INTEREST RATES, TARGET MARKETS AND SUSTANABILITY IN MICROFINANCE

Transcription

1 INTEREST RATES, TARGET MARKETS AND SUSTANABILITY IN MICROFINANCE By Jacinta Nwachukwu Salford Business School University of Salford, UK address: Telephone: +44 (0)

2 INTEREST RATES, TARGET MARKETS AND SUSTANABILITY IN MICROFINANCE Abstract: This paper presents evidence on the role of interest rates and market targeting in helping MFIs to realize financial self-sufficiency. It pools data from 426 institutions in 41 developing countries from 2004 to The results of an ordered-logistic regression strongly support an inverted U-shaped function for the relationship between interest rates and sustainability. Additionally, a shift away from the low-end market does not significantly improve the possibility of being more profitable after controlling for other relevant covariates. MFIs may not, therefore, be forced to drift away from their original goal of serving the poorest borrowers in pursuit of sustainability. JEL classification: G21; L11; L25; N20 Key words: Microfinance, interest rates, sustainability, target markets, developing countries 2

3 INTEREST RATES, TARGET MARKETS AND SUSTANABILITY IN MICROFINANCE Introduction In March 2002, the world leaders at the United Nation s Monterrey conference on finance and development adopted a consensus recognizing the link between the provision of microfinancial services and the attainment of internationally agreed development goals, including the millennium declaration to halve extreme poverty by To further support this view, the United Nations General Assembly declared 2005 to be the International Year of Microcredit. The contribution of microfinance to economic development has also caught the attention of academics. Many of their studies have argued that microfinance institutions (MFIs) will only be able to meet the expected enormous increase in the demand for microcredit over the mediumto long-term by becoming sustainable commercial ventures. Such a movement toward commercialisation will require that they charge cost-covering interest rates. The general suspicion has been that ongoing subsidisation of MFIs will ultimately lead to a rise in loan default rates, higher costs and a diversion of credit to politically-favoured non-poor borrowers. This paper contributes to the debate on the promise of microfinance in three ways. First it explores the role that interest charges play in determining the financial selfsufficiency of microfinance institutions in developing countries. We utilize data on 426 MFIs in 41 countries across South Asia, East Asia, Eastern and Central Europe, Latin America, the Middle East and Sub-Saharan Africa (SSA) 1 that consistently reported to the Microfinance Information exchange, Inc (MIX) 2 from 2004 to It seems likely that these MFIs are the most successful in terms of transparency and accountability and there are probably a much larger number of institutions whose operations go unrecorded and which are probably less 1 The regional classifications are used by Mix and other international agencies including the UN and World Bank 2 3

4 successful. Nevertheless, we believe that those MFIs which opened their accounts to scrutiny by MIX persistently over a minimum of four years present us with the shape of things to come with regard to the broader industry s commitments in achieving poverty alleviation with profit. Second, this article tests the validity of the agency theory that abusively high interest rates will undermine the overall profitability of the microcredit providers. This is primarily due to an anticipated escalation in the problems of adverse selection and moral hazard with associated higher rates of loan delinquency. Consequently, we employ the non-linear U-shaped function between interest rates and profitability described by Cull et al (2007). This empirical model introduces a quadratic term for the interest rate variable. Third, the paper analyses whether the relationship between interest rates and financial self-sufficiency varies across the different groups of borrowers targeted by microcredit providers. According to information from MIX, microfinance institutions orientate their services toward clients at three different socioeconomic levels. They are: (i) low-end borrowers with an average loan size as a percentage of the respective country s Gross National Income (GNI) per capita of less than 20 percent; these are the poorest clients who generally borrow very small amounts to finance informal microenterprises, (ii) broad-end borrowers who obtain average loan balances that are between 20 and 150 percent of GNI per capita for use in their small-scale enterprises and (iii) high-end customers who receive average loans greater than 150 percent of GNI per capita for investment in medium-scale businesses. Such target market categorisation enables us to examine whether there is a trade-off between financial self-sufficiency and the depth of outreach to the poorest populations a phenomenon commonly referred to as mission drift. In addition, we are able to compare the sensitivity of credit demand to the interest charged to the different target groups. If MFIs operating in different target markets show significant divergence in interest rate elasticity and financial performance, this could have important implications for achieving the MDG goal of poverty alleviation. Apart from the article 4

5 by Cull et al (2007) which sub-divided microbanks into categories by lending method, we know of no other econometric study which explicitly estimates these interest rate elasticities for the different socioeconomic groups of borrowers. The remainder of the paper is organised as follows: Section 1 describes the trend in our data on sustainability, interest rates and average loan sizes used as indicators of target markets for profitable and loss-making institutions across our six developing regions from 2004 to Section 2 specifies the empirical model and the characteristics of the estimation method used in the analysis. Section 3 presents the results and Section 4 summarizes our findings and recommends policies for improving the financial self-sufficiency of MFIs. 1. Data and Descriptive Statistics This section initiates our empirical investigation by looking for the basic trends in interest rates and the profitability of our sample of MFIs by type of borrowers across six developing regions from 2004 through Our data is collected from the MIX website and consists of those 426 institutions that provided information on their portfolio yield, financial revenue and expenses as well as costs arising from loan loss provisions and operations for at least three out of the five years of study. This results in an unbalanced panel in the sense that some time series data were still missing for some of the MFIs. Nonetheless, our calculations here are derived from a dataset comprising between 1673 and 2130 observations, depending on the variable of interest as shown in Appendix Table 1. A summary of these statistics is organised under: (i) financial self-sufficiency, (ii) microcredit interest rates and (iii) target markets for MFIs. 5

6 1.1. The financial self-sufficiency of MFIs: Financial sustainability is concerned with whether an institution is able to generate enough revenue to cover its full costs without recourse to donor subsidies (Khandker et, al, 1995; Ledgerwood (1999); Murdoch 1999; Zeller and Meyer, 2002; Cull et al 2007; Rosenberg et al, 2009; Campion, 2010). Thus, those MFIs which have crossed the financial viability hurdle are able to raise all their required funds in the capital market, repay their creditors, earn profits and/or accumulate reserves for their equity holders. To measure the ability of MFIs to achieve financial self-sufficiency, empirical analysts frequently categorise the institutions in terms of those which are: (i) Operationally unsustainable, (ii) Operationally sustainable and (iii) financially sustainable 3 (Morduch, 1999, Cull et al, 2007). The MFIs classified as operationally unsustainable have operational selfsufficiency ratios lower than 100 percent. This means that earnings from their financial activities are lower than their costs. By contrast, institutions categorized as operationally sustainable have self-sufficiency ratios higher than 100 percent, but equal to, or lower than a selected threshold. The MFIs within the financially sustainable sub-group have operational sustainability ratio in excess of the chosen cut-off point. For this study, we have chosen a sustainability cut-off point of 110 percent a year. Our choice of an implied annual average profitability threshold of 10 percent is based on the approach suggested by Yunus (2007) and sensitivity tested by Gonzalez (2010). These papers classified MFIs that earn up to 10 percent profit as poverty focused, meaning that they are able to achieve their ultimate goal of serving as many impoverished borrowers as possible in a sustainable manner. On the other hand, MFIs with profit rates in excess of this threshold were regarded as commercial enterprises with the primary objective of maximizing returns for their 3 The operational self sufficiency ratio (OPSR) is equal to adjusted financial revenue divided by the sum of adjusted financial expense, loan loss provision expense and operating expense. All definitions are abstracted from page 53 of the Microbanking Bulletin, Issue 19, December

7 shareholders rather than using credit to overcome poverty. We do not intend to justify this proposed 10 percent profitability threshold beyond its simplicity in categorizing our sample of MFIs into clearly defined zones based on interest and fee income from financial and nonfinancial operations. Nonetheless, we test the sensitivity of our results to this suggested profitability cut-off point using an ordered-logit model in Section 3. Figure 1 illustrates how the trend in the average value of the operational self-sufficiency ratio varied across our six regions between 2004 and It shows that, on average, our group of MFIs, regardless of the region in which they are located, earned financial revenues that more than covered their total costs. MIX data confirms that, on average, a developing country MFI in our sample of 426 institutions earned roughly US$76 million a year from loans and other financial services including interest fees, penalties, investment income and commissions. These financial activities were related to an annual average expense of US$64 million arising from interest and fees on 7

8 deposits and borrowed funds, salary and other administrative costs as well as changes in the provisioning for potential losses from loan defaults, subject to the adjustments for subsidies by MIX. This resulted in a positive average net income of US$12 million a year, equivalent to 18 percent of total financial expenses. Across our six regions, financially sustainable MFIs in Eastern Europe and Central Asia (ECA) were, on average, the most profitable in this classification, generating a positive return of 54 percent of total financial expenses a year. Data shows that the high profitability of MFIs in the ECA region stems primarily from their relatively low operating and loan loss provision expenses. A plausible reason for this region s comparatively low cost structure is that most of the ex-communist Eastern European states have a fairly good infrastructure, including road and rail networks, health services and education systems, which lowers the costs of administering loans, collecting repayments and following up on defaults. But despite their comparatively low operating costs, financially sustainable ECA MFIs charged a nominal average interest rate of 38 percent per annum, three percentage points higher than the annual average rate of 35 percent charged by their counterparts in our overall sample. We may therefore infer that monopoly profit formed a significant proportion of the positive net return reported for our ECA MFIs. 2.2: Microcredit Interest Rates A plethora of reports document the rate of interest on micro-loans to low-income borrowers in developing and transition countries (Morduch, 1999; Rosenberg et al, 2009; Campion et al 2010). On the one hand, there is overwhelming evidence that in order to make their operations fully financially sustainable, MFIs tend to charge considerably higher interest rates than conventional banks do for their more affluent borrowers. The claim is that MFIs have high operating expenses arising particularly from the cost of administering small sized loans. Thus, charging interest rates that are high enough to cover these expenditures are necessary to ensure 8

9 that MFIs continue to expand the diversity and quality of their products without a continuing demand for subsidy. To overcome the difficulties of comparing quoted rates across MFIs, analysts normally use portfolio yield as a proxy for the effective interest rate charged. The MIX market calculates portfolio yield by dividing an MFI s actual cash income from loans, fees and commissions in nominal terms by its average gross loan portfolio over the same period. The statistics in Figure 2 show that the nominal average portfolio yield across our overall sample of 426 MFIs was 34 percent, although there was significant regional variation. The subgroup of South Asian MFIs had the lowest average portfolio yield at 23 percent and East Asia the highest rate at 39 percent. A possible explanation for the higher rates in East Asia lies with personnel and infrastructure costs. The rate of growth of their per capita incomes is a good deal higher than that in the other regions and so will be the investment in staff training, salaries and equipment such as motor vehicles and computers by their MFIs. Such will enhance the quality and range of products offered to their increasingly better-off customers. 9

10 Disaggregating the data by level of profitability of MFIs revealed that the rates charged by both sustainable and loss making institutions in any particular region were broadly the same, with the exception of Latin America. Contrary to expectations, the group of unsustainable MFIs in this region charged the highest average interest rate of 44 percent vis-à-vis the 36 percent charged by a typical financially self-sufficient MFI in that area. This may be because the unprofitable MFIs loaned small individual amounts in relation to per capita GNI to the very poor land cultivators in the North East of Brazil and on the low-yield slopes of the Andean republics which the large land owning haciendadoes allow their feudal peasantry to farm only. On the other hand, MFIs in South Asia, profitable or otherwise, charged on average a nominal interest rate of 23 percent a year. Potential reasons for the relatively low interest rate reported for our South Asian MFIs may include the fact that the microcredit markets in this region are mature and comprise the oldest and most experienced providers, like the Grameen bank in Bangladesh. Moreover, most South Asian MFIs are run as cooperatives, solidarity groups or not-for profit NGOs with a stated mission of offering subsidized credit to as many impoverished households as possible (The Mix Market, 2011). Additionally, according to Morduch (1999), these long-established South Asian MFIs offer a diverse range of services, including deposits of micro-savings, insurance and leasing, to more than 200 million customers. Cross-subsidization using earnings from these other services might have kept the actual cash costs to borrowers lower than they otherwise would have been. Besides, in line with the comments by Morowczyinski (2009) and Campion et al (2010), the region s high average population density, combined with good telecommunication networks, including the internet and mobile phones, may well have enabled our sub-sample of financial self-sufficient MFIs in particular to reduce their transaction costs per client. 10

11 1.3. The target market for MFIs Microfinance institutions offer services to clients that differ significantly in a number of ways, including borrower income levels, gender, location, education and marital status together with scale and type of business. However, the literature on clientele targeting in microfinance has focused primarily on the implications for the operating costs, outreach and profitability of lending to the poorest vis-à-vis the better-off among the poor as approximated by the size of loans extended to them. A common assertion is that to be financially self-sufficient, MFIs must target the more affluent among their poor borrowers who are able to absorb larger average loan sizes equivalent to a minimum of two to three times the country s per capita GNI. Such enables the MFIs to capture economies of scale and lower total administrative costs on each unit loaned (Navajas et al, 2000; Attanasio et al, 2005; Goldberg, 2005; Lafourcade et al, 2005). Representations in Figure 3 which measure the average loan sizes extended by our MFIs in terms of the per capita GNI of their respective countries reveal that the implications for profitability of this predicted trade-off between outreach to the poorest borrowers and operating 11

12 costs is far from clear-cut for providers in different regional markets. For example, as expected, loss-making MFIs extended loans equivalent to 54 percent of their countries per capita GNIs compared to the 73 percent reported for our sub-group of financially self-sufficient MFIs. A breakdown by region, results from the sub-sample of ECA MFIs indicated a potential nonlinear inverted U-shape function in the relationship between sustainability and increases in average loan size. We observed that those institutions which loaned almost three times the corresponding per capita GNI of the region failed to raise their net returns above our 110 percent sustainability threshold, whereas those MFIs which extended loans of just over twice the region s per capita GNI obtained profits above this cut-off point. However, those MFIs which loaned less than two times the regional per capita GNI were unprofitable. This finding is consistent with the quadratic U-shaped association between average loan balances and operating costs reported by Cull et al (2007) and Campion et al (2010). As noted by Campion (2010), the implied non-linear relationship between average loan balances and operational efficiency is presumably due to the fact that the non-repayment rates with the costs associated with chasingup and prosecuting defaulters increase markedly beyond a certain size of loans. The discussion here demonstrates that the relationship between profitability and its major determinants is complicated and deserves further analysis. For instance, with respect to nominal interest rates and customer orientation, there does not appear to be any significant pattern in the raw data for our overall sample of sustainable and unsustainable MFIs. There are, however, some variations between profitable and loss-making MFIs within a region. Such implies a possible interaction between MFI characteristics and the varying circumstances within the countries in which they operate. The basic features of an econometric model which we use to capture the relevance of these complexities are discussed in the next section. 12

13 2. Empirical Methodology The aim of this section is threefold. First, it specifies the empirical model used in our analysis. Second, it describes the control variables which enter the model independently in addition to our measure of interest rates and target markets. Third, it outlines the basic features of the estimation technique used. We may deal with each in turn. 3.1: Model specification The analysis in this paper takes advantage of the quadratic regression model proposed by Cull et al (2007). However, we have modified this basic model to focus specifically on the effect of interest rates paid by the low-end, broad-end and top-end borrowers normally targeted by MFIs. The reduced-form equation which we use is expressed as follows: ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Where is a dummy variable which takes values of 0, 1, 2 if the yearly operational self-sufficiency ratio (OPSR) for the MFI in year is less than or equal to 100, greater than 100 but less than or equal to 110 and in excess of 110 percent respectively. As we noted in the previous section, the choice of the sustainability cut-off point merely simplifies the task of creating an ordered partitioning of the observed continuous data on the operational selfsufficiency ratio of our MFIs. The symbol is a dummy variable capturing the effects of those 13

14 unobserved characteristics which are unique to a particular MFI in our sample and which do not vary over time. These institution-specific dummies are treated as either fixed or random parameters depending on the outcome of a test proposed by Hausman (1978). The notation is the white noise error term. The parameter coefficient is a constant term representing the annual average value for the operational self-sufficiency ratio for the sub-group of unsustainable MFIs since they are the category left out of specification. The symbol ( ) is nominal gross portfolio yield. As we mentioned in the prior section, this variable is defined by MIX as the sum of an MFI s actual income from interest charges, fees and commission on loans in nominal terms divided by its outstanding gross loan portfolio, averaged over the same period. According to agency theorists, institutions that charge relatively low interest rates are expected to attract high-quality borrowers that repay their loans on time with an associated reduction in informational asymmetric problems (Stiglitz, 1990; Banerjee et al, 1994; Besley and Coate, 1995; Ghatak and Guinnane, 1999; Gangopadhyay et al, 2005; Armendáriz de Aghion and Morduch, 2005). The higher repayment rate should give rise to an increase in profitability, other things being equal. Moreover, comparatively low interest rates may increase the demand for total credit substantially where the demand for it is elastic and with it profits. We anticipate that the coefficient in equation 1 will have a positive sign where the probability of achieving financial sustainability is concerned. To capture the implications of informational asymmetry for MFIs which charge excessively high interest rates, we include in equation 1 the square of the nominal portfolio yield variable (YLDSQ). It is hypothesised that the relationship between overly high interest charges and profitability is negative due to an aggravation in the problems of adverse selection and moral hazard with related costs of screening, monitoring, enforcing contracts and prosecuting defaulters. These additional expenses could undermine the lender s financial selfsufficiency prospects. Such means that the coefficient is predicted to have a negative sign. 14

15 The variables ( ) and ( ) are intercept dummies included to illustrate the effect of client targeting on the sustainability of MFIs. The notation is coded as 1 if the MFI is classified as focusing on the broad-end borrowers and 0 otherwise. The symbol is a dummy that takes a value of 1 if the MFI is categorised as serving the top-end customers and 0 if not. The theoretical prediction is that MFIs which re-orientate their services from the poorest to wealthier borrowers have higher profit levels because they are able to extend larger loan sizes to individuals with savings, ability to repay and, perhaps, collateral assets as well. Moreover, the more affluent borrowers are likely to be educated and to be able to keep records, evaluate the viability of enterprises for which they might borrow and to fill out loan application forms without assistance. These characteristics should confer cost advantages on the MFIs that cater for our sub-sample of broad and top-end borrowers. It is therefore anticipated that the and differential intercept coefficients will have a positive sign. On the other hand, Olomola, (2000) pointed out that educated individuals have better employment opportunities and tend to move from rural to urban areas in search of white-collar jobs. Their greater mobility implies that they are considered a bad credit risk because they may effectively evade an MFI s enquiries. A negative sign on the coefficients and will occur if MFIs that serve such educated, wealthier borrowers encounter higher monitoring and enforcement costs than their competitors that target the poorest population. Thus, a priori, the effect of market orientation on sustainability cannot easily be determined. To test whether statistically significant differences exist in the elasticity of credit demand with respect to interest rates, we created a set of interaction variables by multiplying the dummies for the broad and top-end target markets with the linear and quadratic interest rates. The resulting interaction terms ( ) ( ) ( ) and ( ) were included simultaneously in our regression model. It is anticipated that the interest-elasticities of the better-off borrowers are relatively high as they 15

16 may have access to alternative competing sources including commercial banks. Such falls in credit demand with related scale diseconomies could undermine a microbank s profits, if any. Consequently, we expect the slope differential coefficients and to have a significant negative sign Control variables and expected correlations We added a further nine variables to our empirical model in order to control for the effects of MFI characteristics and macroeconomic stability. They were: First, the total operating costs scaled by the size of its assets ( ) is the sum of capital and personnel costs faced by an MFI expressed in terms of the book value of its total assets. It is theorized that higher proportions of operating expenses to assets will force sustainable MFIs to raise their interest rates so as to maintain their expected profit margin. However, as we said earlier, such increases in interest charges above a certain cut-off point will tend to lower profits primarily due to deterioration in loan portfolio quality and a falling credit demand. Accordingly, the slope coefficient is expected to show a negative sign. Second, to investigate the potential association between sustainability and the effectiveness with which MFIs manage their human capital, we included the ratio of borrowers to staff member ( ) in our regression model. A positive relationship suggests that sustainable MFIs are able to provide their loan delivery services, including application reviews, interviews and site visits with a minimal number of employees. This is probably because they are likely to be already mature institutions with staff that have gained experience in serving ever higher numbers of loan clients at lower costs for each. On the other hand, according to Olomola (2000), reducing the number of staff for site visits, loan monitoring and recovery can raise the default rate, adversely affecting profits. The foregoing implies that the effect of borrowers per 16

17 staff member on financial viability as represented by the sign on the coefficient may be either positive or negative depending on the characteristics of borrowers. Third, the ratio of a gross loan portfolio to the book value of total assets( ) is another measure of the efficiency of asset utilisation employed in our study. For our purpose, MFIs with a greater proportion of total assets generated through lending are likely to be more experienced and effective in directing assets to their most productive use. Consequently, we anticipate a positive sign for the coefficient in equation 1. Fourth, another important consideration in our empirical analysis is the relationship between institution size and sustainability. We measure size as the book value of total assets ( ). If economies of scale exist in the microfinance industry, we would expect a positive relationship between profitability and our size variable. It follows that larger MFIs are able to spread their costs by delivering services to a larger group of clients and/ or by extending larger sized loans to individual borrowers. Such helps lower the lender s costs per unit loaned. Moreover, larger MFIs are more likely to have the resources to undertake the investment in technology and human capital needed to enhance efficiency and profits. Theoretically, we predict that the coefficient will have a positive sign where sustainability is concerned. Fifth, the degree to which our MFIs are able to generate enough revenues to cover costs is also related to the quality of their outstanding loan portfolios. We use the share of a loan portfolio that remains unpaid after at least thirty days past due date ( ) as a proxy for portfolio quality and the risk of loan delinquency faced by each of our MFIs. Agency theorists predict that sustainable MFIs maintain relatively high loan portfolio quality (i.e., low values of PAR30). They are therefore less likely to have to write-off bad loans from their books or refinance by extending their term. For that reason, MFIs with higher proportions of loan portfolio at risk of default are expected to charge higher interest rates with a concomitant 17

18 decrease in customer base and possible profitability outlook. We therefore expect that the coefficient to bear a negative sign. Sixth, the scale of profitability of MFIs is also affected by regulation and the quality of public institutions, including the courts and capital markets in the countries in which they operate. A sound regulatory environment, especially one that encourages transparency in accounts and share purchases and which also protects the rights of investors, will help MFIs attract the long-term equity capital they need to expand their product portfolios. To analyse these potential associations, Sinkey and Carter (2000) suggested the inclusion of the ratio of equity to total assets ( ) as a measure of the capacity of an institution to meet the requirements of the regulatory authorities. A positive correlation would indicate that the likelihood for financial self-sufficiency is enhanced when MFIs have sufficient long term capital to meet both their regulatory obligations and to expand physical and intellectual assets. A negative relationship, on the other hand, suggests that, on average, our MFIs used the proceeds from newly issued equity capital to recapitalise by repurchasing outstanding debt in order to meet any required capital adequacy ratio and so reduced their likelihood of bankruptcy. This means that the sign on the coefficient could be positive or negative depending on the initial debt position and the consequent threat of liquidation faced by the reporting MFI. Seventh, the institutional mission of MFIs is the other major driver of the level profitability often cited in the microfinance literature. We proxy for the commitment of MFIs to achieve deeper outreach to the poorest borrowers in terms of their average loan size per borrower relative to the per capita GNI of the country in which the institution is located ( ). It is assumed that MFIs with very strong commercial objectives will make larger loans to their customers. The results from Cull et al (2007) and Campion et al (2010) have shown that such increases in loan sizes are related to lower operating expenses with associated profits. Consequently, we expect the coefficient to have a positive sign. 18

19 Eight, another variable commonly used in empirical models to demonstrate the dedication of MFIs to meet the pledge of providing credit to the poorest members of society is the share of loans extended to women borrowers( ). Evidence shows that MFIs with strong social mission make more loans to women. However, women, by comparison with men, are less literate. Moreover, this group of poor borrowers are more likely to live in rural areas which have infrastructure deficiencies and greater client dispersion. The combined effect on sustainability of MFIs of the implied extra cost of making loans to a large number of female borrowers is represented by a potential negative sign on the coefficient, although evidence from the literature about this is inconclusive (Pitt and Khandker, 1998, Brau and Woller, 2004). Ninth, inflation ( ) is one of the important external factors influencing the profitability of MFIs. It is expected that sustainable microfinance institutions will regularly adjust the nominal interest rates they charge borrowers to reflect the changing impact of anticipated inflation on the cost of funds. Thus, a rise in anticipated inflation rates, with a subsequent increase in the nominal portfolio yield of MFIs, will lead to a lowering in the demand for microcredit and related profits. We may therefore infer that the sign on the coefficient will be negative. Items 2 to 10 in Appendix Table 1 provide the summary statistics for all the explanatory variables used in our empirical model. The data is reported for both our overall group of 426 MFIs and for our sub-samples of unsustainable, operationally sustainable and financially sustainable MFIs from 2004 through As expected, the group of financially self-sufficient MFIs have higher averages for loan balances per GNI. They also have higher asset utilisation ratios measured either by borrowers per staff member or gross loan portfolio relative to total assets. Further, our sub-group of financially sustainable MFIs faced lower default problems as approximated by the share of loans which were unpaid after thirty days of the due date. The overall effect of these favourable factors is to reduce the average operating expense to asset ratio 19

20 for the sub-group of financially-viable MFIs to roughly half of the 29 percent per year reported for their loss-making rivals. But, the importance of size, equity to capital ratio and lending to women in influencing the level of profitability of our sample of MFIs is not readily apparent. Appendix Table 2 displays the contemporary pairwise association between the variables in our overall sample dataset based on the Kendall's tau-b rank correlation coefficient which adjusts for outliers and ties in the orderings of the data. The results are generally similar to those obtained from ordinary Pearson correlations, though the tau-b statistics are somewhat lower than their counterparts. Expected relationships that are evident include a statistically significant positive correlation between the operational self-sufficiency ratio and the ratios of equity to assets, borrowers per staff member, gross loans to assets and the overall size variable. As envisaged, changes in the operational self-sufficiency ratio are inversely correlated with a rise in the percentage of loans that are delinquent, the ratio of operating costs to assets and with the proportion of loans extended to women at a five percent confidence level. Interestingly, the relationship between portfolio yield and financial viability is not statistically different from zero. The extent to which these outcomes are biased by the omission of the MFI attributes used as explanatory variables in our regression model in equation 1 will be explored in the next section Research methodology Our empirical investigation employs a non parametric test for the differences in means and a regression. We estimate equation 1 using an ordered-logit model since the dependent variable ( ) is a dummy coded as 0, 1 and 2 for our three different categories of MFIs. { } 20

21 where ( ) is the sustainability classification dummy assigned to MFI in time ; ( ) is the corresponding continuous data on the ratio of operational self-sufficiency for the MFI in time. The notations are the threshold parameters with values of 100 and 110 in that order. As we said earlier, these sustainability cut-off points merely simplify the partitioning of our sample of MFIs and there is no statistical significance between these unit distances and the set of observed values of operational self-sufficiency ratios. As part of our estimation process, we checked the reduced-form regression model in equation 1 for possible violations of the assumptions of the classic linear regression model. Researchers normally test for five potential errors in a panel data framework. Therefore, first, we examined the correlation matrix of the regressors in Appendix Table 2 for evidence of potential multicollinearity. The statistics showed that the explanatory variables are moderately correlated with one another with the exception of the association between portfolio yield and operating costs which is around Nonetheless, the results suggest that the degree of multicollinearity in our regression model is minimal. Second, we carried out a Hausman test for endogeneity using the auxiliary regression method proposed by Davidson and MacKinnon (1989, 1993). The estimated Wu-Hausman F-statistic rejects the null hypothesis that our explanatory variables are jointly exogenous. To correct for this misspecification bias, we estimated what is essentially a two-stage simultaneous equation model by replacing all the stochastic regressors in equation 1 with their corresponding predicted values which were uncorrelated with the error term. Further, the use of such fitted values helps minimize the potential problem of heteroskedasticity. The fitted values were retrieved from an OLS regression of each random explanatory variable on all of the other regressors in equation 1 lagged two years. Third, we checked the appropriateness of the fixed versus random effect estimators using the Hausman s chi squared statistic suggested by Greene (2002). A small and statistically insignificant Hausman statistic was generated as part of regression output. Such favours the random effect model. Fourth, to improve model efficiency, we used the natural logarithm of all the regressors at level. The result 21

22 of a J-test of a non-nested hypothesis proposed by Davidson and MacKinnon (1993) rejected the linear model against the log-linear specification at the five percent significance level. Fifth, a test for a lack of first order serial correlation in the disturbances ( ) is not rejected. This is presumably because of the structure of our panel dataset which comprises a large cross section of 426 MFIs with a relatively short-term time period of five years. Also, the pooling of large cross-sectional data with a small time series dimension means that our empirical results are less prone to the problem of non-stationarity or unit roots in data. 3. The Empirical Results and Discussion This section presents the results of our empirical analysis. Our argument here is conducted under: (i) The differences in means and (ii) An ordered-logit regression model. Explanations in these paragraphs are based on an informed, but speculative filigree of cause and effect. They may, or may not trace what actually happened. 3.1.The differences in means Appendix Table 3 presents the differences between our three sub-sample means for the variables employed in our empirical model in equation 1. We also report the corresponding F-test statistic obtained using a single factor Analysis of Variance (ANOVA) method under the null hypothesis that the sub-groups have the same mean and variance. Column 1 compares the average values for loss-making and operationally sustainable MFIs. It shows that MFIs with profits below our cut-off point of ten percent have average assets that are four times larger than those of their unprofitable opponents. They have more loans outstanding relative to assets with a lower proportion of these loans at risk of non-repayment. 22

23 The other variables with differences in means that are statistically significant at the five percent level include the lower ratios to assets of operating expenses and equity capital observed for the operationally profitable group. Contrary to expectations for the group of unprofitable MFIs, the differences in the means of interest rates, average loan balances relative to per capita GNI, the share of women clients and the borrowers per staff member are not statistically significant. Column 2 shows differences between group means for unsustainable and financially selfsufficient MFIs. Seven of the nine explanatory variables have statistics that are significant at the five percent level. As expected, they show that our sub-sample of financially successful MFIs with net returns in excess of our ten-percent threshold have more assets with higher average loan balances, a higher quality of loan, a lower operating expense ratio, a higher proportion of loans outstanding relative to assets and more borrowers per staff member. However, contrary to theoretical beliefs, our group of financially sustainable MFIs have lower proportions of equity in their total assets, compared to their loss making rivals. Column 3 shows the differences in group means for operationally versus financially sustainable MFIs. The pattern that emerges from the statistics shows that, on average, the asset size of the sub-group of financially self-sufficient MFIs is similar to that of the operationally sustainable category, even if the former has significantly more equity capital in their total assets. Moreover, financially viable MFIs are more efficient in utilizing their assets as illustrated by their significantly higher average ratios of gross loan portfolio to total assets, lower percentages of loan at the risk of default and lower operating expense to asset ratios. This was in spite of the fact that they charged similar interest rates, extended the same per capita GNI loan balances and served the same proportions of women borrowers. Taken together, the statistics in Appendix Table 3 indicate that financial self-sufficiency was related to cost advantages arising from a larger size of average loan balances relative to per capita GNI, a higher borrower per staff member and a higher ratio of equity to asset ratio up to a 23

24 certain optimal level. Finally, we find no support for the assertion that profitable MFIs are more likely to charge higher interest rates and serve fewer women borrowers compared with their socially-minded loss making competitors. 3.2.Ordered-logit regression results Appendix Tables 4 and 5 show the results of random effects ordered-logit probability models estimated using our overall sample of 2130 observations with different combinations of the aforementioned explanatory variables. The choice of model was based on the diagnostic test statistics provided as part of a regression output generated by using LIMDEP. These statistics included the McFadden Pseudo R-squared and the Chi-squared statistic as well as the Akaike and Bayesian information Criteria. A discussion of these outputs is conducted under: (i) the financial sustainability probability and (ii) marginal effects The financial sustainability probability: Appendix Table 4 presents the estimates and asymptotic t-statistics obtained from random effects ordered-logit models. In LIMDEP, the dataset is normalised so as to determine the probability of observing the largest value of the coded dependent variable in the overall sample. We may therefore infer that the estimated coefficients in this table reflect the association between the regressors and the likelihood that a randomly chosen MFI in our sample will belong to the financially sustainability category as opposed to the other two sub-groups the unsustainable and operationally sustainable subsamples. A statistically significant positive coefficient indicates that the explanatory variable is correlated with a greater probability of being in the financially self-sufficient sub-group. A significantly negative coefficient, on the other hand, suggests that the regressor is related to a lower possibility of belonging to the financially sustainable category. In Appendix Table 4, we 24

25 present six separate regressions, although the argument here is limited to Model 3 in Column 3 which has the highest Chi-squared statistic and McFadden Pseudo R-squared, as well as the lowest values for the Akaike and Bayesian Information Criteria. The results in Column 3 show that, as hypothesised, the linear portfolio yield variable YLD has a positive and significant coefficient while its squared covariate YLDSQ is negatively significant. These estimates indicate that MFIs which charge reasonably low interest rates have a greater chance of being classified within our financially-sustainable sub-sample than their counterparts with extremely high portfolio yields. Interestingly, the coefficients on the broad-end and top-end target market dummy variables (BEMTRK and TEMTRK) and their interactions terms (BEMKT*YLD, TEMKT*YLD, BEMKT*YLDSQ and TEMKT*YLDSQ) are not statistically different from zero. This demonstrates that MFIs that oriented their loans towards the better-off customers are no more likely to achieve financial self-sufficiency than those which lend to the poorest. This is contrary to received opinion, but it was also supported by Cull et al (2007). A possible reason for it is that the economies which arise from jointliability group lending and other innovations employed by MFIs which serve the very poor have removed the disadvantages of targeting this group of borrowers versus their richer compatriots. With regard to our nine extra control variables, the results show that, as expected, relatively high gross loan portfolio-to-assets ratios (GLPR), equity capital-to-assets ratios (ECAR) and average loan balances scaled by per capita GNI (ALPBP) significantly improve an institution s possibility of being financially successful. By contrast, high operating expense-toassets ratio (TOCR), portfolio at risk after thirty days (PAR30) and the inflation rate (INFL) significantly lower the chances of an MFI becoming financially successful. The sign and significance of these regressors are robust across all the models, whereas the result for the borrowers per staff member indicator (BPSM) is sensitive to the addition of the asset size variable in the model. This outcome highlights the link between asset size and productivity. The 25

26 inference is that only those mature MFIs with a sufficiently large scale of activities have the ability to fund the capital investment in the training, technology and equipment needed to maximise services with minimal resources including staff. The variable for the proportion of women to total active borrowers (WBR) is insignificantly positive. This result suggests that the relatively high illiteracy rate among women, coupled with the extra cost of reaching them in the rural areas where the majority reside, counteract the gains that MFIs might otherwise make from their comparatively high repayment rates : The marginal effects: Appendix Table 5 presents the results of the marginal effect of a one percentage change in the regressors on the probability that our randomly chosen MFI will be within the financially sustainable sub-sample. Once again, our discussion concentrates on Model 3 in Column 3. The classification table shows that this model correctly identified percent of the 783 observations in our sample with operational self-sufficiency ratio in excess of the 110 threshold as belonging to the financial sustainable sub-group ( ). The results show that the coefficients for the linear and squared interest rates persisted with their respective positive and negative signs reported in Appendix Table 4. Moreover, the absolute value of the estimated marginal effect of the linear interest rate continued to dominate the impact of the quadratic variable. We calculated an average interest rate of 76 percent a year as the cut-off point at which the positive linear coefficient was completely cancelled out by the negative quadratic effect. The presumption is that the profitability outlook for a typical MFI in our sample will rise with increases in the interest rates, but at a declining rate. The opportunities for further profit will be fully exhausted when nominal yield reaches an average of 76 percent a year. Beyond this threshold, additional returns from higher interest charges by lenders will be outweighed by the extra costs of making, monitoring and recovering these further loans and any consequent loss of borrowers and scale economies. Our findings are comparable to those of Cull 26

27 et al (2007) who reported a threshold value of 60 percent per annum for individual loans based on the coefficient estimates of a simple OLS regression model. The results show that the marginal impact of four of our nine control variables on the probability of our MFI being financially profitable are significantly more than 0.5 percent. First, a one percentage increase in the ratio of operating costs to total assets (TOCR) will reduce the possibility for achieving financial sustainability by 0.51 percent. Second, a unit change in gross loan portfolios relative to total assets increases the profitability prospect of our typical MFI by 0.62 percent. Third, a rise in the share of equity in total assets raises the odds in favour of financial success by 0.54 percent, although the magnitude of this coefficient is somewhat sensitive to model specification. Fourth, a one percentage increase in the rate of inflation reduces the likelihood that an average MFI in our overall sample will be in the financially sustainable sub-group by 0.7 percent. The implication is that MFIs seeking to achieve financial self-sufficiency should regularly update the premium for inflation and allow for it in full within the interest which they charge. The difficulty is that borrowers may not see the direct link between the rise in the interest rate which they pay and the added inflation premium and may reduce their demand for loans accordingly. Conclusions and Policy Implications The primary objective of the empirical analysis conducted in this paper was to provide evidence on the role of interest rates and the poverty level of target customers in determining the financial sustainability of MFIs. It used pool annual time series data from 426 institutions in 41 countries across the Third World from 2004 to The overall findings may be summarized as follows: First, the results strongly support an inverted U-shaped association between interest rates and financial self-sufficiency. Our estimation of the unrestricted slope coefficients on the linear 27