regression on the group means (we
denote this specification GROUP MEANS), as follows: This section presents the results of our empirical analyses. We
start by presenting summary statistics about the prevalence and

geographic distribution of Islamic MFIs, based on our newly con-
yij ¼ a0 þ a1 ISMFIij þ c Z ij þeij ; ð1Þ
ducted survey. Then we offer new summary statistics regarding
the most common forms of financial contracts used by Islamic
where yij denotes measures of either the outreach or financial per- MFIs. Finally, we present the regression results related to the
formance of an MFI i operating in country j; a0 is a constant; and eij impact of Islamic versus conventional MFIs on outreach and finan-
indicates mean-zero errors. In Eq. (1), all variables refer to means cial sustainability.
across the period from 1999 to 2016.
Then we employ a second cross-sectional approach, with two- 4.1. Global expansion of Islamic microfinance: mapping exercise
step Fama and MacBeth (1973) regressions (FM). The first step esti-
mates the following cross-sectional regressions for each year in our The values reported in this subsection derive from Table 1. We
sample, from 1999 (period 1) to 2016 (period T), Eq. (2) : classify roughly 15.7% of the responding MFIs as Islamic: 101 MFIs
in 33 countries, spread across all world regions, report that they
yij;1 ¼ a0; 1 þ a1;1 ISMFIij þ c1 Z ij;1 þ eij;1 offer Sharia-compliant products, whereas 543 MFIs exclusively
yij;2 ¼ a0;2 þ a1;2 ISMFIij þ c2 Z ij;2 þ eij;2 offer conventional products. Of the 543 conventional MFIs that
; ð2Þ responded to our survey, 129 provide interest-free products, such
..
. as grants and loans.15 Sudan (13 Islamic MFIs) and Pakistan (13 Isla-
yij;T ¼ a0; T þ a1;T ISMFIij þ cT Z ij;T þ eij;T mic MFIs) host the highest numbers of Islamic microfinance service
providers,16 followed by Bangladesh (9), Indonesia (7), and Palestine
(7). Table 1 also shows a rough estimate of the projected number of
where all variables are defined as previously. The second step takes
averages of the estimated a and c coefficients computed in the first 13
The first-stage results for the other outreach and financial performances
step and uses Fama and MacBeth (1973) adjusted t-statistics to test indicators are very similar and can be obtained on request. Because our endogenous
for the significance of the coefficients. variable ISMFI is time-invariant, we cannot use a first-difference generalized method
In addition, we estimate the effect of interest using panel of moments (GMM) approach to identify ISMFI. In a system GMM approach, it would
be possible to include (and identify) ISMFI. However, the first difference of ISMFI
regressions. For the time-invariant variable ISMFIij , we estimate a
cannot be used as an instrument (it would disappear), so we cannot treat ISMFI as
random effects model with the following form, Eq. (3): endogenous variable according to a system GMM approach. The only solution thus is
to treat ISMFI as exogenous. Treating it as endogenous would require finding external
yij;t ¼ a0 þ a1 ISMFIij þ cZ ij;t þ cij þ eij;t ; ð3Þ instruments, as in our IV regression methods. Therefore, we prefer to use our method
with external instruments, rather than GMM-based estimation techniques.
14
The weighted instrumental regressions are estimated without random effects
where cij is the individual unobserved (random) effect of MFI i in (normal OLS), because no available STATA programs can estimate weighted random
effects models with instruments. However, we cluster standard errors at the MFI
country j, and eij;t is a mean-zero error term. level. Because we are mainly interested in the coefficient for Islamic MFIs, this choice
As we have argued, the Islamic MFI treatment variable is does not change the main results.
15
likely endogenous, because MFIs choose to offer Islamic products. In our survey, we directly asked about interest-free or Islamic microfinance
Thus, we also use a random effects instrumental variable estima- product offerings, which helped distinguish Islamic interest-free MFIs from non-
Islamic interest-free MFIs. We identified 230 interest-free MFIs, of which 129 are non-
tor (REIV). We leverage the percentage of Muslims (PMP) and
Islamic and 101 are Islamic MFIs. The non-Islamic MFIs that provide interest-free
whether Islam is a state religion (Islstate) in the country where products are not considered Islamic for this analysis. The full questionnaire is
the MFI operates as instruments for ISMFI. These two variables available on request.
16
should correlate closely with the extent to which an MFI This finding is not surprising: In Sudan, the entire financial sector is required to be
operates in line with Islamic lending principles. Yet it is unlikely Sharia-compliant by national law. Whereas it hosted few MFIs in 2006, serving only
9500 clients, Sudan more recently supported over 400,000 customers via Islamic MFIs
that these variables influence the outreach or financial perfor- (El-Zoghbi & Tarazi, 2013). Unlike in Sudan, both Islamic and conventional MFIs
mance of MFIs directly, other than through the ISMFI variable. operate in Pakistan. However, since 2007, the State Bank of Pakistan has increased
The first-stage results for the IV regression for the ‘‘Average Loan institutional support for Islamic microfinance.
Table 1
Worldwide Distribution of Islamic, Conventional, and Future Islamic MFIs.
Eastern Europe and Central Asia Latin America and the Caribbean South Asia East Asia and the Pacific Sub-Saharan Africa Middle East and North Africa
Countries No. of MFIs Countries No. of MFIs Countries No. of MFIs Countries No. of MFIs Countries No. of MFIs Countries No. of MFIs
ISMFIs CMFIs FISMFIs ISMFIs CMFIs FISMFIs ISMFIs CMFIs FISMFIs ISMFIs CMFIs FISMFIs ISMFIs CMFIs FISMFIs ISMFIs CMFIs FISMFIs
Albania 2 Argentina 2 Afghanistan 6 6 Cambodia 1 9 2 Benin 1 9 5 Bahrain 1 1
Armenia 3 Bolivia 10 Bangladesh 9 31 17 China 1 2 2 Burkina Faso 16 7 Egypt 2 2 3
Azerbaijan 5 4 Brazil 5 Bhutan 1 Indonesia 7 5 12 Burundi 3 Iraq 6 5 9
Bosnia and 2 6 2 Chile 2 India 1 40 6 Laos 2 Cameroon 1 11 7 Jordan 2 1 3
Herzegovina
Bulgaria 1 Colombia 19 1 Nepal 10 1 Malaysia 1 1 Central African 1 Lebanon 2 1 2
Republic
Croatia 1 Costa Rica 8 Pakistan 13 11 17 Myanmar 2 1 Congo, D.R. 6 2 Palestine 7 2 7
Georgia 4 Dominican 8 Sri Lanka 1 8 3 Guinea 2 1 Cote d’Ivoire 1 8 5 Saudi 2 1 2
Republic Arabia
Kazakhstan 6 Ecuador 16 Philippines 2 16 4 Ethiopia 2 6 6 Syria 1 1 2
Kosovo 2 3 2 El Salvador 7 Vietnam 4 Gabon 1 1 Tunisia 1
Kyrgyzstan 1 10 4 Guatemala 11 1 Ghana 19 3 Yemen 6 6
Macedonia 1 Haiti 3 Guinea 1 1
S. Ahmad et al. / World Development 136 (2020) 105130

Moldova 2 Honduras 16 Kenya 6 2
Mongolia 3 Mexico 1 9 1 Madagascar 4
Montenegro 2 Nicaragua 7 1 Malawi 7 1
Poland 2 Panama 3 Mali 8 1
Romania 3 Paraguay 2 Niger 1 4 3
Russia 8 Peru 1 14 1 Nigeria 1 13 1
Serbia 2 1 Uruguay 1 Rwanda 6
Tajikistan 9 5 Senegal 2 11 8
Uzbekistan 1 Sierra Leone 1
Somalia 1 1
South Africa 1
South Sudan 3 1
Sudan 13 13
Tanzania 8 4
Togo 10
Uganda 6 2
Total 5 74 18 2 143 5 30 101 50 12 42 14 23 169 74 29 14 35
Notes: This table shows the geographical distribution of Islamic, conventional, and future Islamic MFIs (denoted ISMFIs, CMFIs, and FISMFIs, respectively) across six global regions. To identify Islamic MFIs, we use three data
sources; the main source was the survey conducted in 2016, but the Sanabel (2012) and the Islamic Banking Database (2014) were also consulted. The identification of conventional MFIs comes from our survey. For the group of
future Islamic MFIs, we added available ISMFIs to CMFIs that indicated they intend to provide Islamic financial products in the future in our survey.
7
Table 2 Table 3
Islamic financial products. Comparing Islamic and conventional MFIs.
Islamic Financial Percentage Percentage of Clients Using Islamic Conventional p-Value

Products Provision Products
Clients
Murabaha 75.8 47.6 Farmers 45.6 34.2 0.031
Mudaraba/Musharaka 17.2 22.6 Salaried persons 12.3 17.2 0.25
Qard e hasan 58.6 23.1 Micro-entrepreneurs 40.9 53.5 0.043
Sala’m 20.7 18.2 Women 64.4 63.6 0.87
Other 24.1 73.5 Clients below poverty line 45.2 43.4 0.78
Lending classification
Rural 69.1 60.7 0.14
Islamic MFIs for our sample, calculated as a sum of the number of all Group 53.1 48.9 0.59
current Islamic MFIs plus all conventional MFIs that indicated they Sources of funds for MFIs
intended to provide Islamic financial products in the future. These Donor agencies 48.2 33.8 0.13
projections show that the shares of MFIs offering Islamic microfi- Philanthropic donations 7.4 6.6 0.87
nance products are expected to grow across all regions (average of Charities 3.7 3.6 0.98
Government support 37 16.4 0.01
1.3%), with the largest growth expected in Eastern Europe and Cen-
Waqf 3.7 4.4 0.86
tral Asia (average of 2.3%).17 Zakat Fund 7.4 0.4 <0.001
This data set also highlights the growth achieved already by
Notes: This table presents the difference between Islamic and conventional MFIs in
Islamic MFIs. Global market share (represented by financial rev-
terms of targeted clients, lending classification, and sources of funds, according to
enue) has increased from US$1 million in 1999 to US$325 million an equality of means test. Data come from our survey.
in 2016, and the market size (represented by total assets) increased
from US$9 million to US$1,827 million in the same period.18 To put
such growth in perspective, we compare this market size with that of
both types attract a predominantly female client base, such that
the total assets of one big U.S. bank, JP Morgan Chase (2017): At the
64.4 percent of Islamic MFI and 63.6 percent of conventional MFI
end of December 2017, it amounted to more than US$2.5 trillion.
clients are female on average. On average, 17.2 percent of clients
That is, on a global scale, the Islamic MFI sector is still small and con-
of conventional MFIs are employed as salaried workers, compared
centrated, mostly in the Middle East. However, our survey results
with 12.3 percent for Islamic MFIs, which aligns with the higher
suggest the number of Islamic MFIs (and conventional MFIs offering
average percentage of poor members in Islamic MFIs’ customer
Islamic products) will grow quickly, so this market share also is
bases (45.2% versus 43.4%). Neither of these differences is statisti-
likely to increase considerably in the near future.
cally significant though. Other differences appear more substantial,
such that 45.6 percent of Islamic MFIs’ client bases engage in farm-
4.2. Characterization of the provision of Islamic financial products ing, versus only 34.2 percent for conventional MFIs (p = .031). In
addition, 53.5 percent of conventional MFIs’ customer base
Our survey results reveal some key features about the popular- includes micro-entrepreneurs, compared with 40.9 percent for
ity, clientele, and funding base of the different instruments. The Islamic MFIs (p = .043). Table 3 reveals no statistically significant
term ‘‘average” in this discussion refers to an average across MFIs. differences in terms of the MFIs’ reliance on rural lending, group
First, equity instruments are relatively uncommon. As Table 2 lending, or donor agencies, yet a significantly higher percentage
shows, mudaraba and musharaka products are issued19 by 17.2 per- of Islamic MFIs report government support (37 percent) and Zakat
cent of Islamic MFIs, and an average of 22.6 percent of Islamic MFI funds21 (7.4 percent) as main sources of funding.
clients use these products.20 This preference might reflect the high
risk associated with equity products and the difficulties associated 4.3. Regression results: outreach and financial sustainability
with determining a project’s yield, which would imply the need
for costly monitoring. Instead, debt instruments, and murabaha 4.3.1. Data and descriptive statistics
and qard e hasan in particular, are substantially more common than Table 4 describes the variables used in the regression analyses
equity instruments. In Table 2, 75.8 percent of Islamic MFIs provide in detail. In particular, for our outreach dependent variable, we
murabaha, 58.6 percent provide qard e hasan, and 20.7 percent pro- include measures of both breadth (serving many people, even if
vide sala’m products. On average, 47.6 percent of Islamic microfi- they are somewhat less poor) and depth (serving the poorest seg-
nance clients use murabaha, 23.1 percent use qard e hasan, and ments of the population) (Brau & Woller, 2004; Schreiner, 2002).
18.2 percent use sala’m. Thus, our results establish that Islamic MFIs The measure of breadth of outreach uses the logarithm of the num-
and their clients mainly rely on murabaha and qard e hasan. ber of active borrowers (LNNAB) of MFI i operating in country j at
Second, there are some notable similarities and differences time t (Cull, Demirgüç-Kunt, & Morduch, 2009; Louis et al., 2013;
between Islamic and conventional MFIs, in terms of their clientele, Mersland & Strøm, 2009); the measure of depth reflects the ratio
lending techniques, and funding sources. According to Table 3, of the average loan size of MFI i at time t to the gross national
income per capita at time t (ALBGNI) of the country j in which it
17
For each region, we calculate the average shares of current Islamic MFIs and of operates. Greater depth implies smaller values of ALBGNI. Although
expected future Islamic MFIs. Then, the expected growth rate of the share of Islamic we note the ongoing discussion about the validity of the outreach
MFIs per region is calculated as (average share of future Islamic MFIs – average share
of current Islamic MFIs)/average share of current Islamic MFIs.
measures (see Section 2.2), more appropriate indicators have not
18
For more information, please see contextual details in Appendix A.1.1. been established yet, so we maintain standard outreach indicators,
19
We calculate the percentage of Islamic MFIs providing a certain Islamic financial in line with prior literature (Cull et al., 2007; Fan et al., 2019;
product as (number of Islamic MFIs offering a particular Islamic product/total number Hermes et al., 2011; Mersland & Strøm, 2010; Quayes, 2012), even
of Islamic MFIs) 100. while acknowledging that they can only give an indication of the
20
We calculate the percentage of clients using Islamic products as (number of
clients using the offered product/total number of clients) 100 for each MFI that
21
offers a particular product. The values in the text represent simple averages of this Zakat is a charitable contribution, mandatory for Muslims who seek to satisfy
percentage, across all MFIs that offer the product. criteria related to their wealth.
Table 4
Variable definitions and sources.
Variable Abbreviation Definition

Main Variable of Interest
Islamic microfinance institution ISMFI Dummy variable, 1 if the MFI uses Islamic lending techniques; 0 otherwise. Source: Survey.
Country-Specific Variables
GDP growth GDPgrowth How fast the economy is growing, calculated by comparing one year of the country’s GDP to the
previous year. Source: http://data.worldbank.org
Instruments
Percent of Muslim population PMP Percentage of Muslims residing in the country. Source: Kettani (2010)
Islamic state Islstate Dummy variable equal to 1 if the official state religion of the country is Islam and 0 otherwise.
Source: Barro and McCleary (2005)
Dependent Variables
Log of number of active borrowers LNNAB Log number of entities with currently outstanding loan balances with the MFI or that are primarily
responsible for repaying any portion of the gross loan portfolio. Entities with multiple loans with an
MFI are counted as a single borrower.
Average loan balance per ALBGNI Average deposit balance per depositor, relative to local GNI per capita, which provides an estimate
borrower/GNI per capita of the coverage of the low-income population achieved through deposits. The indicator, calculated
in national currencies, is converted to U.S. dollars at official exchange rates to enable comparisons
across economies. To smooth fluctuations in prices and exchange rates, a special Atlas method of
conversion is used, as suggested the World Bank.
Return on assets ROA Measure of how well an MFI manages its assets to optimize its profitability. This ratio is net of
income taxes and excludes donations and non-operating items. It is calculated as net operating
income (less taxes) relative to average assets.
MFI-Specific Variables
Market share Mktshare Market concentration of an MFI in terms of earning revenue. We take the fraction of financial
revenue earned by an MFI in a given year with respect to total financial revenues in a given year
earned by all MFIs in the country.
Market size Mktsize Market size is proxied by the log of the assets, or the total value of resources controlled by MFI as a
result of past events, from which future economic benefits are expected to flow. For the calculation,
assets are the sum of each individual asset account listed.
Portfolio at risk >30 days PAR Portion of the loan portfolio ‘‘contaminated” by arrears and at risk of not being paid back. It
represents the outstanding balance in arrears over 30 days in addition to restructured loans,
divided by total outstanding gross portfolio.
Capital-to-assets ratio CAR Representing institutional solvency, it is calculated as total capital divided by risk-weighted assets.
Yield on gross loan portfolio YGLP Earning performance of an MFI, according to how effectively the MFI matches the maturities of its
assets and liabilities. It is calculated as financial revenue from the loan portfolio, divided by average
gross loan portfolio.
Gross loan portfolio-to-assets ratio GLP/assets The relation of an MFI’s loan portfolio to total assets, calculated as gross loan portfolio divided by
total assets.
Maturity in Age DumAge The length of duration since the MFI’s establishment. The dummy variable equals to 1 if the age of
an MFI is over 8 years and 0 otherwise.
Regulated DumReg Dummy variable equal to 1 if the MFI is regulated by some supervisory authority and 0 otherwise.
Notes: Unless otherwise noted, the source for these variables is MIX Market.
social performance of MFIs. We measure financial performance as 0.9 percent for conventional MFIs. The purpose of our regression
the return on assets (ROA), or the ratio of net operating income to analysis is to explore these differences in a more robust manner.
total assets (Ahlin, Lin, & Maio, 2011; Armendáriz & Morduch, The regression analysis also includes a vector of the following
2010; Mersland & Strøm, 2009; Servin et al., 2012; Strøm et al., control variables: (1) market share, which reflects the market con-
2014),22 which varies across MFIs and time. centration of an MFI in terms of earning revenue (Mktshare); (2)
Table 5 provides the means, standard deviations, and ranges of market size proxied by the log of assets (Mktsize), which is MFI-
all the variables in our main regression, including the three depen- and time-specific; (3) whether the portfolio at risk is greater than
dent variables. Foreshadowing our regression analysis, we list 30 days (PAR), which is MFI- and time-specific; (4) the capital-to-
these summary statistics separately for Islamic and conventional assets ratio (CAR), which is MFI- and time-specific; (5) the yield
MFI subsamples (i.e., for which we compare the conditional means on the gross loan portfolio (YGLP); (6) the ratio of the gross loan
in our regression analysis). The log of the number of borrowers is portfolio to assets (GLP/assets); (7) age (AgeDummy) as a dummy
similar across Islamic and conventional MFIs, suggesting a similar variable, where 1 = mature MFI and 0 otherwise, which is MFI-
breadth of outreach. But the average loan balance per borrower and time-specific; and (8) the regulatory status of the MFI
(scaled by gross national income per capita) is much smaller for (RegDummy), another dummy variable, where 1 = MFI is regulated
Islamic than for conventional MFIs (US$0.4 and 0.6, respectively), and 0 otherwise, which is MFI-specific. We also include a country-
so the depth of outreach appears higher for Islamic MFIs. In terms and time-specific control variable GDP growth (GDPgrowth) to cap-
of financial performance, Islamic MFIs underperform; their ture general business cycle variation in the dependent variables in
mean ROA value is lower, equal to 0.6 percent, compared with our panel models. This variable, when used in our cross-sectional
specifications, controls for the average growth rate of a country
in the sample period. Finally, we include time dummies in the
22
In banking studies, it is common to use return on equity (ROE) to measure panel regressions. As noted, the definitions, abbreviations, and
financial performance, but in microfinance studies, it is more common to use ROA, sources of the variables used in our regression analyses are in
because ROE depends on the firm’s capital structure and equity. Our sample includes
non-profit MFIs, which lack equity capital for earnings purposes. As an advantage of
Table 4. Then Table 5 shows that, with the exception of PAR
ROA, it evokes the same interpretation in all categories of MFIs, which facilitates (clearly higher for Islamic MFIs), the control variables are similar
comparisons. across the two groups of MFIs.
Table 5
Descriptive Statistics.
Variable Full sample (response = 1) Islamic MFIs Conventional MFIs

Obs. Mean SD Min. Max. Obs. Mean SD Min. Max. Obs. Mean SD Min. Max.
Dependent Variables
ALBGNI 5,045 0.61 1.27 0 30.67 491 0.43 0.47 0.01 3.35 4,554 0.63 1.33 0 30.67
LNNAB 5,086 9.47 1.90 2.49 15.92 499 9.55 1.92 4.58 14.06 4,587 9.46 1.90 2.49 15.92
ROA 4,407 0.01 0.13 3.45 1.01 414 0.01 0.15 1.45 0.26 3,993 0.01 0.13 3.45 1.01
Independent Variables
GDPgrowth 5,365 4.98 3.39 28.10 54.16 525 4.96 4.71 28.10 54.16 4,840 4.98 3.21 20.49 34.50
Mktshare 4,975 0.12 0.22 0.48 2.15 490 0.17 0.27 0.21 1 4,485 0.11 0.21 0.48 2.15
Mktsize 5,209 15.87 2.01 5.43 22.69 512 15.76 1.76 9.49 19.95 4,697 15.88 2.03 5.43 22.69
PAR 5,330 0.05 0.12 0 5.48 522 0.08 0.27 0 5.48 4,808 0.05 0.09 0 1.05
CAR 5,188 0.34 0.31 4.13 7.12 504 0.42 0.38 1.87 1 4,684 0.33 0.31 4.13 7.12
YGLP 5,312 0.24 0.25 1.33 11.48 520 0.24 0.54 0 11.48 4,792 0.23 0.20 1.33 1.52
GLP/assets 5,181 0.76 0.34 0 11.95 506 0.72 0.23 0 3.68 4,675 0.76 0.35 0 11.95
DumAge 5,366 0.69 0.46 0 1 526 0.60 0.49 0 1 4,840 0.70 0.46 0 1
DumReg 5,366 0.67 0.47 0 1 526 0.61 0.49 0 1 4,840 0.67 0.47 0 1
Instrumental Variables
PMP 5,366 33.86 39.64 0.01 99.98 526 73.20 32.61 0.01 99.98 4,840 29.58 37.95 0.01 99.77
Islstate 5,366 0.21 0.41 0 1 526 0.68 0.47 0 1 4,840 0.16 0.36 0 1
Notes: This table lists summary statistics for key variables for the full sample of respondents, Islamic MFIs, and conventional MFIs. Obs is the number of observations for each
variable; SD is standard deviation; and Min and Max are the minimum and maximum values for the variables, respectively.
Table 6
Regression results for outreach: ALBGNI.
Variables Group Means Fama MacBeth Random Effects Random Effects with IV Inverse Probability with IV
ISMFI 0.257* 0.225*** 0.276*** 1.212*** 1.120***
(0.132) (0.030) (0.103) (0.431) (0.329)
GDPgrowth 0.184 0.018 0.009** 0.010** 0.008
(0.137) (0.010) (0.004) (0.004) (0.011)
Mktshare 1.757*** 0.704*** 0.241** 0.153 0.704***
(0.667) (0.080) (0.120) (0.111) (0.253)
Mktsize 0.058 0.078*** 0.084*** 0.084*** 0.049**
(0.050) (0.011) (0.023) (0.020) (0.023)
PAR 0.029 0.594 0.181 0.172** 0.156**
(0.872) (0.413) (0.115) (0.081) (0.077)
CAR 0.319 0.130*** 0.075 0.066 0.092
(0.234) (0.040) (0.080) (0.075) (0.130)
YGLP 0.001 0.642*** 0.230*** 0.181* 0.612***
(0.635) (0.114) (0.079) (0.094) (0.191)
GLP/assets 0.892* 0.420*** 0.028 0.017 0.204*
(0.467) (0.112) (0.048) (0.036) (0.110)
DumAge 0.111 0.151*** 0.009 0.024 0.188**
(0.174) (0.037) (0.044) (0.048) (0.089)
DumReg 0.270*** 0.298*** 0.400*** 0.383*** 0.293***
(0.104) (0.040) (0.144) (0.144) (0.070)
Constant 1.228 0.400* 0.742** 0.656** 0.457
(5.352) (0.215) (0.314) (0.278) (0.385)
Time dummies Yes Yes Yes Yes Yes
Observations 4,680 4,680 4,680 4,680 4,637
R-squared 0.177 0.137 0.037 0.029 0.035
Number of MFIID/ groups 571 18 571 571 563
Weak Identification Test
F-statistic of excluded instruments 29.30*** 13.49***
Stock-Yogo critical values (TSLS bias) 11.57 11.57
Stock-Yogo critical values (TSLS size) 11.59 11.59
Overidentification Test of All Instruments
Hansen statistic 0.874 3.822
p-Value Hansen test 0.349 0.051
Notes: Standard errors for Group means and RE estimates are based on a bootstrapping method; for Random Effects with IV and Inverse Probability with IV, the standard
errors are robust clustered standard errors (with MFI as the cluster). Group means refers to a between-subjects estimator (based on Group means); Fama MacBeth refers to
results using the Fama-MacBeth method; Random Effects refers to a random effects panel estimate; Random Effects with IV refers to a random effects with instruments panel
estimate; and Inverse Probability with IV refers to panel estimate that combines inverse probability weighting and instruments. The Stock-Yogo TSLS bias critical values are
critical values for the weak instrument test based on TSLS bias (5% significance). The critical value is a function of the number of included endogenous regressors (in our case,
1), the number of instrumental variables (in our case, 2), and the desired maximal bias of the IV estimator relative to ordinary least squares (in our case, 1%). The critical value
is obtained from Skeels and Windmeijer (2018); Stock and Yogo (2005) do not present relative bias tables for two instrumental variables. The Stock-Yogo TSLS size critical
values are critical values for the weak instrument test based on TSLS size (5% significance). The critical value is a function of the number of included endogenous regressors (in
our case, 1), the number of instrumental variables (in our case, 2), and the desired maximal size (in our case, 15%) of a 5 percent Wald test where b = b0.
* Significant at 0.10%. ** Significant at 0.05%. *** Significant at 0.01%.
4.3.2. Regression results The coefficients for the Islamic MFI dummy variable in Table 7
We present the regression results in Table 6 (depth of outreach, are positive and highly statistically significant across almost all
dependent variable AVLNGNI), Table 7 (breadth of outreach, depen- regressions (cf. RE); Islamic MFIs exhibit greater breadth of out-
dent variable LNNAB), and Table 8 (financial performance, depen- reach than conventional MFIs, ceteris paribus. Apparently, potential
dent variable ROA). In each table, we provide results obtained liquidity problems created because Islamic MFIs cannot borrow
with the five estimation methods we introduced previously (Group from conventional MFIs (see Section 2) do not restrict their breath
means, FM, RE, REIV, and IPWIV). Supplementary regressions for of outreach. Larger MFIs, measured by asset size, also indicate
the first-stage are provided in Appendix C, Table C.1 and C2. greater breadth, which contrasts with the results for depth. But
In Table 6, the coefficients for the Islamic MFI dummy variable in line with our depth results, we find that older MFIs indicate
are negative and highly statistically significant in all specifications. greater breadth in most specifications. That is, older MFIs seem
Therefore, and in line with our first hypothesis, Islamic MFIs to have greater outreach, in terms of both depth and breadth. Reg-
appear to exhibit greater depth of outreach than conventional MFIs, ulated MFIs, according to most specifications, exhibit less breadth
controlling for macroeconomic growth, the size of the MFI proxied and depth. According to the coefficients on GDP growth in most
by the log of assets (Mktsize), the age of the MFI, other relevant MFI specifications, countries that grew faster on average in the sample
characteristics (e.g., market concentration), being regulated, and period also featured greater breadth of outreach. The portfolio at
time dummies. Most of the regression models show that MFIs with risk and capital-to-assets ratio coefficients suggest that higher val-
greater market shares (Mktshare) and bigger, more regulated MFIs ues are associated with less breadth, though not always
(Mktsize and DumReg, respectively) offer less depth of outreach. significantly.
Somewhat surprisingly, older MFIs reveal greater depth. We also The size of the coefficient for ISMFI is similar for the Group
note some indications that periods of higher GDP growth are asso- means, FM, and RE estimates, which offers some reassurance
ciated with less depth of outreach. Only the random-effects speci- regarding the robustness of the results. However, the (absolute
fications RE and REIV show significant results. Finally, the value of the) coefficient for ISMFI is bigger for both IV estimates
coefficients for the portfolio at risk and capital-to-assets ratio are (REIV and IPWIV). The relative magnitude of the estimates, and
mostly statistically insignificant in Table 6. the ability of the IV estimates to correct for endogeneity bias,
Table 7
Regression Results for Outreach: LNNAB.
ISMFI 0.456*** 0.356*** 0.162 2.726*** 3.059***
(0.142) (0.084) (0.142) (0.597) (0.793)
GDPgrowth 0.128*** 0.057*** 0.006* 0.006* 0.055***
(0.041) (0.015) (0.003) (0.003) (0.017)
Mktshare 0.877** 0.370*** 0.391*** 0.378*** 0.353
(0.386) (0.092) (0.099) (0.094) (0.387)
Mktsize 0.746*** 0.699*** 0.747*** 0.750*** 0.821***
(0.044) (0.032) (0.030) (0.028) (0.037)
PAR 1.668** 1.468*** 0.043 0.044 0.538**
(0.703) (0.250) (0.112) (0.097) (0.209)
CAR 0.371 0.402*** 0.105 0.108 0.483**
(0.246) (0.061) (0.083) (0.077) (0.240)
YGLP 0.085 0.548** 0.326*** 0.302*** 0.739**
(0.522) (0.213) (0.092) (0.104) (0.296)
GLP/assets 0.619*** 0.664*** 0.445*** 0.446*** 0.593***
(0.206) (0.102) (0.167) (0.144) (0.157)
DumAge 0.470*** 0.449*** 0.066 0.070* 0.348***
(0.157) (0.102) (0.040) (0.042) (0.132)
DumReg 0.206* 0.059 0.265** 0.235* 0.170
(0.108) (0.059) (0.108) (0.131) (0.146)
Constant 4.725 2.547*** 2.535*** 2.877*** 3.493***
(5.217) (0.488) (0.404) (0.384) (0.586)
Observations 4,693 4,693 4,693 4,693 4,650
R-squared 0.629 0.619 0.588 0.769 0.517
is obtained from Skeels and Windmeijer (2018), Stock and Yogo (2005) do not present relative bias tables for two instrumental variables. The Stock-Yogo TSLS size critical
our case, 1), the number of instrumental variables (in our case, 2), and the desired maximal size (in our case, 15%) of a 5 percent Wald test where b = b0.
* Significant at 0.10%. ** Significant at 0.05%. *** Significant at 0.01%.
Table 8
Regression results for financial performance: ROA.
ISMFI 0.013 0.022* 0.013 0.066 0.043
(0.025) (0.012) (0.020) (0.054) (0.036)
GDPgrowth 0.002 0.001 0.001 0.001 0.001
(0.002) (0.001) (0.001) (0.001) (0.001)
Mktshare 0.005 0.001 0.001 0.003 0.018
(0.023) (0.012) (0.036) (0.034) (0.019)
Mktsize 0.015*** 0.011*** 0.027*** 0.024*** 0.010***
(0.005) (0.002) (0.005) (0.004) (0.002)
PAR 0.216 0.161*** 0.064 0.067* 0.009
(0.191) (0.038) (0.046) (0.039) (0.023)
CAR 0.097*** 0.064*** 0.065*** 0.067*** 0.088***
(0.030) (0.016) (0.024) (0.023) (0.012)
YGLP 0.061 0.101 0.124*** 0.104** 0.018
(0.088) (0.060) (0.043) (0.044) (0.031)
GLP/assets 0.068*** 0.090*** 0.041*** 0.040*** 0.046***
(0.025) (0.018) (0.011) (0.011) (0.008)
DumAge 0.001 0.032*** 0.005 0.007 0.016*
(0.028) (0.008) (0.007) (0.007) (0.008)
DumReg 0.018 0.010 0.011 0.012 0.012
(0.014) (0.012) (0.015) (0.015) (0.008)
Constant 0.188 0.276*** 0.426*** 0.379*** 0.180***
(0.276) (0.057) (0.084) (0.061) (0.059)
Observations 4,379 4,379 4,379 4,379 4,338
R-squared 0.142 0.215 0.065 0.092 0.083
is obtained from Skeels and Windmeijer (2018); Stock and Yogo (2005) do not present relative bias tables for two instrumental variables. The Stock-Yogo TSLS size critical
our case, 1), the number of instrumental variables (in our case, 2), and the desired maximal size (in our case, 15%) of a 5 percent Wald test where b = b0. * Significant at 0.10%.
** Significant at 0.05%. *** Significant at 0.01%.
may seem to suggest that the actual effect size is greater than selection problems that arise when Islamic MFIs use PLS contracts
implied by our OLS estimates, but this conclusion would be unwar- do not lead to significantly lower ROA. However, more research is
ranted. In the OLS regression, the independent variable ISMFI is needed to explain the precise reasons for this outcome. Arguably, it
binary (0, 1), but the IV procedure transforms it into a probability. could arise because Islamic MFIs receive funding from donations.
Because the probability of being an ISMFI is always less than 1, the In addition, in terms of control variables, larger MFIs and those
coefficient would need to be larger to produce the same effect size. with higher capital-to-assets ratio display significantly higher ROA.
Therefore, the IV estimates in our case, where the endogenous vari- In summary, the results in Tables 6 and 7 strongly suggest that
able is a dummy, likely overestimate the actual effect size.23 outreach by Islamic MFIs is better than that by conventional MFIs,
Finally, Table 8 provides weak evidence in support of our sec- in terms of both depth and breadth, so we confirm our first hypoth-
ond hypothesis that Islamic MFIs financially underperform con- esis. However, Table 8 only provides weak evidence for our
ventional MFIs. The point estimate of the coefficient for IMFI is hypothesis that Islamic MFIs underperform financially: The coeffi-
negative for all regressions but statistically insignificant in almost cient is negative but almost always insignificant.
all cases. Apparently, the higher operational costs associated with As we noted in the previous section, two approaches (REIV and
debt-like contracts (e.g., murabaha) and the potential for adverse IPWIV) use instruments to control for the possible endogeneity
bias.24 Valid (external) instruments must correlate with the endoge-
nous variable (ISMFI) but not the error term. Thus, appropriate exclu-
23
In an effort to make IV and OLS estimates comparable, de Jong (2016) suggests
correcting the coefficient of the IV estimate by multiplying it by the mean value of the
first-step regression fitted values. In our case, we could approximate this value by
24
taking the percentage of Islamic MFIs in the data set (i.e., percentage of 1 values for Although the data for most of the control variables come from the Mix Market
the endogenous dummy). When we do so, recalling that the percentage of Islamic data set, information about two instruments (percentage of Muslim population and
MFIs is about 11 percent, the coefficients of the IV estimates become comparable to being an Islamic state) comes from other sources, which theoretically could introduce
those of the OLS estimates. a bias.
sion restrictions need to hold, and the instruments cannot affect out- provide a comparative analysis of their performance, relative to
come variables unless they go through ISMFI. We use the Hansen J that of conventional MFIs.
statistic tests for joint instrument validity to test the null hypothesis The mapping exercise identifies 101 institutions in 33 countries
that the instruments are valid and uncorrelated with the error term. offering Sharia-compliant microfinance services. South Asia
Lower p-values indicate stronger evidence that the instruments are accounts for the largest number of Islamic MFIs, with 30 (13 in
not valid, so rejecting the null hypothesis of the validity of the Pakistan, 9 in Bangladesh). The Middle East and North Africa also
instruments would need to be reconsidered. However, not rejecting feature many Islamic MFIs, with 29 providers, and sub-Saharan
the null hypothesis does not mean the instruments are valid. We Africa accounts for 23 Islamic MFIs. In terms of the availability of
also test for the failure of the relevance condition and weak instru- different types of Islamic financial products, murabaha, a debt-
ments. The first-stage F-statistic of the excluded instrument is, in based financing product, is the most widespread, and 64.7 percent
case of a single endogenous variable (as in our case), equal to the of Islamic MFIs in our sample provide it. Due to the high risk asso-
Kleinbergen-Paap F statistic; we compare it with Stock and Yogo ciated with equity-based products such as musharaka, they are
(2005) critical values. According to the relevant IV test statistics in offered by relatively fewer MFIs.
Tables 6–8, for all models, the null hypothesis of weak instruments When we test whether Islamic MFIs are better able to reach the
is rejected, as also is true for small instrumental biases relative to socially oriented goal of increased outreach, in line with our pre-
the OLS estimator and small IV sizes. Moreover, the Hansen J statistic dictions, we find that Islamic MFIs perform better. In particular,
suggests that we cannot reject the null prediction of instrumental they serve more and poorer borrowers. In terms of financial results,
validity for all but two specifications (ALBGNI in inverse probability we find some weak evidence that Islamic MFIs underperform con-
with IV and ROA in random effects with IV). ventional MFIs, though in most cases, the differences are not statis-
tically significant.
Some caveats must be noted. The results clearly indicate that
4.3.3. Robustness checks Islamic MFIs perform better in terms of outreach, but we reiterate
In addition to the regression results in Tables 6–8, we present the caution that we cannot address all the potential endogeneity
three sets of alternative regressions in Appendix D, see Table D1, issues. As a limitation of our analysis, we also cannot identify
D.2.1, D.2.2, D.2.3, D.3.1, D.3.2, D.3.3, D.4.1, D.4.2 and D.4.3. First, whether the positive results reflect differences in the production
we consider a limited set of control variables, such that we exclude function between the two types of MFIs or differences in the clien-
time dummies. Second, we include a control variable for fees and tele. In other words, we are not able to robustly identify whether
commission in another set of regressions, because Islamic MFIs the increase in outreach is due to involuntary or voluntary financial
might charge fees.25 Controlling for fees in the estimates might be exclusion. Islamic clients, for religious reasons, may refuse to deal
relevant, because excluding them could bias financial performance with conventional MFIs, so our results could be partly explained by
measures. Although we have limited information about these fees demand effects and voluntary exclusion. Yet several studies assert
and thus cannot control fully for this potential bias, MixMarket pub- that Muslim consumers are not solely interested in Islamic
lishes, for a limited group of MFIs, information about a variable it calls finance—Dar (2004) claims that that only 5 percent of U.K. Muslims
‘‘Fee and commission income on loan portfolio,” which includes seek Islamic financial services—so we reason that our results are
‘‘penalties, commissions, and other fees earned on the loan portfolio, more likely to be (mainly) explained by differences in production
other than penalty fees for late payment. It also includes revenues functions. Still, this conclusion requires caution, and our results
under Islamic finance methods.‘‘ Therefore, we add this partial should be interpreted as robust correlations rather than evidence
control (see Table D.1) and provide the regression results in of causality. We hope these findings in turn prompt continued
Appendix D.26 Third, another set of regressions (Tables D.4.1 to comparative studies of the performance of Islamic and conven-
D.4.3) uses the binary dummy for Sharia-compliant MFIs as the main tional MFIs along other dimensions, as well as studies that delve
independent variable; it identifies Islamic MFIs that only offer Islamic deeper into the reasons for their varying social performance.
microfinance products (whereas in the main analysis, we also
included Islamic MFIs that offer both conventional and Islamic
microfinance products). We thus can consider whether Islamic MFIs CRediT authorship contribution statement
offering conventional products differ from those offering only Islamic
products. In Appendix D, the three sets of additional regressions Syedah Ahmad: Conceptualization, Methodology, Formal anal-
provide results very similar to our main regressions, in support of ysis, Investigation, Writing - original draft, Data curation. Robert
the robustness of our results. Lensink: Conceptualization, Methodology, Formal analysis, Inves-
tigation, Writing - original draft, Supervision. Annika Mueller:
Conceptualization, Methodology, Investigation, Writing - original
5. Conclusion draft, Supervision.
This study has pursued two main aims. First, we sought a

clearer picture of the global expansion and performance of Islamic Declaration of Competing Interest
MFIs by offering, to the best of our knowledge, the first compre-
hensive mapping of their global presence. Second, we attempt to The authors declare that they have no known competing finan-
cial interests or personal relationships that could have appeared
25
to influence the work reported in this paper.
The service fee usually reflects the MFIs’ administrative costs, incurred during
loan disbursements. Generally, MFIs that provide qard e hasan (small, interest-free
loans) charge less than 1 percent as a one-time fee (Maazullah, 2017). Although
Acknowledgements
Islamic MFIs with equity-based contracts might charge higher fees, we lack detailed
information about this feature in our data set.
26
In addition, we have conducted estimates, for which, following Dorfleitner, Leidl, We thank participants of seminars at SOAS University of Lon-
Priberny, and von Mosch (2013), we added a lending rate LR of MFIs calculated don, and the Faculty of Economics and Business, University of
as;LR ¼ GLPð1WORÞ
Income
þ Fees
GLP where Income is interest/profit rate, Yield is yield on gross Groningen, as well as participants of The European Research Con-
portfolio, GLP is gross loan portfolio, fees is fee and commission earned and WOR is
ference on Microfinance for valuable comments. We are also grate-
the write-off ratio. The results are similar to the main results and can be obtained on
request. ful for very helpful comments from three anonymous referees.
Appendix A. Data construction Table A.1.1

Highlights of non-profit versus for-profit Islamic MFIs.
A.1. Constructing the dataset
Variables Non-Profit For-Profit
We construct our dataset by combining our online survey with Islamic MFIs Islamic MFIs
the existing data sources and published reports. Below, we provide (in Millions) (in Millions)
a detailed description of the data sources which are employed in Market size 1337 149
our study. Market share 296 26
Gross Loan portfolio 1113 82
A.1.2. Details for online survey Number of borrowers 4 0.085
Number of depositors 3 0.285
– Survey format and implementation: For receiving maximum Notes: The table represents the data for Islamic MFIs in 2016 and
responses we designed a short survey that can be completed the values are presented in Million. Total assets are taken for repre-
within 5 min on average. For achieving this, we had to make a senting market size; financial revenue is taken for representing
choice between amassing detailed information from Islamic market share. Data source: Author’s survey and MIX market data.
microfinance institutions or gathering a more general informa-
tion from both Islamic and non-Islamic microfinance institu-
tions. To make the most use of a short survey, we collect A.1.2. Existing databases
information from both Islamic and non-Islamic microfinance MIX Market database: For MFI-level variables, we use the data
institutions, about the indicators which are not available at (financial and social indicators) on the MFIs, provided by MIX
MIX market database. Market.
We organized our survey into three key sections: (1) identifica- World Bank: For country-level variables, we collected data from
tion of Islamic microfinance service providers; (2) provision of World Bank database and it includes information about country’s
Islamic microfinance products; and (3) information about clien- growth.
tele, lending methodologies and sources of funding. We organized
the questions in an order so that the information was first col- A.1.3. Published reports
lected on the type of financial products offered i.e., interest- We consulted some published reports for constructing our
based, interest-free or both types. Following the response, we instrumental variables, Kettani (2010) and Barro and McCleary
then asked specific questions each related to adherence with the (2005).
Islamic principles or future plans of providing Islamic financial
products. Appendix B. Variable’s information/details
With responses to Section 1, we identify those MFIs who at pre-
sent provide microfinance products in line with Islamic principles B.1. Percentage of Muslim population (PMP)
as well as those MFIs who plan to provide such products in the
future. Additionally, we identify those non-Islamic MFIs that are We use the data from Kettani (2010) because this is the only
providing interest-free microfinance products. Section 2 was reference available that provides data covering the entire period
answered by Islamic MFIs only because we used skip and display of our study. Kettani provides the percentage of Muslims’ popula-
logic in our survey design. Subject to the completion of the survey, tion by using official census and reliable national sources of data.
Section 3 was filled by all responding MFIs. The only other available dataset is the Pew Research Center
dataset (2015) which is based on the data from the study con-
– Data collection mode: We administered our online survey ducted in 2009–2010. This source provides data for the last period
through Qualtrics; a web service used for conducting online (i.e., 2010 and 2020), but not for earlier period of our study. In
surveys. We programmed Qualtrics to display one survey ques- order to avoid using two datasets for the same variable we pre-
tion at a time allowing the direct entry of the responses, while ferred using the Kettani dataset for the entire period.27
employing the display and skip logic checks. It also facilitates a Both available datasets use official census as the primary source
user friendly survey design so that the survey can be accessed for attaining percentage of Muslims which, for any country the offi-
via mobile phone by the respondent. For quality controls and cial population census, is after every ten years. Thus, both of the
to avoid multiple responses by the same respondent, the system available datasets have a time lapse of ten years. There is another
assigns only one specified web-link to each respondent email ID caveat about Pew’s report that it does not provide information
for survey participation. After the survey completion, the data about the growth rate of Muslims making it impossible for achiev-
can be downloaded from the server in a format compatible for ing the yearly percentage of the Muslim population. While Kettani
analysis. documents annual population growth rate (APGR) which we
– Contextual details: Following details are for the calendar year employ to estimate the yearly percent of Muslims.
2016 that is the period when the online survey identifying Since our study period is 1999–2016, we project the estimate
Islamic MFIs is conducted. In our data, most of the Islamic n
for each year as PopulationFuture ¼ PopulationPresent xð1 þ iÞ , where i
MFIs are found to be not-profit MFIs. Concerning the market is the growth rate and n is the number of years, which is 10 years
size in our data, total assets for non-profit Islamic MFIs are for our study.
1337 Million and for profit Islamic MFIs are149 Million. In
terms of Active Borrowers, the end of the year 2016 shows
non-profit Islamic MFIs to be the largest provider with over
4 Million clients and a market share of 296 Million. The
share of non-profit Islamic microfinance segment remained
27
stagnant at 3 Million in terms of number of depositors from We also compared the projections provided in the Kettani dataset with that of
Pew Research Center dataset for 2010 and 2020. The correlation between the two
2015 to 2016. A quick highlight is provided in Table A.1.1
datasets was extremely high (correlation coefficient is 0.998), so we don’t have any
below. reason to believe that the use of the Kettani dataset would bias our results.
Appendix C. Supplementary tables for first stage regressions. C.2.1 (continued)
Response Coeff. Std. z P>z [95% Interval]

Table C.1 Err. Conf.
Average Loan Balance per borrower per GNI (Instrumental random 2007 0.418 0.218 1.920 0.055 0.844 0.009
effects). 2008 0.428 0.218 1.960 0.050 0.855 0.001
2009 0.373 0.218 1.710 0.087 0.800 0.054
ISMFI Coeff. Robust Std. Err. t P>t
2010 0.350 0.218 1.610 0.108 0.777 0.077
GDPgrowth 0.0002 0.0001 1.3 0.194 2011 0.314 0.218 1.440 0.150 0.741 0.113
Mkrshare 0.014 0.005 2.59 0.010 2012 0.269 0.220 1.220 0.221 0.700 0.162
Mktsize 0.001 0.002 0.69 0.491 2013 0.234 0.221 1.060 0.290 0.667 0.200
PAR 0.003 0.002 1.22 0.224 2014 0.166 0.220 0.750 0.450 0.597 0.265
CAR 0.003 0.004 0.8 0.424 2015 0.216 0.222 0.970 0.330 0.650 0.218
YGLP 0.015 0.012 1.25 0.213 2016 0.145 0.225 0.640 0.521 0.587 0.297
GLP/assets 0.000 0.001 0.01 0.992
DumAge 0.003 0.002 2.08 0.037 Legal status of MFIs
DumReg 0.010 0.027 0.35 0.723 CU 0.264 0.098 2.690 0.007 0.071 0.456
PMP 0.001 0.001 1.68 0.094 NBFI 0.288 0.081 3.540 0.000 0.128 0.447
Islstate 0.238 0.062 3.86 0.000 NGO 0.791 0.094 8.420 0.000 0.607 0.975
Constant 0.028 0.026 1.07 0.285 Other 0.610 0.190 3.220 0.001 0.238 0.981
Year Dummies Yes Rural Bank 0.579 0.152 3.820 0.000 0.876 0.282
Observations 4,680 GDPgrowth 0.0004 0.007 0.060 0.955 0.014 0.015
Mktshare 0.759 0.148 5.140 0.000 0.470 1.049
Notes: This table presents first stage results for instrumental Mktsize 0.093 0.014 6.520 0.000 0.065 0.122
random effects, where Coeff represents coefficient estimates, Std. PAR 0.399 0.193 2.070 0.038 0.777 0.022
Err shows the standard error term, and t represents t-statistics. CAR 0.119 0.071 1.690 0.092 0.019 0.258
ISMFI (a dummy variable which is 1 when an MFI provides Islamic YGLP 0.855 0.115 7.440 0.000 0.630 1.080
financial products/services and 0 when it provides conventional GLP/assets 0.015 0.012 1.250 0.211 0.039 0.009
financial products whether or not it charges interest rate from DumAge 0.014 0.050 0.270 0.786 0.084 0.111
the clients) is the dependent variable for the first stage. GDPgrowth DumReg 0.439 0.059 7.450 0.000 0.323 0.554
measures how fast the economy is growing and is calculated by Constant 2.755 0.362 7.600 0.000 3.465 2.045
comparing one year of the country’s GDP to the previous year. Mkt- Observations 15,531
share is the fraction of financial revenue earned by an MFI in a Log 3180.13
given year with respect to total financial revenues in a given year likelihood
earned by all MFIs in the country. Mktsize is the log of total assets. Pseudo R2 0.165
PAR is portfolio at risk and is that portion of loan portfolio which is
‘‘contaminated” by arrears and is at the risk of not being paid back Notes: The dependent variable is a dummy, equal to 1 if the MFI
to MFI. CAR is capital to assets ratio representing the institutional returned the questionnaire. Coeff represents coefficient estimates,
solvency and is calculated as total capital divided by risk weighted Std. Err shows the standard error term, and z represents z-
assets. YGLP represents financial revenue from loan portfolio statistics. As independent variables, we use dummies for the fiscal
divided by average gross loan portfolio. GLP/assets measure the year. Legal status of the MFIs include; Bank is dummy variable iden-
relation of an MFI’s loan portfolio to the total assets. DumAge is a tifying an MFI as 1 if statuary status is bank; 0 otherwise, CU is credit
dummy variable that equals to 1 if the age of an MFI is over 8 union/cooperative and is a dummy variable identifying an MFI as 1 if
years; 0 otherwise. DumReg is a dummy variable identifying an statuary status is a credit union or a cooperative; 0 otherwise, NBFI is
MFI as 1 if it is regulated by some supervisory authority; 0 other- Non-Banking Financial Intermediary and dummy variable identify-
wise. PMP is the percent of Muslim population in a country where ing an MFI as 1 if statuary status is a credit union or a cooperative;
MFI is located. Islstate is 1 if the state religion of the country is 0 otherwise, NGO is non-governmental organization and is a dummy
Islam; 0 otherwise. variable identifying an MFI as 1 if statuary status is non-
governmental organization; 0 otherwise and Other is a dummy vari-
C.2. Weighted instrumental regressions able identifying an MFI as 1 if statuary status is other; 0 otherwise.
GDPgrowth measures how fast the economy is growing and is calcu-
C.2.1. Logistic regression for weighting the variables (inverse lated by comparing one year of the country’s GDP to the previous
probability weighting) year. Mktshare is the fraction of financial revenue earned by an
MFI in a given year with respect to total financial revenues in a given
Response Coeff. Std. z P>z [95% Interval] year earned by all MFIs in the country. Mktsize is the log of total
Err. Conf. assets. PAR is portfolio at risk and is that portion of loan portfolio
which is ‘‘contaminated” by arrears and is at the risk of not being
Fiscal Year
paid back to MFI. CAR is capital to assets ratio representing the insti-
2000 0.023 0.253 0.090 0.928 0.473 0.519
tutional solvency and is calculated as total capital divided by risk
2001 0.055 0.237 0.230 0.817 0.520 0.410
weighted assets. YGLP represents financial revenue from loan portfo-
2002 0.224 0.226 0.990 0.320 0.667 0.218
lio divided by average gross loan portfolio. GLP/assets measure the
2003 0.416 0.220 1.900 0.058 0.847 0.014
relation of an MFI’s loan portfolio to the total assets. DumAge is a
2004 0.389 0.217 1.790 0.074 0.815 0.037
dummy variable that equals to 1 if the age of an MFI is over 8 years;
2005 0.418 0.216 1.940 0.053 0.841 0.005
0 otherwise. DumReg is a dummy variable identifying an MFI as 1 if it
2006 0.417 0.216 1.930 0.054 0.841 0.007
is regulated by some supervisory authority; 0 otherwise.
Table C2 shows the standard error term, and t represents t-statistics. ISMFI
Average Loan Balance per GNI (Inverse Probability with Instru- (a dummy variable which is 1 when an MFI provides Islamic finan-
mental variables). cial products/services and 0 when it provides conventional financial
products whether or not it charges interest rate from the clients) is
ISMFI Coeff. Robust Std. Err. t P>t the dependent variable for the first stage. GDPgrowth measures how
GDPgrowth 0.0004 0.002 0.180 0.860 fast the economy is growing and is calculated by comparing one
Mkrshare 0.101 0.058 1.730 0.083 year of the country’s GDP to the previous year. Mktshare is the frac-
Mktsize 0.009 0.006 1.410 0.158 tion of financial revenue earned by an MFI in a given year with
PAR 0.138 0.027 5.080 0.000 respect to total financial revenues in a given year earned by all MFIs
CAR 0.000 0.032 0.000 0.999 in the country. Mktsize is the log of total assets. PAR is portfolio at
YGLP 0.001 0.044 0.020 0.986 risk and is that portion of loan portfolio which is ‘‘contaminated”
GLP/assets 0.010 0.009 1.130 0.260 by arrears and is at the risk of not being paid back to MFI. CAR is
DumAge 0.042 0.017 2.400 0.017 capital to assets ratio representing the institutional solvency and
DumReg 0.004 0.028 0.150 0.879 is calculated as total capital divided by risk weighted assets. YGLP
PMP 0.0001 0.0004 0.280 0.781 represents financial revenue from loan portfolio divided by average
Islstate 0.312 0.067 4.680 0.000 gross loan portfolio. GLP/assets measure the relation of an MFI’s loan
Constant 0.116 0.098 1.190 0.236 portfolio to the total assets. DumAge is a dummy variable that
Year Dummies Yes equals to 1 if the age of an MFI is over 8 years; 0 otherwise. DumReg
Observations 4,637 is a dummy variable identifying an MFI as 1 if it is regulated by
some supervisory authority; 0 otherwise. PMP is the percent of
Notes: This table presents first stage results for instrumental ran- Muslim population in a country where MFI is located. Islstate is 1
dom effects, where Coeff represents coefficient estimates, Std. Err if the state religion of the country is Islam; 0 otherwise.
Appendix D. Robustness checks.
Table D1
Variable definitions and descriptive statistics.
Variable name Definition N Mean SD

Fees and commission Fees and commission earned by an MFI refers to penalties, commissions, and other 3,519 7,82,187 3,008,610
earned (Fees) fees earned on the loan portfolio, excluding penalty fees for late payment. Under
Islamic finance methods, this may also include revenues earned by an MFI.
Sharia compliant Sharia compliant is a binary variable and indicates Islamic MFIs that only offer 5,366 0.041 0.198
Islamic microfinance products.
Notes: This table reports variable definitions, number of observations (N), means and standard deviations (Std. Dev.) of the dependent vari-
ables used for data quality estimates. For fees, the data is taken from MIX Market and Sharia compliant stems from the authors’ survey.
D.2. Results with smaller set of controls
Table D.2.1
Variables Group Means Fama MacBeth Random Effects Random Effects Inverse Probability
with IV with IV
ISMFI 0.317*** 0.150*** 0.333*** 1.226*** 0.949***
(0.108) (0.034) (0.107) (0.416) (0.306)
GDPgrowth 0.130 0.014 0.009*** 0.010*** 0.010
(0.130) (0.009) (0.003) (0.004) (0.010)
Mktsize 0.167*** 0.112*** 0.043** 0.049** 0.078***
(0.065) (0.011) (0.021) (0.020) (0.022)
PAR 0.979 0.811 0.179* 0.169** 0.144
(0.733) (0.479) (0.100) (0.079) (0.099)
CAR 0.331* 0.087*** 0.072 0.066 0.055
(0.198) (0.026) (0.074) (0.066) (0.116)
DumAge 0.379* 0.190*** 0.050 0.068 0.193**
(0.204) (0.045) (0.043) (0.045) (0.089)
DumReg 0.301* 0.309*** 0.424*** 0.406*** 0.335***
(0.179) (0.035) (0.156) (0.139) (0.071)
Constant 2.368* 1.297*** 0.193 0.176 0.692**
(1.318) (0.175) (0.317) (0.285) (0.335)
Table D.2.1 (continued)

with IV with IV
Observations 4,894 4,894 4,894 4,894 4,848
R-squared 0.064 0.095 0.036 0.018 0.022
Weak identification test

Stock-Yogo critical values (TSLS Bias) 11.57 11.57
Stock-Yogo critical values (TSLS Size) 11.59 11.59
Overidentification test of all instruments

Hansen Statistic 0.683 5.204
P-value Hansen test 0.409 0.023
Notes: * refers to significant at 0.10%; ** significant at 0.05%; *** significant at 0.01%. Standard errors for Group Means and RE estimates are
based on a bootstrapping method; For Random Effects with IV and Inverse Probability with IV standard errors are robust clustered standard
errors (with MFI as cluster). Group Means refers to a between estimator (results based on group means); Fama MacBeth refers to results using
the Fama MacBeth method; Random Effects refers to a random effects panel estimate; Random Effects with IV refers to a random effects with
instruments panel estimate and Inverse Probability with IV refers to panel estimate that combines inverse probability weighting and instru-
ments. The Stock-Yogo TSLS Bias critical values are critical values for the Weak Instrument Test Based on TSLS Bias: Significance level is 5%;
The critical value is a function of the number of included endogenous regressors (in our case 1), the number of instrumental variables (in our
case 2), and the desired maximal bias of the IV estimator relative to OLS (in our case 1%). The critical value is obtained from Skeels and
Windmeijer (2018) as Stock-Yogo do not present relative bias tables for 2 instrumental variables (it present bias tables only for 3 or more
instrumental variables).The Stock-Yogo TSLS Size critical values are critical values for the Weak Instrument Test Based on TSLS Size: Signif-
icance level is 5%; The critical value is a function of the number of included endogenous regressors (in our case 1), the number of instrumen-
tal variables (in our case 2), and the desired maximal size (in our case 15%) of a 5% Wald test of b = b0.
Table D.2.2
Regression results for outreach: LNNAB
with IV with IV
ISMFI 0.357** 0.280*** 0.144 2.454*** 2.883***
(0.142) (0.079) (0.159) (0.566) (0.790)
GDPgrowth 0.154*** 0.066*** 0.001 0.001 0.054***
(0.036) (0.014) (0.003) (0.003) (0.015)
Mktsize 0.704*** 0.681*** 0.689*** 0.688*** 0.766***
(0.037) (0.029) (0.021) (0.021) (0.033)
PAR 1.892*** 1.604*** 0.076 0.077 0.537**
(0.598) (0.303) (0.141) (0.115) (0.258)
CAR 0.359* 0.442*** 0.059 0.060 0.334
(0.198) (0.076) (0.099) (0.105) (0.232)
DumAge 0.751*** 0.477*** 0.056 0.058 0.320**
(0.141) (0.104) (0.047) (0.045) (0.130)
DumReg 0.218** 0.104* 0.227** 0.201 0.160
(0.104) (0.050) (0.098) (0.123) (0.147)
Constant 2.769*** 1.628*** 1.508*** 1.779*** 3.288***
(0.593) (0.436) (0.325) (0.319) (0.531)
Observations 4,914 4,914 4,914 4,914 4,868
R-squared 0.607 0.606 0.588 0.472 0.504


Table D.2.3
Regression results for Financial Performance: ROA.
with IV with IV
ISMFI 0.016 0.027** 0.021 0.092* 0.054
(0.024) (0.012) (0.021) (0.054) (0.038)
GDPgrowth 0.002 0.002* 0.001 0.001 0.002*
(0.003) (0.001) (0.001) (0.001) (0.001)
Mktsize 0.011*** 0.011*** 0.018*** 0.018*** 0.008***
(0.004) (0.002) (0.003) (0.003) (0.002)
PAR 0.254 0.152*** 0.074 0.075* 0.014
(0.171) (0.033) (0.052) (0.043) (0.025)
CAR 0.086*** 0.063*** 0.077*** 0.078*** 0.092***
(0.030) (0.016) (0.020) (0.020) (0.011)
DumAge 0.026 0.029*** 0.001 0.003 0.015**
(0.019) (0.006) (0.008) (0.007) (0.007)
DumReg 0.025* 0.005 0.015 0.014 0.011
(0.015) (0.014) (0.015) (0.015) (0.009)
Constant 0.237*** 0.200*** 0.325*** 0.305*** 0.165***
(0.073) (0.052) (0.055) (0.053) (0.040)
Observations 4,390 4,390 4,390 4,390 4,349
R-squared 0.077 0.163 0.061 0.049 0.050


D.3. Results after adding fees and commission on loans
Table D.3.1
Regression results for outreach: ALBGNI
Variables Group means Fama MacBeth Random Effects Random Effects Inverse Probability
with IV with IV
ISMFI 0.242** 0.217*** 0.176** 1.024*** 1.056***
(0.117) (0.026) (0.079) (0.294) (0.315)
GDPgrowth 0.027 0.018 0.005 0.005 0.003
(0.024) (0.012) (0.004) (0.004) (0.008)
Fees 0.000 0.000* 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Mktshare 1.566** 0.769*** 0.164 0.056 0.950***
(0.628) (0.120) (0.163) (0.175) (0.316)
Mktsize 0.034 0.081*** 0.104*** 0.099*** 0.058**
(0.043) (0.009) (0.023) (0.020) (0.025)
PAR 0.439 0.223* 0.432*** 0.399*** 0.280
(0.823) (0.118) (0.128) (0.121) (0.199)
CAR 0.366* 0.080 0.061 0.052 0.035
(0.219) (0.048) (0.073) (0.060) (0.129)
YGLP 0.084 0.536*** 0.144 0.099 0.443**
(0.540) (0.143) (0.111) (0.101) (0.180)
GLP/assets 0.296 0.300** 0.004 0.013 0.274**
(0.247) (0.111) (0.068) (0.053) (0.111)
DumAge 0.109 0.151*** 0.056 0.074 0.236***
(0.138) (0.041) (0.052) (0.056) (0.085)
DumReg 0.156 0.197*** 0.205 0.197 0.231***
(0.128) (0.032) (0.146) (0.134) (0.072)
Constant 0.092 0.480* 1.119*** 0.945*** 0.062
(0.807) (0.232) (0.341) (0.303) (0.421)
Observations 3,348 3,348 3,348 3,348 3,324
R-squared 0.197 0.153 0.052 0.027 0.067


Notes: * refers to significant at 0.10%; ** significant at 0.05%; *** significant at 0.01%. Standard errors for Group means and RE estimates are
errors (with MFI as cluster). Group means refers to a between estimator (results based on Group means); Fama MacBeth refers to results using
Table D.3.2
Regression results for outreach: LNNAB
with IV with IV
ISMFI 0.317** 0.423*** 0.095 2.300*** 3.025***
(0.162) (0.066) (0.165) (0.522) (0.745)
GDPgrowth 0.163*** 0.064*** 0.008** 0.008** 0.067***
(0.031) (0.019) (0.003) (0.003) (0.019)
Fees 0.000 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Mktshare 0.592 0.456*** 0.351*** 0.334*** 0.587
(0.423) (0.123) (0.111) (0.097) (0.443)
Mktsize 0.826*** 0.762*** 0.731*** 0.732*** 0.859***
(0.047) (0.021) (0.028) (0.029) (0.047)
PAR 1.183* 1.249*** 0.026 0.026 0.743*
(0.693) (0.212) (0.149) (0.148) (0.447)
CAR 0.192 0.520*** 0.134 0.136* 0.449*
(0.296) (0.077) (0.088) (0.079) (0.266)
YGLP 0.223 0.848*** 0.298** 0.271** 0.830**
(0.582) (0.177) (0.128) (0.133) (0.365)
GLP/assets 0.469** 0.400** 0.598*** 0.599*** 0.843***
(0.237) (0.139) (0.116) (0.088) (0.135)
DumAge 0.453*** 0.235*** 0.055 0.061 0.319**
(0.141) (0.073) (0.043) (0.042) (0.155)
DumReg 0.270** 0.119*** 0.188* 0.156 0.121
(0.108) (0.026) (0.101) (0.127) (0.154)
Constant 4.762*** 3.341*** 2.793*** 3.085*** 5.925***
(0.946) (0.400) (0.484) (0.483) (0.839)
Observations 3,355 3,355 3,355 3,355 3,331
R-squared 0.668 0.630 0.590 0.754 0.515


Table D.3.3
Regression results for financial performance: ROA
with IV with IV
ISMFI 0.013 0.003 0.000 0.058 0.034
(0.020) (0.008) (0.022) (0.053) (0.034)
GDPgrowth 0.003 0.000 0.001 0.001 0.001
(0.002) (0.001) (0.001) (0.001) (0.001)
Fees 0.000 0.000 0.000 0.000** 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Mktshare 0.009 0.019 0.043 0.043 0.031*
(0.023) (0.011) (0.045) (0.045) (0.017)
Mktsize 0.013*** 0.016*** 0.035*** 0.030*** 0.011***
(0.004) (0.002) (0.006) (0.005) (0.003)
PAR 0.332 0.159*** 0.157*** 0.162*** 0.128***
(0.202) (0.023) (0.032) (0.033) (0.036)
CAR 0.102*** 0.099*** 0.074** 0.077** 0.101***
(0.023) (0.020) (0.036) (0.031) (0.013)
YGLP 0.067 0.032 0.175** 0.144* 0.021
(0.099) (0.034) (0.075) (0.077) (0.042)
GLP/assets 0.067** 0.087*** 0.027 0.026** 0.028***
(0.027) (0.016) (0.017) (0.012) (0.010)
DumAge 0.024 0.019*** 0.000 0.002 0.018*
(0.033) (0.005) (0.006) (0.006) (0.010)
DumReg 0.020 0.004 0.007 0.007 0.009
(0.013) (0.004) (0.017) (0.016) (0.009)
Constant 0.276*** 0.355*** 0.657*** 0.573*** 0.239***
(0.092) (0.055) (0.123) (0.090) (0.039)
Observations 3,228 3,228 3,228 3,228 3,203
R-squared 0.164 0.281 0.074 0.135 0.098


D.4. Results for MFIs offering only Sharia compliant microfinance products
Table D.4.1
with IV with IV
Sharia compliant 0.107 0.107** 0.124 1.697*** 1.363***
(0.237) (0.045) (0.121) (0.637) (0.465)
GDPgrowth 0.184 0.017 0.009** 0.010** 0.006
(0.138) (0.010) (0.004) (0.004) (0.011)
Mktshare 1.721** 0.689*** 0.242** 0.152 0.679***
(0.675) (0.079) (0.120) (0.111) (0.245)
Mktsize 0.058 0.076*** 0.084*** 0.081*** 0.044*
(0.051) (0.011) (0.023) (0.020) (0.023)
PAR 0.021 0.557 0.181 0.175** 0.187**
(0.867) (0.417) (0.115) (0.082) (0.083)
CAR 0.339 0.157*** 0.075 0.059 0.148
(0.232) (0.037) (0.079) (0.074) (0.132)
YGLP 0.013 0.619*** 0.231*** 0.164* 0.544***
(0.607) (0.111) (0.078) (0.098) (0.175)
GLP/assets 0.885* 0.403*** 0.028 0.017 0.194*
(0.466) (0.113) (0.048) (0.036) (0.107)
DumAge 0.099 0.147*** 0.009 0.026 0.198**
(0.175) (0.038) (0.044) (0.048) (0.089)
DumReg 0.271*** 0.301*** 0.401*** 0.364** 0.276***
(0.104) (0.039) (0.145) (0.145) (0.069)
Constant 1.317 0.393* 0.766** 0.633** 0.364
(5.328) (0.212) (0.314) (0.278) (0.383)
Observations 4,680 4,680 4,680 4,680 4,637
R-squared 0.175 0.134 0.036 0.026 0.038


case 2), and the desired maximal bias of the IV estimator relative to OLS (in this case 1% and 5%, respectively). The critical value is obtained
from Skeels and Windmeijer (2018) as Stock-Yogo do not present relative bias tables for 2 instrumental variables (it present bias tables only
for 3 or more instrumental variables).The Stock-Yogo TSLS Size critical values are critical values for the Weak Instrument Test Based on TSLS
Size: Significance level is 5%; The critical value is a function of the number of included endogenous regressors (in our case 1), the number of
instrumental variables (in our case 2), and the desired maximal size (in our case 20%) of a 5% Wald test of b = b0.
Table D.4.2
Regression results for outreach: LNNAB.
with IV with IV
Sharia compliant 0.459** 0.264*** 0.033 3.827*** 4.115***
(0.203) (0.073) (0.175) (0.905) (1.227)
GDPgrowth 0.130*** 0.058*** 0.006* 0.006* 0.060***
(0.041) (0.015) (0.003) (0.003) (0.017)
Mktshare 0.859** 0.354*** 0.392*** 0.372*** 0.297
(0.386) (0.091) (0.099) (0.094) (0.336)
Mktsize 0.749*** 0.702*** 0.747*** 0.755*** 0.839***
(0.045) (0.032) (0.030) (0.027) (0.038)
PAR 1.586** 1.408*** 0.043 0.041 0.679***
(0.712) (0.264) (0.112) (0.095) (0.214)
CAR 0.336 0.373*** 0.105 0.099 0.309
(0.242) (0.061) (0.083) (0.076) (0.236)
YGLP 0.088 0.512** 0.328*** 0.275** 0.541**
(0.490) (0.210) (0.092) (0.112) (0.268)
GLP/assets 0.615*** 0.645*** 0.445*** 0.447*** 0.573***
(0.206) (0.103) (0.167) (0.144) (0.151)
DumAge 0.468*** 0.452*** 0.066 0.072* 0.391***
(0.159) (0.103) (0.041) (0.042) (0.132)
DumReg 0.204* 0.058 0.267** 0.188 0.114
(0.108) (0.060) (0.108) (0.136) (0.147)
Constant 4.497 2.564*** 2.517*** 2.884*** 3.826***
(5.351) (0.485) (0.405) (0.382) (0.572)
Observations 4,693 4,693 4,693 4,693 4,650
R-squared 0.626 0.616 0.587 0.766 0.547


case 2), and the desired maximal bias of the IV estimator relative to OLS (in this case 1% an d5% respectively). The critical value is obtained
for 3 or more instrumental variables).The Stock-Yogo TSLS Size critical values are critical values for the Weak Instrument Test Based on TSLS
Table D.4.3
Regression results for Financial Performance: ROA
with IV with IV
Sharia compliant 0.030 0.021* 0.035 0.095 0.057
(0.046) (0.012) (0.038) (0.077) (0.052)
GDPgrowth 0.002 0.001 0.001 0.001 0.001
(0.002) (0.001) (0.001) (0.001) (0.001)
Mktshare 0.003 0.001 0.001 0.003 0.019
(0.024) (0.012) (0.036) (0.034) (0.018)
Mktsize 0.015*** 0.010*** 0.027*** 0.023*** 0.010***
(0.005) (0.003) (0.005) (0.004) (0.002)
PAR 0.218 0.176*** 0.064 0.067* 0.007
(0.190) (0.044) (0.046) (0.040) (0.024)
CAR 0.097*** 0.056*** 0.065*** 0.066*** 0.086***
(0.030) (0.017) (0.024) (0.023) (0.012)
YGLP 0.059 0.098 0.124*** 0.106** 0.022
(0.088) (0.059) (0.042) (0.044) (0.031)
GLP/assets 0.068*** 0.092*** 0.041*** 0.040*** 0.047***
(0.025) (0.018) (0.011) (0.011) (0.008)
DumAge 0.000 0.035*** 0.005 0.006 0.015*
(0.028) (0.009) (0.007) (0.007) (0.008)
DumReg 0.018 0.009 0.011 0.011 0.012
(0.014) (0.012) (0.015) (0.015) (0.008)
Constant 0.189 0.263*** 0.424*** 0.374*** 0.178***
(0.274) (0.061) (0.085) (0.060) (0.059)
Observations 4,379 4,379 4,379 4,379 4,338
R-squared 0.143 0.214 0.066 0.094 0.082


case 2), and the desired maximal bias of the IV estimator relative to OLS (in this case 1% and 10% respectively). The critical value is obtained
for 3 or more instrumental variables). The Stock-Yogo TSLS Size critical values are critical values for the Weak Instrument Test Based on TSLS
References Fama, E. F., & MacBeth, J. D. (1973). Risk, return and equilibrium: Empirical tests.
Journal of Political Economy, 81(3), 607–636.
Fan, Y., John, K., Liu, F. H., & Tamanni, L. (2019). Security design, incentives, and
Abdelkader, I. B., & Salem, A. B. (2013). Islamic vs conventional microfinance
Islamic microfinance: Cross country evidence. Journal of International Financial
institutions: Performance analysis in MENA countries. International Journal of
Markets, Institutions and Money, 62(C), 264–280.
Business and Social Research, 3(5), 219–233.
Field, E., Pande, R., Papp, J., & Rigol, N. (2013). Does the classic microfinance model
Abdul Rahman, A. (2007). Islamic microfinance: A missing component in Islamic
discourage. American Economic Review, 2196–2226.
banking. Kyoto Bulletin of Islamic Area Studies, 1(2), 38–53.
Garcia, A., & Lensink, R. (2019). Microfinance-Plus: A Review and Avenues for
Abedifar, P., Molyneux, P., & Tarazi, A. (2013). Risk in Islamic banking. Review of
Research. In M. Hudon, M. Labie, & A. Szafarz (Eds.), A Research Agenda for
Finance, 17(6), 2035–2096.
Financial Inclusion and Microfinance (pp. 111–125). Cheltenham, UK: Edward
Adhikary, S., & Papachristou, G. (2014). Is there a trade-off between financial
Elgar Publishing Limited.
performance and outreach in south Asian Microfinance institutions?. Journal of
Garcia, A., Lensink, R., & Voors, M. (2020). Does microcredit increase aspirational
Developing Areas, 48(4), 381–402.
hope? Evidence from a group lending scheme in Sierra Leone. World
Ahlin, C., Lin, J., & Maio, M. (2011). Where does microfinance flourish? Microfinance
Development, 128.
institution performance in macroeconomic context. Journal of Development
Hansen, N., Huis, M., & Lensink, R. (2020). Microfinance Services and Women’s
Economics, 95(2), 105–120.
Empowerment. In L. San-Jose, J. Retolaza, & L. van Liedekerke (Eds.), Handbook
Ahmed, H. (2002). Financing microenterprises: An analytical study of Islamic
on Ethics in Finance. International Handbooks in Business Ethics. Springer,
microfinance institutions. Islamic Economic Studies, 9(2), 27–64.
Cham.
Armendáriz, B., & Labie, M. (2011). The handbook of microfinance. World Scientific
Hartarska, V. (2005). Governance and performance of microfinance institutions in
Publishing.
Central and Eastern Europe and the Newly Independent States. World
Armendáriz, B., & Morduch, J. (2010). The economics of microfinance. London: The
Development, 33(10), 1627–1643.
MIT Press.
Hartarska, V., Nadolnyak, D., & Mersland, R. (2014). Are women better bankers to
Armendáriz, B., & Szafarz, A. (2011). On mission drift in microfinance institutions. In
the poor? Evidence from rural microfinance institutions. American Journal of
B. Armendáriz, & M. Labie (Eds.), The handbook of microfinance (pp. 341-66).
Agricultural Economics, 96(5), 1291–1306.
London and Singapore: World Scientific Publishing.
Hartarska, V., Shen, X., & Mersland, R. (2013). Scale economies and input price
Balkenhol, B. (2007). Microfinance and public policy: Outreach, performance and
elasticities in microfinance institutions. Journal of Banking and Finance, 37(1),
efficiency. UK: Palgrave Macmillan.
118–131.
Banerjee, A., Karlan, D., & Zinman, J. (2015). Six randomized evaluations of
Hermes, N., & Hudon, M. (2018). Determinants of the performance of microfinance
microcredit: Introduction and further steps. American Economic Journal:
institutions: A systematic review. Journal of Economic Surveys, 32(5),
Applied Economics, 7(1), 1–21.
1483–1513.
Baquero, G., Hamadi, M., & Heinen, A. (2018). Competition, loan rates, and
Hermes, N., & Lensink, R. (n.d.). Microfinance and Development. In J. Y. Abor, C.
information dispersion in nonprofit and for-profit microcredit markets.
Adjasi, & R. Lensink (Eds.), Contemporary issues in development finance (p. in
Journal of Money, Credit and Banking, 50(5), 893–937.
press). Routledge.
Barro, R. J., & McCleary, R. M. (2005). Which countries have state religions? The
Hermes, N., Lensink, R., & Meesters, A. (2011). Outreach and efficiency of
Quarterly Journal of Economics, 120(4), 1331–1370.
microfinance institutions. World Development, 39(6), 938–948.
Beck, T., De Jonghe, O., & Schepens, G. (2013). Bank competition and stability: Cross-
Hudon, M., & Sandberg, J. (2013). The ethical crisis in microfinance. Business Ethics
country heterogeneity. Journal of Financial Intermediation, 22(2), 218–244.
Quarterly, 23(4), 561–589.
Boatright, J. R. (2014). Ethics in finance. Sussex, UK: Wiley-Blackwell.
Islamic Banking Database. (2014). Global report on Islamic finance. World Bank.
Brau, J. C., & Woller, G. M. (2004). Microfinance: A comprehensive review of the
JP Morgan. (2017). JP Morgan Chase & Co. Annual Report.
existing literature. Journal of Entrepreneurial Finance, 9(1), 1–28.
Karim, N., Tarazi, M., & Reille, X. (2008). Islamic microfinance: An emerging market
Bulte, E., Lensink, R., & Vu, N. (2017). Do gender and business trainings affect
niche. Focus Note No. 49. CGAP.
business outcomes? Experimental Evidence from Vietnam. Management Science,
Kettani, H. (2010). World Muslim population: 1950–2020. International Journal of
63(9), 2885–2902.
Environmental Science and Development, 1(2).
Caudill, S. B., Gropper, D. M., & Hartarska, V. (2009). Which microfinance
Khan, M. S., & Mirakhor, A. (1992). Islam and the economic system. Review of Islamic
institutions are becoming more cost effective with time? Evidence from a
Economics, 2(1), 1–29.
mixture model. Journal of Money, Credit and Banking, 41(4), 651–672.
Kleynjans, L., & Hudon, M. (2016). A study of codes of ethics for Mexican
Chong, B. S., & Liu, M.-H. (2009). Islamic banking: Interest-free or interest-based?.
microfinance institutions. Journal of Business Ethics, 134(3), 397–412.
Pacific-Basin Finance Journal, 17(1), 125–144.
Lensink, R., & Bulte, E. (2019). Can we improve the impact of microfinance? A survey
Churchill, S. A. (2019). Microfinance financial sustainability and outreach: Is there a
of the recent literature and potential avenues for success. In A. N. Berger, P.
trade-off?. Empirical Economics, 1–22.
Molyneux, & J. S. Wilson (Eds.), Oxford handbook of banking (pp. 404-431).
Cull, R., Demirgüç-kunt, A., & Morduch, J. (2007). Financial performance and
Oxford University Press.
outreach: A global analysis of leading microbanks. The Economic Journal, 117
Louis, P., & Baesens, B. (2013). Do for-profit microfinance institutions achieve better
(517), F107–F133.
financial efficiency and social impact? A generalised estimating equations panel
Cull, R., Demirgüç-Kunt, A., & Morduch, J. (2009). Microfinance meets the market.
data approach. Journal of Development Effectiveness, 359–380.
Journal of Economic Perspectives, 23(1), 167–192.
Louis, P., Seret, A., & Baesens, B. (2013). Financial efficiency and social impact of
Dahal, M., & Fiala, N. (2020). What do we know about the impact of microfinance?
microfinance institutions using self-organizing maps. World Development, 46,
The problems of power and precision. World Development, 128 104773.
197–210.
Dar, H. A. (2004). Demand for Islamic Financial Services in the UK: Chasing a
Maazullah, & Bedi, A. S. (2017). Returns to Islamic microfinance: Evidence from a
Mirage?. Economics Research Paper, no. 04-11. Loughborough University
randomized experiment in Pakistan. IZA Institute of Labor Economic, Discussion
Institutional Repository.
Paper Series, IZA DP No. 10965.
de Jong, E. (2016). A simple procedure for estimating the effect size in IV
Mersland, R., & Strøm, R. Ø. (2009). Performance and governance in microfinance
regressions.
institutions. Journal of Banking and Finance, 33(4), 662–669.
de Quidt, J., Fetzer, T., & Ghatak, M. (2018a). Commercialization and the decline of
Mersland, R., & Strøm, R. Ø. (2010). Microfinance mission drift?. World Development,
joint liability microcredit. Journal of Development Economics, 134, 209–225.
38(1), 28–36.
de Quidt, J., Fetzer, T., & Ghatak, M. (2018b). Market structure and borrower welfare
Mersland, R., DÉspallier, B., & Supphellen, B. (2013). The effects of religion on
in microfinance. Economic Journal, 128(610), 1019–1047.
development efforts: Evidence from the microfinance industry and a research
Dehejia, R., Montgomery, H., & Morduch, J. (2012). Do interest rates matter? Credit
agenda. World Development, 41, 145–156.
demand in the Dhaka slums. Journal of Development Economics, 97(2), 437–449.
Mohamed, S., Goni, A., & Hasan, S. (2018). Islamic finance development report.
DÉspallier, B., & Goedecke, J. (2020). Social performance measurement in
Thomson Reuters.
microfinance. In H. Marek, M. Labie, & A. Szafarz (Eds.), A Research Agenda
Morduch, J. (1999). The microfinance promise. Journal of Economic Literature, 37(4),
for Financial Inclusion and Microfinance (pp. 62–75). Cheltenham, UK: Edward
1569–1614.
Elgar Publishing Limited.
Morduch, J. (2000). The microfinance Schism. World Development, 28(4), 617–629.
Dhumale, R., & Sapcanin, A. (1999). An application of Islamic banking principles to
Morduch, J. (2016). How statistics shaped microfinance. New York University and
microfinance. Technical Note. Washington: Rural Finance Learning Center,
Institute for Advanced Study.
World Bank Group.
Navajas, S., Schreiner, M., Meyer, R. L., Gonzalez-Vega, C., & Rodriguez-Meza, J.
Dorfleitner, G., Leidl, M., Priberny, C., & von Mosch, J. (2013). What determines
(2000). Microcredit and the poorest of the poor: Theory and evidence from
microcredit interest rates?. Applied Financial Economics, 23(20), 1579–1597.
Bolivia. World Development, 28(2), 333–346.
El-Gamal, M., El-Komi, M., Karlan, D., & Osman, A. (2014). Bank-insured RoSCA for
Obaidullah, M. (2008). Introduction to Islamic microfinance. New Delhi, India: IBF Net
microfinance: Experimental evidence in poor Egyptian villages. Journal of
(P) Limited.
Economic Behavior and Organization, 103, S56–S73.
Quayes, S. (2012). Depth of outreach and financial sustainability of microfinance
El-Zoghbi, M., & Tarazi, M. (2013). Trends in Sharia-Compliant Financial Inclusion.
institutions. Applied Economics, 44(26), 3421–3433.
Focus Note No. 84. Washington, DC: CGAP.
Reichert, P. (2018). A meta-analysis examining the nature of trade-offs in
Fall, F., Akim, A.-M., & Wassongma, H. (2018). DEA and SFA research on the
microfinance. Oxford Development Studies, 46(3), 430–452.
efficiency of microfinance institutions: A meta-analysis. World Development,
Robinson, M. S. (2001). The microfinance revolution: Sustainable finance for the
107(C), 176–188.
poor. The Wold Bank.
Sanabel. (2012). Islamic Micro and Small Medium Enterprise (MSME) finance Tchakoute-Tchuigoua, H. (2012). Active risk management and loan contract terms:
survey. New Cairo: Sanabel. Retrieved from https://www.sanabelnetwork.org/ Evidence from rated microfinance institutions. The Quarterly Review of
home/. Economics and Finance, 52(4), 427–437.
Schreiner, M. (2002). Aspects of outreach: A framework for discussion of the social Tulchin, D. (2003). Microfinance’s double bottom line. Boston: MicroCapital Institute.
benefits of microfinance. Journal of International Development, 14(5), 591–603. Ugi, S. (2018). Riba and interest in Islamic finance: Semantic and terminological
Seaman, S. R., & White, I. R. (2013). Review of inverse probability weighting for ıssue. International Journal of Islamic and Middle Eastern Finance and
dealing with missing data. Statistical Methods in medical Research, 22(3), Management, 11(1), 131–138.
278–295. Valette, C., & Fassin, B. (2018). Microfinance and profitabilities. Paris: Convergences.
Servin, R., Lensink, R., & van den Berg, M. (2012). Ownership and technical efficiency Visser, H. (2013). Islamic finance: Principles and practice. Cheltenham, UK: Edward
of microfinance institutions: Empirical evidence from Latin America. Journal of Elgar.
Banking and Finance, 36, 2136–2144. Weill, L. (n.d.). Islamic microfinance. In M. Hudon, M. Labie, & A. Szafarz (Eds.), A
Shahinpoor, N. (2009). The link between Islamic banking and microfinancing. research agenda for financial inclusion and microfinance (pp. 99-110).
International journal of social economics, 36(10), 996–1007. Cheltenham: Edward Elgar Publishing Limited.
Skeels, C. L., & Windmeijer, F. (2018). On the Stock-Yogo tables. Econometrics, 6(4), Widiarto, I., & Emrouznejad, A. (2015). Social and financial efficiency of Islamic
44. microfinance institutions: A Data Envelopment Analysis application. Socio-
Stock, J. H., & Yogo, M. (2005). Testing for weak instruments in linear IV regression. Economic Planning Sciences, 50, 1–17.
In D. K. Andrews, & J. H. Stock (Eds.), Identification and inference for Wilson, R. (2007, December). Islamic finance in Europe. RSCAS Policy Papers No.
econometric models: Essays in Honor of Thomas Rothenberg (pp. 80-108). 2007/02, pp. 1-29. Retrieved from http://cadmus.eui.eu/bitstream/handle/
New York: Cambridge University Press. 1814/7739/RSCAS_PP_2007_02.pdf?sequence=1&isAllowed=y.
Strøm, R. Ø., DÉspallier, B., & Mersland, R. (2014). Female leadership, performance, World Bank. (2013). Global financial development report 2014: Financial inclusion.
and governance in microfinance institutions. Journal of Banking and Finance, 42, Washington, DC: World Bank Group. doi:10.1596/978-0-8213-9985-9.
60–75.

World Development: Syedah Ahmad, Robert Lensink, Annika Mueller

Uploaded by

Copyright:

Available Formats

World Development: Syedah Ahmad, Robert Lensink, Annika Mueller

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

World Development: Syedah Ahmad, Robert Lensink, Annika Mueller

Uploaded by

Copyright:

Available Formats

World Development 136 (2020) 105130

Contents lists available at ScienceDirect

The double bottom line of microfinance: A global comparison between

1. Introduction tices adopted by Islamic MFIs, such as interest-free forms of finan-

S. Ahmad et al. / World Development 136 (2020) 105130

Islamic Financial Percentage Percentage of Clients Using Islamic Conventional p-Value

Variable Abbreviation Definition

Variable Full sample (response = 1) Islamic MFIs Conventional MFIs

This study has pursued two main aims. First, we sought a

Appendix A. Data construction Table A.1.1

Appendix C. Supplementary tables for first stage regressions. C.2.1 (continued)

Response Coeff. Std. z P>z [95% Interval]

Appendix D. Robustness checks.

Variable name Definition N Mean SD

D.2. Results with smaller set of controls

Table D.2.1 (continued)

Weak identification test

Overidentification test of all instruments

Weak identification test

Overidentification test of all instruments

Weak identification test

Overidentification test of all instruments

D.3. Results after adding fees and commission on loans

Weak identification test

Overidentification test of all instruments

Weak identification test

Overidentification test of all instruments

Weak identification test

Overidentification test of all instruments

Weak identification test

Overidentification test of all instruments

Weak identification test

Overidentification test of all instruments

Weak identification test

Overidentification test of all instruments

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.