Mission Efficiency Analysis of For-Profit Microfinance Institutions with Categorical Output Variables

Abstract

The primary objective of microfinance institutions (MFIs) is to provide financial services to low-income clients and underprivileged women. As such, evaluating the efficiency of MFIs should take into account categorical output factors such as outreach and financial intermediation, rather than using the same metrics applied to traditional banks and credit unions. However, under adverse economic situations, one can expect the phenomenon of mission drift of for-profit MFIs such as microfinance banks and credit unions. When a mission drift occurs, MFIs intend to entertain wealthier clients to maximize profits, crowding out the poor ones. This paper empirically examines if such a phenomenon was observed during the global financial crisis period in Latin America and the Caribbean region using categorical Data Envelopment Analysis (DEA) data that have not been considered for the analysis of MFI efficiency. In addition, we conducted two-limit Tobit regression to find significant factors for MFI efficiency. We confirm that for-profit MFIs did not experience mission drift during the adverse economic situation while country, disclosure requirements, institutions’ age, and scale affected the efficiency of the for-profit MFIs. This indicates that for-profit MFIs in Latin America and the Caribbean region performed well in terms of their missions for micro-finance such as outreach, financial intermediation, as well as profit. The financially underprivileged faced a lack of household and business capital under the economic crisis. Based on the results, we conclude that support policies for younger and non-traditional MFIs to help the socially disadvantaged should be actively established for their sustainability in adverse economic situations.

1. Introduction

Many low-income wage-earners are excluded from the formal banking system [1] and have difficulties obtaining traditional financial services despite their need to improve their status [2]. Microfinance is a prime example of the effort to provide financial services for underprivileged people [3,4]. The main objective of microfinance is to help poor clients or underprivileged women who cannot access the formal banking system [5,6]. Poor clients can strengthen their positions within the household when they receive financial support from a microfinance institution (MFI). The national economy can also be activated when they become the main agents of economic activity [4,7,8,9,10,11,12].

The types of organizations that serve as microfinance institutions (MFIs) include microfinance banks, credit unions, non-bank financial institutions (NBFI), and non-governmental organizations (NGOs). The mission of MFIs is to provide loans to very poor individuals, but when an MFI leaves the poor customer group to improve their financial performance, “mission drift” is said to occur [13,14,15]. Mission drift theory states that organizations lose quality services overall to increase their profit [16]. Armendariz and Szafarz [17] indicated that this phenomenon occurs when MFIs seek to increase their profitability by reaching out to wealthier individuals while crowding out poor customers. If MFIs focus on profitability, this prevents them from fulfilling their main mission of poverty reduction [18,19].

Our main concern with mission drift by MFIs is especially related to microfinance banks and credit unions, as their main functions are not microfinance activities. Their mission drift can increase when they are exposed to an adverse economy [20]. As relatively fewer poor customers lead to relatively high profits, for-profit MFIs such as microfinance banks and credit unions tend to forsake the poorest customers when the economy enters a downturn.

Our research question is if the efficiency of for-profit MFIs in terms of their microfinance activities is influenced by the economy. For example, under an adverse economy such as the global financial crisis, governments may try to support underprivileged people to increase the level of economic stability. As part of this effort, the government may provide favors for MFIs, thus causing the efficiency of the microfinance activities of for-profit MFIs to increase even given such an economy. On the other hand, the microfinance activities of for-profit MFIs may decrease during an economic crisis because these institutions want to avoid the risk of bankruptcy.

Many studies have attempted to assess the efficiency of MFIs with multiple input and output variables using DEA (data envelopment analysis) [10,21,22,23,24]. DEA was developed to assess the relative efficiency of decision-making units (DMUs) or economic entities in the public and private sectors [25]. However, none of the previous studies focused on the efficiency of for-profit MFIs in terms of mission drift from microfinance activities.

In particular, previous DEA studies did not consider categorical social performance variables that indicate their mission of financial support for poor and low-income customers, such as outreach and financial intermediation, as provided by the MIX (microfinance information exchange). Outreach reflects the total number of loan applicants receiving enterprise training, education courses, and female empowerment training. It is measured in three levels: small, medium, and large. Financial intermediation (FI) is the proportion of the total assets funded by voluntary savings and is categorized into three levels of non FI, low FI, and high FI. These variables should be considered in DEA to measure the efficiency of microfinance institutions. In this way, the result of DEA with the performances of outreach and financial intermediation can reflect the efficiency of MFIs in terms of their original mission of MFI.

The main purpose is to investigate if mission drift occurs in-profit MFIs under adverse economic situations. We consider categorical output variables of MFIs such as outreach and financial intermediation for DEA. In addition to the economic situation, we aim to understand the influence of various environmental factors on the efficiency.

In order to investigate the efficiency of MFIs, we used the MIX database, one of the largest sources of integrated performance information for MFIs [11,26]. MIX offers financial and social performance data of various types of MFIs across the globe. We analyzed MFIs located in Latin America and the Caribbean region covering 16 countries (Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominica Republic, Ecuador, El Salvador, Honduras, Mexico, Nicaragua, Panama, Paraguay, Peru, and Venezuela) [27]. They recorded double-digit growth of microfinance in recent years and are recognized as having tremendous growth potential [28].

To analyze the efficiency of MFIs under an adverse economy, we used data from 2008 to 2010, as this dataset encompasses the recent international financial crisis. According to Ocampo [29], the economic boom of Latin America ended in 2007. Latin America had been considerably affected by the global financial crisis and was the region hardest hit among the developing counties [30].

We used categorical DEA data, and the result of the DEA was used as the response variable for a two-limit Tobit regression analysis in which year was used as one of the environmental factors that influenced the efficiency of the MFIs in Latin America and the Caribbean region [31,32]. Tobit regression was used to determine the relationship between the efficiency scores from DEA and environmental factors unrelated to the inputs of DEA [33,34].

The structure of this research is as follows. In Section 2, we summarize existing DEA studies utilizing the MIX market database. In Section 3, we introduce our categorical DEA and the two-limit Tobit regression model. Lastly, in Section 4, we summarize our observations and offer suggestions for the directions of future study.

2. Related Literature

In this section, we examine and summarize the existing DEA studies related to the efficiency of microfinance institutions (MFIs). Data envelopment analysis (DEA) was developed to measure the inefficiencies of decision-making units (DMUs) and to suggest areas for improvement [35,36,37,38,39,40]. This method is also commonly used for efficiency analyses of MFIs. The related studies based on the MIX market database and the inputs and outputs used in each of the previous studies are summarized in

Table 1. Inputs and Outputs used in DEA for MFIs.

Authors	Input	Output
Hassan and Sanchez [21]	Total Financial Expenses	Gross Loan Portfolio (In USD)
	-Financial Expenses
	-Loan Loss Provisions Expense
	Operating Expenses	Total funds
	Labor	Financial Revenue
		Number of Active borrowers
Haq et al. [10]	Production Approach
	Labor	Number of borrowers per staff member
	Cost per borrower	Number of savers per staff member
	Cost per saver
	Intermediation Approach
	Total number of staff members	Gross loan portfolio
	Operating/administrative expenses	Total savings
Qayyum and Ahmad [22]	Production Approach
	Number of employees	Gross loan portfolio
	Capital expenditures
	Intermediation Approach
	Labor	Loans
	Capital cost	Financial investments
	Interest payable on deposits
	Assets Approach
		Assets of financial institutions
	Credit officers (number)	Interest and fee income
	Operating expenses (USD thousands)	Gross loan portfolio
		Number of loans outstanding
Gutiérrez-Nieto et al. [41]	Credit officers (number)	Interest and fee income
	Operating expenses (USD thousands)	Gross loan portfolio
		Number of loans outstanding
Bassem [42]	Number of employees	Percentage of female borrowers
	Total amount of the MFI assets	Return on Assets (ROA)
Wu et al. [23]	Deposits	Loans
	Fixed assets	Investments
	Number of employees	Claims on other banks

.

Hassan and Sanchez [21] investigated the efficiency of MFIs in three regions, Latin America, the Middle East and North Africa (MENA), and South Asia in 2005. The authors compared the levels of efficiency according to the region and type of MFI, i.e., whether they were microfinance banks, credit unions, non-government institutions, or non-financial institutions. In addition, the authors investigated statistical differences among MFIs in terms of type and region, finding interaction effects between these two factors using a two-way ANOVA. The authors found that microfinance banks and credit unions had higher levels of efficiency compared to non-government institutions and non-financial institutions. Furthermore, South Asian MFIs were more efficient than their counterparts, but no interaction effects were observed between the regions and types of MFIs.

Haq et al. [10] confirmed the efficiency of 39 MFIs across Africa, Asia, and Latin America using the 2004 MIX data. They identified which MFIs were most efficient in terms of minimizing costs, finding that NGO MFIs qualified as most efficient, though microfinance bank outperformed the other types due to the fact, according to the authors, that microfinance banks are financial intermediaries and can therefore easily access the capital market.

The objective of a DEA study performed by Qayyum and Ahmad [22] was to conduct an efficiency analysis of MIX in South Asian countries. They analyzed the correlation between MFI efficiency and the external environmental factors of the MFI’s age, the value of their total assets, the quality of financial management, their degree of operational self-sufficiency (OSS), and the rate of return on assets (RONA). They also performed a linear regression analysis of the efficiency scores to determine the significance of the environment variables used in the correlation analysis. Their results suggested that MFI efficiency had a positive relationship with the value of the total assets, the level of OSS, RONA, and the age of the entity. From the results of the regression analysis, they also found that the value of the total assets was positively related to the efficiency of the MFI.

Gutiérrez-Nieto et al. [41] used data from 30 Latin American MFIs, as obtained from the Microrate web page for 2003. The authors conducted DEAs for 21 combinations of two inputs and three outputs. Their results showed that two MFIs were efficient under most conditions. The authors also applied a principal component analysis to the efficiency values obtained from the DEA for the 21 combinations of all inputs and outputs. As a result, they obtained four principal components of efficiency, representing the overall efficiency, NGO status, the input choices as credit officers and operating expenses, and the output choice as the gross loan portfolio.

Bassem [42] conducted an efficiency analysis of 35 MFIs that were located in the Mediterranean area based on their activities during 2004–2005. Data were obtained from the MIX database. Their results showed that only five MFIs were efficient according to the CCR model. In the BCC model [42,43], nine MFIs were efficient in 2004 and eight MFIs were efficient in 2005. In addition, according to the DEA, they determined that the medium-sized MFIs were more efficient than their smaller counterparts.

Wu et al. [23] evaluated and compared efficiency levels using thirteen commercial microfinance banks and three rural financial institutions from 2004 to 2007. The MFIs were grouped into twos for two different pairwise comparisons of (1) commercial microfinance banks and rural financial institutions (including rural credit cooperatives) and (2) rural credit cooperatives and microfinance institutions. The authors found that commercial microfinance banks had higher efficiency scores than rural financial institutions. In addition, rural credit cooperatives and one of the selected microfinance institutions were efficient, while the other MFIs were inefficient.

For the efficiency study of MFIs based on DEA, various studies have been conducted. Mersland and Beislad [44] suggested an MFI rating that considered size, profitability, efficiency, risk, social performance, and solvency. The authors indicated that MFI size and profitability were the main factors for rankings. Rizkiah [45] studied the effect of social outreach on the financial performance of MFI based on the regression model. The author found that the breadth of social outreach positively influenced financial performance. Quayes and Hasan [46] confirmed the relationship between financial disclosure and MFI’s performance based on a three-stage least squares model.

As summarized in Table 1, all DEAs employed continuous inputs and outputs. None of them considered categorical outputs, such as outreach or financial intermediation reflecting the MFIs’ missions. In this study, we adopted categorical outputs as well and applied input-oriented DEAs for the efficiency comparison of MFIs, consisting of microfinance banks and credit unions. Next, a two-limit Tobit regression was used to identify the environmental characteristics of the efficient MFIs.

3. Empirical Analysis

This section describes the categorical output-based DEA and summarizes the input and output variables in our research model. We defined both outreach and financial intermediation as categorical variables. These variables need to be converted into dummy variables. The efficiency was calculated comparing various inputs to outputs that were considered in previous research. Afterward, the relationship between the environmental variable and the efficiency score was analyzed.

3.1. Categorical Output-Based DEA

In the MIX data, ordinal categorical variables such as outreach and financial intermediation represent the goals of microfinance. With a general DEA, categorical inputs and outputs cannot be considered. A categorical variable-based DEA was introduced by Banker and Morey [35] to cover two cases: a categorical input-based DEA and a categorical output-based DEA. In this paper, we considered two categorical outputs; therefore, we applied the latter. Kamakura [47] pointed out that the Banker and Morey model was flawed due to an incorrectly specified constraint, and they proposed a revised model. Kamakura [47] was able to reduce the integer problem to a more conventional linear programming approach [48]. In this study, we applied an input-oriented DEA to provide inefficient MFIs with feedback in terms of input adjustments. Kamakura’s input-oriented model is as follows:

$$\begin{array}{c}\min {\mathsf{\Phi }}_0-ϵ\left({\displaystyle {\displaystyle \sum }_{r=1}^{\mathrm{K}}}{\mathrm{S}}_r^{-}+{\displaystyle {\displaystyle \sum }_{l=1}^{\mathrm{M}}}{\mathrm{S}}_i^{+}+{\displaystyle {\displaystyle \sum }_{l=1}^{\mathrm{L}}}{\mathrm{t}}_l\right)\hfill \\ \mathrm{s}.\mathrm{t}.{\displaystyle {\displaystyle \sum }_{j=1}^{\mathrm{N}}}{\mathsf{\lambda }}_j{\mathrm{Y}}_{rj}-{\mathrm{S}}_r^{-}={\mathrm{Y}}_{rj0}\left(r=1,2,\dots ,\mathrm{K}\right),\hfill \\ {\displaystyle {\displaystyle \sum }_{j=1}^{\mathrm{N}}}{\mathsf{\lambda }}_j{\mathrm{X}}_{ij}+{\mathrm{S}}_i^{+}={\mathsf{\Phi }}_0{\mathrm{X}}_{ij0}\left(i=1,2,\dots ,\mathrm{M}\right),\hfill \\ {\displaystyle {\displaystyle \sum }_{j=1}^{\mathrm{N}}}{\mathsf{\lambda }}_j{\mathrm{W}}_{lj}-{\mathrm{t}}_l={\mathrm{W}}_{lj0}\left(l=1,2,\dots ,\mathrm{L}-1\right),\hfill \\ {\displaystyle {\displaystyle \sum }_{j=1}^{\mathrm{N}}}{\mathsf{\lambda }}_j=1,\hfill \\ {\mathrm{t}}_l-{\mathrm{t}}_{l-1}\le {\mathrm{w}}_{l-1,j0}-{\mathrm{w}}_{l,j0}\left(l=2,\dots ,\mathrm{L}\right),\hfill \\ \hspace{1em}\hspace{1em}\hspace{1em} {\mathrm{S}}_r^{-},{\mathrm{S}}_i^{+},{\mathsf{\lambda }}_j,{\mathsf{\Phi }}_0\ge 0.\hfill \end{array}$$

(1)

Here, DMUs (MFIs) are indexed by means of j = 1, 2, …, j₀, …, N to assess the relative performance of the j₀th DMU’s. The rth type of output for the jth DMU is denoted as {r = 1, 2, …, R}, and the ith input (to be conserved) for the jth DMU is denoted as {I = 1, 2, …, M}. ${\mathsf{\Phi }}_0$ represents the j₀th DMU’s efficiency score. $ϵ$ is a non-Archimedean value designed to enforce strict positivity as regards the variable. ${\mathsf{\lambda }}_j$ is the weight of the output and input of each DMU. ${\mathrm{S}}_r^{-}$ is the slack variable of the ith input, ${\mathrm{S}}_r^{+}$ is the surplus variable of the rth output, and ${\mathrm{t}}_l$ is the binary slack. The categorical output in L levels is coded via L−1 “dummy” variables W_lj such that the lowest level of this output for any DMU_j is represented by W_lj = 0 (l = 1, 2, …, L−1), the second lowest level by W_lj = (1, 0, 0, …, 0), and the highest level of W_l = 1(l = 1, 2, …, L−1).

3.2. DMU, Inputs, and Outputs

In this study, we obtained data from the MIX market database. The total number of MFIs in 110 countries during the 1995–2011 period was 10,223. The raw data can be found on the World Bank’s MIX market website (https://databank.worldbank.org/source/mix-market, access on 15 January 2023). According to Gutierrez-Nieto et al. [41], Latin American MFIs mobilized more savings and loans than did Asian MFIs. Moreover, MFIs in Latin America recorded the largest volume per transaction, although their rural outreach remained low. Therefore, we considered 2813 MFIs from Latin America and the Caribbean region.

Among them, we kept the 1059 MFIs that were operating during 2008–2011 in order to observe the roles of MFIs after the economic deterioration of 2008. We excluded the data where the MFI’s disclosure requirement level was less than two out of five levels, as these levels did not provide information about outreach and financial intermediation. In addition, 85 MFIs in 2011 were also excluded due to the fact that their disclosure requirement levels were lower than 2. Therefore, 963 MFIs from 2008–2010 remained. Among them, we included only 248 MFIs with the then-current legal status of a bank or credit union.

Credit unions are non-profit organizations, along with NBFIs and NGOs, but the former were set up by members wanting to benefit their community. Instead of comparing banks with all other types of MFIs, we compared banks and credit unions, which are non-traditional MFIs.

After excluding 83 MFIs with missing values, the final number of MFIs was 165, with 57 banks and 108 credit unions. These were used as decision-making units (DMU) in our DEA.

Next, we chose the inputs and outputs (based on those summarized in Table 1) with the MIX data. Because some inputs and outputs did not exist in the 2008–2010 MIX data, we excluded those factors. After correlation tests between the inputs and outputs, we selected the inputs as cost per borrower, personnel, and total expenses/assets and the outputs as borrowers per loan officer, the percentage of female borrowers, and profit margin, which had correlations of less than 0.3. In addition, we considered two categorical output variables: outreach and financial intermediation. The quantitative criteria used to categorize these variables are summarized in

Table 2. The Quantitative Criteria.

Variables	Levels	Criteria
Outreach	Small	Number of borrowers is less than 10,000
	Medium	Number of borrowers between 10,000 and 30,000
	Large	Number of borrowers exceeds 30,000
Financial Intermediation	Non FI	No voluntary savings
	Low FI	Voluntary savings is less than 20% of total assets
	High FI	Voluntary savings exceeds 20% of total assets

.

In summary, we used three continuous input variables, three continuous output variables, and two categorical output variables (

Table 3. Inputs and Outputs.

	Variables	Initial
Input	Cost per Borrower (Operating Expense/Average Number of Active Borrowers)	X1
	Personnel	X2
	Total Expenses/Assets	X3
Output	Outreach (Categorical variable)	C1
	Financial Intermediation (Categorical variable)	C2
	Borrowers per Loan Officer	Y1
	Percentage of Female Borrowers	Y2
	Profit Margin	Y3

).

3.3. Data Envelopment Analysis and Two-Limit Tobit Regression Model

We conducted an input-oriented DEA with the categorical output variables using Equation (1). The categorical output variables Outreach and Financial Intermediation have three levels, and we created two dummy variables for each categorical output. We calculated the efficiency scores using Excel Solver and Visual Basic for Application (VBA). The results of the DEA are displayed in

Table 4. The Result of the DEA.

Input/Output	Variables	Mean	Std.	Min.	Max.	Result of DEA (with Categorical Output)
Input	Cost per borrower (USD)	288.38	345.70	48.41	4046.87	Efficient DMUs: 50 Inefficient DMUs: 115
	Personnel (USD)	636.16	1334.58	4	9773
	Total expense/Asset (%)	0.18	0.05	0.05	0.47
Output	Borrowers per loan officer (persons)	414.24	248.95	71.87	1651.27
	Percentage of female borrowers (%)	0.51	0.16	0.02	0.98
	Profit margin (%)	0.06	0.41	−4.68	0.49
Categorical output	Outreach (3 levels)	Large: 70 (frequency) Medium: 25 (frequency) Small: 70 (frequency)
	Financial Intermediation (3 levels)	High FI: 144 (frequency) Low FI: 12 (frequency) Non FI: 9 (frequency)

. A total of 50, consisting of 33 banks (29.8) and 17 credit unions (30.6), out of the 165 DMUs were efficient. Thus, overall, the banks and credit unions did not show significant differences in terms of efficiency.

Although the legal status of an MFI was found to be insignificant in terms of efficiency variation, other environmental variables had the potential to influence this variation. We subsequently employed a two-limit Tobit regression model to identify the significant environment factors associated with the MFI efficiency variation (

Table 5. Environmental Variables.

Variables	Levels	Criteria
Country	16 countries	-
Current legal status	Bank, Credit Union/Cooperative	-
Diamonds (Disclosure Requirements)	Level 3	General information, outreach data, and financial data
	Level 4	Level 3 and audited financial statements
	Level 5	Level 3–4 and rating or other due diligence report
Regulated	No	The institution is not regulated by a state banking supervisory agency
	Yes	The institution is regulated by a state banking supervisory agency
Age	New	1 to 4 years
	Young	5 to 8 years
	Mature	Longer than 8 years
Scale (Gross Loan Portfolio in USD)	Large	More than 15 million
	Medium	Between 4 million and 15 million
	Small	Less than 4 million
Sustainability	Non OSS	Operational Self-Sufficiency < 100%
	OSS	Operational Self-Sufficiency = 100%
Target Market(Depth = Average Loan Balance per Borrower 1/GNI per Capita)	Low end	Depth < 20%
	Broad	Depth between 20% and 149%
	High end	Depth between 150% and 250%
	Small Business	Depth over 250%

¹ Gross Loan Portfolio/Number of Active Borrowers.

and

Table 6. The Result of the Two-limit Tobit Regression for MFI Efficiency.

Parameter	Level of Each Variable	Coefficient	S.E.	$\mathbf{Wald} {\mathit{\chi }}^2$	p-Value
Intercept		0.5412	0.155781	3.47	0.0005
Fiscal Year	2008 **	0.148538	0.062671	2.37	0.0178
	2009 **	0.156629	0.064934	2.41	0.0159
	2010	0
Country	Argentina	−0.215133	0.223918	−0.96	0.3367
	Bolivia	−0.051205	0.073597	−0.7	0.4866
	Brazil	−0.063999	0.076386	−0.84	0.4021
	Chile **	−0.539552	0.12153	−4.44	<0.0001
	Colombia **	−0.155134	0.065543	−2.37	0.0179
	Costa Rica	0.165435	0.186063	0.89	0.3739
	Dominica * Republic	−0.157912	0.089807	−1.76	0.0787
	El Salvador	0.189502	0.126309	1.5	0.1335
	Honduras	0.079721	0.135236	0.59	0.5555
	Mexico	−0.127892	0.084621	−1.51	0.1307
	Nicaragua	−0.013749	0.096015	−0.14	0.8861
	Panama *	0.287112	0.160129	1.79	0.073
	Paraguay **	−0.215908	0.096891	−2.23	0.0259
	Peru	−0.105021	0.067484	−1.56	0.1197
	Venezuela **	−0.523199	0.111526	−4.69	<0.0001
	Ecuador	0
Current legal status	Credit Union/Cooperative	−0.041797	0.046796	−0.89	0.3718
	bank	0
Diamonds(Disclosure Requirements)	3	0.12906	0.090872	1.42	0.1555
	4 **	0.096154	0.03992	2.41	0.016
	5	0
Regulated	No	−0.109367	0.072711	−1.5	0.1325
	yes	0
Age	New **	0.116882	0.050233	2.33	0.02
	Young **	0.270135	0.08184	3.3	0.001
	Mature	0
Scale	Large	−0.0246	0.070255	−0.35	0.7262
	Medium **	−0.150538	0.049798	−3.02	0.0025
	Small	0
Sustainability	Non-OSS	−0.048631	0.058722	−0.83	0.4076
	OSS	0
Target Market	Broad	0.033066	0.111949	0.3	0.7677
	High end	−0.142198	0.113583	−1.25	0.2106
	Small Business	0.167996	0.125577	1.34	0.181
	Low end	0

*: significant at 0.10; **: significant at 0.05; Optimization Method: Quasi-Newton; Log Likelihood: −37.11661; AIC: 136.23323; Maximum Absolute Gradient: 0.00059.

). A two-limit Tobit regression was used after DEA to analyze the relationship between environmental factors and DEA efficiency bounded from 0 to 1 [49]. The Tobit regression model is an alternative to ordinary least squares regression (OLS) and is used when the dependent variable is censored [33]. Figure 1 shows the distribution of MFI’s efficiency with categorical output. Hoff [50] investigated two alternative approaches to second-stage DEA, that is, a two-limit Tobit model and OLS regression model. In this paper, the authors stated that a two-limit Tobit regression model was sufficient in representing the second-stage DEA models in most cases.

As shown in Table 6, the current legal status, regulation, sustainability (operational self-sufficiency), and target market variables were not significantly related to efficiency variation in the MFIs. Conversely, the factors fiscal year (2008 and 2009 versus 2010), country (Chile, Colombia, Dominica, Paraguay, and Venezuela versus Ecuador), diamond (level 4 versus level 5), age (new and young versus mature), and scale (medium versus large) showed significant differences from their reference groups in terms of MFI efficiency.

The bank and credit union types of MFIs from 2008 and 2009 were more likely to be efficient than those from 2010. Based on this result, we can infer that for-profit MFIs in Latin America and Caribbean countries efficiently supported underprivileged people under an adverse economy.

MFIs in Chile, Colombia, Dominica, Paraguay, and Venezuela tended to be less efficient than those in Ecuador, and MFIs in Panama were better than Ecuadorian MFIs, while the remaining 10 countries were not significantly different from Ecuador. The fourth-level disclosure requirements of MFIs were more likely to be efficient than those with the fifth-level requirements. In addition, if it had been less than 8 years since the establishment of an MFI, the level of efficiency was high compared to a period longer than 8 years. The MFIs whose scale was between USD 4 million and USD 15 million were more likely to be efficient than those less than USD 4 million.

4. Conclusions

Missions of Microfinance Institutions (MFIs) differ from those of traditional financial institutions, as the former provide financial services to low-income and female customers [51,52]. Many DEA studies have analyzed MFIs’ efficiency while utilizing the MIX database. However, categorical variables have not yet been considered as the inputs or outputs in DEAs of MFIs. Moreover, previous studies have not focused on the efficiency of non-traditional MFIs, such as microfinance banks and credit unions.

In this paper, we conducted a DEA with categorical output variables reflecting outreach and financial intermediation, two characteristics of MFIs, using the activities of Latin America and Caribbean MFIs from 2008 to 2010.

The results showed that approximately one-third of the MFIs consisting of banks and credit unions were efficient. Subsequently, based on the results of the DEA with categorical outputs, we performed a two-limit Tobit regression analysis to identify significant environment variables associated with the efficiency variations of non-traditional MFIs.

The results of the two-limit Tobit regression analysis indicated the following: (1) Microfinance banks and credit unions did not show significant differences in terms of MFIs’ efficiency. (2) Microfinance banks and credit union types of MFIs during the financial crisis were more efficient than those during the economic recovery period. This indicates that non-traditional MFIs in Latin America and the Caribbean region performed well in terms of their missions of microfinance. (3) We infer that the national economic situation influences the efficiency of MFI. (4) Younger, non-traditional MFIs showed better efficiency, as relatively young banks and credit unions were more flexible in their non-traditional roles. Additionally, (5) when the gross loan portfolio of MFIs was small or large, then their efficiency was higher than those with a medium scale.

To summarize the results, microfinance banks and credit unions did not show significant differences in terms of MFIs’ efficiency, and microfinance banks and credit union MFIs during the adverse economic situation were more efficient than those during the economic recovery period. This result is similar to that of previous studies that focused on other countries such as Indonesia, India, Mexico, Kenya, and Bangladesh [53]. Therefore, during an economic crisis, it is necessary to establish policies to support MFIs to help the socially disadvantaged.

The results of our study show the distribution of the efficiency of microfinance banks and credit union MFIs in Latin America and Caribbean countries. These results contribute to the effective management of MFIs; however, limitations also exist, as we only considered microfinance banks and credit unions. This coverage can be extended to non-bank financial institutions (NBFIs) and non-governmental organizations (NGOs) as well. More extensive efficiency analyses can be addressed in further studies as well.