Microfinance institutions and efficiency

Download Microfinance institutions and efficiency

Post on 11-Sep-2016




1 download


  • Omega 35 (2007) 131142www.elsevier.com/locate/omega

    Micronance institutions and efciency

    Begoa Gutirrez-Nietoa,, Carlos Serrano-Cincaa, Cecilio Mar MolinerobaDepartamento de Contabilidad y Finanzas, Universidad de Zaragoza, Gran Va 2, 50005 Zaragoza, Spain

    bKent Business School, University of Kent, UK

    Received 26 January 2005; accepted 26 April 2005Available online 23 June 2005


    Micronance Institutions (MFIs) are special nancial institutions. They have both a social nature and a for-prot nature.Their performance has been traditionally measured by means of nancial ratios. The paper goes beyond simple nancial ratiosusing a data envelopment analysis (DEA) approach to measure the efciency of MFIs.

    Special care is taken in the specication of the DEA model. We take a methodological approach based on multivariateanalysis. We rank DEA efciencies under different models and specications; e.g. particular sets of inputs and outputs. Thisserves to explore what is behind a DEA score.

    The results show that we can explain MFIs efciency by means of four principal components of efciency, and this way weare able to understand differences between DEA scores. It is shown that there are country effects on efciency; and effectsthat depend on non-governmental organization (NGO)/non-NGO status of the MFI. 2005 Elsevier Ltd. All rights reserved.

    Keywords: DEA; Efciency; Banking; Operations management; Micronance; Microcredit

    1. Introduction

    Microcredit is the provision of small loans to very poorpeople for self-employment projects that generate income.It is a new approach to ght poverty. In its heart are newnancial institutions, often non-prot organisations, whoseaim is to serve those people who would not have access toa loan from a traditional trading bank.

    The fact that Micronance Institutions (MFIs) tend not tooperate in the same way as traditional banks does not meanthat they are not interested in protability and efciency

    This paper was processed by A.E. Adenso-Diaz. Corresponding author. Tel.: +34 976 761791;

    fax: +34 976 761769.E-mail addresses: bgn@unizar.es (B. Gutirrez-Nieto),

    serrano@unizar.es (C. Serrano-Cinca),C.Mar-Molinero@kent.ac.uk (C. Mar Molinero).

    0305-0483/$ - see front matter 2005 Elsevier Ltd. All rights reserved.doi:10.1016/j.omega.2005.04.001

    issues. However, existing tools to assess the performanceof traditional banking institutions may not be appropriatewithin this new context.

    How can we assess if a MFI is efcient? How should wecompare MFIs? How far is existing knowledge on traditionalnancial institutions appropriate in order to understand thebehaviour of MFIs? These are the issues that are addressedin the current paper.

    The paper starts with a discussion of microcredit and itsrole in the ght against nancial exclusion. Existing tools forthe assessment of performance in MFIs are next reviewedand some lessons are drawn from this review. It is suggestedthat data envelopment analysis (DEA) is an appropriate toolfor the assessment of MFI performance. There is, however,an issue to be resolved: how should the DEA model be spec-ied? Which inputs and which outputs should it contain? Amethodological approach based on multivariate analysis isapplied in order to select appropriate model specications,understand the way in which the relative efciency of a

  • 132 B. Gutirrez-Nieto et al. / Omega 35 (2007) 131142

    MFI is determined by the choice of model, and to producea ranking of MFIs in terms of efciency. The methodologyis applied to the analysis of 30 Latin American microcreditinstitutions. The paper ends with a concluding section thatlists and discusses the ndings.

    2. Microcredit and micronance institutions

    It has long been argued that commercial banks have notprovided for the credit needs of relatively poor people whoare not in a condition to offer loan guarantees but whohave feasible and promising investment ideas that can resultin protable ventures [1]. Meeting this need is of interestto governments, charitable institutions, and socially respon-sible investors. New nancial institutions have arisen thatare in touch with the local community, that can obtain in-formation about the loan taker at low cost, and that oftenare not only interested in prot but also on the creation ofjobs, womens employment, development, and green issues.These new nancial intermediaries, the MFIs, provide smallloans to poor people who can offer little or no collateralassets. But the provision of such microcredit is not limitedto not-for-prot organisations. Traditional nancial institu-tions can, and often do, make loans to the deprived as partof a socially responsible investment policy.

    The best known innovation arising from micronance pro-grams is peer group loan methodology, in which membersaccept joint liability for the individual loans made. This jointresponsibility approach results in low levels of default, butthere are other reasons for successful repayment rates: dy-namic incentives, regular repayment schedules and collat-eral substitutes [2].

    Microcredit institutions have mushroomed in countrieswith less developed nancial systems. The Microcredit Sum-mit Campaign formed by donors, policymakers and morethan 2500 MFIs, claimed to have helped 41.6 million of thepoorest people around the world by 31 December 2002 [3].Their goal is to reach 100 million of the worlds poorestfamilies by 2005. Moreover, the United Nations declared2005 as the Year of Microcredit.

    According to Von Pischke [4], modern microcreditevolved from its origins in the mid 1970s to the presentday from some organisations that offered loans and sav-ings to individuals at the margins of the nancial markets.Some examples of microcredit initiatives are: FINCA andACCION International, two US organisations whose areaof activity is Latin America; the rural units of Bank RakyatIndonesia (BRI), one of the few institutions that receive nosubsidies; and Grameen Bank in Bangladesh, now acting inmore than 50 countries.

    3. Assessing microcredit institutions

    Microcredit emerges as a new approach to ght poverty.But, is the money lent by MFIs efciently managed? Thereis much literature on bank efciency, but very little on mi-

    cronance efciency. Should we assess micronance insti-tutions efciency the way banks do, taking into account -nancial inputs and outputs? This tends not to be the case:Morduch [2] observes that discussions on microcredit per-formance almost ignore nancial matters.

    Yaron [5] suggested a framework, based on the dual con-cepts of outreach and sustainability, that has become popularin the assessment of MFIs performance. Outreach accountsfor the number of clients serviced and the quality of theproducts provided. Sustainability implies that the institutiongenerates enough income to at least repay the opportunitycost of all inputs and assets; [6]. It is difcult to think of asustainable MFI with poor nancial management; [7]. Sus-tainability has two levels: operational and nancial (see, forexample [8]).

    Micronance industry evolution stresses more and morethe importance of nancial viability. A set of performanceindicators has arisen, and many of them have become stan-dardised, but there is by no means general agreement on howto dene and calculate them. A consensus group composedof micronance rating agencies, donors, multilateral banksand private voluntary organisations agreed in 2003 to someguidelines on denitions of nancial terms, ratios and ad-justments for micronance [8]. The ratios fall into four cat-egories: sustainability/protability, asset/liability manage-ment, portfolio quality, and efciency/productivity. Thesemeasures derive from the nancial ratio analysis imple-mented in conventional nancial institutions. In what fol-lows, we will concentrate on efciency ratios. Table 1 showsa list of 21 ratios issued by Microrate, used to assess the per-formance of MFIs and their denitions. These are groupedin terms of portfolio quality, efciency and productivity, -nancial management, protability, productivity and others.

    The efciency/productivity ratios reect how efcientlyan MFI is using its resources, particularly its assets andpersonnel [8]. Thus, efciency ratios compare a measureof personnel employed with a measure of assets. Institutionscan choose as assets either average gross loan portfolio, oraverage total assets, or average performing assets. CGAP de-scribes as performing assets loans, investments, and otherassets expected to produce income. Personnel may be de-ned as the total number of staff employed or the numberof loan ofcers. In this paper we are going to use a differentdenition of efciency, based on DEA, as dened by themicro economic theory of production functions.

    4. DEA efciency and nancial institutions

    The efciency with which nancial institutions conducttheir business has long been studied. Efciency assessmentis based on the theory of production functions. The stan-dard denition of efciency is due to Pareto-Koopman;see [9]. There are two main approaches to efciencyassessment: parametric frontiers and Data EnvelopmentAnalysis (DEA). Berger and Humphrey [10] provide acomprehensive review of methods and models up to 1997.

  • B. Gutirrez-Nieto et al. / Omega 35 (2007) 131142 133

    Table 1The 21 ratios issued by Microrate and their definitions

    PQ1 Portfolio at risk = portfolio at risk/gross loan portfolioPQ2 Provision expense ratio = loan loss provision expense/average portfolioPQ3 Risk coverage ratio = loan loss reserves/portfolio at riskPQ4 Write-off ratio = write offs/average portfolioEP1 Operating expense ratio = operating expenses/gross loan portfolioEP2 Cost per client = operating expenses/average number of clientsEP3 Personnel productivity = number of borrowers per staffEP4 Credit ofcer productivity = number of active borrowers/number of credit ofcersFM1 Funding expense ratio = interest and fee expense/average gross portfolioFM2 Cost of funds ratio = interest and fee expenses on funding liabilities/average funding liabilitiesFM3 Debt/equity ratio = total liabilities/total equityP1 Return on equity = net income/average equityP2 Return on assets = net income/average assetsP3 Portfolio yield = cash nancial revenue/average gross portfolioPrd1 Personnel expense/average gross portfolioPrd2 Credit ofcers/total personnelPrd3 Incentive pay as % of base salaryPrd4 Percent of staff with < 12 monthsO1 Average loan balance per clientO2 Current assets/current liabilitiesO3 Equity/assets

    PQ: Portfolio Quality; EP: Efciency and Productivity; FM: Financial Management; P: Protability; Prd: Productivity; O: Other.

    This subject has continued to interest researchers up to thepresent date; some recent papers on efciency and nancialinstitutions are [1115]. The literature continues to growall the time.

    One advantage of DEA (nonparametric) over parametricapproaches to measure efciency is that this technique canbe used when the conventional cost and prot functions can-not be justied; Berger and Humphrey [10]. DEA performsmultiple comparisons between a set of homogeneous units.For an introduction to the theory of DEA see [9,16,17].

    For the purposes of this paper, it will be useful to make adistinction between model and specication in a DEA con-text. Different philosophical approaches as to what a nan-cial institution does, and what is meant by efciency leadto different models; see [18] for a full discussion. Two ba-sic models are prevalent in the literature: intermediation andproduction; [11]. Specication will refer to a more restrictedconcept: the particular set of inputs and outputs that enterinto model denition.

    Under the intermediation model, nancial institutions col-lect deposits and make loans in order to make a prot. De-posits and acquired loans are considered to be inputs. Insti-tutions are interested in placing loans, which are traditionaloutputs in studies of this kind; see, for example [19]. Underthe production model, a nancial institution uses physicalresources such as labour and plant in order to process trans-actions, take deposits, lend funds, and so on. In the produc-tion model manpower and assets are treated as inputs andtransactions dealt withsuch as deposits and loansaretreated as outputs. See, for example [20,21].

    We notice that the selection of inputs and outputs is deter-mined by our understanding of what a nancial institutiondoes. Deposits provide an extreme example: they are inputsfrom an intermediation point of view, and outputs from aproduction point of view. The specication of what is aninput and what is an output is crucial in the modelling pro-cess. In our particular case we do not need to ponder aboutthe way in which deposits should be treated, since micro-nance institutions do not always collect them, and had tobe excluded as a possible variable in the data set since thetechnique to be applied, DEA, requires homogeneous datafor all the MFIs. Many MFIs obtain funds from the market(loans) or receive grants. Other issues become relevant inthe selection of inputs and outputs. For example, some MFIreceive subsidised loans at an interest rate that is below themarket.

    It follows that the selection of inputs and outputs is crucialin the nancial institution modelling. Berger and Humphrey[10] suggest that one could assess efciency under a vari-ety of output/input specications, and see the way in whichcalculated efciencies change as the specication changes.This is sensible, but they do not provide guidelines on how tochoose between specications. In fact, specication searchesare common in the modelling of nancial institutions; ex-amples are [20,22].

    A major problem with the selection of inputs and outputsin a DEA model is that there is no statistical framework onwhich signicance tests can be based. The neat approachof variable selection that is used in regression, based on t-statistic values, has no parallel in DEA. One may be tempted

  • 134 B. Gutirrez-Nieto et al. / Omega 35 (2007) 131142

    to use as many inputs and outputs as one may think to berelevant, but some of them will be correlated, perhaps highlyso. Parkin and Hollingsworth [23] review the problems thatvariable selection creates in DEA. Jenkins and Anderson[24] warn against the use of correlated inputs and outputsin a DEA model. An important issue is that the number of100% efcient units increases with the number of inputs andoutputs in the model, and adding irrelevant variables maychange the results obtained [25]. Specication search meth-ods in DEA have been proposed by Norman and Stocker[26], Pastor et al. [27], and Serrano-Cinca and Mar Molinero[28].

    Here we will use the model specication methodologysuggested by Serrano-Cinca and Mar Molinero [28]. This,in essence, consists in calculating efciencies for every pos-sible combination of inputs and outputs. A two way tableis obtained in which the columns are output/input speci-cations and the rows are decision units (MFIs). The entriesin the table are the efciencies obtained under each differ-ent model for each MFI. The rows of this table are treatedas cases and the columns as variables in a bivariate statis-tical analysis which throws light on the similarity betweenmodels, extreme observations, and the reasons why a par-ticular MFI achieves a particular level of efciency with aparticular specication. This will be discussed in detail inthe empirical example presented below.

    5. Micronance in Latin America

    Most of the research on banking efciency has concen-trated on US and developed countries. So far, neither DEAnor other parametric or non-parametric frontier techniqueshave been used to evaluate the efciency of micronance in-stitutions. Here we depart from this trend, and analyse thirtyLatin American MFIs from Bolivia, Colombia, DominicanRepublic, Ecuador, Mexico, Nicaragua, Peru and Salvador.Some of them are for prot institutions and others are notprot oriented. Some MFIs are just specialised banking in-stitutions, while others are non-governmental organisations(NGOs). The question arises whether this difference inu-ences efciency, or the way in which efciency is achieved.

    According to Miller [29], some of the most experienced,developed, and diverse MFIs around the world can be foundin LatinAmerica. Using 2001 and 2002 data from 124 world-wide MFIs (provided by the MicroBanking Bulletin), almosthalf of them from Latin America, the author draws severalconclusions: MFIs from this region have more assets, aremore leveraged, and make use of an increasingly growingshare of commercial funds than institutions from other re-gions. Lapenu and Zeller [30] complete this vision: compar-ing African, Asian and Latin America MFIs, they nd thatthe number of institutions and the number of clients remainsmall in Latin American MFIs compared to Asian. How-ever, Latin American MFIs mobilise a good amount of sav-ings and loans in comparison to Asian MFIs. Finally, Latin

    America records the largest volume per transaction althoughrural outreach remains low.

    For the purposes of this paper, data was obtained fromMicrorate web page for the year 2003, and completed withthe Technical Guide prepared by Jansson et al. [31]. All thedata is measured in monetary units (thousand of dollars),except the number of credit ofcers and the number of loansoutstanding.

    6. Selection of inputs and outputs

    The selection of inputs and outputs in the model wasbased onYarons [5] outreach and sustainability framework.The number of loans outstanding (output) and the gross loanportfolio (output) were selected as measures of outreach.The two aspects of sustainability, operational and nancial,guided the selection of a further input and output. Interestand fee income (output) was taken as an indicator of oper-ational sustainability, as a MFI that fails to collect enoughincome is not viable in the long term. Financial sustainabil-ity was captured through operating expenses. In essence, thecollection of fee and interest income is necessary for sur-vival, but such survival cannot be long lasting if this incomeis collected at high cost. In common with other similar stud-ies, the number of credit ofcers was also used as an input.

    The inputs selected in this study are credit ofcers andoperating expenses. A production model would suggest theinclusion of the rst input, while the second input is con-sistent with an intermediation model. Jansson et al. [31] de-ne loan ofcers as personnel whose main activity is di-rect management of a portion of the loan portfolio. Ourchoice of input could have been total staff, but this wouldhave included people whose activity is unrelated to the MFIactivity. The number of employees has been proposed as aninput by Berger and Humphrey [10], Leon [32], and Tortosa-Ausina [33] among others. Operating expensesor similarinputs have been suggested by Berger and Humphrey [10],Pastor [34] and Worthington [35]. Operating expenses areexpenses related to the operation of the institution, includ-ing all the administrative and salary expenses, depreciationand board fees [31].

    The selection of outputs is also consistent with the pro-duction and intermediation models. Interest and fee incomeand the gross loan portfolio are associated with an interme-diation orientation, whereas the number of loans outstand-ing is associated with a production orientation. We wishto emphasize that the gross loan portfolio and the num-ber of loans outstanding appeared as components of MFIefciency ratios in Table 1. Interest and fee incomes areused by Pastor [34]. Gross loan portfolio or similar mea-sures are often mentioned: Berger and Humphrey [10], Leon[32], Tortosa-Ausina [33], and Worthington [35]. Finally,the number of loans outstanding is mentioned by Bergerand Humphrey [10], and Tortosa-Ausina [33]. As there issome difculty in getting data for the number of loans

  • B. Gutirrez-Nieto et al. / Omega 35 (2007) 131142 135

    Table 2Inputs and outputs included in the DEA model, together with their units of measurementa

    Inputs Outputs

    A. Credit ofcers (number) 1. Interest and fee income ($ thousands)B. Operating expenses ($ thousands) 2. Gross loan portfolio ($ thousands)

    3. Number of loans outstanding (number)aSome of the data had to be deduced from the Microrate source as follows:

    A: Credit ofcers.Credit ofcers = Number of clients outstanding/Number of clients per credit ofcer.B: Operating expense.Operating expense = (Total operating expense/average gross portfolio) average gross portfolio.To obtain the average gross portfolio, we take the gross portfolio data from adjusted comparison table 2002 and 2003.Outputs data was directly taken from the adjusted comparison table.

    Table 3List of MFIs and the value of inputs and outputs

    DMU Input A Input B Output 1 Output 2 Output 3(Credit (Operating (Interest and (Gross loan (Number of loansofcers) expenses) fee income) portfolio) outstanding)

    Adopem 92 1483.273 3341 7597 39,717Andes 195 9098.855 16,238 70,058 52,954Bancosol 173 10,816.344 18,082 82,984 41,317Calpia 130 9190.205 12,038 52,550 46,856C-Arequipa 211 10,017.945 26,015 78,985 85,929Cr-Arequipa 67 1157.664 2045 5035 7053C-Cusco 66 3910.601 10,020 34,954 28,506C-Ica 78 2322.093 4470 14,102 18,534Compartamos 525 17,726.376 40,115 48,605 166,580Cona 82 3667.626 8042 18,723 24,320Conanza 23 1201.438 2217 5890 7233C-Sullana 223 5293.925 11,300 31,843 56,343C-Tacna 111 3012.012 6191 18,464 21,327Cr-Tacna 27 818.522 1366 3892 4756C-Trujillo 347 8436.381 16,838 59,047 81,571Diaconia-Frif 38 957.577 1908 7206 15,495D-Miro 20 751.709 1099 2607 8415Edycar 92 5254.613 8862 24,216 25,201Fie 114 3955.857 7967 36,317 28,910Finamerica 72 3040.092 4555 15,414 20,287Fincomun 82 5113.527 4754 6317 11,027Findesa 23 2627.744 5371 12,894 11,243Nieborowski 40 896.714 2792 6449 9619Proempresa 25 1680.174 2931 6491 8031Pro-Mujer 65 1676.766 1762 4682 34,973W-Bogota 39 1355.444 2055 6095 19,466W-Bucaramanga 60 1737.249 3101 8201 37,789W-Cali 118 3121.965 8229 27,423 63,463W-Medellin 39 922.768 1792 4971 17,979W-Popayan 85 1505.178 5454 14,270 61,341

    processed in a given period, we use instead the stock ofloans. Table 2 shows the inputs and outputs included in theDEA model, together with their units of measurement. Table3 gives the values of inputs and outputs for the MFIs in thesample.

    7. Specications and DEA efciencies

    Notation is needed to simplify the discussion of the vari-ous specications. Inputs are referred to by means of capitalletters, in such a way that the rst input (credit ofcers) is

  • 136 B. Gutirrez-Nieto et al. / Omega 35 (2007) 131142

    Table 4The 30 MFIs Efficiency results under the 21 specifications

    DMU A1 A12 A123 A13 A2 A23 A3 AB1 AB12 AB123 AB13 AB2 AB23 AB3 B1 B12 B123 B13 B2 B23 B3

    Adopem 16 16 60 60 15 60 60 62 62 66 66 54 66 66 62 62 66 66 54 66 66Andes 36 64 64 48 64 64 38 66 85 85 66 85 85 38 49 81 81 49 81 81 14Bancosol 45 86 86 47 86 86 33 67 90 90 67 90 90 33 46 81 81 46 81 81 9Calpia 40 72 73 60 72 73 50 55 75 78 60 75 78 50 36 60 60 36 60 60 13C-Arequipa 53 67 76 71 67 76 56 97 97 97 97 87 88 56 72 83 83 72 83 83 21Cr-Arequipa 13 13 18 18 13 18 15 49 49 49 49 46 46 15 49 49 49 49 46 46 15C-Cusco 65 95 95 80 95 95 60 100 100 100 100 100 100 60 71 94 94 71 94 94 18C-Ica 24 32 42 39 32 42 33 64 66 66 64 66 66 33 53 64 64 53 64 64 20Compartamos 33 33 52 52 16 45 44 78 78 78 78 30 45 44 62 62 62 62 29 29 23Cona 42 42 53 53 41 52 41 81 81 81 81 56 57 41 61 61 61 61 54 54 16Conanza 41 46 57 55 46 57 44 70 70 70 70 55 60 44 51 52 52 51 52 52 15C-Sullana 22 26 41 40 26 41 35 66 66 66 66 65 65 35 59 63 63 59 63 63 26C-Tacna 24 30 35 33 30 35 27 66 67 67 66 66 66 27 57 65 65 57 65 65 17Cr-Tacna 22 26 32 31 26 32 25 56 56 56 56 52 52 25 46 50 50 46 50 50 14C-Trujillo 21 30 41 37 30 41 32 62 75 75 62 75 75 32 55 74 74 55 74 74 24Diaconia-Frif 22 34 63 59 34 63 56 63 81 81 63 81 81 56 55 79 79 55 79 79 40D-Miro 24 24 61 61 23 61 58 52 52 61 61 38 61 58 40 40 40 40 37 37 27Edycar 41 47 52 50 47 52 38 65 65 65 65 51 56 38 47 49 49 47 49 49 12Fie 30 57 57 43 57 57 35 70 100 100 70 100 100 35 56 97 97 56 97 97 18Finamerica 27 38 50 45 38 50 39 55 56 56 55 56 56 39 41 53 53 41 53 53 16Fincomun 25 25 26 26 14 22 19 37 37 37 37 14 22 19 26 26 26 26 13 13 5Findesa 100 100 100 100 100 100 68 100 100 100 100 100 100 68 56 56 56 56 52 52 11Nieborowski 30 30 41 41 29 41 33 94 94 94 94 77 77 33 86 86 86 86 76 76 26Proempresa 50 50 59 59 46 58 44 71 71 72 72 48 60 44 48 48 48 48 41 41 12Pro-Mujer 12 13 74 74 13 74 74 33 33 74 74 30 74 74 29 29 51 51 29 51 51W-Bogota 23 28 72 70 28 72 69 53 53 72 70 49 72 69 42 47 47 42 47 47 35W-Bucaramanga 22 24 87 87 24 87 87 59 59 87 87 51 87 87 49 50 53 53 50 53 53W-Cali 30 41 82 78 41 82 74 84 95 95 84 95 95 74 73 93 93 73 93 93 50W-Medellin 20 23 65 65 23 65 64 60 60 65 65 58 65 64 54 57 57 54 57 57 48W-Popayan 28 30 100 100 30 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100

    The column in bold is the specication containing all the inputs and all the outputs.

    represented by the letter A, and the second input (operatingexpenses) by the letter B. Outputs are referred to by meansof numbers. The rst output (interest and fee income) is as-sociated with number 1, the second output (gross loan port-folio) with number 2, and the third output (number of loansoutstanding) with number 3. In this way a specication thattreats a MFI as an institution whose credit ofcers (inputA) take interest and fee income (output 1) and place a num-ber of loans in the market (output 3) would be labeled A13.If this specication is augmented with operating expenses(input B) and gross loan portfolio (output 2), the speci-cation becomes AB123. An intermediation model would bedescribed by a specication such as B2. Under the speci-cation B2, a MFI is an institution that spends money tobuild a loan portfolio. Of course, this is just a performanceindicator, EP1 in Table 1, relating operating expenses togross loan portfolio, contained in the list recommended bythe consensus group of rating agencies, donors, banks, andvoluntary organizations.

    Other views of the way in which a MFI operates can begenerated by using different combinations of inputs and out-

    puts. Efciency ratios are a particular case obtained whenonly one input and only one output enter into the specica-tion. It is, of course, possible to think of all possible combi-nations of inputs and outputs. The total number of possiblespecications with two inputs and three outputs is 21. Thecomplete list of specications can be seen in Table 4.

    DEA efciencies for each MFI were calculated using theCCR model of constant returns to scale [36]. The results aregiven in Table 4.

    Visual examination of Table 4 reveals some important fea-tures. Two MFIs (W-Popayan, an NGO and Findesa, a non-bank nancial institution) are 100% efcient under manyspecications. On the other hand, some MFI achieve lowscores under most specications. No MFI is efcient underall specications, highlighting the fact that the selection ofinputs and outputs and, therefore, the view of what consti-tutes efciency in this sector is a matter of importance. If wetake, for example, W-Popayan, we nd that it is 100% ef-cient under 18 specications, meaning that it is an excellentinstitution, but its efciency drops below 30% under A1,A2 and A12. We conclude that W-Popayan is good in any

  • B. Gutirrez-Nieto et al. / Omega 35 (2007) 131142 137

    specication that contains either input B or output 3, indicat-ing that this MFI is good at generating lots of loans with lowoperating expenses. A counter example is Fie, a non-banknancial institution, whose scores tend to be low, but be-comes 100% efcient under 4 specications: AB12, AB123,AB2, AB23. This indicates that, although Fie can take ac-tion to improve its efciency, it has some strong points thatdeserve further attention.

    In summary, the level of efciency achieved by a particu-lar MFI depends on the specication chosen, indicating thatspecication search is delicate and important. In addition, iftwo MFIs achieve the same efciency score under a givenspecication they may do so following very different pat-terns of behaviour: there is no single path to efciency inMFI. Exploring what is behind a DEA score is the objectiveof the next sections.

    8. Multivariate analysis of DEA efciency results

    Serrano-Cinca and Mar Molinero [28] propose a speci-cation search methodology based on treating the data inTable 4 as a multivariate data set. Other examples of the useof this approach are [37,38]. This involves treating speci-cations as variables and MFIs as cases in a principal com-ponents analysis (PCA). For an account of PCA see, forexample [39].

    The rst principal component, accounting for 57% of thevariance, has an associated eigenvalue of 12.1; the secondcomponent accounts for a further 18% of the variance withan associated eigenvalue of 3.8; the third component, inturn accounts for 15% of the variance with an eigenvalueof 3.1; nally, there is only one more eigenvalue greaterthan 1, at 1.3, accounting for 6.4% of the variance. In total,the rst four principal components account for 97% of thevariance. This suggests that only four numbers (components)are required to explain why a particular MFI achieves acertain level of efciency under all specications.

    Component correlations are shown in Table 5. It can beseen that the rst principal component (PC1) is positivelyand highly correlated with efciency under all specications,suggesting that it provides an overall measure of efciencythat could be seen as an average over all specications. Themeaning of the remaining components could be assessed inthe same way, just looking at the values in the columns inTable 5, but we prefer a more graphical approach to inter-pretation based on component scores. Each MFI is associ-ated with four components, and this forces us to work withprojections on to pairs of components. Component scoresfor each MFI in principal components 1 and 2 can be seenin Fig. 1, and component loadings in principal components2 and 3 can be seen in Fig. 2.

    If we look at Fig. 1 while taking into account the numbersin Table 4, some interesting features appear. W-Popayan,Findesa, C-Cusco, that are efcient under many specica-tions, appear at the right-hand side of the gure. At the other

    Table 5DEA component loadings matrix

    Model PC1 PC2 PC3 PC4

    AB123 0.946 0.041 0.059 0.008AB23 0.914 0.028 0.058 0.316AB12 0.883 0.394 0.038 0.136AB2 0.879 0.352 0.064 0.218AB13 0.854 0.188 0.080 0.396B123 0.843 0.245 0.415 0.163AB1 0.832 0.216 0.031 0.497B12 0.823 0.341 0.407 0.112B23 0.818 0.206 0.377 0.349B2 0.811 0.312 0.361 0.298A23 0.796 0.387 0.413 0.178A123 0.788 0.395 0.426 0.147B13 0.738 0.134 0.521 0.380B1 0.736 0.015 0.515 0.416A13 0.696 0.609 0.361 0.065A2 0.621 0.476 0.599 0.116AB3 0.578 0.800 0.117 0.054A3 0.584 0.793 0.129 0.055B3 0.376 0.775 0.458 0.080A1 0.516 0.323 0.697 0.345A12 0.589 0.490 0.626 0.048


    -4 -2 0 2 4 PC1








    Calpia C-ArequipaCr-Arequipa C-Cusco











    FinamericaFincomun Findesa






























    Global efficiency

    Fig. 1. PC1 versus PC2. Pro-Fit lines.

    extreme of the gure we nd MFIs such as Cr-Arequipaand Fincomun, which achieve low levels of efciency un-der most specications. This is in line with our observationthat the rst principal component provides an overall ratingin terms of efciency. We could approach the understand-ing of the remaining components in a similar vein. For ex-ample, the second component appears to be associated withNGO status, as all the MFIs with a positive score in this

  • 138 B. Gutirrez-Nieto et al. / Omega 35 (2007) 131142

    Fig. 2. PC2 versus PC3. Pro-Fit lines. NGOs are shadowed.

    component are NGOs, and all the MFIs with a negative valueof the component, with the exception of Nieborowski, arenon-NGOs. Towards the top of Fig. 2 we nd MFIs whoseefciency is higher under specications that contain input A(credit ofcers) than under specications that contain inputB (operating expenses). The most extreme example is Find-esa. Findesa is 100% efcient under all models that containinput A, but its efciency drops considerably when this in-put is excluded. This would suggest that the third principalcomponent is associated with the efcient use of input Aversus the efcient use of input B. However, it is dangerousto perform this type of labelling exercise without the help ofa formal tool. In order to interpret the meaning of the com-ponents and in order to highlight the information containedin the gures, we resort to the technique of property tting(Pro-Fit).

    Pro-Fit is a regression-based technique that draws linesin the gures in much the same way in which NorthSouthdirections are drawn in order to orient a geographical map.A particular characteristic of a MFI is taken as a property. Aline is drawn pointing in the direction towards the value ofthe property increases. For example, in Fig. 1, if we calculatethe efciency of the various MFIs under specication B3,we nd that W-Popayan is associated with the highest value,while Fincomun and Bancosol show the lowest values. B3efciency takes intermediate values in the remaining MFIs,increasing as we approach W-Popayan and decreasing aswe approach Bancosol. Thus, a line from the origin towardsW-Popayan, and away from Bancosol, would provide anindication of how B3 efciency changes within Fig. 1. Agood introduction to Pro-Fit can be found in [40]. For someexamples of the use of Pro-Fit within a management sciencecontext see [38,41].

    Pro-Fit lines have been calculated for all the specicationsand displayed in Figs. 1 and 2. Goodness of t statistics

    associated with the Pro-Fit lines is given in Table 6. Figs. 1and 2 will now be interpreted in the light of the informationcontained in the directional vectors.

    The rst principal component has already been identiedas an overall measure of efciency that summarises all themodels. This can be clearly seen in Fig. 1, where all thelines associated with the different specications are at acuteangles with the horizontal axis, indicating positive correla-tion between the value of the rst component score for eachMFI and efciency, in whatever specication efciency ismeasured. In Fig. 1, the label global efciency has beenattached to the rst component.

    The second principal component has been already inter-preted as being related to NGO status, and this is clear inFig. 2 where the shaded area contains all the MFIs withNGO status.

    We observe in Fig. 2 that specications that contain inputA in their denitions are associated with directional vectorsthat point upwards, while specications that contain inputB in their denition are associated with downward pointingdirectional vectors. The third principal component clearlyreects the different strategies followed by MFIs in theirsearch for efciency, opposing those that follow a policyof being efcient in the use of credit ofcerspositivevalues of the third principal componentand those thatfollow a policy of being efcient in their operatingexpensesnegative values of the third principal compo-nent. In Fig. 2 we also see that Findesa can be consideredto be a discordant observation. Indeed, Findesa is an ex-treme case of performance related pay, since 99% of creditofcers salary is due to incentive pay, and this is reectedin our results.

    Principal Component 4 was found to be associated withoutput 2- gross loan portfolio. Specications that containoutput 2 in their denition produce vectors that point to-wards the negative end of the fourth principal component,while specications that exclude this output produce vec-tors that point towards the positive side. This is sending themessage that the inclusion or exclusion of this output affectsefciency values.

    In summary, when describing a MFI from the point ofview of efciency, we need to refer to at least four char-acteristics, or principal components of efciency. The rstprincipal component refers to an overall assessment of ef-ciency under all possible models, and gives a ranking ofMFIs. The second component refers to the NGO status. Thethird principal component is associated with inputs and re-veals which MFIs have an approach to efciency based oncredit ofcers, and which ones approach efciency by con-centrating on operating expenses. The fourth principal com-ponent is associated with the inclusion or exclusion of anoutput in the model: gross loan portfolio.

    Returning to the difference betweenW-Popayan and Find-esa, that was earlier mentioned, we are now in a positionto see in which way these two institutions are different. InFig. 1 we see that both W-Popayan and Findesa are at the

  • B. Gutirrez-Nieto et al. / Omega 35 (2007) 131142 139

    Table 6Pro-Fit analysis. Linear regression results

    Model Directional cosines F Adj R21 2 3 4

    A1 0.09 0.06 0.12 0.06 243.19 0.971(16.30)** (10.20)** (22.01)** (10.91)**

    A12 0.14 0.11 0.15 0.01 330.95 0.978(21.64)** (18.00)** (22.99)** (1.76)

    A123 0.17 0.08 0.09 0.03 307.00 0.977(27.90)** (13.98)** (15.07)** (5.21)**

    A13 0.14 0.12 0.07 0.01 691.59 0.990(36.79)** (32.20)** (19.09)** (3.46)*

    A2 0.15 0.11 0.14 0.03 398.28 0.982(24.98)** (19.15)** (24.09)** (4.68)**

    A23 0.17 0.08 0.09 0.04 432.07 0.983(33.34)** (16.20)** (17.30)** (7.44)**

    A3 0.12 0.16 0.03 0.01 620.98 0.988(29.28)** (39.71)** (6.48)** (2.74)

    AB1 0.15 0.04 0.01 0.09 466.47 0.987(36.18)** (9.39)** (1.34) (21.61)**

    AB12 0.17 0.08 0.01 0.03 132.07 0.948(20.77)** (9.26)** (0.89) (3.20)*

    AB123 0.16 0.01 0.01 0.00 55.93 0.883(14.91)** (0.64) (0.93) (0.13)

    AB13 0.13 0.03 0.01 0.06 80.48 0.916(15.91)** (3.50)* (1.49) (7.37)**

    AB2 0.20 0.08 0.01 0.05 112.06 0.939(19.11)** (7.65)** (1.39) (4.74)**

    AB23 0.17 0.01 0.01 0.06 97.85 0.930(18.65)** (0.58) (1.18) (6.46)**

    AB3 0.12 0.16 0.02 0.01 690.00 0.990(30.52)** (42.22)** (6.18)** (2.87)**

    B1 0.11 0.00 0.08 0.06 307.43 0.977(26.07)** (0.54) (18.24)** (14.74)**

    B12 0.16 0.07 0.08 0.02 211.64 0.967(24.29)** (10.05)** (12.01)** (3.32)*

    B123 0.15 0.04 0.08 0.03 193.16 0.964(23.79)** (6.91)** (11.73)** (4.60)*

    B13 0.11 0.02 0.08 0.06 264.74 0.973(24.28)** (4.40)* (17.14)** (12.50)**

    B2 0.17 0.07 0.08 0.06 244.50 0.971(25.69)** (9.89)** (11.44)** (9.44)**

    B23 0.17 0.04)** 0.08 0.07 258.94 0.973(26.65)** (6.72)** (12.28)** (11.38)**

    B3 0.08 0.16 0.09 0.02 142.26 0.951(9.15)** (18.89)** (11.16)** (1.95)

    Signicant at the 0.05 level;Signicant at the 0.01 level.

    extreme right-hand side of the rst principal component,indicating that both are fully efcient in an overall assess-ment. W-Popayan, is towards the top of this gure, at theextreme of vector A3, indicating that W-Popayan places ahigh number of loans per credit ofcer, while Findesa is atthe extreme of vector B1, indicating that with little oper-ating expenses obtains a great deal of interest and fee in-

    come. But it is in Principal Component 3 where the differ-ence appears most clearly. W-Popayan is at the bottom ofFig. 2 indicating efcient use of operating expenses, whileFindesa is located towards the top of the same gure, in-dicating efcient use of credit ofcers. Both W-Popayanand Findesa achieve similar scores with respect to PrincipalComponent 4.

  • 140 B. Gutirrez-Nieto et al. / Omega 35 (2007) 131142

    9. Non-governmental organisations and country effect

    Two aspects of MFIs will now be examined: their countryof operation, and their non-governmental (NGO) status. Wewill start with the NGO status.

    Given the aims and objectives of MFIsthe ght againstpoverty, self-help, and the promotion of womens statusitis not surprising to discover that many of them are NGOs. Infact, very often an organisation starts as an NGO, and when itbecomes well established in the micronance world, changesinto a non-banking nancial institution. But are NGOs moreor less efcient than non-NGOs MFIs? Is there anything inthe way they achieve efciency that distinguishes them?

    A region has been highlighted in Fig. 2. This region con-tains only NGO institutions and does not contain any insti-tution that is not NGO. MFIs outside this region are all non-NGOs. It is clear that, from the point of view of efciencythere is something that distinguishes a NGO MFI. Lookingfurther into Fig. 2, we see that the Pro-Fit line B3 pointsdirectly towards the cluster of NGO MFIs and away fromthe rest of the MFIs. This suggests that NGOs try to makea large number of loans and operate as cheaply as possible.This is very much in tune with this type of organisation,since they tend to be operated by volunteers to keep costsdown, and aim at supporting as many individuals as possi-ble. The specications that are most in tune with non-NGOinstitutions are A1, A12, andA2. Non-NGOs, therefore, relyon their specialised staff to build a protable portfolio ofloans, very much like commercial banks would do. The dif-ference is not in the way they view the nancial businessbut in their attitude towards obtaining guarantees for theirloans and, indeed, in the average size of loans. It is to benoticed that the most extreme point in the non-NGO regionof Fig. 2 is Bancosol, a commercial bank that is involved inthe micronance business.

    We now turn our attention to the country effect. There isa country effect, best seen in principal component 4. Fig. 3plots component scores in principal component 1 versusprincipal component 4. The names of the MFIs have beenreplaced with the names of the countries in which MFIs op-erate. We can see that there is very little overlap betweenthe countries. From top to bottom, all Nicaraguan MFIs ap-pear together; all but one Peruvian MFIs appear together; allbut one Colombian MFIs appear together; and all BolivianMFIs appear together. Nothing can be said about Salvador,Ecuador, and the Dominican Republic, since these countriesare represented by just one MFI each. There is no right toleft grouping of countries in Fig. 3, indicating that countryof origin and overall efciency are unrelated. Remember-ing that Principal Component 4 is associated with output 2(gross loan portfolio), one would conclude that efciency ofMFIs in Bolivia is associated with building large portfolios,while efciency of MFIs in Nicaragua has to be assessed interms of the number of loans or the amount of interests andfees collected by the MFI. In fact, Bolivia has one of themore developed micronance markets, where margins are


    PC1-4 -2 0 4










    Peru Peru











































    Global efficiency

    Efficiency in building large portfolios

    Fig. 3. PC1 versus PC4 with Pro-Fit lines. Country effect.

    narrowing and this is resulting in mergers and acquisitionswithin the MFI industry [42].

    10. Conclusions

    DEA has long been applied to the measurement of nan-cial institutions efciency. Here we have used it to assessefciency of MFIs, which have a banking side and a so-cial side. We have suggested a methodological approach thatgoes behind a DEA measure and explains the scores ob-tained under different choices of models and specications.

    We have obtained DEA efciencies for every combinationof inputs and outputs of 30 Latin American MFIs. Thisway, we can see that the level of efciency achieved by aMFI depends on the specication chosen. So the choice ofa particular model or specication is relevant for efciencyassessment.

    We have then followed a multivariate approach onefciencies obtained through DEA: we have combinedprincipal component analysis with property tting. Wehave obtained four principal components of efciency, eachone related to a different issue: overall efciency, NGOstatus, input choice and output choice. This way we canunderstand why a MFI achieves a level of efciency undera given specication, or which are the paths to efciencyfollowed by a group of MFIs.

    Finally, there is no reason why we should be fanatic be-lievers in a DEA efciency world, but the converse is alsotrue. Efciency and productivity ratios that have emergedfrom the deliberations of a committee need not be associ-ated with efciency nor with productivity. We have shownthat our approach to efciency analysis not only produces

  • B. Gutirrez-Nieto et al. / Omega 35 (2007) 131142 141

    an overall ranking of MFIs in terms of the use they make ofinputs and outputs, but also reveals features that distinguishNGOs from non-NGO institutions, that we can explain thereasons why some MFIs are or are not efcient, and thatthere are country effects in the data.

    We nish by encouraging analysts, rating agencies, andusers to go beyond ratio analysis in MFIs and incorporatemeasures of efciency based on Data EnvelopmentAnalysis.


    The work reported in this paper was supported by GrantSEJ2004-04748/ECON of the Spanish Ministry of Educa-tion and Science, and the European Regional DevelopmentFund (ERDF) under the title Management Efciency of theSocially Responsible Investment organizations.


    [1] Hollis A, Sweetman A. Microcredit: what can we learn fromthe past? World Development 1998;26(10):187591.

    [2] Morduch J. The micronance promise. Journal of EconomicLiterature 1999;37:1569614.

    [3] Daley-Harris S. State of the Microcredit Summit CampaignReport 2003, Microcredit Summit Campaign, Washington DC,USA, 2003.

    [4] Von Pischke JD. Micronance in developing countries. In:Carr JH, Tong ZY, editors. Replicating micronance in theUnited States. Washington DC, USA: Woodrow Wilson CenterPress; 2002. p. 6596.

    [5] Yaron J. Successful rural nancial institutions. The WorldBank Research Observer 1994;9(1):4970.

    [6] Chaves R, Gonzlez-Vega C. The design of successfulnancial intermediaries: evidence from Indonesia. WorldDevelopment 1996;24(1):6578.

    [7] Johnson S, Rogaly B. Micronance and poverty reduction.Oxford and London, UK: Oxfam and ACTIONAID; 1997.

    [8] CGAP. Micronance consensus guidelines. Denitionsof selected nancial terms, ratios and adjustments formicronance, 3rd ed. Washington DC, USA: ConsultativeGroup to Assist the Poorest; 2003.

    [9] Thanassoulis E. Introduction to the theory and applicationof data envelopment analysis. Dordrecht, The Netherlands:Kluwer Academic Publishers; 2001.

    [10] Berger AN, Humphrey DB. Efciency of nancial institutions:International survey and directions for future research.European Journal of Operational Research 1997;98(2):175212.

    [11] Athanassopoulos AD. Service quality and operating efciencysynergies for management control in the provision of nancialservices: evidence from Greek bank branches. EuropeanJournal of Operational Research 1997;98(2):30013.

    [12] Bala K, Cook D. Performance measurement with classicationinformation: an enhanced additive DEA model. Omega2003;31(6):43950.

    [13] Hartman TE, Storbeck JE, Byrnes P. Allocative efciency inbranch banking. European Journal of Operational Research2001;134:23242.

    [14] Pastor JM, Perez F, Quesada J. Efciency analysis in bankingrms: an international comparison. European Journal ofOperational Research 1997;98(2):395407.

    [15] Seiford LM, Zhu J. Protability and marketability ofthe top 55 U.S. commercial banks. Management Science1999;45(9):127088.

    [16] Charnes A, Cooper WW, Lewin YA, Seiford ML. Dataenvelopment analysis: theory, methodology, and applications.Dordrecht, The Netherlands: Kluwer Academic Publishers;1994.

    [17] Cooper WW, Seiford LM, Tone K. Data envelopment analysis:a comprehensive text with models, applications, references andDEA-solver software. Dordrecht, The Netherlands: KluwerAcademic Publishers; 2000.

    [18] Berger AN, Mester LJ. Inside the black box: what explainsdifferences in the efciencies of nancial institutions? Journalof Banking and Finance 1997;21:895947.

    [19] Berger AN, Humphrey DB. The dominance of inefcienciesover scale and product mix economies in banking. Journal ofMonetary Economics 1991;28:11748.

    [20] Vassiloglou M, Giokas D. A study of the relative efciency ofbank branches: an application of data envelopment analysis.The Journal of the Operational Research Society 1990;41:5917.

    [21] Soteriou A, Zenios SA. Operations, quality and protabilityin the provision of banking services. Management Science1999;45(9):122138.

    [22] Lovell CAK, Pastor JM. Target setting in a bankbranch network. European Journal of Operational Research1997;98(2):2909.

    [23] Parkin D, Hollingsworth B. Measuring production efciencyof acute hospitals in Scotland, 199194: validity issues in dataenvelopment analysis. Applied Economics 1997;29:142533.

    [24] Jenkins L, Anderson M. A multivariate statistical approach toreducing the number of variables in data envelopment analysis.European Journal of Operational Research 2003;147:5161.

    [25] Pedraja Chaparro F, Salinas Jimenez J, Smith P. On thequality of data envelopment analysis model. Journal of theOperational Research Society 1999;50:63645.

    [26] Norman M, Stocker B. Data envelopment analysis: theassessment of performance. Chichester, UK: Wiley; 1991.

    [27] Pastor JT, Ruiz Gomez JL, Sirvent I. A statisticaltest for nested radial DEA models. Operations Research2002;50(4):72835.

    [28] Serrano-Cinca C, Mar Molinero C. Selecting DEAspecications and ranking units via PCA. Journal of theOperational Research Society 2004;55:5218.

    [29] Miller J. Benchmarking Latin America micronance [on line]micronancegateway.org http://www.micronancegateway.org/content/article/detail/3820 (16/01/04), 2003.

    [30] Lapenu C, Zeller M. Distribution, growth, and performanceof the micronance institutions in Africa, Asia and LatinAmerica: a recent inventory. Savings and Development2002;36(1):87111.

    [31] Jansson T, von Stauffenberg D, Kenyon N, Barluenga-BadiolaMC. Performance indicators for micronance institutions.

  • 142 B. Gutirrez-Nieto et al. / Omega 35 (2007) 131142

    Technical guide. 3rd ed, Washington DC, USA: Microrate andInter-American Development Bank; 2003.

    [32] Leon J. Eciencia en costos en los bancos comerciales deMxico: una aplicacin de la aproximacin no paramtricaDEA. Unpublished paper presented at the Latin AmericanEconometric Association Meeting, Cancun, Mxico, 1999.

    [33] Tortosa-Ausina E. Exploring efciency differences overtime in the Spanish banking industry. European Journal ofOperational Research 2002;139:64364.

    [34] Pastor JM. Efciency and risk management in Spanishbanking: a method to decompose risk. Applied FinancialEconomics 1999;9(4):37184.

    [35] Worthington A. The determinants of non-bank nancialinstitution efciency: a stochastic cost frontier approach.Applied Financial Economics 1998;8(3):27987.

    [36] Charnes A, Cooper WW, Rhodes E. Measuring the efciencyof decision making units. European Journal of OperationalResearch 1978;2:42944.

    [37] Serrano-Cinca C, Mar Molinero C, Chaparro F. Spanishsavings banks: a view on intangibles. Knowledge ManagementResearch and Practice 2004;2(2):10317.

    [38] Serrano-Cinca C, Fuertes Callen Y, Mar Molinero C.Measuring DEA efciency in Internet companies. DecisionSupport Systems 2005;38(4):55773.

    [39] Chateld C, Collins AJ. Introduction to multivariate analysis.London, UK: Chapman & Hall; 1980.

    [40] Schiffman SS, Reynolds ML, Young FW. Introduction tomultidimensional scaling: theory, methods and applications.London, UK: Academic Press; 1981.

    [41] Mar Molinero C, Serrano-Cinca C. Bank failure: amultidimensional scaling approach. European Journal ofFinance 2001;7:16583.

    [42] Silva S. Micronance institutions win a coveted seal ofapproval. Microenterprise Americas 2003; 127.

    Microfinance institutions and efficiency62626262IntroductionMicrocredit and microfinance institutionsAssessing microcredit institutionsDEA efficiency and financial institutionsMicrofinance in Latin AmericaSelection of inputs and outputsSpecifications and DEA efficienciesMultivariate analysis of DEA efficiency resultsNon-governmental organisations and country effectConclusionsAcknowledgementsReferences


View more >