microfinance and poverty—a macro perspective

World Development Vol. 40, No. 8, pp. 1675–1689, 2012� 2012 Elsevier Ltd. All rights reserved.

0305-750X/$ - see front matter

www.elsevier.com/locate/worlddevhttp://dx.doi.org/10.1016/j.worlddev.2012.04.013

Microfinance and Poverty—A Macro Perspective

KATSUSHI S. IMAIUniversity of Manchester, UK

RAGHAV GAIHAMassachusetts Institute of Technology, USA

University of Delhi, India

GANESH THAPAInternational Fund for Agricultural Development, Italy

and

SAMUEL KOBINA ANNIM *

University of Cape Coast, GhanaUniversity of Central Lancashire, UK

Summary. — We test the hypothesis that microfinance reduces poverty at the macro level using cross-country and panel data which areconstructed by the Microfinance Information Exchange data on Microfinance Institutions (MFIs) and the World Bank data. Takingaccount of the endogeneity associated with MFIs’ loans, we show that a country with higher MFIs’ gross loan portfolio per capita tendsto have lower levels of poverty indices. Contrary to recent micro evidence, our results suggest that microfinance significantly reducespoverty at macro level and thus reinforce the case for channeling funds from development finance institutions and governments of devel-oping countries into MFIs.� 2012 Elsevier Ltd. All rights reserved.

Key words — microfinance, poverty, loan portfolio, macro and global

* This study is funded by IFAD (International Fund for Agricultural D-

evelopment). We are grateful to Thomas Elhaut, Director of Asia and the

Pacific Division, IFAD, for his support and guidance throughout this

study. The first author thanks generous research support from RIEB,

Kobe University, during his stay in 2010, and valuable comments from

Shoji Nishijima, Takahiro Sato, and seminar participants at Kobe Univ-

ersity. The second author would like to thank Bish Sanyal for the invit-

ation to work at MIT’s Department of Urban Studies, where a first draft

was prepared. We have also benefited from the comments of Thankom

Arun and M.D. Azam. The views expressed are, however, personal and

not necessarily of the organizations to which we are affiliated. Final rev-

1. INTRODUCTION

Most of the recent studies of the impact of microfinance onpoverty or income have relied on micro-level evidence basedon household data or entrepreneurial data (e.g., Hulme &Mosley, 1996; Imai, Arun, & Annim, 2010a, 2010b; Khand-ker, 2005; Mosley, 2001). Due to the scarcity of reliable macrodata on microfinance, macro-level studies of the impact ofmicrofinance on poverty are rather limited. However, thereare a few recent works that investigate the relationship be-tween the macro economy and microfinance activities and/orperformance, such as Ahlin, Lin, and Maio (2011), Ahlinand Lin (2006) and Kai and Hamori (2009), among others.The thrust of these studies is either to examine the environ-mental context in which microfinance operates, or investigatethe potential effect of microfinance on key macroeconomicvariables, such as gross domestic product or inequality. Thefindings of a significant relationship between operations ofMicrofinance Institutions (MFIs) and the macro economy cor-roborate the recent evidence based on household data setswhich posits that microfinance has a poverty reducing effect(e.g., Gaiha & Nandhi, 2009; Imai et al., 2010a, 2010b;Khandker, 2005).

This study redirects the attention to macro studies given themixed results of microfinance impact studies at the micro levelin recent years. As the separation of causal effects of micro-credit from selection effects is unsatisfactory in many of themicro-level studies, Armendariz and Morduch (2005) pointedto a potential bias arising in the impact of microfinance in

1675

these studies. In view of this, studies that have recentlyemerged have used one of the following three approaches: (i)randomized control trials (Banerjee, Duflo, Glennerster, &Kinnan, 2009; Feigenberg, Field, & Pande, 2010; Karlan &Zinman, 2010); (ii) financial diaries/portfolios of the poor(Collins, Morduch, Rutherford, & Ruthven, 2009) and (iii)use of other variants of quasi-experimental estimation tech-niques such as treatments effect and propensity score matchingin both cross-sectional and panel data setting (Imai et al.,2010a, 2010b). Evidence from such studies remains mixeddue to different microfinance outcome measures and/or differ-ent methodologies adopted by these studies, leading to the per-ception that microfinance is likely to have little impact onpoverty. 1 Our econometric analysis points to robust povertyreducing effects of microfinance, as elaborated below.

ision accepted: February 29, 2012.

http://dx.doi.org/10.1016/j.worlddev.2012.04.013

1676 WORLD DEVELOPMENT

The challenges for empirical macro studies of microfinanceinclude: (a) identifying an appropriate measure of microfi-nance activities, in terms of “availability” or “intensity”; (b)identifying the effects of “performance,” distinguished from“presence” and “scale” of microfinance on macro indicators;and (c) examining the robustness of coefficient estimates re-lated to microfinance. Building on the small but emerging lit-erature on analyzing the impacts of microfinance from amacro perspective, the present study aims to examine the rela-tionship between MFI’s gross loan portfolio per capita andFGT class of poverty indices. 2 The results would be usefulfor development agencies, governments, and other investors,as there are important implications for microfinance’s poten-tial role in reducing poverty at macro level. Our counterfactualsimulations illustrate the possible effects on aggregate povertyexpected from the decrease in MFIs’ gross loan portfolio percapita, GDP per capita, or domestic credit which may becaused as a consequence of global recession or financial crisis.

Drawing upon econometric estimations of the cross-countrydata—including a panel—we find consistently that a countrywith higher MFIs’ gross loan portfolio per capita tends tohave lower levels of FGT class of poverty indices, which cor-roborates the poverty reducing role of microfinance. It is nota-ble that microfinance loans per capita are negativelyassociated with not only the poverty headcount ratio, but alsowith the poverty gap and squared poverty gap, implying thateven the poorest benefit from them.

The rest of the paper is organized as follows. The next sec-tion provides a brief explanation of the data which the presentstudy draws upon. Econometric specifications are discussed inSection 3. The main results and simulations are given in Sec-tions 4 and 5, respectively. The final section offers concludingobservations.

2. DATA

The present study analyzes the role of microfinance—vol-ume/scale of activities (not performance/quality)—on poverty,using cross-sectional data covering 48 countries in the devel-oping regions for 2007. 3 The cross-sectional data are supple-mented by a two-period (2003 and 2007) panel covering 61countries. 4 This is based on the data generated by Microfi-nance Information Exchange (2010) or MIX and the WorldDevelopment Indicators 2011 (World Bank, 2011). Launchedin 2002, MIX provides industry, country, and regional leveldata on microfinance outreach and financial performance indi-cators. The information marketplace collates reports of (self-reporting) MFIs annually, based on: (i) pre-determined for-mats and reporting standards; (ii) validation of informationreceived with both internal and external cross checking sys-tems; and (iii) standardization of data to facilitate comparison.Interpreting and further processing the data, however, requirea great deal of caution because (i) MIX may not be able tosend questionnaires to all the MFIs in the country—in partic-ular if MFIs are small and recent, (ii) not all the MFIs sur-veyed by MIX respond (sample selection bias), and (iii)“self-reporting” may be a source of measurement errors evenif MIX imposes the careful cross-checking systems and thereliability of the data is ranked by MIX (Ahlin et al., 2011).It is not possible to know the extent of the errors arising fromthese factors, but the MIX data provide the most comprehen-sive and largest data on microfinance activities (Cull, Demi-rguc�-Kunt, & Morduch, 2011) and the data cover a largefraction of microfinance customers worldwide (Cull, Demi-rguc�-Kunt, & Morduch, 2007). 5

To cross-check the potential bias arising from self-selection,the current study estimates variants of econometric modelswith different sub-samples, based on several criteria such as ex-tent of validity of data submitted by MFIs (see Ahlin et al.,2011 for details). We have found that the results are broadlysimilar and consistent irrespective of which sub-samples are in-cluded. Also, we have compared the gross loan portfolio ofMFI per capita (based on the MIX data)—a dependent vari-able in our models—with three variables on microfinanceactivity at country levels available from WDI 2011; (a)branches, microfinance institutions (per 100,000 adults); (b)deposit accounts, microfinance institutions (per 1000 adults);and (c) loan accounts, microfinance institutions (per 1000adults). We have consistently found that a pair-wise correla-tion is positive and significant for all the three variables 6

and hence surmise that the MIX data aggregated at countrylevels represent actual performance of MFIs.

It is noted that relatively few studies have used a measure ofmicrofinance operations (volume/scale) in a country based onthe MIX data. Also, the present study uses the World Bankpoverty estimates, released in 2008 (Chen & Ravallion, 2008;Ravallion, Chen, & Sangraula, 2008). These poverty estimatesare based on the poverty line of US$1.25 (based on PPP—Pur-chasing Power Parity) per day in 2005, and cover a widerrange of countries than the previous estimates (based on apoverty line of $1.08 on 1993 PPP). While there are manystudies based on the latter, those based on the more recentpoverty estimates are still few. 7 Also, as noted earlier, weuse the FGT class of poverty indices.

With a view to measuring microfinance activities in a coun-try, we rely mainly on Gross Loan Portfolio (GLP) (divided bythe total population) given that it measures actual funds dis-bursed to households. Total GLP of MFIs aggregated for eachcountry is adjusted for write-offs and inflation. This is a bench-mark indicator generated by MIX. Standardization of rawdata facilitates meaningful comparison of benchmark indica-tors (MIE, 2010). Other variables in the poverty equation in-clude gross domestic product per capita, domestic credit as ashare of GDP, and regional dummies. 8 While a robust inverserelationship between poverty and GDP per capita is confirmedin extant literature, share of domestic credit in GDP has amore complex role partly because financial development isboth a cause and result of growth. It is, however, plausiblethat when financial development is low there may be a mutu-ally reinforcing relationship between financial developmentand microfinance. Finally, as poverty is conditioned on manyunobservable regional characteristics (e.g., vulnerability tonatural shocks), regional dummies are used.

3. SPECIFICATION OF MODELS AND ESTIMATION

Our analysis is based on the data for 2007 (for cross-sec-tional estimations), and 2003 and 2007 (for panel data estima-tions), not only because extensive and reliable historical dataon microfinance do not exist 9 but also because internationalpoverty estimates are available only for one or two specificyears for most of the countries. 10 As a result, country paneldata of poverty are unbalanced, as shown in Appendix 4.

We apply both OLS (Ordinary Least Squares) and IV(Instrumental Variable) model or 2SLS (Two Stage LeastSquares) to estimate the effect of gross loan portfolio per capi-ta of MFIs on poverty. 2SLS involves two stages: gross loanportfolio per capita of MFIs is estimated by an instrumentalvariable and other covariates in the first stage and in the sec-ond poverty head count ratio is estimated by the predicted

Figure 2. Trends and Patterns of Number of Active Borrower. Source:

Authors’ compilation from MIX Data.

MICROFINANCE AND POVERTY—A MACRO PERSPECTIVE 1677

gross loan portfolio per capita and other covariates. The useof IV is necessary because gross loan portfolio per capita ofmicrofinance is likely to be endogenous in the poverty equa-tion. Here the endogeneity is associated with the bi-casualrelationship between gross loan portfolio per capita and pov-erty levels in a country. This reverse causality from poverty togross loan portfolio per capita may arise, for example, if pov-erty-oriented development partners and governments providemore funds to MFIs located in poorer countries.

Given the difficulty in finding a valid instrument that satis-fies “an exclusion restriction,” that is, correlates with grossloan portfolio per capita but does not have a direct causal ef-fect on poverty, this papers uses two kinds of instrument, costof enforcing contract and a lag of 5-year average of gross loanportfolio weighted by the number of MFIs for each coun-try. 11 The unit of analysis for the econometric analysis iscountry.

Eqns. (1) and (2) describe, respectively, the structural andreduced form of least squares used in estimating the relation-ship between gross loan portfolio per capita and poverty.

Povi ¼ b0 þ b1GLP i þ b2GDPPCi þ b3Domestic Crdi

þ b4REGi þ ui ð1Þ

GLP i ¼ p0 þ p1enfconti þ p2lag5yrsGLP l þ p3Xi þ ti ð2Þwhere “Pov” indicates poverty head count ratio (or povertygap; squared poverty gap); “GLP” represents gross loan port-folio; “GDPPC” denotes gross domestic product per capita (at2000 constant USD prices); “Domcred” indicates domesticcredit of banks as a proportion of GDP; “REG” is a vectorof regional dummies with Latin America and Caribbean beingthe reference region; Eqn. (2) is the reduced form which teststhe presence of endogeneity and suitability of our instruments.“enfcont” represents cost of enforcing contracts at the countrylevel and “lag5yrsGLP” is the weighted 5-year average lag ofgross loan portfolio (which is weighted by the number of MFIs

Sub-

1500000

2000000

2500000

3000000

3500000

2005 2006 2007 2008 2009

East Asian and Pacific

1000000

2000000

3000000

4000000

2005 2006 2

Eastern Europe a

4000000

6000000

8000000

10000000

12000000

14000000

2005 2006 2007 2008 2009

Middle East and North Africa

0

2000000

4000000

6000000

8000000

2005 2006

Y

Sou

Figure 1. Trends and Patterns of Real Gross Loan Portf

for each country) and X is the vector of all the other explana-tory variables considered in Eqn. (1). The respective indepen-dently and identically distributed (i.i.d.) error terms for thetwo equations are denoted by “u” and “t.”

In addition to the cross-sectional estimations, we generatepanel data for 2003 and 2007 for all the variables 12 and esti-mate linear panel models. This construction enables us toexamine the robustness of our coefficients as the panel dataestimation takes account of changes of variables over timeand unobservable country or regional-level effects. Our aimis to further examine the hypothesis that higher gross loanportfolio per capita leads to poverty reduction at macro level.

One of the important limitations pointed out by micro-stud-ies is that microfinance or microcredit does not necessarilyreach the poorest of the poor (e.g., Morduch, 1999). To fur-ther investigate this from the macro perspective, we examinethe effects of gross loan portfolio per capita on poverty gap(which measures depth of poverty) and squared poverty gap(which measures severity of poverty).

007 2008 2009

nd Central Asia

3000000

4000000

5000000

6000000

7000000

2005 2006 2007 2008 2009

Latin America and the Caribbean

2007 2008 2009

ear

th Asia

1000000

2000000

3000000

4000000

2005 2006 2007 2008 2009

Sub-Saharan Africa

olio. Source: Authors’ compilation from MIX Data.

Table 1. Descriptive statistics of poverty headcount correlates

Regions Povertyheadcount

Poverty gap Squaredpoverty gap

Gross loanper capita

No. of MFIs Domesticcredit

Gross domesticproduct per capita

2003 2007 2003 2007 2003 2007 2003 2007 2003 2007 2003 2007 2003 2007

EAP No. 3 4 3 4 3 4 4 5 4 5 5 5 5 5Mean 37.46 24.29 10.66 5.83 115.7 37.13 5.1 12.11 10.5 16.8 53.61 57.64 652.33 880.63Median 40.05 24.12 11.2 5.34 125.44 28.7 6.29 0.03 8 15 49.2 40.6 473.42 617.12

ECA No. 19 17 19 17 19 17 19 20 19 20 18 19 19 19Mean 11.13 5.97 2.98 1.51 24.05 5.55 10.43 56.32 7.63 11.5 26.15 37.67 1693.21 2212.38Median 3.09 2 0.66 0.5 0.44 0.25 3.9 24.09 7 7.5 25.16 35.02 1389.97 1945.64

LAC No. 15 14 15 14 15 14 14 15 14 15 15 15 15 15Mean 14.48 8.85 5.8 3.04 74.63 14.57 7.76 40.94 10.14 21.2 47.78 48.86 2638.87 3110.35Median 13.73 7.97 4.77 2.54 22.75 6.55 5.55 24.36 6.5 16 39.98 45.25 2324.45 2692.17

MENA No. 2 2 2 2 2 2 2 3 2 3 2 2 3 2Mean 2 2 0.5 0.5 0.25 0.25 4.54 6.14 5.5 7.33 98.18 102.27 1284.7 2033.5Median 2 2 0.5 0.5 0.25 0.25 4.54 1.54 5.5 7 98.18 102.27 1469.94 2033.5

SA No. 2 3 2 3 2 3 5 5 5 5 4 5 4 4Mean 35.88 48.8 9.98 14.54 153.48 225.62 1.98 2.71 23 32 47.3 43.38 508.01 628.02Median 35.88 49.64 9.98 13.08 153.48 171.09 0.9 0.46 14 31 45.52 49.19 439.51 563.47

SSA No. 16 8 16 8 16 8 19 21 19 21 22 18 22 21Mean 57.88 50.69 24.54 20.28 730.8 508.21 2.34 5.1 6.89 8 26.52 24.75 515.21 548.2Median 56.95 55.38 21.48 22.21 463.01 502.89 0.69 2.38 7 8 16.03 11.88 290.89 342.18

Total No. 57 48 57 48 57 48 63 69 63 69 66 64 68 66Mean 27.07 18.3 10.34 6.21 244.28 108.12 6.2 28.12 9.3 14.23 36.73 40.69 1356.42 1684.62Median 16.07 9.02 6.3 2.99 39.69 8.92 2.02 8.41 7 10 30.66 36.21 821.24 1081.74

Source: Authors’ compilation from MIX and WDI datasets.

Table 2. Results based on cross-sectional regressions (dependent variable: poverty headcount ratio)

Explanatory variables OLS IV

Without Regions With regions Without Regions With regions(1) (2) (3) (4)

Log of GLP per capita �3.00 �1.40 �3.80 �3.25[�4.04]** [�1.70]� [�2.58]** [�2.31]*

Log of GDP per capita �12.83 �7.30 �12.12 �9.02[�8.33]** [�4.01]** [�6.63]** [�4.46]**

Domestic credit �0.09 �0.08 �0.11 �0.11[�1.65] [�1.77]� [�1.62] [�1.87]�

MENA – �7.44 – �9.54– [�1.90]� – [�1.69]�

EAP – 3.53 – �4.46– [0.69] – [�0.55]

ECA – �8.85 – �11.38– [�3.49]** – [�4.01]**

SA – 20.94 – 10.31– [3.34]** – [1.21]

SSA – 20.12 – 9.67– [1.83]� – [0.91]

Constant 119.58 75.48 117.44 96.81[10.20]** [4.40]** [10.31]** [4.77]**

N 47 47 46 46Adj. R2 0.728 0.864 0.717 0.835F-statistic 33.50 195.24 30.59 104.70Under identification test – – 7.65(0.02) 6.72(0.00)Weak identification test – – 10.63(0.00) 11.76(0.00)Over identification test – – 1.13(0.28) 0.51(0.48)Hausman test – – 0.69(0.87) 4.98(0.76)

t-values are in parenthesis.* Significant at 5%.** Significant at 1%.� Significant at 10%.


Table 3. Results based on panel data regressions (dependent variable: poverty head count ratio)

Explanatory variables Pool OLS Fixed effect Random effectsh Random effectst

(1) (2) (3) (4)

Log of GLP per capita �1.31 �0.37 �0.91 �1.59[�2.53]* [�0.44] [�1.76]� [�3.35]**

Log of GDP per capita �9.43 �13.02 �16.93 �10.17[�5.45]** [�1.96]� [�8.89]** [�7.64]**

Domestic Credit �0.07 0.01 0.00 �0.08[�2.40]* [0.08] [0.09] [�1.95]�

2007 Year Dummy �1.28 – – –[�0.64] – – –

MENA �10.25 – – –[�3.22]** – – –

EAP 3.74 – – –[0.78] – – –

ECA �10.96 – – –[�5.32]** – – –

SA 14.19 – – –[1.75]� – – –

SSA 22.56 – – –[3.73]** – –

Constant 92.16 113.73 143.53 100.58[6.67]** [2.55]* [11.67]** [10.53]**

Hausman – – 13.94(0.00) 43.60(0.00)Theta – – 0.76 0.84N 99 99 99 99Adj. R2 0.851 �0.996 – –F-statistic 51.81 4.37 – –

t-values are in parenthesis.* Significant at 5%.** Significant at 1%.� Significant at 10%.H Country effect.t Regional effect.


4. RESULTS

Figures 1 and 2 describe the patterns and trends in size andoutreach of the microfinance industry using real gross loanportfolio (after adjusting for inflation), number of MFIs andactive borrowers. Figure 1 shows the trend and patterns of realgross loan portfolio for different regions. Overall, the com-pound growth rate of the median gross loan portfolio in-creases for all regions over the period 2005–09. However,there are variations (steep and gentle) in the year-by-year up-ward slopes, while in one instance (Eastern Europe and Cen-tral Asia), a downward trend is observed. In particular, theslope for 2007–08 is either gently increasing or sloping down-wards. An interpretation of the trend over this period will needto take cognizance of the potential adverse effect of the globalfinancial crisis on the microfinance industry. Until 2007, thelargest MFIs were located in Latin America and the Carib-bean (LAC). However, in 2008, MFIs in Middle East andNorth Africa (MENA) experienced a sharp increase in theirgross loan portfolio.

Figure 2 compares the patterns and trends of gross loanportfolio of different regions and shows that over the yearsSouth Asia has experienced a greater and sharp increase inthe number of active borrowers than other regions. First, itmight be argued that by virtue of population size of countriesin this region, MFIs may have reached out to more clients(scale of outreach). Second, the MFIs in South Asia (e.g., Ban-gladesh or India) might have placed more emphasis on in-crease in the scale of outreach (e.g., in terms of number ofclients) or poverty reduction as their central mission thanthose in other regions.

Table 1 provides a summary statistics of the variablesused in the regression analyses. In view of the heterogeneityof the size of MFIs (gross loan portfolio), outreach (numberof active borrowers), and a country’s output per head(GDPPC), it is always prudent to observe the dispersionof the data. Table 1 indicates that the median in some in-stances is about either a hundredth (East Asia and the Pa-cific (EAP)) or a tenth (MENA) of the mean. This suggeststhat the raw data for the mean are likely to be affected byextreme values.

From the perspective of both numbers of active borrowersand MFIs, microfinance activities in SA countries are more in-tense than in the other regions. At the lower end, MFI activ-ities in Sub-Saharan Africa (SSA) countries tend to show thelowest values for the number of active borrowers (as a proxyfor MFI operations). As observed from the trends (Figures 1and 2), variations in these indicators over time and across dif-ferent regions suggest the need to develop a meaningful indexthat pulls together all three variables.

In terms of the macro indicators, SSA is the poorest regionfor both periods irrespective of the measure (incidence, depth,and severity) in question. Over the period 2003–07, both thepoverty headcount and poverty gap showed a decline of about7 and 4 percentage points, respectively. 13 South Asia is anexception as our sample showed that poverty levels rose inthe region over this period. 14 Among the less “worse off”regions, 15 MENA recorded the lowest poverty headcount ra-tio while LAC showed the highest output per head (GDPPC)in both years. The poverty headcount ratio continued to below (2%) in MENA, while it substantially decreased from14.5% to 8.9% in LAC.

Table 4. Results based on cross-sectional regressions (dependent variable: poverty gap)


Without regions With regions Without regions With regions(1) (2) (3) (4)

Log of GLP per capita �1.32 �0.87 �2.34 �2.08[�3.07]** [�2.08]* [�2.69]** [�2.45]*

Log of GDP per capita �4.42 �2.56 �3.56 �3.59[�5.76]** [�2.87]** [�3.43]** [�3.58]**

Domestic credit �0.06 �0.04 �0.09 �0.07[�2.31]* [�1.69]� [�2.30]* [�1.77]�

MENA – �2.53 – �3.84– [�1.23] – [�1.19]

EAP – �2.40 – �7.49– [�0.95] – [�1.53]

ECA – �4.07 – �5.58– [�3.48]** – [�3.87]**

SA – 3.41 – �3.35– [1.09] – [�0.74]

SSA – 8.41 – 1.81– [1.52] – [0.32]

Constant 43.08 28.46 40.61 41.67[7.12]** [3.50]** [6.94]** [4.16]**




The results of multivariate regressions are given in Tables 2–7 and simulation outcomes in Table 8. With a view to examin-ing the hypothesis of a relationship between gross loan portfo-lio per capita and poverty (incidence, depth and severity),eight different cases (i.e., four for cross-sectional regressionsand four for panel data regression) are examined for each pov-erty measure. Four different cases using cross-sectional datafor 2007 are given in Tables 2, 4 and 6 and other four casesare given in Tables 3, 5 and 7, using two-period (2003 and2007) panel data. In Tables 2, 4 and 6, OLS is applied in col-umns (1) and (2) and IV in columns (3) and (4). The estima-tions in columns (1) and (2) are robust (corrected forheteroscedasticity), and examine cases either using GLP percapita with and without regional dummies. In a similar fash-ion, the IV (columns (3) and (4) of Tables 2, 4 and 6) and panel(Tables 3, 5 and 7), respectively, test the poverty reducing ef-fect of GLP per capita hypothesis.

Three measures of the FGT class of poverty indices areused. Thus, Tables 2 and 3 contain all the estimations (OLS,IV, Fixed Effects (FE), and Random Effects (RE)) for the pov-erty headcount ratio; Tables 4 and 5 examine the case for thepoverty gap (depth) and Tables 6 and 7 investigate the severityof poverty (squared gap).

In column (1) of Table 2, all three specifications using thecross-sectional data show that GLP per capita is negativelyand significantly associated with the poverty headcount ratio,which is consistent with our hypothesis that micro loans re-duce poverty. For instance, because MFI loan per capita is de-fined in log, we observe that a 10% increase in MFI loan percapita reduces poverty by about 0.325% in the case of theIV estimation (in column (4) of Table 2). 16 The coefficient esti-

mate of log of gross loan portfolio per capita of MFI is nega-tive and significant at the 5% level in this case. As expected,GDP per capita is negative and shows a 1% statistical signifi-cance irrespective of the specification or the estimation methodchosen. Furthermore, consistent with the finance-poverty liter-ature, we find that the coefficient estimate of share of domesticcredit to GDP is negative and significant in some cases (col-umns (2) and (4) of Table 2).

Columns (2) and (4) explore the potential effect of regionaldummies on incidence of poverty. We observe that loan percapita of MFIs remain statistically significant after the inclu-sion of the regional dummies. Inclusion of regional dummiesin the poverty headcount equation reveals that ECA, withLAC as the reference case, has a significant negative coefficient(at the 1% level in both the OLS and IV cases). Also, SA andSSA dummies are positive in the OLS estimation. This impliesthat SA and SSA have higher poverty headcount levels relativeto LAC. These regression results are consistent with the sum-mary statistics of Table 1 in which both poverty levels andGDPPC show that SA and SSA trail LAC.

Columns (3) and (4) present the IV estimation with the aimof resolving the potential endogeneity of microfinance vari-ables in the poverty headcount equation, that is, gross loanportfolio per capita. As discussed earlier, the endogeneitymay be due to a bi-causal relationship between poverty andgross loan portfolio per capita. In terms of a bi-causal rela-tionship between gross loan portfolio per capita and povertyheadcount, we allude to the fact that investors who are in-clined to poverty reduction might direct their financial re-sources to countries and regions where poverty is high.Appendices 2 and 3 show the correlation matrix and the first

Table 5. Results based on panel data regressions dependent variable: poverty gap


(1) (2) (3) (4)

Log of GLP per capita �0.85 �0.04 �0.35 �0.96[�2.86]** [�0.09] [�1.23] [�3.42]**

Log of GDP per capita �3.22 �5.03 �6.70 �3.52[�3.20]** [�1.41] [�6.37]** [�4.50]**

Domestic credit �0.05 0.01 �0.01 �0.06[�2.74]** [0.20] [�0.37] [�2.47]*

2007 year dummy �0.42 – – –[�0.40] – – –

MENA �3.08 – – –[�1.62] – – –

EAP �1.85 – – –[�0.67] – – –

ECA �5.53 – – –[�4.51]** – – –

SA 1.42 – – –[0.41] – – –

SSA 10.41 – – –[2.88]** – – –

Constant 34.31 43.19 56.95 36.83[4.23]** [1.80]� [8.37]** [6.68]**




stage IV estimation which offer a justification for the validityof our instruments.

We use two kinds of instrument, that is, cost of enforcingcontract and weighted 5-year lag of average GLP. The formeris intuitively justified on the ground that the decision of micro-finance commercial investors, especially international funders,on whether to invest in a particular country is likely to dependon the extent to which the country has a good institution (e.g.,represented by a low cost of enforcing contracts) that wouldfacilitate economic activities. In this context, the cost ofenforcing contracts is supposed to have a significant and neg-ative correlation with loan per capita of MFIs (see Appendix2). While a direct effect of the cost of enforcing contracts onpoverty might not be strong as most poor households operatein the informal sector, the higher enforcing costs may excludethe poor from the formal sector employment or credit and per-petuate poverty in the long run. In view of this, we also useweighted 5-year lag of average GLP as an additional instru-ment.

It is worth mentioning that we observe a poverty reducingeffect for MFIs’ gross loan per capita when we use only thecost of enforcing contracts as an instrument. However, sincethe weak identification test is significant at only 10% in thiscase, we find the need to augment the instrumentation. Usingboth instruments yields a much higher Kleibergen-Paap rkWald F-statistic (weak identification test), as shown in col-umns 3 and 4 of Table 2. This does not compromise the Sar-gan’s over-identification test as we fail to reject the nullhypothesis that the instruments are valid, that is, uncorrelatedwith the error term. Also, the p-values (0.02 and 0.00) of the

under identification test in columns (3) and (4) of Table 2 al-low us to reject the null hypothesis that the model is underidentified. These specification tests validate IV estimations. 17

Tables 3, 5 and 7 show the results based on panel estima-tions for the poverty headcount ratio, depth, and severity,respectively. In Table 3, the poverty headcount ratio is esti-mated by the same set of explanatory variables. The numberof observation is 99. It is noted that this estimation is basedon unbalanced panel data and thus the results have to be inter-preted with caution (see Appendix 4 for the list of countriesand frequencies of observations). A similar pattern of resultsis observed, that is, gross loan portfolio per capita of microfi-nance institutions is negatively associated with incidence ofpoverty, after controlling for the effects of other covariatesand unobserved heterogeneity.

In Tables 4 and 5, we have replicated both the cross-sec-tional and panel regressions by replacing the poverty head-count ratio with the poverty gap. To avoid cluttering thetext, we summarize the key findings. All four cases in Table4 show significant negative effects of log of gross loan portfolioper capita of MFIs, implying the potential of microfinance inreducing the depth of poverty. The other explanatory variablesshow expected signs too. This hypothesis is further corrobo-rated by the pooled OLS and random effects model in columns(1) and (4) of Table 5. Loan per borrower shows a statisticallysignificant negative effect. Both the Hausman and theta statis-tics favor random effects over fixed effects.

Tables 6 and 7 report the cases where the dependent variableis squared poverty gap. An examination of the poverty gapand squared poverty gap results shows consistent results in

Table 6. Results based on cross-sectional regressions (dependent variable: squared poverty gap)


Without regions With regions Without regions With regions(1) (2) (3) (4)

Log of GLP per capita �42.48 �33.03 �89.49 �76.74[�2.63]* [�2.04]* [�2.51]* [�2.22]*

Log of GDP per capita �97.31 �48.22 �58.69 �83.16[�3.50]** [�1.50] [�1.49] [�2.17]*

Domestic credit �2.41 �1.55 �3.88 �2.30[�2.41]* [�1.53] [�2.28]* [�1.63]

MENA – �5.93 – �51.60– [�0.09] – [�0.45]

EAP – �120.59 – �301.59– [�1.31] – [�1.56]

ECA – �75.96 – �127.49– [�2.20]* – [�2.53]*

SA – �5.00 – �244.60– [�0.05] – [�1.35]

SSA – 288.95 – 55.60– [1.41] – [0.26]

Constant 997.99 584.47 888.65 1044.34[4.31]** [2.00]� [4.01]** [2.66]**




terms of sign and significance for GLP per capita. Also, in thecase of squared poverty gap, the Hausman test favors randomeffects. 18 Broadly, the results imply that GLP per capita ofMFIs benefits not just the poor but also the poorest. Insum, gross loan portfolio per capita of MFIs is negativelyassociated with the incidence, depth, and severity of poverty.

5. SIMULATIONS

That microfinance is impervious to the global recession fol-lowing the financial crisis is debatable. 19 Some have arguedthat the slowdown of the global economy will impact nega-tively on microfinance as MFIs are now more closely linkedto global financial markets than before. So there will be: (i)a funding or liquidity impact, with greater refinancing risksfor MFIs and (ii) an economic impact, with financial perfor-mance affected by lower lending volumes, higher costs of fund-ing, tighter net interest margins, and greater volatility inforeign exchange losses/gains. Magnoni and Powers (2009),for example, point out that, over 2009–10, the sector-widemicrofinance will grow by some $28 billion less than antici-pated before the crisis. Others are more optimistic. For exam-ple, Littlefield and Kneiding (2009) argue that themicrofinance sector will survive the setbacks because of thestrong foundations and vast untapped market of creditworthyclients. The effects on the poor through MFIs are, however,largely anecdotal. An attempt is made below to simulate thelikely effects of the global recession.

In the context of the recent global recession, we examine thepossible slowdown of poverty reduction 20 as a result of con-

traction: (i) in gross loan portfolio per capita of MFIs and(ii) GDP per capita. Table 8 reports the simulation resultson the basis of the IV estimation reported in column (4) of Ta-ble 2. The calculations 21 are premised on the assumption thatthe coefficient estimate of each variable is the same across dif-ferent regions in the regression and thus the simulated effectsof key variables on poverty are also the same. The third andfourth rows of Table 8 show that a fall in GLP per capita ofMFIs and GDP per capita will tend to increase poverty head-count. While this is plausible in view of the comparative con-tribution of MFI in an economy, the poverty increasing effectof 0.167% (or 0.343%) due to the 5% (or 10%) decline in GLPper capita points to the need for sustainable flow of on-lendingfunds. Although the overall rise in poverty is low, it must beemphasized that this is additional to other cumulative effectsof the global slowdown on poverty (investments, for example,continue to be sluggish).

6. CONCLUDING OBSERVATIONS

Recent assessments of impact of microfinance—based lar-gely on randomized trials—have led to questioning of claimsof women’s empowerment and poverty reduction. The so-called “magic” of microfinance has thus come under deepscrutiny and the findings of little or weak impacts are begin-ning to turn the tide against it. The faltering global economyhas also raised serious concerns about the immunity of themicrofinance sector and its potential for poverty reduction.From this perspective, the preceding analysis centered on thehypothesis that microfinance reduces poverty. We carried

Table 7. Results based on panel data regressions (dependent variable: squared poverty gap)


(1) (2) (3) (4)

Log of GLP per capita �35.68 10.47 �11.18 �39.13[�2.80]** [0.47] [�0.83] [�2.89]**

Log of GDP per capita �56.62 �165.80 �189.12 �69.82[�1.53] [�0.94] [�4.03]** [�1.86]�

Domestic credit �2.73 0.59 �1.02 �2.89[�2.72]** [0.28] [�0.78] [�2.55]*

2007 year dummy �8.93 – – –[�0.19] – – –

MENA 36.98 – – –[0.48] – – –

EAP �83.00 – – –[�0.78] – – –

ECA �128.15 – – –[�2.83]** – – –

SA �8.98 – – –[�0.08] – – –

SSA 420.29 – – –[2.79]** – – –

Constant 707.54 1306.96 1595.16 849.04[2.37]* [1.10] [5.28]** [3.29]**



Table 8. Effect of decrease in GLP per capita, GDP per capita and domestic credit on poverty

Coefficients Means 5% 10%(1) (2) (3) (4)

Effect of decrease in GLP per capita on poverty �3.252 2.060 0.167 0.343Effect of decrease in GDP per capita on poverty �9.018 7.095 0.463 0.950

Net effect 0.523 1.192

All developing countries Mean poverty 18.400 18.400Final poverty 18.923 19.592

East Asia and the Pacific Mean poverty 24.293 24.293Final poverty 24.816 25.485

Eastern Europe and Central Asia Mean poverty 5.969 5.969Final poverty 6.492 7.161

Latin America and the Caribbean Mean poverty 8.846 8.846Final poverty 9.369 10.038

Middle east and North Africa Mean poverty 2.000 2.000Final poverty 2.523 3.192

South Asia Mean poverty 48.801 48.801Final poverty 49.324 49.993

Sub-Saharan Africa Mean poverty 50.685 50.685Final poverty 51.208 51.877


out tests using cross-country data for 2007 and a panel for2003 and 2007. Taking account of the endogeneity associatedwith loans per capita from Microfinance Institutions (MFIs),our econometric results consistently confirm that microfinance

loans per capita are significantly and negatively associatedwith poverty, that is, a country with a higher MFIs’ gross loanportfolio per capita tends to have lower poverty after control-ling for the effects of other factors influencing it. The negative


relationship remains unchanged when the poverty headcountratio is replaced by the poverty gap and squared povertygap. These results suggest that microfinance not only reducesthe incidence of poverty but also its depth and severity. Thepanel results further corroborate these findings. Other factorsthat contribute to poverty reduction include GDP per capitaand share of credit in GDP (as a measure of financial develop-ment of an economy). Besides, there are significant regionaleffects.

Our simulations point to worsening of poverty in a mildrecession scenario with small reductions in gross loan portfolio

per capita, GDP per capita, and share of credit in GDP. Thesesimulations are helpful in adding precision to anecdotal evi-dence about how setbacks to MFIs hurt the poor. Indeed, sus-tained flows to MFIs may help avert to some extentaccentuation of poverty as a consequence of the slow and fal-tering recovery of the global economy.

In conclusion, assertions that microfinance is “oversold”and lacks the “magic” associated with it are widely off themark, if not largely mistaken.

NOTES

1. Some have even questioned the impact in terms of poverty reduction,promotion of gender equality, and reduction in child mortality. Indeed aview that microfinance is oversold is gaining credibility (Rosenberg, 2010).

2. The Foster–Greer–Thorbecke (FGT) class of poverty indices is ageneralized measure of poverty in an economy (e.g., a country) and it isdenoted as pa ¼ 1

n

Pqi¼1

z�yiz

� �awhere z is the poverty line (e.g., reflecting

nutritional requirement), yi is the ith lowest income, n is the population inthe economy, and as q is the number of people whose income is below z,and a is a measure of poverty aversion where a larger a gives greateremphasis to the poorest of the poor (Foster, Greer, & Thorbecke 1984).This comprises the headcount ratio when a = 0 (the fraction of thepopulation which lives below the poverty line), the average income povertygap when a = 1 (the population average of the distances between poorpeople’s income and the poverty line or the extent to which a poor fallsbelow the poverty line on average) and the squared poverty gap (P2) indexwhen a = 2 (a distributionally sensitive poverty index as a weighted sumof poverty gaps where the weights are the proportionate poverty gaps)(Foster et al., 1984). The present study focuses on all three poverty indices.

3. See Appendix 1 for the list of the countries.

4. Appendix 4 lists the countries included in the panel data estimations.Because we include the countries only with data in 2003, the panel covers alarger number of countries.

5. Cull et al. (2007) did not provide any direct evidence for this. Theirclaim was based on Honohan (2004) who used the data on MFIs providedby the Microcredit Summit Organization similar to the MIX data andfound that “the largest 30 microfinance firms account . . . for more than90% of the clients served worldwide by the 234 top firms (and hence formore than three-quarters of those served by all of the 2572 firms reportingto the MicrocreditSummit)” (Honohan, 2004, p. 4). Cull et al. then arguedthat “Honohan’s evidence suggests that . . . the banks here (covered by theMIX data) served over half of all microfinance customers worldwide”

(Cull et al., 2007, p. F111, the phrase in brackets added by authors).

6. For instance, the correlation coefficient between loan accounts peradult and aggregated gross loan portfolio per capita is 0.483 andstatistically significant for 2009. This does not change much when weexclude the outliers, such as Bangladesh.

7. Exceptions include Imai et al. (2010a, 2010b).

8. See Table 1 for descriptive statistics of these variables. These will bediscussed in Section 4.

9. MIX data date back to 1994, but not until 2002 most MFIs were notkeen on submitting their records for public use.

10. This is because the construction of international poverty estimateshas to rely on nation-wide household expenditure or income surveys whichare carried out once or twice in a decade for most of the developingcountries.

11. This index passes the statistical validity of a valid instrument as itshows a high correlation with gross loan portfolio and a low correlationwith the poverty headcount ratio (with the coefficient of correlation being0.8 for the former and 0.1 for the latter, respectively).

12. Due to data constraints on the international poverty data, we tookaverages of poverty for the period 2000–03 and 2004–07.

13. Poverty data for the panel were constructed by taking averages for2000–03 and 2004–07. The data are given in Appendix 5.

14. It is noted that an overall increase in poverty in South Asia is due tothe highly unbalanced nature of our panel data of poverty. As shown inAppendix 5, while Bangladesh has two observations, India, Nepal, and SriLanka have only one either for 2003 or 2007. Despite the fall in povertyheadcount for Bangladesh from 57.8% in 2003 to 49.6% in 2007, includingcountries with high poverty (e.g., Nepal—55.1%) in 2007 accounts for thehigher figure in 2007.

15. Most of the countries used in the study are either transitional ordeveloping countries.

16. This is based on the formulae to interpret the coefficient of semi-logspecification (level-log): DY = (b/100)%DX (Wooldridge, 2009, p.46).Note that poverty headcount ratio is defined in percentages.

17. The relationship between GLP per capita and poverty head countratio may be different according to the countries’ income level. We haverepeated the same regressions (OLS and IV) for three income groups ofcountries, that is, “low income,” “lower middle income,” and “uppermiddle income.” Appendix 6 shows the results of IV. It is observed thatthe negative association between GLP per capita and poverty is the largestand most highly significant for low income countries and becomes smallerand statistically insignificant as the income goes up, while it is significantin case of middle income countries as a whole. A similar pattern of resultsis observed in case of OLS and for other measures of poverty.

18. In addition to the Hausman test, we calculate theta (measure of theextent of biasedness of random effects’ model) and we observe values closeto one (greater than 0.75) for each of three estimations. This supports thefact that country/regional effects over time are important and cannot beignored. Use of theta, which is preferred due to the strong finiteassumption underlying the Hausman test, further supports our choice offixed effects model.


19. For a comprehensive review, see Llanto and Badiola (2010).

20. Imai et al. (2010a, 2010b) report a slowdown in the rate of povertyreduction post 2000 relative to the 1990-decade.

APPENDIX 1. LIST OF REG

Low income countries Lower middle incomecountries

Country Region Country Region

Afghanistan SAa Armenia ECABangladesh SA Cameroon SSABenin SSAb Congo, Rep. SSABurkina Faso SSA Egypt, Arab Rep. MENACambodia EAPc El Salvador LACCongo, Dem. Rep. of SSA Georgia ECAEthiopia SSA Ghana SSAGambia, The SSA Guatemala LACGuinea SSA Honduras LAVGuinea-Bissau SSA India SAHaiti LACd Indonesia EAPKenya SSA Iraq MENA

21. The simulated effects of GLP per capita and GDPPC are calculatedwith the semi-log (or “level-log”) specification (DY = (b/100)%DX), whilethe effect of domestic credit is based on the “level-level” specification(DY = bDX) (see Wooldridge, 2009, p. 46 for details).

REFERENCES

Ahlin, C., & Lin, J. (2006). Luck or skill? MFI performance inmacroeconomic context, bureau for research and economic analysis ofdevelopment. BREAD Working Paper, No. 132. Cambridge, MA:Centre for International Development, Harvard University.

Ahlin, C., Lin, J., & Maio, M. (2011). Where does microfinance flourish?Microfinance institution performance in macroeconomic context.Journal of Development Economics, 95(2), 105–120.

Armendariz, de Aghion, & Morduch, J. (2005). The economics ofmicrofinance. Cambridge: MIT Press.

Banerjee, A., Duflo, E., Glennerster, R., & Kinnan, C. (2009). The miracleof microfinance? Evidence from a randomised evaluation. Cambridge,MA: Department of Economics, MIT, Mimeo.

Chen, S., & Ravallion, M. (2008). The developing world is poorer than wethought, but no less successful in the fight against poverty. PolicyResearch Working Paper, WPS 4703. Washington, DC: World Bank.

Collins, D., Morduch, J., Rutherford, S., & Ruthven, O. (2009). Portfoliosof the poor: How the world’s poor live on $2 a day?. Princeton, NJ:Princeton University Press.

Cull, R., Demirguc�-Kunt, A., & Morduch, J. (2007). Financial perfor-mance and outreach: A global analysis of leading microbanks. TheEconomic Journal, 117(517), F107–F133.

Cull, R., Demirguc�-Kunt, A., & Morduch, J. (2011). Does regulatorysupervision curtail microfinance profitability and outreach?. WorldDevelopment, 39(6), 949–965.

Feigenberg, B., Field, E. M., & Pande, R. (2010). Building social capitalthrough microfinance. RWP 10-019. Cambridge, MA: Kennedy School,Harvard University.

Foster, J., Greer, J., & Thorbecke, E. (1984). A class of decomposablepoverty measures. Econometrica, 52, 761–766.

Gaiha, R., & Nandhi, M. A. (2009). Microfinance, self-help groups andempowerment in Maharashtra. In R. Jha (Ed.), The Indian economysixty years after independence. London: Palgrave Macmillan.

Honohan, P. (2004). Financial sector policy and the poor: Selected findingsand issues. World Bank Working Paper, No. 43. Washington, DC:World Bank.

Hulme, D., & Mosley, P. (1996). Finance against poverty (Vol. 1). London:Routledge.

Imai, S. K., Arun, T., & Annim, S. K. (2010a). Microfinance andhousehold poverty reduction: New evidence from India. WorldDevelopment, 38(12).

Imai, S. K., Gaiha, R., & Thapa, G. (2010b). Is the millenniumdevelopment goal of poverty still achievable? Role of institutions,finance and openness. Oxford Development Studies, 38(3),309–337.

Kai, H., & Hamori, S. (2009). Microfinance and inequality. Research inApplied Economics, 1(1: E14), 1–14.

Karlan, D., & Zinman, J. (2010). Expanding credit access: Usingrandomised supply decisions to estimate the impacts. Review ofFinancial Studies, 23(1), 433–464.

Khandker, S. R. (2005). Micro-finance and poverty: Evidence using paneldata from Bangladesh. The World Bank Economic Review, 19(2),263–286.

Littlefield, E., & Kneiding, C. (2009). The global financial crisis and itsimpact on microfinance. 52 CGAP Focus Note 3, February 2009.<http://india.microsave.org/sites/default/files/CGAP_Global_Cri-sis%20_%20Impact_on_MF.pdf/> Accessed January 2012.

Llanto, G. M., & Badiola, J. A. R. (2010). The impact of the globalfinancial crisis on rural and microfinance in Asia. Mimeo, Rome: IFAD.

Magnoni, B., & Powers, J. (2009). Will the bottom of the pyramid hitbottom? The effects of the global credit crisis on the microfinance sector.Micro Report No. 150. Washington, DC: USAID.

Microfinance Information Exchange (2010). Regional benchmarking LatinAmerica and the Caribbean 2009 benchmarks. Washington, DC: MIXMarket.

Morduch, J. (1999). The role of subsidies in microfinance: Evidencefrom the Grameen Bank. Journal of Development Economics, 60,229–248.

Mosley, P. (2001). Microfinance and poverty in Bolivia. Journal ofDevelopment Studies, 37(4), 101–132.

Ravallion, M., Chen, S., & Sangraula, P. (2008). Dollar a day revisited.World Bank Policy Research Working Paper, No. 4620. Washington,DC: World Bank.

Rosenberg, R. (2010). Does microcredit really help poor people? FocusNote, No. 59. Washington, DC: CGAP.

Wooldridge, J. (2009). Introductory econometric: A modern approach (4thed.). Mason, Ohio: South Western Educational Publishing.

World Bank (2011). World development indicators 2011. Washington, DC:Oxford University Press.

IONS AND COUNTRIES

Upper middle income countries High incomecountries

Country Region Country Region

Albania ECA Poland ECAAzerbaijan ECA Croatia ECABosnia and Herzegovina ECA

f Brazil LACBulgaria ECAChile LACChina EAPColombia LACCosta Rica LACDominican Republic LACEcuador LACJordan MENA

(continued on next page)

http://india.microsave.org/sites/default/files/CGAP_Global_Crisis%20_%20Impact_on_MF.pdf

http://india.microsave.org/sites/default/files/CGAP_Global_Crisis%20_%20Impact_on_MF.pdf

APPENDIX 1—Continued

Low income countries Lower middle income countries Upper middle income countries High incomecountries

Country Region Country Region Country Region Country Region

Kyrgyz Republic ECAe Kosovo ECA Kazakhstan ECAMadagascar SSA Lao P.D.R. EAP Macedonia, FYR ECAMalawi SSA Moldova ECA Mexico LACMozambique SSA Mongolia ECA Panama LACNepal SA Nicaragua LAC Peru LACRwanda SSA Papua New Guinea SSA Romania ECASierra Leone SSA Paraguay LAC Russia ECATajikistan ECA Sri Lanka SA South Africa SSATanzania SSA Swaziland SSA Turkey ECAUganda SSA Ukraine ECA

Uzbekistan ECAVietnam EAPZambia SSA

a SA: South Asia.b SSA: Sub-Saharan Africa.c EAP: East Asia and Pacific.d LAC: Latin America and the Caribbean.e ECA: Eastern Europe and Central Asia.f MENA: Middle East and North Africa.

APPENDIX 2. CORRELATION MATRIX

Variables Povertyheadcount

Povertygap

Squaredpoverty gap

Log of GLPper capita

Log of GDPper capita

Domesticcredit

Cost ofcont. enf.

LogW. GLP

Poverty headcount 1Poverty gap 0.96** 1Squared poverty gap 0.86** 0.95** 1Log of GLP per capita �0.43** �0.41** �0.40** 1Log of GDP per capita �0.80** �0.73** �0.60** 0.30* 1Domestic credit �0.36* �0.36* �0.36* �0.07 0.49** 1Cost of contract enforcement 0.55** 0.55** 0.53** �0.28* �0.44** �0.32* 1Log. W. GLP �0.17 �0.28� �0.36* 0.60** 0.30* 0.16 �0.13 1

W.—weighted 5-year lag average.* Significant at 5%.** Significant at 1%.� Significant at 10%.

APPENDIX 3. FIRST STAGE REGRESSION (DEPENDENT VARIABLE: LOG OF GLP PER CAPITA)

Explanatory variables Coefficients

Cost of contract enforcement �0.02[�2.01]�

Log of weighted 5-year lag of average GLP 0.58[4.23]**

Log of GDP per capita �0.80[�2.33]*

Domestic credit �0.01[�1.14]

MENA �0.77[�0.62]

EAP �3.61[�3.54]**

ECA �0.59[�0.97]

SA �5.37[�4.71]**



Explanatory variables Coefficients

SSA �3.42[�3.07]**

Constant �0.17[�0.04]

N 46Adjusted R2 0.620F-statistic 9.15Log-likelihood �76.30


APPENDIX 4. LIST OF COUNTRIES FOR PANEL DATA

No. Countries Freq. No. Countries Freq.

1 Albania 2 32 Kazakhstan 22 Armenia 2 33 Kenya 13 Azerbaijan 2 34 Kyrgyz Republic 24 Bangladesh 2 35 Macedonia, FYR 25 Benin 1 36 Madagascar 26 Bosnia and Herzegovina 2 37 Malawi 17 Brazil 2 38 Mexico 28 Bulgaria 1 39 Moldova 29 Burkina Faso 1 40 Mongolia 210 Cambodia 1 41 Mozambique 111 Cameroon 1 42 Nepal 112 Chile 2 43 Nicaragua 213 China 2 44 Panama 114 Colombia 2 45 Paraguay 215 Congo, Dem. Rep. of 1 46 Peru 216 Costa Rica 2 47 Poland 217 Croatia 2 48 Romania 218 Dominican Republic 2 49 Russia 219 East Timor 1 50 Rwanda 120 Ecuador 2 51 Sierra Leone 121 Egypt, Arab Rep. 2 52 South Africa 122 El Salvador 2 53 Sri Lanka 123 Ethiopia 2 54 Swaziland 124 Georgia 2 55 Tajikistan 225 Guatemala 2 56 Tanzania 126 Guinea 1 57 Turkey 227 Haiti 1 58 Uganda 228 Honduras 2 59 Ukraine 229 India 1 60 Vietnam 230 Indonesia 1 61 Zambia 231 Jordan 2

Source: Authors’ compilation from MIX data.

APPENDIX 5. INTERNATIONAL POVERTY DATA—UNBALANCED PANEL

No. Year Countries Poverty HC No. Year Countries Poverty HC

1 2003 Albania 2.00 54 2007 Kenya 19.722 2007 Albania 2.00 55 2003 Kyrgyz Republic 34.033 2003 Armenia 12.20 56 2007 Kyrgyz Republic 12.624 2007 Armenia 3.65 57 2003 Macedonia, FYR 2.315 2003 Azerbaijan 6.32 58 2007 Macedonia, FYR 2.00

(continued on next page)



No. Year Countries Poverty HC No. Year Countries Poverty HC

6 2007 Azerbaijan 2.00 59 2003 Madagascar 76.347 2003 Bangladesh 57.82 60 2007 Madagascar 67.838 2007 Bangladesh 49.64 61 2007 Malawi 73.869 2003 Benin 47.33 62 2003 Mexico 4.2810 2003 Bosnia and Herzegovina 2.00 63 2007 Mexico 2.4011 2007 Bosnia and Herzegovina 2.00 64 2003 Moldova 25.0512 2003 Brazil 10.40 65 2007 Moldova 5.2613 2007 Brazil 8.00 66 2003 Mongolia 15.4714 2003 Bulgaria 2.32 67 2007 Mongolia 22.3815 2003 Burkina Faso 56.54 68 2003 Mozambique 74.6916 2007 Cambodia 33.02 69 2007 Nepal 55.1217 2003 Cameroon 32.81 70 2003 Nicaragua 19.4218 2003 Chile 2.00 71 2007 Nicaragua 15.8119 2007 Chile 2.00 72 2007 Panama 9.3420 2003 China 28.36 73 2003 Paraguay 17.2321 2007 China 15.92 74 2007 Paraguay 7.8822 2003 Colombia 16.07 75 2003 Peru 13.8423 2007 Colombia 16.01 76 2007 Peru 7.9424 2007 Congo, Dem. Rep. of 59.22 77 2003 Poland 2.0025 2003 Costa Rica 4.52 78 2007 Poland 2.0026 2007 Costa Rica 2.19 79 2003 Romania 3.0927 2003 Croatia 2.00 80 2007 Romania 2.0028 2007 Croatia 2.00 81 2003 Russia 2.0029 2003 Dominican Republic 5.27 82 2007 Russia 2.0030 2007 Dominican Republic 4.46 83 2003 Rwanda 76.5631 2003 East Timor 52.94 84 2003 Sierra Leone 53.3732 2003 Ecuador 10.49 85 2003 South Africa 26.2033 2007 Ecuador 7.24 86 2003 Sri Lanka 13.9534 2003 Egypt, Arab Rep. 2.00 87 2003 Swaziland 62.8535 2007 Egypt, Arab Rep. 2.00 88 2003 Tajikistan 36.2536 2003 El Salvador 13.73 89 2007 Tajikistan 21.4937 2007 El Salvador 8.70 90 2003 Tanzania 88.5238 2003 Ethiopia 55.58 91 2003 Turkey 2.0039 2007 Ethiopia 39.04 92 2007 Turkey 2.6540 2003 Georgia 12.80 93 2003 Uganda 57.3741 2007 Georgia 13.44 94 2007 Uganda 51.5342 2003 Guatemala 14.99 95 2003 Ukraine 2.0043 2007 Guatemala 11.70 96 2007 Ukraine 2.0044 2003 Guinea 70.13 97 2003 Vietnam 40.0545 54.9 Haiti 54.90 98 2007 Vietnam 22.8246 2003 Honduras 18.10 99 2003 Zambia 64.6047 2007 Honduras 20.19 100 2007 Zambia 64.2948 2003 India 41.6449 2007 Indonesia 25.4250 2003 Jordan 2.0051 2007 Jordan 2.0052 2003 Kazakhstan 3.4253 2007 Kazakhstan 2.00

Source: Authors’ compilation based on 2011 world development indicators.

APPENDIX 6. DISAGGREGATION BY INCOME GROUPS (IV) (DEPENDENT VARIABLE: POVERTY HEADCOUNTRATIO)

Explanatory variables (a) (b) (c) (d)Low income Lower middle income Upper middle income Middle income (b) + (c)

Log of GLP per capita �10.13 �2.13 0.53 �2.16[�3.59]** [�1.58] [0.79] [�2.09]*

Log of GDP per capita 4.45 �18.72 �1.34 �11.09[0.51] [�3.38]** [�0.56] [�5.22]**

Share of domestic credit of GDP �0.49 �0.11 0.07 �0.05[�2.41]* [�1.03] [1.66]� [�0.78]



Explanatory variables (a) (b) (c) (d)Low income Lower middle income Upper middle income Middle income (b) + (c)

Constant 43.20 158.38 10.69 101.98[0.99] [4.17]** [0.55] [6.36]**

N 11 15 19 35Adj. R2 0.705 0.470 . 0.469F-statistics 5.68 4.70 0.79 10.18

t statistics in brackets.* p < .05.** p < .01.� p < .10


microfinance and poverty—a macro perspective

Documents