The welfare impact of microcredit on rural households in China

Download The welfare impact of microcredit on rural households in China

Post on 28-Oct-2016




5 download


  • The Journal of Socio-Economics 40 (2011) 404411

    Contents lists available at ScienceDirect

    The Journal of Socio-Economics

    journa l homepage: www.e lsev ier .com/ locate /soceco

    The welfare impact of microcredit on rural house

    Xia Li1, CFaculty of Com ity, Ca

    a r t i c l

    Article history:Received 20 AReceived in reAccepted 6 Ap

    JEL classicatioO10O12O17

    Keywords:MicrocreditHousehold weDifference-in-Rural China


    poverty reduction) of Chinas microcredit programmes. 2011 Elsevier Inc. All rights reserved.

    1. Introduction

    Like mospopulationmajor challsupport hasfarmers proing their livColeman, 1nancial insthe non-secoffered as coseek for infneeds, whicever, the exfarmers intrapped in p

    The failuers and theproviding c

    CorresponE-mail add

    Christopher.G1 Tel.: +64 3

    ments in many low-income countries (LICs) including China to

    1053-5357/$ doi:10.1016/j.t Asian developing countries, the majority of the poorin China dwell in rural areas and poverty remains oneenge to the society. Inability to acquire formal creditbeen often argued as a crucial constraint in expandingduction, which largely restrains farmers from improv-ing conditions in China (e.g., Gale and Collender, 2006;999). The lending terms and conditions set by formaltitutions usually exclude the poor farmerswho farmonured lands and possess little tangible assets that can bellateral for formal loans. As analternative, poor farmersormal loans to meet their consumption or productionh are typically offered at higher interest rates. How-

    ploitive interest rate of informal loans have exacerbateddebtedness and further kept most of the of formal nancial institutions to serve small farm-great success of the Grameen Bank in Bangladesh inredit service to rural poor have inspired the govern-

    ding author. Tel.: +64 3 25 2811; fax: +64 3 325 3847.resses: (X. Li), (C. Gan), (B. Hu).25 2811; fax: +64 3 325 3847.

    adopt microcredit mechanism to deliver credit at reasonable coststo rural people. Microcredit was introduced into China in the mid-1990s as part of the governments poverty alleviation strategies inthe mid-1990s, attempting to ameliorate rural poverty through anancially sustainable approach. Various organisations have beeninvolved in the delivery of microcredit service in China, includingnon-governmental organisations (NGOs), governmental agencies,and rural nancial institutions such as the Rural Credit Cooperative(RCC).2 Since implementing microcredit programmes in 2000, theRCC has expanded its microcredit activity with an extensive net-work in rural areas and evolved as the largest microcredit providerin China (Du, 2004; Sun, 2003).

    However, unlike other countries such as Bangladesh, where themicrocredits potential in reducing poverty has been thoroughlyexamined (see for example, Khandker, 2005; Morduch, 1998; Pittand Khandker, 1998), few attempts have been made in China totest the efciency ofmicrocredit as a poverty reduction instrument.Studies on Chinas microcredit tend to elaborate the developmentand regulations of microcredit programmes, giving little quanti-tative information on the outcomes of programme participation.

    2 The RCC programme outperforms both the NGO and government programmesin terms of nancial sustainability and programme replicability. For details see Parkand Ren (2001) and Zuo (2001).

    see front matter 2011 Elsevier Inc. All rights reserved.socec.2011.04.012hristopher Gan , Baiding Hu1

    merce, Department of Accounting, Economics and Finance, P.O. Box 84, Lincoln Univers

    e i n f o

    pril 2010vised form 28 February 2011ril 2011



    a b s t r a c t

    Microcredit has gained worldwide accuals (especially the poors) access to poverty reduction and other social dto examine the welfare effects of micin many other countries such as Bangholds livelihood are not well docummicrocredit on household welfare oumation is based on the difference-intackling the selection bias issue in assdataset, including both primary and sOur empirical results favour the wideimprove households welfare such asmicrocredit has changed the rural houof the programme participants are noholds in China

    nterbury, New Zealand

    ce in recent years as a exible mechanism to expand individ-ial services, which is considered as an efcient way to achievepment. A large number of empirical studies have been donedit on the borrowers and such effects are well documentedh. However, the impacts of microcredit on China rural house-. This paper attempts to empirically evaluate the impact ofes such as income and consumption in rural China. The esti-rence approach which is an increasingly popular method ofg the impacts of microcredit. The study uses a two-year panelary data collected through a household survey in rural China.f in the literature that joining microcredit programme helpse and consumption. Despite the optimistic ndings on how

    lds living conditions, our results show that the vast majorityor, which casts some doubts on the social potential (such as

  • X. Li et al. / The Journal of Socio-Economics 40 (2011) 404411 405

    As a result, the impacts of microcredit on China rural householdslivelihood are not well documented.

    Since the lack of credit is regarded as the crucial constraint inimproving the Chinese rural households livelihoods, it is hypoth-esised that microcredit, which targets the rural households for theextension of credit facilities have a positive impact on householdswelfare such as increasing the households income and/or con-sumption. This paper aims to empirically examine this hypothesisby focusing on the microcredit programme operated by the RCC.The rest of the paper is organised as follows: Section 2 discusses theresearchmeempirical the paper.

    2. Method

    2.1. Identifyanalytical fr

    Assessinparing outca householdwhen he/shindicator ofhouseholdthe value ofin the progoutcome wprogrammeby i, can b(Sarangi, 20

    i = Yi1 YTwo majorprogrammeing data probserved insame timeadit programwords, oneEq. (1) ismiovercome tof treatmendata on indular groupwhich meaoutcome ofhave experiPerry and Mtrue prografollowing e

    = E(Yi1|wwhere E()E(Yi0|wi = 1ipants hadHeckman, 1

    This, hoE(Yi1|wi = 1

    3 There aretreatment effenon-treated, wparticipants ifdetails.

    1) cannot. This problem is addressed by constructing counterfac-tuals based on a treatment/control framework, where a group ofprogramme non-participants are selected as a control group andthe observed outcomes of this control group are supposed to serveas counterfipants (treatreatment/c() and the

    = E(Yi1|w

    * ioldspatese in

    gates thringptio



    well incotota

    pass, nonptio

    ion aviework aon bhe imtiongrameasutcomrrects leaentoldsrved

    an, 1ringrrowgenes)wtire cto t

    trolleadoptheferenr tecmmehe Dgatest groho

    mmeof thcalleeive

    Bertrthodology, followed by data collection in Section 3. Thendings are discussed in Section 4. Section 5 concludes

    ology to impact assessment

    ing the impact of microcredit programme amework

    g the impact of microcredit programmes requires com-omes (e.g., household income and consumption) whenparticipates in the programme to the same outcomes

    e does not participate. For example, let wi be a binaryprogrammeparticipation equal one for participationbyi and zero for non-participation. Further let Yi1 denotethe outcome of interest when household i participates

    ramme and Yi0 denote the potential value of the samehen it is in the state of non-participation. Thus the trueimpact on the outcome of household i, representede quantied by the difference between Yi1 and Yi0, as07; Perry and Maloney, 2007):

    i0 (1)

    problems should be addressed to identify the trueimpact: missing data and unobservability. The miss-

    oblem arises because the same household cannot beboth participation and non-participation states at thend thus the true impact of participation in themicrocre-me on a certain outcome cannot be observed. In otheror the other component of the difference expressed inssing (Heckman, 1997;RosenbaumandRubin, 1983). Tohis problem, group statistics such as the average effectt on the treated (ATT) is adopted to replace the missingividual subject (Heckman, 1997). ATT is the most pop-statistics widely used in impact evaluation literature,sures the extent to which the programme change thea group of participants compared to what they wouldenced in the absence of participation3 (see for example,aloney, 2007; Nguyen, 2007; Nguyen et al., 2007). The

    mme impact measured by ATT can be expressed by thequation:

    i = 1) E(Yi0|wi = 1) (2)signies expectation in the population. Specically,) represents the counterfactual outcome for partic-they not participated (Dehejia and Wahba, 2002;997).wever, gives rise to the problem of unobservability,) can be estimated while the counterfactual E(Yi0|wi =

    other statistics used in impact assessment, such as local averagect, marginal treatment effect, or the effect of non-treatment on thehich measures the impact the programme would have on the non-they had participated in the programme. See Heckman (1997) for

    wherehousehparticioutcominvestiassessecompaconsumand no

    2.2. Em

    Theannuato theencomcultureconsumsumpt

    Areical wselectiity of tevaluadit proto unmthe ounot cosourceplacemhousehunobseColemcompanon-boheteroteristicthe encomesuncon

    Webias inthe difpopulaprogradata. TinvestiThe rholds wprogratiationgroup,not rec2006;actuals to theobservedoutcomesof programmepartic-tment group). Accordingly, the ATT measured with thisontrol framework is used to estimate the true impactbasic idea of this approach is described as follows:

    i = 1) E(Yj0|wj = 0) (i /= j N) (3)

    s the estimation of , i, and j denote two differentin a chosen sample of N households where household iin theprogrammewhile household jdoesnot;Yi1 is thevestigated of household i and Yj0 is the same outcomed of household j (Sarangi, 2007). Specically, the papere average welfare impact of microcredit programme bythe average household outcomes (such as income andn) between borrowing households (treatment group)rrowing households (control group).

    cal model and estimation strategy

    fare indicators used in this paper include householdme and annual consumption. Household income refersl income earned by all household members, whiches the income from all possible sources, such as agri--agriculture, self-employment, and wages. Householdn in this paper is measured by the sum of food con-nd non-food consumption within a household.of the impact evaluation literature suggests that empir-ssessing microcredits impact must carefully addressias issue, which otherwise would reduce the reliabil-pact estimation. Selection bias in microcredit impact

    arises when the households participation in microcre-me or households receipt of microcredit is relatedred (or unobserved) factors that simultaneously affectes of their credit use but these unobserved factors arely accounted for in the impact assessment. Two mainding to such selection bias are non-random programmewhich is caused by unmeasured village attributes and self-selection into the programme which stems fromhousehold characteristics (for details see Islam, 2007;

    999; Pitt and Khandker, 1998). It can be argued thatwelfare outcomes between borrowing households anding households without accounting for unobservedity (such as unobserved household and village charac-ould produce biased results since it will wrongly ascribehange (improvement or deterioration) in welfare out-he programme impact, which partly arises from thed and unobserved factors.t panel data model to mitigate the potential selection

    impact estimation. The empirical analysis is built uponce-in-difference (DD) method, which is an increasinglyhnique in economics for identifying causal effects ofs or treatments in the absence of pure experimentalD estimation framework requires that the outcomes

    d be observed for two groups over two time periods.up, referred as the borrowing group, consists of house-receive microcredit in the period after the start of the(i.e., post-programme period) but not prior to the ini-e programme (i.e., pre-programme period); the secondd non-borrowing group comprises households who domicrocredit during either period (Athey and Imbens,and et al., 2004).

  • 406 X. Li et al. / The Journal of Socio-Economics 40 (2011) 404411

    The standard DD method can be illustrated by the followingregression equation (Angrist and Krueger, 1999; Meyer, 1995):

    Yit = 0 + 0d2t + 1Pi + Mit + it (4)where Yit is the natural logarithm of household outcome inves-tigated (i.e., household annual income or consumption) forhousehold i at period t; d2t is a time dummy variable equal to 1for t=2 (poperiod); PihouseholdMit is an intcates the prborrowed m(i.e., particitures time igroup; 1 call averagesof interest ming (treatmindependenwith mean

    Thekeyamon trendthe programin the outcodifferent bethere werethis assumpthe followin

    sdd = YBE(

    where the standaracross housborrowingaverage diffdifferencesstrategy enarising fromin post-pronent differeunrelated tisons over tbe the resuprogrammeand Card, 1

    The stanmicrocreditditional onThe randomthat mighttime are banon-experihouseholdsanced in thto the outcthe outcomhouseholdsestimate of2005a,b).

    4 Pre-prograteristics of hou

    Table 1Description of variables used in welfare impact analyses.

    Variables Type of variables Description of variables




    imprd to eobsel varioupsunm

    d in soldc comsele

    per leobseme. Ationtrate

    0 +

    Yit,housatmet capusehold

    hat rme ae methe

    me, is, 20elp cof th;Meyfferereditgramme in period t=2; and (2) a binary variable equal tohousehold i receives loans from the programme in periodd equal to zero otherwise. Compared to the binary treat-ariable, the cumulative borrowing (continuous) variable isften used in assessing the impact of microcredit since itrepresents the extent of a households programme partic-(see for example, Nguyen et al., 2007; Nino-Zaraza, 2007;

    ker, 2005; Alexander, 2001). The parameter thus measuresrageprogramme impactwhenMit is a binary treatment vari-d measures the average programme impact of additional

    amount on borrowing households where Mit represents thet of loans borrowed by the household (Nguyen et al., 2007;der, 2001). The variables used in thewelfare impact analysesed in Table period) and 0 for t=1 (pre-programmeis a group dummy variable and takes a value of one ifi belongs to the borrowing group and zero otherwise;eraction term of the product of d2t and Pi, which indi-ogramme participation and is equal to 1 if household ioney and the observation occurs in the second period

    pating in the programme), and zero otherwise; 0 cap-nuence suffered by both treatment group and controlaptures the potential time-invariant difference in over-between the two groups; is the primary parametereasuring the average programme impact on borrow-

    ent) group; it is the idiosyncratic error assumed to bet and identically distributed over households and time,...


View more >