Drivers of MFI success

Drivers of MFI success

Addressing three pillars of determinants in context of the microfinance schism

Jasper N. Quak^‡

Supervisor: dr. M. (Michalis) Zaouras^§ Thesis: MSc. Economics and MSc. Finance

June, 2016

A B S T R A C T

This paper studies the effects of the macroeconomic context, the macro-institutional environment and MFI-specific characteristics on the performance of MFIs. These effects are analysed in the context of the ongoing debate among microfinance advocates, characterised as the microfinance schism. A thorough understanding of the effects of the three sets of variables on the performance of MFIs can be essential in the formation of policy on the development of the microfinance sector. Providing the results both from a welfarist and an institutionalist point of view can even further enhance understanding on maximising MFI performance. The data used consist of 666 MFIs in 60 countries over a period of 11 years. It is found that economic determinants primarily affect the welfarist measure of success of MFIs, while the institutional context mainly affects the institutionalist measure. MFI characteristics are found to especially have a significant effect on the welfarist measure of success, implying that effective regulation of microfinance the sector might improve performance.

Keywords: microfinance, outreach, financial sustainability JEL classification: G21, G28

‡ The author can be contacted through jnquak@gmail.com. Student number: s1891243

§ The author would like to thank dr. Zaouras, without whose excellent help and insightful comments this thesis would not have been as it is.

1. Introduction

“Provide a man not with a fish, yet with a fishing rod”

This well-known and extensively used metaphor has provided many with an argument against the traditional shape of developmental aid. Developmental aid has been subject to critique over the past decades, with many doubting its effectiveness in the long run. The notion of aid dependency is closely tied to the metaphor above, as one might argue that merely receiving a final good prevents the recipient from actually gaining the skills and knowledge to fend for him- or herself. Thus, providing one with the metaphorically fishing rod, might induce aid’s recipients to actually be able to live without developmental aid in the future. Emergency aid set aside, the notion intuitively seems to hold. Taking this line of reasoning one step further – or rather back – leads to the following extended metaphor as put by Rajesh Mirchandani^**:

“Forget the Fish. Forget the Fishing Rod. Give a Man Some Capital.”

With this metaphor Mirchandani took the debate one step further. The solution for poverty around the world would not lie in providing people with in-kind benefits, yet with allowing them to create their own business, thus providing entire communities or even countries with a chance on economic development. The problem with providing credit to the poor however, lies in their poverty itself. Any traditional financial institution would not willingly provide the poorest of poor with a loan, as there simply was no collateral to back the loan.

A solution that does provide credit to the poor is found in microfinance, that allows the poor to borrow and deposit funds in small amounts without having large – or any – collateral.

Microfinance has gained much ground in the years since its foundation in the 1980s, developing steadily over the first decades of its existence and expanding rapidly during the 2000s. The momentum gained by microfinance becomes quite clear when looking at figure 1, where the number of served clients and the total amount of outstanding loans are depicted over time. Figure 1 is constructed using data on all MFIs that report to MIX, an organisation that collects global microfinance data and represents over 70 percent of all MFIs according to Honohan (2004). From figure 1 it becomes apparent that not only did microfinance reach an increasingly large number of borrowers, yet also constitutes a large sector in monetary terms.

** http://www.cgdev.org/blog/forget-fish-forget-fishing-rod-give-man-some-capital-podcast-chris- blattman

Figure 1. Developments in total gross loan portfolio and outreach over time. Data extracted from MIX Database

Apart from experiencing rapid growth, microfinance nowadays has evolved a lot;

whereas the initial concept employed by the Grameen Bank – the bank founded by Yunus and the first official microfinance institution – essentially revolved around microcredit (the provision of small loans), modern microfinance also includes concepts such as microsavings and microinsurance. Both microsavings and microinsurance pursue the same goal as microcredit; financial inclusion of the poor, yet obviously the means are different.

With the Grameen bank, built on Yunus’ foundations, being regarded as generally successful, its progress over the years has been used as an example for MFIs. What however, generally is not taken into account, is the context of that success and, perhaps, more importantly, the measurement of that success. In this paper, the context of MFIs is divided into three pillars consisting of macroeconomic indicators, institutional determinants and MFI characteristics. Success of MFIs is also differentiated and assessed twofold; through assessing the financial performance and through assessing the number of served borrowers.

0 20 40 60 80 100 120

2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Outstanding Loans (Bln. US $) Outreach (Mln. Borrowers)

2. Literature review

The field of microfinance is currently characterised by the so-called microfinance schism (Morduch, 2000). This schism comprises the ongoing discussion between two opposing camps: welfarists and institutionalist. Welfarists propagate the importance of microfinance as a poverty reducing tool and believe dependency of MFIs on external funding is not an issue.

Institutionalists on the other hand, believe that MFIs should aim at financial self-sufficiency, not denying the importance of poverty reduction. Primarily, this paper attempts to provide insight in the effects of not only the institutional environment, yet also the economic environment of countries and characteristics of MFIs on the success of microfinance.

Secondarily, this research attempts to unite both approaches (i.e.: welfarist and institutionalist) using the same data, thus providing the discussion with a way of meaningfully comparing the effects of several sets of variables on both the outreach – dubbed most important by the welfarists – of microfinance and the operational self-sufficiency – considered the appropriate measure of success by institutionalists. Outreach is measured as the number of active borrowers relative to the population. Operational self-sufficiency (OSS) is depicted by equation 1 and presents a meaningful insight in the financial performance of an MFI. The information provided by the OSS ratio can be interpreted quite straightforward; a ratio below 1 indicates the concerned institute is not able to continue without external funding (depending on its reserves) for a long time, whereas ratios over 1 indicate financial sustainability.

𝑂𝑆𝑆 = 𝑂𝑝𝑒𝑟𝑎𝑡𝑖𝑛𝑔 𝑟𝑒𝑣𝑒𝑛𝑢𝑒

𝐹𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙 𝑒𝑥𝑝𝑒𝑛𝑠𝑒+𝑙𝑜𝑎𝑛 𝑙𝑜𝑠𝑠 𝑝𝑟𝑜𝑣𝑖𝑠𝑖𝑜𝑛 𝑒𝑥𝑝𝑒𝑛𝑠𝑒+𝑜𝑝𝑒𝑟𝑎𝑡𝑖𝑛𝑔 𝑒𝑥𝑝𝑒𝑛𝑠𝑒 (1)

2.1. Previous scope of research on microfinance

The focus of previously written papers – especially early work before the 2000s – on microfinance lies mainly on the impact of different ranges of products (Brau and Woller, 2004). Questions on how the introduction of microfinance in a community influences its economic agents can be found answered in many papers, ranging from focus on type of entrepreneurial activity (Shaw, 2004) to consumption and income smoothing (Zeller, 2001).

One criticism on the existing literature was expressed by Banerjee et al. (2013), which entails the focus on relatively specific groups (i.e. one community or institution). Their criticism mainly entails the fact that many research on microfinance has been focused on case studies, while claiming external validity. The literature continues to assess the impact of microfinance based on borrowers’ characteristics on which Gomez and Santor (2001) have contributed significantly by combining microfinance outcomes (in terms of occupation and consumption) with characteristics of the borrowers (in terms of neighbourhood characteristics). A more recent scope of research in the field of microfinance can be found to lie on the sustainability of the concept, mainly explained by the self-sufficiency of MFIs (i.e.: the independence of donors) (Brau and Woller, 2004). The idea of sustainable microfinance is in line with the original idea of microfinance filling the void left by formal credit institutions for the poor, as financially sustainable MFIs should be able to start the movement towards not only sustainable financial inclusion, yet also ignite the formation of a larger, more accessible formal financial system.

In the literature on microfinance is, however, a void: according to Hardy et al. (2002) and Hermes and Meesters (2011), the large differences in performance of MFIs across the world have not been sufficiently explained. By assessing the combined effects of the macroeconomic environment, the institutional environment and MFI-specific characteristics on the performance of MFIs, this paper attempts to fill the above-mentioned void where possible, while at the same time contributing to the ongoing debate in the literature.

2.2. Macroeconomic environment

In recent years the belief that macroeconomic conditions could very well have a significant effect on MFI performance has gained ground. The literature however, has been lacking behind even though some authors were committed to researching this relationship. Over the course of the past decades few researchers were committed to assess the effects of the macro- environment on MFI success (Hermes and Meesters, 2011), however the literature started developing after a paper by Christen et al. (1995) as part of a USAID report on the current state of microfinance. They were basically the first to assess a relationship between performance, which they proxied by outreach, of microfinance in a country and that country’s economy. In their research Christen et al. include two main economic indicators to assess the relationship:

interest rates and GDP growth, of which they conclude the former to have a rather large

impact on MFI outreach and the latter none whatsoever. These results however, were criticised on grounds of the small sample, only 11 MFIs, that was used. The view of Christen et al., entailing that macroeconomic variables can influence microfinance outcomes is also in line with conclusions drawn by Honohan (2004), who concludes financial sector development in general is highly dependent on the macroeconomic environment. Christen et al.’s paper also led to the renowned case study by Patten and Johnston (2001) who study the effects of the Asian monetary crisis on several MFIs and small-scale entrepreneurs. In their paper Woller and Woodworth (2001) assert the importance of macroeconomic stability for small-scale entrepreneurs, which is directly linked to MFI performance, as MFIs lend to exactly those small-scale entrepreneurs. The mechanism through which the macroeconomic environment influences the success of MFIs is well explained by Hermes and Meesters (2011) in their paper contributing to The Handbook of Microfinance. They state the effects of the macroeconomic environment can be considered rather ambiguous: a stable growing economy will lead to higher demand for investments on all levels in the economy, thus raising demand for microfinance products. At the same time however, a growing economy can also cause a decline in the demand for microcredit, due to the fact that in such an economy agents are more likely to have access to formal sources of credits, such as regular banks, effectively rendering microfinance obsolete to some extent. In case of a slowing economy the same line of reasoning as above can be followed, which results, again, in ambiguous expectations of the reaction of MFIs to the macroeconomic environment. As a third potential outcome to consider, Hermes and Meesters state that it might also very well be the case the MFI performance does not depend on the macroeconomic environment as its borrowers were located in the informal part of the economy to begin with. It thus becomes apparent that the effects of the macroeconomic environment are hard to predict ex ante. Nevertheless, Ahlin et al (2008) and Vanroose (2008) find significant results, especially for the GDP per capita.

2.3. Institutional environment

Apart from the macroeconomic environment, it is believed that the institutional environment also has an impact on the success of the financial sector, including the microfinance sector, in a country. Fisman and Svensson (2007), for instance, conclude that corruption (taking many forms) greatly affects development among Ugandan business. Apart from corruption, several institutional factors have been attributed to financial development.

Even though these factors were not directly linked to the performance of the microfinance sector in a country, they can be assumed to be very well applicable to that sector as well, as Morduch (1999) stated.

Huang (2005) proposes differentiation of these institutional factors into three groups:

geographic, policy and the legal/regulatory framework. Even though Huang (2005) and Fisman and Svensson (2007) focus on financial performance of entrepreneurial activity in general, their conclusions can be applied to the field of microfinance as well (Morduch, 1999).

These subcategories are also partly used by Vanroose (2008) in her paper on performance differences of MFIs. Ahlin et al. (2011) on the other hand, choose to apply the framework proposed by Kaufman et al. (2009). Kaufman et al. compiled the Worldwide Governance Indicators and employ a set of six subcategories to assess the institutional environment, termed governance, in a country. The subcategories, proposed by Kaufman et al., utilised by both Ahlin et al. and Hermes and Meesters (2011), include voice and accountability, political stability and absence of violence/terrorism, government effectiveness, regulatory quality, rule of law, and control of corruption. According to Hermes and Meesters the above subcategories can be described as formal institutions that in turn, according to Hubka and Zaidi (2005), facilitate the dire need for small-scale enterprises in the informal sector to be formalised.

Hubka and Zaidi at the same time also note the importance of the general business environment, which is acknowledged by Ahlin et al. who include several indicators from the Doing Business project from the World Bank. Another insightful paper on the effects of the institutional environment on the success of MFIs by Fogel et al. (2011), in which they, similar to Hermes and Meesters (2011), differentiate between formal and informal institutions with the addition of using the Hofstede five cultural dimensions (masculinity, uncertainty avoidance, collectivism, power distance and long-term orientation) as informal institutions.

2.4. MFI characteristics

Including MFI characteristics in research to the success of an MFI is described to be the best practice literature (Brau and Woller, 2004). The best practice for MFIs can be described to be rather heterogeneous in nature, as different offerings and structures of MFIs match with different target communities (Dunford, 2000). Bhatt and Tang (2001) conclude that even though the best practice for MFIs has to design thus, that it is able to be adapted to serve every specific need, yet they also assert the dominant factors among an MFI’s characteristics

that influence its impact and success. Among these identified factors are whether to lend to groups or individuals, loan sizes and commercialisation of MFIs. The commercialisation factor proposed by Bhatt and Tang is further elaborated on by Woller (2002), who finds significant differences between commercial and not-for-profit MFIs in terms of financial sustainability.

Among microfinance best practice research, several papers can be found to investigate the importance and effects of the target group (Conning, 1999; Hollis and Sweetman, 1998; Perry, 2002); all of which conclude that indeed the target group can influence sustainability and outreach to a high extent.

In the papers by Ahlin et al. (2011), Hermes and Meesters (2011) and Vanroose (2008), all mentioned in the previous two sections, either few MFI characteristics have been included in their models as control variables or few macroeconomic variables have been included.

Generally, the literature on the effects of institutions and/or the macroeconomic environment– however scarce – choose to either use the outreach of microfinance or the self- sufficiency of an MFI as dependent variable. This paper aims at using both in order to provide a meaningful contribution to the debate between the welfarists, researchers who believe in the poverty lending approach, and institutionalists, who support the financial systems approach (Brau and Woller, 2004).

3. Data

In this paper, success of MFIs is explained by different sets of variables. These different sets were compiled using various sources of data. Data on MFI characteristics were extracted from MIX, a web based platform for microfinance data. MIX is widely used and especially useful, since it provides information on the trustworthiness of data, by assigning ‘diamonds’ to the observations on microfinance institutions. In the analysis provided in this paper, only MFIs with 4 and 5 diamonds are included, as financial data on these observations have been audited by third-party accounting firms. One potential downside of the MIX dataset that is acknowledged in this paper, is the fact that MFIs provide their records on a voluntary basis to MIX. At the same time however, according to Honohan (2005), the MIX database comprises the largest amount of reliable data on the world’s microfinance institutions. One can then reasonably expect that even more of the 4 and 5 diamonds institutions are represented as these institutions already have their records readily available. The sample used for the empirical analysis is limited to data from the years 2003-2013. This limitation is imposed due

to availability of data on microfinance institutions; for the years 2003-2013 more observations in the MIX dataset are complete than for previous years. The observations for MFIs are not consistent in that sense that for many MFIs there might be only one observation, while others are included in the dataset for each year. This drawback is dealt with by including only MFIs in the sample that have at least three observations available over time. Note, that this does not mean the panel is now balanced. Furthermore, Zimbabwe is dropped from the sample, due to extreme outliers found for that country. The presence of hyperinflation in Zimbabwe during the sample period, caused data on inflation and the lending interest rate to be extremely divergent. The complete sample, after imposing the limitations on the time period and exclusion due to unpaired observations, includes observations on 666 MFIs, in 60 countries over 11 years.

3.1. Dependent variables

This paper attempts to show the dependence of MFI success on several factors, which will be addressed below. The success of microfinance is assessed through two different approaches; the institutionalist approach and the welfarist approach. Both approaches are generally not used in research at the same time, yet it is the believe of the author that including both not only provides better insight in the overall success of MFIs, yet also contributes to the ongoing debate in the field of microfinance between the institutionalists and welfarists. By making use of the same data, while employing the analysis twice (i.e.: once for the welfarist approach and once for the institutionalist approach), a fair comparison can be made which potentially provides new insights for the ongoing discussion.

3.1.1. Institutionalist approach

As described in the previous section on the relevant literature, the institutionalist approach focuses on the financial sustainability of an MFI, i.e.: the degree to which an institution is financially viable regardless of external funding, such as donations. Following the most widely accepted measure of financial sustainability of MFIs, this papers employs the so- called operational self-sustainability (OSS), the calculation of which is depicted in equation 1.

3.1.2. Welfarist approach

The welfarist approach, contrary to the institutionalist approach, does not focus on the financial performance of an MFI. Instead, the welfarists believe the sole goal of microfinance should be to reach as many of the poor as possible, thus alleviating larger groups of people from poverty. In qualitative approaches the welfarist point of view is often described as the need to aim for both wide (i.e.: as many different target groups as possible) and deep (i.e.:

penetrate the market to the poorest of the poor) outreach of microfinance (Brau and Woller, 2004). In papers of a more quantitative nature, outreach in general is used. This paper will partly follow the generally accepted way, as stated by Vanroose (2008) and use the number of active clients as a first proxy for outreach. This measure is then extended by using the relative values of outreach by dividing the number of active clients by the population. This relative measure is also proposed by Vanroose (2008). One note of caution should be added to the use of this measure, as clients who have taken on loans with multiple MFIs will be attributing to the outreach variable multiple times.

3.2. Explanatory variables

As explained in the literature section of this paper, the explanatory variables in this research are divided into three pillars: economic indicators, institutional variables and MFI characteristics. In this sub-section the variables of all pillars will be discussed and accompanied by their respective descriptive statistics.

3.2.1. MFI characteristics

MFIs face risks and challenges similar to ‘formal’ financial institutions, yet the extent and depth of those risks and challenges are relatively more severe. To overcome these challenges and deal with the large risks that lending to the poor inevitably embodies, MFIs have had to innovate their lending products. These innovations take many forms and can to some extent be identified through quantitative data on their respective characteristics.

Data on the number of borrowers per loan officer are collected from MIX. MFIs do not serve as traditional financial institutions in a sense, that generally their aim to a more or

lesser extent is to alleviate people from poverty. In order to achieve their goal merely offering credit to people excluded from the formal financial market is not all that is required. Many MFIs have been trying to develop methods to increase repayments and success of microenterprises of their borrowers. Probably the most obvious of these methods, is guiding borrowers more intensively than a more traditional financial institution would have.

According to Karlan and Valdivia (2011), increasing guidance and increasing training, will increase the rate of success of microenterprises, thus raising repayment rates. At the same time however, MFIs are believed to attain economies of scale when the number of borrowers per loan officer is higher, effectively lowering wage costs for the MFI. The effects of the number of borrowers per loan officer on the outreach of an MFI are less straightforward. In that relation higher success of microenterprises does not necessarily mean more borrowers. In fact, more stringent guidance might discourage people in which case a positive relationship between outreach and number of borrowers per loan officer is expected to be observed. A word of caution regarding the use of the number of borrowers per loan officer is required.

Since the number of loan officers per borrower also depends on the number of borrowers, a potential endogeneity problem arises. It is however expected, that the number of loan officers is chosen carefully and reflects the effort of guidance rather than the total number of borrowers.

Profit status is a dummy variable that takes on a value of 1 when the concerned MFI is a for-profit institution and 0 when it is non-profit. According to Jansson et al. (2004), non- profit MFIs can be expected to underperform in terms of efficiency compared to for-profit MFIs, which potentially results in lower OSS ratios. The effects of an MFI’s profit status on outreach is expected to exhibit a negative sign, as non-profit MFIs are believed to operate from a more developmental point of view.

In recent decades the microfinance sector is becoming increasingly regulated. These regulations apply to many MFIs, yet not all, and are mainly installed in order to ensure high outreach of microfinance (Hishisgsuren, 2006). On the other hand, Hartarska and Nadolnyak (2007) found no evidence that operational self-sufficiency is being affected by being regulated or not. The dummy regulated takes on a value of 1 if the MFI in question is regulated by either a formal banking regulator or some other financial services regulator.

Many MFIs have chosen to offer microsavings next to their microcredit products. These microsavings products are either offered on a voluntary basis, or are mandatory for their borrowers. In the latter case, a borrower is obliged to deposit some funds with the MFI in order to increase the probability of repayment of the loan, for example. The data do not allow

for differentiation between mandatory and voluntary savings, however including a dummy for microsavings might provide insight in the potential success of offering microsavings products.

This might especially be the case since the effects of offering microsavings have not been assessed in similar research. The dummy microsavings is included in the analysis to represent the offering of any microsavings products and is constructed by using data on the total deposits of clients with MFIs; for MFIs that report positive levels of savings, and thus assumed to offer microsavings, the dummy variable takes on a value of 1. The effects of microsavings can, ex ante, be seen as rather ambiguous; especially with respect to outreach, as it is impossible to control for a potentially mandatory nature of microsavings. In case of mandatory microsavings, agents might decide to rather take on a loan with different MFIs, not demanding the same. In any case however, agents might enjoy the possibility of a savings account which might increase the outreach. This leaves the expected sign of the coefficient for the dummy variable rather ambiguous in nature. With respect to the operational self- sufficiency, it is expected to observe a positive effect, since offering microsavings products might induce agents to behave more prudently.

Four dummies (target low-end, target broad, target high-end and target small business) are included to indicate the target market, as reported by MIX, of the concerned MFI. This differentiation is created on a basis of the average loan size relative to the GNI per capita. Low-end targeting entails that the average loan size that is lent by an MFI is no higher than 20 percent of GNI per capita, targeting broad constitutes average loan sizes between 20 and 70 percent and high-end targeting implies average loan sizes of over 150 percent of GNI per capita. The targeting of small businesses is not based on the average loan size, yet on the reported preferences of MFIs. The broad target market is excluded from this analysis due to issues of multicollinearity when including all four dummy variables.

In past research on targeting by MFIs, researchers have limited themselves on the targeting of women. In order to further identify the drivers of MFI success, the target markets of MFIs (as reported in the MIX database) are included. It is expected that targeting low-end markets results in higher outreach of MFIs, as demand for microfinance can reasonably be expected to be the largest in that particular target group. At the same time however, targeting low-end consumers is expected to lead to lower operational self-sufficiency, due to higher expected default rates and lower economies of scale. Targeting high-end consumers or small businesses on the other hand, is expected to especially have a positive effect on the operational self-sufficiency, while outreach is expected to be negatively affected. This is in line with expectations and findings on the average loan size, that is used to compile the target

market variable. According to Hermes and Meesters (2011), the average loan size shows a significant and positive relationship with MFI efficiency. That is, the higher the loan size, the higher the chance MFIs can reach many of the targeted poor while still operating efficiently.

This entails that indeed targeting high-end would results in higher operational self- sufficiency. It could be argued the average loan size, on which the target market variable is partly based, exhibits endogeneity problems with regards to the outreach variable, as higher number of borrowers might effectively lower the average loan size. It is however argued by Mosley (1996), that the average loan size is in fact a policy tool for MFIs and allows them to choose carefully between outreach and financial sustainability, with lower average loans resulting in higher outreach and higher average loans in higher self-sufficiency.

The legal type of an MFI is generally carefully chosen by the institution. The employed dataset allows for differentiation between banks, rural banks, non-bank financial institutions, cooperatives, NGOs and other types of institutions. It should be noted that the sample hardly contains any ‘other’ type institutions. Given the scarcity of the “other” type institutions and singularity issues, they will be excluded in this paper. This exclusion follows the line of reasoning by Hermes and Meesters (2011), who also exclude “other” type MFIs. The differentiation between the MFI types allows for comparison of their performance differences.

According to Hermes and Meesters (2011) these differences potentially arise from the fact that some types of MFI might be more attractive for developmental organisations and donors.

Table 1 represents the descriptive statistics on all MFI characteristics.

Table 1. Descriptive statistics on all MFI characteristics. Data extracted from MIX

VARIABLES N mean sd min max

Dependent Variables

Outreach (share of population) 4,760 0.00257 0.00656 3.12e-07 0.122

Operational self sufficiency 4,760 1.154 0.262 0.263 2.183

Explanatory Variables

Borrowers per loan officer 4,760 304.0 213.8 0.330 1,994

Profit Status 4,760 0.456 0.498 0 1

regulated 4,760 0.658 0.474 0 1

MFI offering Microsavings 4,760 0.516 0.500 0 1

Target Market

Target Small Business 4,760 0.0282 0.165 0 1

Target Low-end 4,760 0.429 0.495 0 1

Target Broad 4,760 0.498 0.500 0 1

Target High-end 4,760 0.0450 0.207 0 1

MFI Type

Bank 4,760 0.117 0.321 0 1

NBFI 4,760 0.372 0.483 0 1

NGO 4,760 0.354 0.478 0 1

Cooperative 4,760 0.0971 0.296 0 1

Rural Bank 4,760 0.0487 0.215 0 1

other 4,760 0.00924 0.0957 0 1

3.2.2. Economic indicators

The economic indicators in this paper were extracted from the World Bank’s World Development Indicators (WDI) dataset. The widely used main indicator for economic development, GDP per capita is also used in this paper. GDP per Capita is in 2011 US dollars and PPP adjusted, thus allowing for meaningful comparisons across countries. Higher levels of income can result in higher repayment rates, and higher profits, yet might also lower the demand for microfinance products. Ahlin et al. (2011) find positive yet insignificant results for income on MFI efficiency. However, it is argued by Ahlin et al. (2011) that despite the insignificance of income it is nevertheless a useful control variable. On the effects of GDP per capita on MFI success, Hermes and Meesters (2011) provide an insightful rationale; they state that ex ante the effects of GDP per capita can be considered rather ambiguous, since high levels of income in a country can be expected to boost entrepreneurial activity and household income, thus resulting in higher demand for microcredit and higher repayment rates of such credit. On the other hand, however, lower levels of GDP per capita might force people away from the formal sector and revert to developmental initiatives such as microfinance.

Lending interest rate represents the prevailing interest rate that is, on average, charged to borrowers in a country at time t. Higher interest rates on loans are expected to increase MFI activity as more people are forced away from the formal financial institutions.

Informal lending facilities such as the reliance on family and friends or even resorting to loan sharks gaining ground, might induce MFIs to experience larger growth. This argument however, depends fully on the assumption that MFIs will not raise their own lending rates in accordance with the prevailing country average. This is a rather strong assumption, since many – especially self-sufficient – MFIs borrow on the formal market in order to fund the demanded credits. In that case, high prevailing lending rates will induce MFIs to find

themselves in a position where they will have to charge the higher rates to their clients.

Another main argument on interest rates has been put forward by Rosenberg et al. (2009), who find that high prevailing interest rate in fact contribute to the microfinance cause, as MFIs generally need to charge high interest rates to cover the relatively high costs of the (generally) small loans. In that case, higher average lending interest rates could result in a smaller gap between formal financial institutions and MFIs.

Data on the inflation in which all selected MFIs are active have been extracted from the MIX database. The variable Inflation is compiled using the Laspeyres index. High inflationary pressure results in decreasing loan value. This can in turn result in higher outreach, due to agents making use of arbitrage opportunities, yet it is more likely to result in lower financial performance by MFIs. In addition, it is argued by Zaidi et al. (2009) in their case study on Pakistani borrowers from MFIs, that high inflation pressures MFIs to a great extent. Even though borrowers might profit directly from high inflation rates, MFIs will have to adjust their prevailing interest rates to inflationary expectations. It would therefore appear that in case MFIs are able to flexibly set their interest rates they can very well cope with high inflationary pressure.

From the WDI database data on the unemployment rates of all countries included in the sample (where available) have been extracted. The WDI database uses the International Labour Organisation’s (ILO) estimates. Unemployment is presented as the share of the total workforce that is involuntarily unemployed (or at least seeking for a job). High unemployment rates potentially cause people to revert to self-employment, thus raising demand for microfinance products. At the same time however, high unemployment might also induce people to take on loans they cannot repay, in which case it would have a negative impact on an MFI’s operational self-sufficiency.

Table 2 represents the descriptive statistics of the economic indicators mentioned in this sub-section.

Table 2. Descriptive statistics on economic indicators. Data extracted from World Bank and MIX

VARIABLES N mean sd min max

Lending Interest Rate 3,683 0.147 0.0725 0.0425 0.600

Adult Unemployment Rate 4,557 0.0688 0.0504 0.00100 0.336

GDP per Capita in US$ 4,712 6,516 4,175 512.9 23,214

Inflation ^{4,741 0.0712 0.0478 -0.132 0.515}

3.2.3. Worldwide Governance Indicators

The World Bank provides the WGI database, compiled by Kaufman et al. (2008), on several indicators on the stability of countries’ governments. These indicators contain the perception of individuals in a country of the quality of institutions and governance in their country. All WGIs are normalised and range from approximately -2.5 to 2.5. The relationship between the governance indicators, also dubbed formal institutions by Hermes and Meesters (2011), and MFI success – seen from both the welfarist and institutionalist points of view – can be seen as rather ambiguous ex ante. This suspected ambiguity is due to the fact that, on the one hand, one might expect stronger institutions (i.e.: stable governments with low corruption and violence) to improve the business environment, while, on the other hand, the lack of corruption controls, for instance, might enable small businesses to circumvent extensive regulations (Hermes and Meesters, 2011). This is asserted by Zeller and Meyer (2002), who find that strong institutions result in a high administrative burden that discourages small-scale entrepreneurs, resulting in lower performance of MFIs. An opposing view is presented by Ledgerwood (1999) who asserts the importance of strong institutions, especially on the field of property rights. Recent research by Ahlin et al. (2011) indicates that stronger institutions result in higher operating costs for MFI, thus lowering their financial sustainability. From above arguments, it becomes apparent that it is generally believed the institutional environment can have an effect on microfinance through two channels. A poorly functioning government that does not enable the formation of successful private sector enterprises forces people into the informal sector. It can then be argued that poorly functioning governments lead to higher success of microfinance, as MFIs potentially serve many people in the informal sector. On the other hand, however, well-functioning governments – in terms of the included governance indicators – that allow people to develop their businesses, can have a great positive influence on entrepreneurial activity. With increased entrepreneurial activity also comes increased formation of small businesses, which can be funded by microcredit. In the latter case, a well- functioning government thus has a positive effect on microfinance success.

The six indicators that are included in the database are all used in the analysis in this paper. Control of corruption indicates the degree to which public power is averted to achieve personal gains. Government effectiveness indicates the perception of the quality of public and civil services. It moreover indicates the believe that government policies can be regarded credible and durable. The rule of law is, arguably, the most important governance indicator for the business environment, as it includes the perception of the extent to which

contracts can be enforced, the quality of property rights and the quality of law enforcement.

Political stability and absence of violence/terrorism measures the perception of the population on the likeliness of the government using violence, the occurrence of politically motivated violence including terrorism and general political stability. The effects of political stability on economic growth were assessed by Looney (2014), who finds, for a sample of African countries, that especially political stability, control of corruption and rule of law are positively associated with higher economic growth. The perception of the extent to which the government promotes private sector development and enables this development through regulation is captured by regulatory quality. Finally, voice and accountability measures the perception of the extent to which people believe they can participate in the selection of their government. This also includes the perception of freedom of speech and media and is often compared to having a system of ‘checks and balances’ in a country. According to Stasavage (2002), voice and accountability is positively related to FDI inflow and entrepreneurial activity.

Table 3. Descriptive statistics of the institutional indicators. Data extracted from WDI and WGI, both datasets from the World Bank.

VARIABLES N mean sd min max

World Governance Index

Control of Corruption 4,760 -0.579 0.385 -1.816 1.564

Government Effectiveness 4,751 -0.413 0.417 -1.677 1.261

Political Stability and Absence of Violence/Terrorism 4,735 -0.775 0.692 -2.812 1.163

Regulatory Quality ^{4,744 -0.308 0.483 -1.681 1.536}

Rule of Law 4,760 -0.610 0.439 -1.956 1.367

Voice and Accountability 4,760 -0.227 0.523 -2.099 1.244

3.2.4. Collinearity

Apart from the evaluation of the descriptive statistics for all variables in order to identify outlier problems, the potential issue of collinearity between variables also has to be taken into account. In order to assess the potential of these collinear relations, a correlation matrix was constructed and can be found in appendix 6.1.

When observing the correlation coefficients, especially the variable government effectiveness stands out. Taking the definition as provided by Kaufman (2010) into account;

“Capturing perceptions of the quality of public services, the quality of the civil service and the degree of its independence from political pressures, the quality of policy formulation and implementation, and the credibility of the government's commitment to such policies.”^††

Given this definition, the potential collinearity can be seen to be quite a problem indeed, since the factors contributing to the government effectiveness can also be attributed to the other variables in the World Governance Indicators dataset. Moreover, a factor such as

“…credibility of the government’s commitment to such policies” looks quite similar to enforceability of contracts or rule of law. The problem however, with the WGI variables is the fact that when government effectiveness is omitted, high correlation coefficients appear between voice and accountability and the remaining WGI variables. After estimation of the models with and without government effectiveness, without observing effects of multicollinearity, it is decided to include it. Moreover, including all WGI indicators is in compliance with the literature that does include these institutions.

Another variable standing out is Target Broad, which seems to be potentially collinear with Target Low-end, the reason for which might be found in the fact that many MFIs report the two rather interchangeably.

Finally, the indicator variable for NGOs, shows a rather high correlation with profit status, which in itself is not surprising, since MFIs tend to choose their ‘type’ of organisation depending on many variables, among which their financial goals.

3.2.5. Variable Transformation

Looking at the histograms in appendix 6.2, it becomes apparent that several variables suffer from a non-normal distribution, which potentially harms the explanatory power of the model. Especially for the lending interest rate, borrowers per loan officer and the average loan size this is the case. In order to deal with this problem, these variables will be transformed by taking their natural logarithms, which will then be used in the estimations. For the GDP per capita, inflation rate and unemployment rate the natural logarithm will also be used for purposes of consistency of the literature.

†† Definition in WGI Working paper 5430

Considering the stationary nature of the data, an augmented Dickey-Fuller test is employed. The null of variables containing a unit-root is firmly rejected for all variables, except for GDP per capita. Taking the log of GDP per capita however, solves the problem of its non-stationary nature. Given the results from the Dickey-Fuller test, no further transformations are required.

4. Methodology

In order to provide a meaningful comparison of the potentially different effects the independent variables have on the two dependent variables, utilising the same model for both analyses is of paramount importance. Although the literature on the effects of the macroeconomic and institutional environment is basically limited to a handful of papers, this paper nevertheless aims at achieving some degree of comparability with the main previous contributions to this area of the microfinance literature. For the sake of comparability, first, a baseline model will be estimated by means of a pooled OLS estimation similar to Ahlin et al.

(2011) and Vanroose (2008) who employ a pooled OLS model to assess macroeconomic and/or institutional effects on their MFI-level variable of interest. After this baseline model estimation, the pooled OLS model will be expanded by including country-level controls.

Finally, considering the panel structure of the dataset, a fixed effect panel model will be estimated to assess the robustness of the results.

4.1. Baseline Model

The pooled OLS basic model is depicted by equation 2 and is similar to the models employed by Ahlin et al. (2011) and Vanroose (2008).

𝑦_𝑖𝑗𝑡 = 𝛼 + 𝛽_𝐶𝐶_𝑖𝑗𝑡+ 𝛽_𝑀𝑀_𝑗𝑡+𝛽_𝑁𝑁_𝑗𝑡+ 𝜖_𝑖𝑗𝑡, 𝜖_𝑖𝑗𝑡~𝐼𝐼𝐷(0, 𝜎²) (2)

Equation 2’s left-hand side consists of a dependent variable 𝑦𝑖𝑗𝑡, that is observed for MFI i in country j at time t. As explained in previous sections, this paper aims at providing a better understanding of the effects of the macroeconomic, institutional and MFI-specific variables on success of an MFI. Since success is explained both through an institutionalist and welfarist point of view, the variable 𝑦𝑖𝑗𝑡 is explained as the outreach of an MFI in the welfarist

regression and the OSS in the institutionalist regression. 𝐶_𝑖𝑗𝑡 consists of a set of all MFI- specific characteristics, which can be found described, in section 2.2.1. These variables are all observed for MFI i in country j at time t. 𝑀𝑗𝑡 depicts a set of macroeconomic variables that are all observed for country j at time t. All included macroeconomic variables are described in section 2.2.2. 𝑁𝑗𝑡 depicts the institutional environment variables, that can be found described in section 2.2.3. The institutional variables are, similar to the included macroeconomic variables, observed for country j at time t. The intercept, 𝛼, is due to the nature of the pooled OLS model of little interest, since this intercept does not differentiate between countries and/or MFIs or over time.

Given the nature of the dataset, i.e.: a panel structure over MFIs, countries and time, simple (pooled) OLS is not expected to produce trustworthy estimates. Given the assumption in pooled OLS of unique and independent observations, this method of estimation without accounting for the panel structure of the data is expected to yield biased estimates. Therefore, equation 2 is expanded on by including robust standard errors, clustered at the MFI level, leading to equation 2.1 below. Using clustered standard errors on the MFI level allows for serial correlation within the MFI observations that may have been caused by MFI-specific shocks or unobserved time-invariant variables. Moreover, using clustered standard errors control for potential heteroscedasticity in the data. This is depicted by the expression for the error term in equation 2.1a, where 𝑢𝑖 depicts the group-level shocks.

𝑦_𝑖𝑗𝑡 = 𝛼 + 𝛽_𝐶𝐶_𝑖𝑗𝑡+ 𝛽_𝑀𝑀_𝑗𝑡+𝛽_𝑁𝑁_𝑗𝑡+ 𝜈_𝑖𝑗𝑡 (2.1) 𝜈_𝑖𝑗𝑡 = 𝑢_𝑖+ 𝜖_𝑖𝑗𝑡, 𝜖_𝑖𝑗𝑡~𝐼𝐼𝐷(0, 𝜎²) (2.1a)

4.1.1. Allowing for country-level controls

The nature of the dataset does not allow for standard errors clustered at the MFI-level and country-level at the same time. It does seem plausible many relevant variables at the country-level have not been taken into account. For this reason, a second pooled OLS model with standard errors clustered at the MFI level will be estimated, yet with the addition of country dummies to allow for unobserved country-specific factors. This model is depicted by equation 3 and will henceforth be referred to as the baseline model.

𝑦_𝑖𝑗𝑡 = 𝛼 + 𝛽_𝐶𝐶_𝑖𝑗𝑡+ 𝛽_𝑀𝑀_𝑗𝑡+𝛽_𝑁𝑁_𝑗𝑡+ ∑^𝑛_𝑧=1𝜑_𝑧𝜇_{𝑧,𝑖𝑗}+ 𝜈_𝑖𝑗𝑡 (3) 𝜈_𝑖𝑗𝑡= 𝑢_𝑖+ 𝜖_𝑖𝑗𝑡, 𝜖_𝑖𝑗𝑡~𝐼𝐼𝐷(0, 𝜎²) (3.1)

In equation 3, the country dummies that are now included to account for unobserved country-specific effects on the success of MFIs are depicted by ∑^𝑛_𝑧=1𝜑_𝑧𝜇_{𝑧,𝑖𝑗}, where the country-dummies 𝜇_{𝑧,𝑖𝑗} equal one if 𝑧 = 𝑗.

With the control for unobserved country effects and clustered standard errors part of the potential bias in the pooled OLS estimator is controlled for. The model depicted by equation 3 however, could potentially be meaningful extended in order to control for unobserved MFI effects by allowing the intercept to vary over MFIs and still controlling for unobserved country-specific variables. This specification however, is highly overfitted and exhibits signs of (near) perfect collinearity. Given these findings, this estimation will not be included and the specification using country dummies will be used as baseline model.

4.2. Extensions

As explained in the previous section, the baseline model will be estimated by pooled OLS.

This method of estimation however, might very well not be the appropriate model of choice given the nature of the data. The introduction of clustered standard errors at the MFI level solves part of this problem, yet effectively utilising all explanatory power of the data requires the incorporation of a panel data model. Moreover, the pooled OLS model can potentially produce biased estimates given the implicit assumption of the pooled OLS model of the explanatory variables being uncorrelated with the error term, 𝐶𝑜𝑣[𝑢_𝑖𝑗𝑡, 𝑥_𝑖𝑗𝑡] = 0. Even though standard errors are clustered at the MFI level, the term 𝜖𝑖𝑗𝑡 in equations 2.1 and 3.1 can still potentially be correlated with the MFI characteristics that are incorporated in the model, in which case the standard error estimates will be biased.

In search for the panel model that fits theory and the nature of the panel structured data, first, a fixed effects model will be estimated. Using a Hausman-test, it has been estimated that the fixed effects model is definitely preferred over a random effects model (i.e.: the null hypothesis of both specifications being appropriate is soundly rejected). Given the Hausman- test outcome and its strong assumptions, the random effects model will not be incorporated in this analysis. The assumption of the explanatory variables being uncorrelated with the randomly distributed unobserved individual effect seems highly unlikely to hold. Due to the rejection of the random effects model, both on theoretical and empirical grounds (Hausman-

test), both the random effects model and the between estimator, which operates under the same assumptions, will not be estimated.

4.2.1. Fixed Effects

Employing a fixed effects model has the attractive feature of a rather weak assumption that allows for correlation between the individual fixed effects and the explanatory variables.

When employing a fixed effects model however, an important problem is likely to arise; a fixed effects model does not allow for time-invariant variables. When only comparing the institutional and macroeconomic variables this would not pose a problem, yet many MFI characteristics will cancel out, as these characteristics are included in the model as dummy variables that do not change over time. One way to deal with this problem is to include interaction effects between time and the time-invariant variables. This allows for including the time-invariant variables, yet harms the interpretation of the results, as the coefficients for the interaction variables can only be compared to their results in the base year (2003) of the model.

Furthermore, looking at the panel summary statistics (appendix 6.3), it becomes apparent that within-variation in the dataset is much smaller than between-variation. Another note of caution should be added regarding the fixed effects model; the employed dataset can be characterised as a rather short, yet broad dataset (i.e.: small T and large N). Even though the sample period contains observations for 11 years, many MFIs do not report all variables over that period. This is partly dealt with, as explained in the data section, by dropping all MFIs that have less than three observations on the focal variables. The short nature of the panel potentially harms the explanatory power of the fixed effects model, as it focuses on the within-variation of the data, effectively utilising the time-series component of the time-variant variables.

For robustness purposes, the fixed effects model will be estimated. Equation 4 represents the specification of the model that will be estimated. The MFI-fixed effects are now captured by the new intercept parameter 𝛼𝑖.

𝑦_𝑖𝑗𝑡 = 𝛼_𝑖+ 𝛽_𝐶𝐶_𝑖𝑗𝑡+ 𝛽_𝑀𝑀_𝑗𝑡+𝛽_𝑁𝑁_𝑗𝑡+ 𝜖_𝑖𝑗𝑡, 𝜖_𝑖𝑗𝑡~𝐼𝐼𝐷(0, 𝜎²) (4)