Becchetti and Castriota_2011_MF After Tsunami Srilanka

8/3/2019 Becchetti and Castriota_2011_MF After Tsunami Srilanka

1/15

Does Microfinance Work as a Recovery Tool After Disasters?

Evidence from the 2004 Tsunami

LEONARDO BECCHETTIUniversita di Tor Vergata, Roma, Italy

and

STEFANO CASTRIOTA *

University of Perugia, Italy

Summary. We evaluate the effectiveness of microfinance as a recovery tool after tsunami by testing the impact of an equity injectionfrom foreign donors which recapitalizes a Sri Lankan MFI and allows it to refinance borrowers seriously damaged by the calamity. Wefind that loans obtained from the MFI after the catastrophic event have a positive and significant effect on the change in real income andin weekly worked hours, and that the impact on performance variables is significantly stronger for damaged than non-damaged borrow-

ers. Results hold after controlling for selection effects and for heterogeneity in both the timing of the intervention and the characteristicsof treatment and control samples. 2011 Elsevier Ltd. All rights reserved.

Key words Asia, Sri Lanka, tsunami, natural catastrophe, crisis recovery, microfinance

1. INTRODUCTION

One of the main obstacles to economic development for thepoor is the lack of access to traditional credit markets due tothe scarce availability of collateral resources and the highscreening, monitoring and enforcement costs incurred by

financial intermediaries when lending to them (Hermes & Len-sink, 2007). Microfinance tries to circumvent these problemswith a mix of solutions.

Assessing the impact of microfinance programs is not easy.First, empirical studies may incur in self-selection bias sincethose who borrow may have better unobservable traits thanthe control sample, mainly as a consequence of the same bankscreening process. Second, undocumented village-level differ-ences could influence the demand for/use of credit, therebyleaving space for placement bias (Hulme & Mosley, 1996).Third, comparing old and new clients might be subject to attri-tion bias, with survived old clients being ofbetter type thannew ones, as underlined by Karlan (2001). Fourth, data collec-tion is difficult and costly, especially when repeated across

time. Nevertheless, a number of studies have found positive ef-fects of microcredit programs on clients income, womenempowerment, contraceptive use and nutrition (for a survey,see Goldberg (2005) and Armendariz de Aghion and Morduch(2005)).

The focus of our research is on the relatively less exploredtopic of the relevance of microcredit as a recovery tool aftera natural catastrophe such as the 2004 Asian Tsunami. 1 Aftera catastrophe occurs, the first financial source used to recoveris self-insurance (savings and accumulated assets). Unfortu-nately, a relevant share of the poorest do not have enough sav-ings after a natural disaster when the latter destroys their fewavailable resources. For this reason, especially in low-incomerural areas, it is common practice to form risk-sharing net-works. The problem with them is that they work at best when

members incomes are uncorrelated or negatively correlated,

generally not the case for developing countries (Fafchamps& Gubert, 2007) and after natural catastrophes (Skees,Varangis, Larson, & Siegel, 2002). This is why an importantrecovery tool is represented by loans provided by traditionalbanks and MFIs. With respect to donations and charity, credithas the advantage that it does not affect income in the mere

short term and that, if the loan is paid back, it perpetuatesthe financial flow and satisfies new investment opportunities.

It is important to notice that natural hazards tend to beaccompanied by liquidity squeezes since, in spite of a boomin credit demand due to the need to restore destroyed anddamaged buildings and economic activities, banks are oftenforced to reduce the supply of loans because of the suddenworsening in the quality of their assets. For this reason recapi-talizing MFIs after calamities may be crucial. The recent his-torical evidence documents that microcredit programscontributed to reduce the vulnerability of the poor by assistingthem to re-build assets and by providing emergency assistanceafter natural disasters. There are many examples of MFIs ac-tive in post-conflict and post-disaster countries whose loans

have been claimed to be a useful recovery tool. The wide-spread diffusion of MFIs in Uganda and Bosnia, among othercountries, in the post-war periods is a clear example. Anothercase refers to Thailand in the post-tsunami period where, as re-ported in the USAIDs web site, 2 a USAID-financed micro-finance fund in a tiny, tsunami-ravaged village has proven so

* We thank Robert Cull, Rafael Di Tella, Rita Ferrer-i-Carbonell, Robert

Lensink, Enrico Longoni, Craig McIntosh, and Bruce Wydick for their

useful comments and suggestions. We wish to thank David Berno, Salv-

atore Morelli, and Francesca Palermo for valuable research assistance and

Laura Foschi, Francesca Lo Re, Marco Santori, and all Etimos team for

their support. The usual disclaimer applies. Final revision accepted: Oct-

ober 30, 2009.

World Development Vol. 39, No. 6, pp. 898912, 2011 2011 Elsevier Ltd. All rights reserved

0305-750X/$ - see front matter

www.elsevier.com/locate/worlddevdoi:10.1016/j.worlddev.2009.10.020

898
http://-/?-http://-/?-http://-/?-http://-/?-http://dx.doi.org/10.1016/j.worlddev.2009.10.020http://dx.doi.org/10.1016/j.worlddev.2009.10.020http://-/?-http://-/?-http://-/?-http://-/?-


2/15

successful that it has been accepted as an associate member ofthe Credit Union League of Thailand. 3

Apart from this anecdotal information, rigorous and soundempirical evidence on the usefulness of microcredit as recoverytool after natural catastrophes is still scarce. Khandker (2007)studies the coping strategies adopted by rural households dur-ing the 1998 flood in Bangladesh and assesses their impact on

welfare. The author concludes that

the presence of micro-credit programs increased the amount of borrowing coping.Household borrowing also increased household welfare byraising both consumption and asset holding (p. 179). Hoque(2008), using data from rural Bangladesh, finds that BRACsmicro-credit program may increase participating householdsabilities to cope with economic hardships following floodsand other natural disasters. Nevertheless, the author concludesthat further research to much more systematic informationneeds to be conducted about micro-credit program before con-clusive results can be reached. The purpose of our research isto contribute to fill in this gap.

Our specific focus is the evaluation of the effects of a donorsintervention aimed at recapitalizing a Sri Lankan MFI 4 whichreported post-tsunami certified losses for 24.4% of its credit

portfolio due to the serious damages to the business of a largepart of its borrowers. This implies that, in case of adoption ofa standard capital adequacy rule of, say, 10%, the MFIs losseswould amount to 250% of its capital. The recapitalization pro-gram was aimed to provide the MFI with the capital necessaryto grant new loans to the damaged borrowers enabling themto start back their activity and proved to be much cheaperfor the donors 5 than more classical aid schemes. Our goal isto evaluate whether such intervention acted as an effectiveliquidity injection for the damaged borrowers enabling themto restore their economic activity and to significantly improvetheir economic conditions with respect to the immediate post-tsunami levels.

The advantage of our framework is that the tsunami event

in fact creates two

randomly selected

groups: a treatmentgroup (borrowers directly hit by the tsunami shock) and a con-trol group (borrowers from the same MFI not affected by it).Our study can, therefore, be assimilated to a (quasi) naturalexperiment in which the exogenous shock makes a differencebetween the two above mentioned groups which are ex antenot significantly different in terms of borrowers quality orseniority characteristics, overcoming the standard selectionbias problem in microfinance impact analyses. We exploit thisunique opportunity by focusing on the effects of post-tsunamiMFI refinancing. More specifically, we evaluate its impact(measured by the size of the loans obtained after the tsunamiscaled on the borrowers post-tsunami pre-refinancingmonthly income) on two performance variables (percentchange in income and in worked hours after refinancing) by

carefully taking into account problems related to heterogene-ity of loan timing and endogenous size and timing of the loan(see Section 3).

The paper is divided into five sections including introductionand conclusions. Section 2 provides details on our survey. Sec-tion 3 describes the dataset, explains the methodology adoptedand provides summary statistics. Section 4 presents the estima-tion approach and comments descriptive and econometric evi-dence. Section 5 concludes.

2. AGRO MICRO FINANCE AND THE SURVEY

Agromart Foundation is a Sri Lankan NGO founded in1989 to carry out grassroots work with a large number of com-

munities in Sri Lanka. The Head Office is located in Colombowith nine other provincial offices in Uva, the Southern, NorthWestern, and Eastern provinces. The core of its mission isstrengthening the competencies of its members through partic-ipatory trainings. In order to achieve this goal AgromartFoundation created self-help groups in rural areas throughthe provision of technical assistance and education. In 1994

the Foundation broadened its activity by working as a micro-credit institution for its clients, but in 2000 it decided to fundAgro Micro Finance (AMF) and to delegate this task to it.Even if the respective fields of action remain separated, thelinks between the two organizations are strong. Agro MicroFinance lends only to members of community based organiza-tions which received for at least six months self-employment,entrepreneur development and literacy trainings from theAgromart Foundation. Seventy-two percent of AMF borrow-ers are women. In March 2005 the MFIs loan portfolio was of295.000.

After the tsunami, Agro Micro Finance and AgromartFoundation certified direct and indirect losses on 620 clientsin the district of Galle, Matara, and Hambantota. The esti-mated corresponding financial needs to cover such losses

amounted to almost 24.4% of the MFI loan portfolio at thetsunami date. This evidence documents the importance of for-eign intervention to avoid the MFI financial distress and theconsequent restriction to credit access for the MFI borrowers.The liquidity provided by foreign institutions allowed AMF toavoid credit restrictions and the risk of default. Support toAMF refinancing needs came from USAID, UNDP, and anItalian MFI (Etimos).

To evaluate the impact of post-tsunami MFI refinancing werandomly selected from the bank records a sample of 305 bor-rowers: 200 with at least one type of damage (which we defineas treatment group) and 105 with no damages (which we de-fine as control group). We created a treatment group largerin size since part of our analysis specifically focuses on subs-

amples of the treatment group which differ by damage typol-ogies in addressing some of the above mentioned researchquestions. A questionnaire was administered to both groupsin April 2007. The interviews were conducted face to face byone of the authors of the paper with the help of two moreresearchers and three translators with economic degree (thequestionnaire was translated in Sinhalese). Some borrowerswere interviewed at their homes, some during the monthlysociety meetings and the remaining during some extra meetingarranged by AMF for this purpose.

Since the tsunami was unexpected we could not organize apanel survey with observations repeated in time, before andafter tsunami, and, therefore, had to rely on a retrospective pa-nel data approach (see McIntosh, Villaran, & Wydick, inpress) specifically tailored for our case. In April 2007 respon-

dents were asked to declare the current and remember the pastlevels of memorable variables by making reference to four dif-ferent periods. We selected periods easy to remember due tothe occurrence of memorable events. The four time windowswe consider are: (P1) the six month interval before the firstmicrofinance loan ever obtained; (P2) the period going fromthe first microfinance loan to the tsunami date (December26, 2004); (P3) the period between the tsunami date and thefirst microfinance loan after tsunami; (P4) the period fromthe first microfinance loan after tsunami to the survey date(April 2007, see Figure 1). Our approach is not free from crit-ical points which we conveniently address.

A first methodological issue in this analysis is the heteroge-neity in time windows of the four different periods since onlytwo time points (the tsunami date, December 26, 2004, and

EVIDENCE FROM THE 2004 TSUNAMI 899
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


3/15

the month in which the survey was administered, April 2007)are common to every interviewed borrower. Consequently,only the first time interval (six months before the first MFIfinancing) is fixed in length, even though not coincident forall respondents. It is, however, important to remark that ourfocus on the effect of post-tsunami refinancing (periods threeand four) limits the heterogeneity of time windows to the refi-nancing date 6 since the other two extremes (beginning of per-iod three and end of period four) are common to allrespondents. Nevertheless, there remains a potential problem

of heterogeneity of time windows which will be controlledfor in two ways. 7 First, we add the length of the windows inmonths as controls in our regressions. Second, and moreimportant, we perform robustness checks of post-tsunami refi-nancing effects by reducing the sample to individuals within aone year difference in monthly length (restriction onlyP4 < 12 months in Tables 7a, 7b, 8, 9, 10). This implies thatwe finally handle a level of heterogeneity not different fromthat of analyses based on yearly data and, by knowing themonthly timing of the refinancing event and controlling forit with the window length, we are actually doing more thanwhat is usually done in such studies.

A second methodological issue is the quality of informationon lagged variables in two respects: the quality of respondentsmemories (events which are more distant in time, such as thefirst loan ever obtained from AMF, are more difficult toremember) and the absence of interview biases. From thispoint of view it is important to underline that part of the infor-mation we use does not come from survey data since the tim-ing and amounts of all loans released by AMF have beenobtained from the MFI official records. Furthermore, we takeinto account the limits, remarked by McIntosh et al. (in press),of memories on indicators such as income by looking at analternative performance indicator such as the more easilymemorable number of weekly worked hours. Our assumptionis that it is easier to remember with precision the length of theworking day (approximated by the discrete number of workedhours) than ones own past income. We preliminarily verifythat there is no significant change in productivity (real income

per hour worked) across the two relevant sample periods and

within subsample groups. The change in worked hours, there-fore, appears to be a good (and more memorable) proxy ofreal income.

3. SUMMARY STATISTICS

With survey data collected on the field we obtain informa-tion on the respondents socio-demographic characteristics,on hours worked and on a series of economic indicators.

Table 1a and b describe the socio-demographic characteristicsand economic variables used in our study. Table 2a reportssummary statistics for selected socio-demographic characteris-tics of AMF sample borrowers.

We can see in this table that slightly less than half of thesample has house and economic activity within one kilometerfrom the coast. Most clients work at home or very close to it,to save money and time (only a minority of families hold amotorbike, almost nobody has a car). Eighty-five percent ofthe sample is composed of women. Most of the clients aremarried and aged around 40 or 50, with complete primaryor incomplete secondary education. Twenty-three percent ofthem (the sum of the men and the widowed women) claimto be the head of the household. Over the four time windows,most borrowers (94%) are self-employed while only 2% areunemployed. A large number of them are involved in trade(46%) and manufacturing (39%), with a significant share(21%) working in the agricultural sector. 8 The average familysize is 4.6, with 2.3 children currently living at home.

Table 2b reports summary statistics for the economic vari-ables. Real and equivalent 9 income exhibits a high variability,while equivalent income in PPP is 5.26$ per day, well abovethe symbolic 1 or 2$ threshold. On average, over the four ana-lyzed periods, 13% of clients had problems to provide dailymeals to their family, with a peak of 26% in the third periodafter the tsunami shock. Most of the families either are unableto save money or save very little. However, a lot of money isoften invested to start or improve a business activity, thusthe actual savings given by investments plus net savings should

be higher.

Figure 1. Time schedule in our tsunami study (P1 = six month interval before first AMF financing; P2 = period ranging from first AMF financing to thetsunami date; P3 = period ranging from the tsunami date to AMF refinancing; P4 = period ranging from the AMF refinancing to the survey date; dotted lines

indicate non-overlapping window borders, continuous lines coincident window borders).

900 WORLD DEVELOPMENT
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


4/15


5/15

Many clients reported damages from Tsunami to raw mate-rials (32%), tools (27%), buildings (25%), and house (19%).Only 4% had family members injured or dead in the catastro-phe. Half of the people in the sample declared that their busi-ness was indirectly damaged by the tsunami because of theworsened macroeconomic situation. One-third of the samplewas forced to use their savings immediately after the tsunamito buy food or repair the damages. After the disaster the gov-ernment, international organizations, and NGOs tried to helpthe populations by providing food, raw materials, medicines,money, etc. In our random sample, only looking at the thirdwindow, 6% of people could rely on remittances from relativesabroad, 32% on governmental subsidies (in most cases a four-month check of Rs. 5,000 per month to buy food), 27% ondonations from international organizations, foreign govern-ments, and NGOs and 3% on other forms of charity or fund-

ing.The last four variables in Table 2b are from bank records,apart from the denominator of the ratio loans (t)/income(t 1) which comes from the survey, and refer to the fourthtime window only, which is the period of interest for oureconometric analysis. MFI borrowers received after the tsu-nami loans equivalent to the income gained in almost ninemonths, with an average number of 3.44 loans obtained sincethe beginning of the relationship with Agro Microfinance.

4. EMPIRICAL ANALYSIS

Our empirical analysis focuses on the effects of MFI refi-nancing on damaged borrowers recovery with preliminary

descriptive evidence on the damage and recovery in the firstplace and, after it, a series of econometric tests on the impactof refinancing which try to solve the usual methodologicalproblems arising in this kind of impact analyses.

(a) Descriptive analysis of the fall and recovery

Even though the focus of the paper is on the role of micro-finance as a recovery tool after calamities and the econometricanalysis will be exclusively run on the period going from thetsunami pre-refinancing to the post refinancing period, we ex-tend the descriptive analysis to the situation before the tsu-nami. We believe that this is a necessary premise to providea description of the background in which the tsunami shockoccurs and to give some insights on the counterfactual (whatwould have happened to the damaged if they were not) in

two ways showing how (i) in the

normal

(shock free) periodsAMF microfinance borrowers tend to improve their economicconditions; (ii) damaged borrowers interrupt the trend whenmoving from the second to the third (post-tsunami pre refi-nancing) period and, after post-tsunami loan refinancing, haveto catch up versus non-damaged who did not suffer the shock.

We start by looking at average changes in the selected vari-ables period by period over the four windows (see Table 3ac).Among the several possible variables, we consider monthlyreal household income (net of loan repayments), daily equiva-lent income in PPP, and the number of hours worked. An indi-rect measure of poverty is created with answers to thequestion: did you have problems in providing daily mealsto your family?. With this respect an important consistencycheck in our data is the strict correspondence between

Table 2. (a) Socio-demographic characteristics and (b) economic variables

(a) Obs. Mean Std. Dev. Min Max (b) Obs. Mean Std. Dev. Min Max

House on coast 1,220 0.44 0.50 0 1 Number of children 1,216 2.38 1.48 0 7Business on coast 1,220 0.46 0.50 0 1 Real income 1,122 19,277 13,540 1080 120,000Galle 1,220 0.31 0.46 0 1 Equivalent real income 1,118 8,351 6,067 327 47,817Matara 1,220 0.52 0.50 0 1 Equivalent real income PPP 1,108 5.26 4.45 0.21 39.02Hambantota 1,220 0.17 0.37 0 1 Standard of living 1,219 2.27 0.96 0 4

Female 1,220 0.85 0.35 0 1 Problems with meals 1,219 0.13 0.34 0 1Age 1,160 48.48 10.15 23 73 Savings 1,200 0.86 1.01 0 4Single 1,220 0.08 0.27 0 1 Van 1,219 0.05 0.21 0 1Married 1,220 0.82 0.39 0 1 Tractor 1,219 0.03 0.17 0 1Widowed 1,220 0.09 0.28 0 1 Motorbike 1,219 0.21 0.40 0 1Divorced 1,220 0.00 0.06 0 1 Bicycle 1,214 0.51 0.50 0 1Separate 1,220 0.01 0.08 0 1 Hours worked 1,220 49.94 27.30 0 100Cohabitant 1,220 0.00 0.00 0 0 Damages to family 305 0.04 0.19 0 1Head of family 1,220 0.23 0.42 0 1 Damages to the house 305 0.19 0.39 0 1No education 1,220 0.35 0.48 0 1 Damages to office buildings 305 0.25 0.43 0 1Primary education 1,220 0.48 0.50 0 1 Damages to working tools 305 0.27 0.45 0 1Secondary education 1,220 0.16 0.37 0 1 Damages to raw materials 305 0.32 0.47 0 1Full time 1,219 0.02 0.13 0 1 Damages to the market 305 0.49 0.50 0 1Part time 1,219 0.02 0.14 0 1 Number of damages 305 1.56 1.67 0 6Self-employed 1,219 0.94 0.23 0 1 Savings withdrawn 300 0.32 0.47 0 1

Unemployed 1,219 0.02 0.15 0 1 Remittances 305 0.06 0.23 0 1Student 1,219 0.00 0.06 0 1 Subsidies 304 0.32 0.47 0 1Retired 1,219 0.00 0.03 0 1 Donations and grants 305 0.27 0.44 0 1Agriculture 1,219 0.21 0.41 0 1 Other charity/fundings 305 0.03 0.18 0 1Fishery 1,219 0.02 0.15 0 1 Etimos 305 0.34 0.48 0 1Manufacturing 1,219 0.39 0.49 0 1 Loan to income ratio 261 8.94 11.17 .12 89.51Trade 1,219 0.46 0.50 0 1 Senioritya 289 3.44 1.25 2 8Other sector 1,218 0.09 0.28 0 1 Lengtha 285 10.51 5.93 1 26People in the house 1,216 4.61 1.57 1 12

a Data from bank records, apart from income which comes from the survey.



6/15

equivalent household income and declared existence of prob-lems in providing daily meals to the family. The average equiv-alent income in PPP for those declaring such problems is 2.74$against a value of 5.58$ for those not declaring them (the dif-ference in means being highly significant).

The first approach we use here is simply a test on the signif-icance of the change in the indicator from a period to the nextone (t-statistics are reported in brackets). In Table 3a we ob-serve, for the overall sample, a (slight) amelioration in severaleconomic indicators in the second period (P2), a fall after thetsunami (P3) and a recovery in the last time window (P4). Thispattern is obviously stronger if we look at the subsample ofdamaged individuals (Table 3c), while the tsunami effectalmost disappears in the non-damaged subsample (Table 3b)since economic indicators of the latter stop growing but donot display any reduction in the third period. 10

The strong impact of the tsunami shock on the full sample isevident from the descriptive findings presented in Table 3a.Real average income falls by Rs. 5,556 (a 25% reduction withrespect to sample average), daily equivalent income in PPP by1.67$ (a 27% reduction) and the probability of having prob-lems in providing daily meals rises by 18% (from 8% in P2to a worrying 26% in P3). Worked hours fall by nine units.The fourth column ofTable 3a provides a test on the statisticalsignificance of the changes from the second to the fourth per-iod for the full sample. This can be considered as a test to ver-ify whether the economic indicators in the fourth period (P4)recovered to pre-tsunami levels (P2). The real income in P4

is not statistically different from that in P2 and the standard

of living in terms of consumption goods improves. Non-dam-aged people significantly improve their economic situation,while their hours worked remain unchanged. Damaged indi-viduals increase the number of hours worked but do not fullyrecover the pre-tsunami purchasing power.

It is also important from a descriptive point of view toexamine not just average values but also the dynamics of theentire distribution of selected well being indicators across thefour time intervals. Figure 2 clearly documents the downwardshift of the cumulative distribution of real household incomeand weekly worked hours in the third period for the wholesample and the subsample of the most damaged people. Thecomparison of cumulative distributions in the four periodsshows that the shock and the catching up effect do not act onlyon the mean of the subsample distribution of the selected eco-nomic indicators, but almost on any point of the distribution,with special reference to its low tail. In fact, it is clear that thepoorest are both the most damaged and those registering themost significant recovery across the tsunami and refinancingperiods, respectively. In the third time interval all points ofthe cumulative distribution function are not above those ofthe second time interval so that first order stochastic domi-nance of the former on the latter is evident just from this pic-ture. By focusing on the subsample of people with at leastthree damages the fall in economic indicators after the tsunamiand the more pronounced drop in the left part of the distribu-tion is even more evident.

Table 4 tests period by period the difference in the mean

of each variable between damaged and non-damaged

Table 3. (a) Changes in mean of selected indicators, full sample. (b) Changes in mean of selected indicators, non-damaged respondents only. (c) Changes inmean of selected indicators, damaged respondents only

Variable P2P1 P3P2 P4P3 P4P2

(a)

Change in real income 4273.118 5556.833 4441.441 1066.862(7.18) (7.04) (7.22) (1.51)

Change in Eq. Real Income PPP 1.3792 1.675444 1.409149 0.2225023

(7.49) (7.07) (6.62) (0.93)Change in hours worked 7.006557 9.203279 11.10164 1.898361

(6.50) (7.13) (8.65) (2.36)Change in problems with meals 0.0263158 0.1836066 0.1704918 0.0131148

(1.89) (7.49) (7.26) (0.78)

(b)

Change in real income 3972.47 1255.463 2908.87 1795.856(5.33) (1.55) (3.00) (1.70)

Change in Eq. real income PPP 1.257967 0.4536367 0.8078155 0.4296515(5.46) (1.80) (3.43) (1.38)

Change in hours worked 8.342857 1.933333 0.7904762 1.142857(4.85) (1.48) (0.58) (0.70)

Change in problems with meals 0.0380952 0.0380952 0.047619 0.0095238(1.65) (1.42) (1.68) (0.38)

(c)Change in real income 4431.439 8037.377 5333.118 2529.337

(5.40) (7.24) (6.79) (2.79)Change in Eq. real income PPP 1.442423 2.381209 1.759619 0.5557458

(5.69) (7.16) (5.77) (1.72)Change in hours worked 6.305 13.02 16.515 3.495

(4.59) (7.28) (9.71) (4.06)Change in problems with meals 0.0201005 0.26 0.235 0.025

(1.16) (7.78) (7.42) (1.15)

Note: Change in monthly real income is expressed in 2007 Sri Lankan Rupees, while the change in equivalent real income in PPP in US Dollars. Thechange in weekly worked hours is the difference between hours of current and previous periods. The variable capturing the change in food problems isequal to one if in the previous period providing daily meals to family members was not a problem while in the current period it was. T-statistics are inparentheses. Variable legend is in Table 1a and b.

http://-/?-http://-/?-http://-/?-


7/15

respondents. Our two main questions here are whether the twogroups were significantly different before and after the tsunamiand whether there has been a complete recovery in the fourthperiod. The homogeneity between damaged and non-damagedborrowers, all clients of the same MFI, is confirmed by the factthat all the indicators were not significantly different amongthe two groups at 5% level before the tsunami (P1 and P2).

In the third period all the means become strongly different,while in the fourth period there is a partial convergence ofdamaged people to the levels of non-damaged ones. In fact,the number of hours worked and the probability of havingproblems in providing daily meals of the two subsamples arenot different at 5% level, while the difference still exists forthe remaining variables. However, it must be underlined that

Figure 2. Cumulative distributions of real income and hours worked. Note: Ventiles of monthly real income and weekly worked hours are calculated withrespect to the pre-microfinance (P1), the pre-tsunami (P2), the tsunami (P3) and the refinancing (P4) time windows.

Table 4. Difference in mean of selected indicators between non-damaged and damaged respondents

Variable P1 P2 P3 P4

Real income 1037.945 566.9911 7139.124 4905.54(0.65) (0.30) (5.03) (2.93)

Eq. real income PPP 0.5749012 0.4018226 2.192877 1.326885(1.08) (0.65) (5.05) (2.29)

Hours worked 0.9738095 1.064048 12.15071 3.57381

(0.29) (0.36) (3.50) (1.15)Problems with meals 0.0007657 0.0183333 0.2402381 0.0528571

(0.02) (0.56) (4.68) (1.52)

Note: Monthly real income is expressed in 2007 Sri Lankan Rupees, while the change in equivalent real income in PPP in US Dollars. Weekly hoursworked are average number of worked hours per week. The variable capturing the food problems is equal to one if the respondent had problems inproviding daily meals. T-statistics are in parentheses. Variable legend is in Table 1a and b.



8/15

the gap has enormously reduced: the size and the statisticalsignificance of the difference in the variables are much lower

in P4 than in P3 documenting a partial, but still incomplete,catch up.

(b) The post-tsunami refinancing effect of microfinance

In order to evaluate the effectiveness of microfinance as arecovery tool after tsunami we test its impact measured bythe loan to income ratio, that is, the amount of the first loanobtained after the tsunami divided by the post-tsunami pre-refinancing monthly income on performance.

As performance indicators we consider two variables: (i) thechange in real income from the post-tsunami pre refinancingto the post refinancing period (from P3 to P4); and (ii) thechange in the number of hours worked in the same interval.

Since we are aware of the criticism on the use of past incomein retrospective panel data (McIntosh et al. (in press)) weintroduce the second indicator in addition to income. Ourassumption is that it is much easier to remember the lengthof the working day (approximated by the discrete number ofworked hours) than ones own past income. 11

Before using the change in hours worked we test whetherthere are significant reductions in productivity (real incomeper hour worked) across the sample period. We indeed findthat differences are not significant also between and withintreatment and control group in post tsunami pre and post refi-nancing periods. We are, therefore, encouraged to considerchanges in hours worked after post tsunami refinancing asgood proxies of changes in income.

(i) Descriptive evidence on the effect of MFI refinancingWe start in Table 5 by evaluating from a descriptive point of

view the correlation between the loan to income ratio and ourperformance variables. The percent change in income from P3to P4 for those with a loan to income ratio below the samplemedian is 21.8% against 86.4 for those above it (and goes up to120% for borrowers with a loan size to income ratio above the75th percentile). In the same direction, the (absolute) changein weekly worked hours for those receiving after tsunami aloan to income ratio below the sample median is 4.9, against15.6 for the complementary sample, the difference being signif-icant at 1% (and up to 18.8 if we consider borrowers with aloan size to income ratio above the 75th percentile).

Since we are arguing that the MFI loan is a useful recoveryinstrument for tsunami victims, as a robustness check we limit

the sample to individuals who reported damages. This helps usalso to control for heterogeneity problems between treatment

and control samples which may be related to the size and theuse of the loan. Our descriptive results on the two performancevariables of interest remain strong and significant. By calculat-ing the new median loan to income ratio for the subsample oftreatment (damaged) group we find that the change in real in-come is around 110% for the individuals whose loan to size ra-tio is above the median against around 31% for those belowthe median. Similarly, the change in (weekly) worked hoursis 21.6 for people above and 9.3 for people below the median.The effect of MFI refinancing described here with simple dif-ferences in means requires to be controlled for composition ef-fects, heterogeneity between treatment and control groups andin time windows, and various selection effects. This is what wewill try to do in the following sections.

(ii) Econometric evidence: the simultaneous pattern of effectsWhen evaluating econometrically the effect of the loan to in-

come ratio on performance indicators we need to correct forconcurring factors affecting it. Intuitively, we want to avoidour findings to be driven by spurious effects related to a thirdunobserved factor, correlated with both the loan to income ra-tio and the performance indicator, which may generate a spu-rious positive correlation between the two. A strong suspectmay be borrowers seniority which affects both timing and sizeof the loan and may proxy unobservable borrowers quality. Asecond concern is about selection effects on the timing and rel-ative size of the loan. With regard to the first, we find that tim-ing is affected by the number of borrowers previous loans withthe MFI and the sum of damages. As for the second, we ob-

serve that the loan to income ratio is affected by three factors:the number of damages, the post-tsunami pre refinancing levelof income (negatively), and the number of previous microfi-nance loans. 12 This implies that, on the one side, the lenderlooks at reputation and previous record of the borrower while,on the other side, he is not insensitive to the tsunami emer-gency and tries to counteract two of its effects (low incomeand number of damages). 13

To sum up, the most damaged and those with longer senior-ity receive first and more with respect to their income. How-ever, as we will see in the next paragraph, while the loan toincome ratio affects performance, timing is not significant.Moreover, both timing and loan to income ratio are not re-lated to proxies of past performance or positively related toproductivity of the pre-tsunami period (the relative size is

Table 5. Percent change in real household income and weekly worked hours for individuals whose loan to income ratio was above and below the median

Variable Full sample Damaged borrowers only Non-damaged borrowersonly

Below themediana

Above themediana

Below themediana

Above themediana

Below themediana

Above themediana

(1) Percent change in real income

Mean 21.8 86.40 30.97 101.96 11.44 29.58Min. Conf. Int. 13.4 56.30 16.72 63.33 0.66 7.90Max. Conf. Int. 30.1 117.20 45.22 140.59 22.22 51.26Obs. 131 130 83 82 48 48

(2) Absolute change in worked hours

Mean 4.9 15.6 9.3 21.6 2.96 7.26Min. Conf. Int. 2.5 11.7 5.0 15.4 2.79 8.27Max. Conf. Int. 7.2 19.5 13.1 29.3 8.71 22.78Obs. 130 130 81 81 49 49

a Average (first post tsunami) loan to (post-tsunami pre-refinancing monthly) income ratio below the median in period P4.



9/15

indeed weakly negatively correlated with such productivityvariable). 14

In order to take into account this complex pattern of linksamong variables we estimate a simultaneous (SUR) 15 threeequation system which takes into account the impact of deter-minants of loan to income ratio, timing of the refinancing andour performance variable and the possible correlation of resid-uals among the three single equations. We, therefore, estimatethe following system:

lnYi;t lnYi;t1 a0 a1 lnLi;t=Ii;t1 a2Lengthi;t1

Xj

a3jX1ji;t e1i;t 1:1

lnL=Ii;t b0 Xk

b1kX2k;it e2i;t 1:2

Lengthi;t c0 Xl

c1lX3l;it e3i;t 1:3

where Y is the real household income (Table 6, columns 1 and2) or, alternatively, weekly worked hours (Table 6, columns 3and 4) and the dependent variable in 1.1 is the log difference ofY, L is the size of the first AMF loan obtained after the tsu-nami, I is the real household income, Length is the length ofthe post-tsunami pre-refinancing time window (the distancein month from the tsunami date to the first MFI refinancing)and X(1), X(2), and X(3) are the other regressors of the threeequations, respectively, as described in Table 6 (and detailed inthe legend of Table 1a and b).

A first concern is whether a strong correlation amongregressors may distort the magnitude of our coefficients. We,however, find that there are no serious multicollinearityproblems in our estimate. The highest VIF 16 factor among

Table 6. SUR estimates

Variable (i) (ii)

Coef. z Coef. z

Eqn. (1) Dep. Var.: log diff. of real income Dep. Var.: log. diff. of worked hours

Galle 0.16 1.77 0.22 1.00Matara 0.06 0.72 0.13 0.62

Agriculture 0.12 1.59 0.11 0.62Fishery 0.05 0.28 0.22 0.49Manufacture 0.03 0.44 0.06 0.40Age 0.00 0.31 0.00 0.46Female 0.02 0.26 0.23 1.14Primary education 0.09 1.43 0.00 0.02Secondary education 0.02 0.26 0.59 2.72Number of children 0.00 0.16 0.02 0.43Number of damages 0.03 1.27 0.15 2.90Remittances 0.13 0.64 0.19 0.39Subsidies 0.10 0.87 0.06 0.19Donations and grants 0.29 1.09 0.57 0.89Other charity 0.04 0.64 0.23 1.47

Ln (Loan to income ratio) 0.25 7.97 0.17 2.25

Length 0.01 0.94 0.01 0.56

Seniority 0.15 2.78 0.13 0.95Constant 0.09 0.45 0.72 1.41

Eqn. (2) Dep. Var.: log of loan to income ratio Dep. Var.: log of loan to income ratio

Female 0.44 3.34 0.45 3.35Number of damages 0.08 2.39 0.07 2.02Real income (t 1) 0.84 12.21 0.79 10.90Seniority 0.77 10.93 0.78 10.67Constant 8.65 12.47 8.18 11.18

Eqn. (3) Dep. Var.: time window length Dep. Var.: time window length

Number of damages 0.51 2.40 0.49 2.12Number of loans 3.72 7.64 3.69 7.31Constant 10.80 11.97 10.83 11.48

Equation Obs. R2 Obs. R2

Eqn. (1) 247 0.22 230 0.13Eqn. (2) 247 0.55 230 0.53Eqn. (3) 247 0.20 230 0.19Residuals correlation Correl. Correl.

Correlation (e1, e2) 0.53** 0.19**

Correlation (e1, e3) 0.28** 0.06**

Correlation (e1, e3) 0.84** 0.84**

Note: SUR regressions make use of heteroskedasticity-robust standard errors.** Significant at 99%. Variable legend: see Table 1a and b.



10/15

variables used as regressors in the three equations is belowthree (in the first equation). 17

(iii) Significance and magnitude of our resultsEstimate results from the SUR in Table 6 show that residu-

als of the three system equations are significantly correlatedthereby justifying the simultaneous estimate. The loan to in-come ratio affects positively and significantly the change inreal income (Table 6, columns 1 and 2) confirming evidencefrom subgroup difference in means shown in Table 5. Themagnitude of its effect is such that, by doubling the loan sizeto income ratio from its average level we obtain a change inincome of 25% in the SUR estimate. Furthermore, the vari-ables affecting timing and size of the post tsunami refinancingare all significant in the second and third equation of thesimultaneous model, confirming the previously explored pat-tern of relationships.

The results in column 1 are paralleled by an analogous sig-nificant effect of the loan to income ratio on the percent

change in worked hours (Table 6, columns 3 and 4) with

17% elasticity (to correctly frame this result consider that themean change from P3 to P4 is 11 hours and the average num-ber of worked hours in P3 is 44.3).

(iv) Robustness checks: lagged income, time window heteroge-neity, and treatment sample only

Since we are confident that all other potential factors sup-porting convergence of damaged to non-damaged borrowersare conveniently captured by our regressors (which includeremittances, grants, etc.) we do not include the lagged valuesof real income and weekly worked hours in our base specifica-tion presented in Table 6. As a first robustness check we, how-ever, correct for the possibility of a convergence process notfully captured by our regressors by including such variablein our specification. As a consequence the impact of the loanto income ratio is still significant but lower in magnitude (Ta-bles 7a and 7b).

A second robustness check is on heterogeneity in the timingof MFI refinancing. When focusing on the effects of post-tsu-

nami refinancing the problem of time window heterogeneity

Table 7a. Robustness checks on the SUR model (dependent variable: log difference of real income)

Check Nr. One year refinancingwindow

Damaged borrowersonly

Lagged real incomeincluded

(From general)to specific

Obs. Pseudo R2 of Eq. (1.1) Loan toincome ratio

Coef. T-stat

1 No No Yes No 247 0.33 0.05 1.962 No Yes Yes No 156 0.43 0.05 1.94

3 No Yes No No 156 0.26 0.30 7.474 Yes No Yes No 180 0.36 0.10 2.175 Yes No No No 180 0.27 0.29 7.86 Yes Yes Yes No 115 0.49 0.05 1.957 Yes Yes No No 115 0.33 0.35 7.628 No No Yes Yes 258 0.28 0.07 5.79 No Yes Yes Yes 164 0.33 0.10 2.7410 Yes No Yes Yes 189 0.32 0.09 2.7411 Yes Yes Yes Yes 121 0.39 0.13 3.21

Note: Robustness checks are built on regression in Table Table 6, column 1, with additional restrictions. The possible restrictions are (i) only individualswhose P4 window is shorter than 12 months, (ii) only people damaged from the Tsunami, (iii) ln of monthly real income of the previous period included ascontrol and (iv) only significant regressors included in the regression. Loan to income ratio: amount of the first loan obtained after the tsunami divided bypost-tsunami pre-refinancing monthly income. The pseudo R2 refers to the first of the three equation of the SUR model. Estimates make use ofheteroskedasticity-robust standard errors.

Table 7b. Robustness checks on the SUR model (dependent variable: log difference of hours worked)

Check Nr. Only P4 < 12months

Damagedborrowers only

Lagged workedhours included

(From general) to specific Obs. Pseudo R2 of Eq. (1.1) Loan toincome ratio

Coef. T-stat

1 No No Yes No 230 0.36 0.09 1.942 No Yes Yes No 140 0.40 0.20 1.903 No Yes No No 140 0.18 0.30 2.404 Yes No Yes No 167 0.39 0.07 1.755 Yes No No No 167 0.16 0.23 2.256 Yes Yes Yes No 103 0.45 0.21 1.977 Yes Yes No No 103 0.24 0.41 2.608 No No Yes Yes 239 0.30 0.12 2.07

9 No Yes Yes Yes 146 0.33 0.17 1.9210 Yes No Yes Yes 174 0.30 0.18 2.2211 Yes Yes Yes Yes 107 0.36 0.26 2.21

Note: Robustness checks are built on regression in Table 6, column 2, with additional restrictions. The possible restrictions are (i) only individuals whoseP4 window is shorter than 12 months, (ii) only people damaged from the Tsunami, (iii) number of worked hours of the previous period included as controland (iv) only significant regressors included in the regression. Loan to income ratio: amount of the first loan obtained after the tsunami divided by post-tsunami pre-refinancing monthly income. The pseudo R2 refers to the first of the three equation of the SUR model. Estimates make use of heteroske-dasticity-robust standard errors.

http://-/?-http://-/?-http://-/?-http://-/?-


11/15

reduces to the timing of the post-tsunami loans since the othertwo time points (tsunami and interview dates) are given andequal for all respondents. To reduce it further we restrict thesample to borrowers receiving the first post-tsunami loan with-in a one year interval (between 10 and 22 months from the tsu-nami) dropping around 27% of the observations.

Consider that, by doing so, we handle a level of heterogene-ity not different from that of analyses based on yearly datawhen events are irregularly spaced during the same year. Fur-thermore, by knowing the monthly timing of the refinancingevent and controlling for it with the window length, we aredoing more than what is done in studies on yearly data. Tables7a and 7b show that the significance of the loan to income ra-tio persists when we consider this restricted sample. The elas-

ticity on the change in income is around 10% when we includelagged income and 29% if we do not include it (Table 7a,checks 4 and 5). The corresponding numbers are 7% and23%, respectively, for the change in worked hours (Table 7b,checks 4 and 5).

Consider further that heterogeneity in the loan to income ra-tio and in the loan purpose between treatment and controlsamples may distort our previous findings. To avoid this we re-strict the sample to those individuals reporting damages andfind that elasticities are up to 30% for both performance indi-cators in the base specification without the lagged dependentvariable (Tables 7a and 7b, checks 3 and 7), while are muchsmaller but still significant when we include it (Tables 7aand 7b, checks 2 and 6). When considering jointly the time

Table 8. The different refinancing effects on damaged versus non-damaged borrowers

Variable Log difference of real income Log difference of worked hours

(i) (ii) (iii) (iv)

Galle 0.1412 0.1315 0.1532 0.2518(1.21) (1.80) (0.94) (0.62)

Matara 0.0223 0.1545 0.1053 0.2052

(0.93) (0.76) (0.92) (0.93)Agriculture 0.1142 0.1261 0.0764 0.1962(1.03) (0.91) (0.98) (1.53)

Fishery 0.0754 0.0064 0.2164 0.5295(0.71) (0.08) (0.56) (1.81)

Manufacture 0.0215 0.0215 0.1048 0.2514(0.42) (0.71) (0.96) (0.35)

Age 0.0013 0.001 0.0021 0.0083(0.62) (0.76) (0.25) (1.01)

Female 0.0051 0.0352 0.1981 0.0732(0.42) (0.91) (0.71) (0.06)

Primary 0.0632 0.0941 0.0062 0.0628(1.21) (1.21) (0.08) (0.52)

Secondary 0.0012 0.0062 0.3625 0.9526(0.04) (0.07) (1.81) (1.71)

Number of children 0.0051 0.0652 0.0752 0.1503(0.11) (0.76) (0.91) (1.03)DAM 0.084 0.0072 0.2514 0.0963

(0.89) (0.62) (1.75) (1.61)Remittances 0.0241 0.0732 0.1051 0.3215

(0.88) (0.73) (0.81) (0.91)Subsidies 0.1517 0.0752 0.0863 0.0521

(1.66) (0.63) (0.81) (0.07)Donations and grants 0.3163 0.2615 0.5262 0.5377

(1.91) (1.62) (1.62) (1.02)Other charity 0.0412 0.0744 0.2512 0.2142

(0.35) (0.88) (1.52) (1.21)Loan to income ratio 0.0891 0.1351 0.0632 0.0632

(2.14) (2.51) (0.61) (0.61)Ln [1 + Loan to income ratio DAM] 0.1564 0.1281 0.1504 0.1841

(2.38) (1.89) (1.94) (2.05)Length 0.0083 0.0072 0.0065 0.0521(1.71) (0.25) (0.51) (0.58)

Seniority 0.052 0.1062 0.0523 0.0352(1.14) (1.36) (0.35) (0.61)

Constant 0.936 0.1062 0.4623 1.2518(0.21) (0.91) (1.41) (1.61)

Only P4 < 12 months No Yes No YesObs. 247 180 230 167(Pseudo) R2 0.2813 0.3105 0.1931 0.1942

Note: OLS regressions with robust standard errors in parenthesis. Ln [1 + Loan to income ratio DAM] is the log of a slope dummy variable, being theproduct of the loan to income ratio (amount of the first loan obtained after the tsunami divided by post-tsunami pre-refinancing monthly income) and adummy variable (DAM) equal to 1 if the respondent was damaged by the tsunami. Variable legend: see Table 1a and b.



12/15

window heterogeneity and treatment-sample-only restrictionswe still have strong and significant results with elasticitiesabove 20% in the specification without lagged dependent var-iable (Tables 7a and 7b, check 7). Finally, given the limitednumber of observations and the large number of potentialcontrols included in our base estimates, as in a general to spe-cific approach we re-estimate our model including only vari-ables whose significance is above the 95 percent threshold tomake the effect of our MFI refinancing variable cleaner. Wedo that for the overall sample and for all the above com-mented robustness checks. In all of these cases the impact ofthe loan to income ratio remains strong and significant (Tables

7a and 7b, checks 811).

(v) The differential effect of the loan to income ratio on damagedversus non-damaged borrowers

To evaluate the impact of microfinance we are finally inter-ested to test whether, ceteris paribus, loans received by dam-aged borrowers were more productive than those received bynon-damaged ones. If the null of no significant difference is re-

jected we can show that the loan is relatively more importantwhen hit by natural calamities. To perform such test we esti-mate with OLS the first equation of the SUR model and intro-duce as additional regressor the size to income variableinteracted with the damage dummy (DAM (L/I)) whereDAMtakes value of one if the borrower suffered from tsunami

damages or zero otherwise.

18

The null of no different impact

of the MFI loan on damaged versus non-damaged borrowersis rejected if the added variable is significant. More specificallywe estimate the following specification:

lnIi;t lnIi;t1 a0 a1 lnLi;t=Ii;t1 a2

ln 1 DAMi Li;t=Ii;t1

a3Lengthi;t Xj

a4jX1ji;t e1i;t 2

for real income (I) or:

lnHi;t lnHi;t1 a0 a1 lnLi;t=Yi;t1 a2

ln 1 DAMi Li;t=Ii;t1

a3Lengthi;t Xj

a4jX1ji;t e1i;t 20

for worked hours (H) and test the following null hypothesisH0 : a2 0. If we reject the null on the positive size we havethat the loan to income ratio affected significantly more theperformance of damaged versus non-damaged borrowers.Such test is almost free from selection problems since we dem-onstrated that the tsunami acted as a random shock which di-vided a homogeneous set of borrowers from the same MFIinto a treatment and a control sample. We know from Table 6that the size of the loan is affected by some determinants, suchas gender, number of damages, seniority, and lagged real

Table 9. Robustness checks on regressions with slope dummy variables for damaged people

Dependent variable Log difference of real income Log difference of worked hours

(i) (ii) (iii) (iv)

Ln [1 + Loan to income ratio DAM] 0.1214 0.1215 0.1351 0.1914

t-statistic 2.16 1.91 2.41 1.92Ln of real income (t 1) Yes Yes

Worked hours (t 1) Yes YesOnly P4 < 12 months No Yes No YesObs. 247 180 230 167R2 0.3315 0.3516 0.3721 0.3806

Note: Results are robustness checks of regressions in Table 8, with lagged values of the log of monthly real income/weekly worked hours and restrictionson the time window length. Ln [1 + Loan to income ratio DAM] is the log of a slope dummy variable, being the product of the loan to income ratio(amount of the first loan obtained after the tsunami divided by post-tsunami pre-refinancing monthly income) and a dummy variable (DAM) equal to 1 ifthe respondent was damaged by the tsunami. T-statistics are calculated according to heteroskedasticity-robust standard errors.

Table 10. Treatment effect models

Robustness check Log difference of real income Log difference of worked hours

(i) (ii) (iii) (iv) (v) (vi) (vii) (viii)

Ln [1 + Loan to income ratio DAM] 0.1275 0.1553 0.1235 0.1311 0.1732 0.1362 0.2159 0.2642t-statistic 2.90 3.16 2.90 3.23 2.52 2.13 2.51 2.05Ln of real income (t 1) Yes No Yes No Yes No Yes NoOnly P4 < 12 months No No Yes Yes No No Yes Yesv2 (1)a 2.09 2.13 2.66 2.17 3.12 1.27 2.61 0.92

Prob > v2 0.15 0.14 0.10 0.16 0.08 0.21 0.14 0.32Obs. 247 247 180 180 230 230 167 167Wald v2 131.2 91.3 117.1 87.6 82.1 35.1 63.3 37.1

Note: Results are from treatment regression estimates where the main equations remain a those shown in (2) and ((20)) in the text while in the selectionequation (see Eqn. (3)), participation to the treatment group (the Damage dummy equal to 1 if the respondent was damaged from the tsunami) is regressedon Agriculture and House on coast, the two above mentioned variables found to be significantly different between the treatment and the control samples.Ln [1 + Loan to income ratio DAM] is the log of a slope dummy variable, being the product of the loan to income ratio (amount of the first loanobtained after the tsunami divided by post-tsunami pre-refinancing monthly income) and a dummy variable (DAM) equal to 1 if the respondent wasdamaged by the tsunami.a LR test of independent equations (correlation of residuals of the two equations = 0).



13/15

income. We also take the point that there may be some omit-ted variables which still affect both loan size and performance(i.e., better business opportunities). For this reason we devisein Table 8 a test which should be immune from the selectionbias generated by the endogeneity of the loan to income ratiosince we measure with the slope dummy (damage dummy lo-an to income ratio) the presence of a significantly different ef-

fect of the loan for damaged versus non-damaged borrowersconditional to the same loan size. Findings from our estimateshow that the base effect with the full sample estimates isaround 9% to which is added an additional 16% for damagedborrowers (Table 8, column 1). The significance of our resultsis confirmed when restricting the sample to borrowers receiv-ing the first post-tsunami loan in a one year interval in orderto correct for time window heterogeneity (Table 8, column2). Results on our second performance variable (the numberof hours worked) are consistent with previous findings show-ing that the elasticity of the impact of the loan to income ratiofor damaged people is now between 15% (Table 8, column 3)and 18% (one year window of loan refinancing, column 4). Fi-nally, Table 9 repeats the exercise by adding lagged income orlagged weekly worked hours to regressions in Table 8, with

findings which remain consistent with those shown in Table 8.

(vi) Robustness checks with treatment regressionEven though the tsunami acts as a random shock on an ex

ante homogenous group of borrowers from the same MFI (seeTable 4), two characteristics (house on the coast and agricul-tural activity) may still in principle discriminate between them.In order to check whether such differences create a selectionproblem we re-estimate the model with a treatment regressionapproach in which main equation and selection equation are

jointly estimated. The main equations remain as those shownin (2) and (20) while in the selection equation, participation tothe treatment group (the Damage dummy) is regressed on thetwo above mentioned variables which we have found to be sig-

nificantly different between the treatment and the control sam-ples:

Damagei b0 b1Agriculturei b2House:on:coasti vi 3

with Agriculture being a dummy for borrowers working inagriculture and House on coast for those living not fartherthan one km from the coast. 19 Note that, in order to meetthe requirement of using selection variables not affecting ourperformance indicator, in both estimates we use regressorswhich revealed themselves not to be correlated with the depen-dent variable in single equation estimates.

Two characteristics of our new results are important(Table 10). First, findings are not substantially different inmagnitude with respect to those in the SUR. Second the nullof no correlation of residuals between the main and the selec-

tion equations is not rejected. As a consequence, the only twofactors which make treatment and control samples different(location of the house on the coast and agricultural activity)do not generate selection effects on performance. This rein-forces the validity of our test on the differential effect of the

loan to income ratio on damaged versus non-damaged bor-rowers documented in Table 8.

5. CONCLUSIONS

Our paper examines the role of MFI loans as an effective

recovery tool after tsunami for a sample of 305 randomly se-lected clients of a Sri Lankan MFI. In order to reconstructtime series we devise an ad hoc retrospective panel data ap-proach by asking interviewed borrowers to declare their cur-rent and remember their past economic levels. We carefullycontrol for the reliability of our data by combining surveydata and bank records, comparing results on income withthose on a more memorable performance indicator like thenumber of weekly worked hours, looking at validating evi-dence across different indicators and controlling for heteroge-neity of time windows and characteristics between treatmentand control samples.

Our main findings are that (i) the post-tsunami loan to in-come ratio has a significant effect on the borrowers recoverymeasured in terms of change in income or in worked hours

and (ii) the effect of the loan to income ratio is significantlystronger for damaged versus non-damaged borrowers. Toevaluate the robustness of our results we control for selectionbias on loan to income size, heterogeneity in loan timing andcharacteristics between treatment and control borrowers.

Based on our findings we identify four elements supportingthe usefulness of the donors intervention on the MFI and ofthe microfinance loans after tsunami in our specific case: (i) thedeterioration of the MFI portfolio after tsunami was such thatit is hard to imagine that it could have continued to operatewithout recapitalization (this starting point is based on theofficial certified losses of the MFI); (ii) the MFI contrastedthe negative effects of tsunami by granting to damaged bor-rowers and to borrowers with lower income higher loans in

proportion to their level of income. This is far from beingobvious (even though for individuals with lower income theloan to income ratio could tend to be higher) since damagesand falls in income negatively affect the collateralizationcapacity and, therefore, the creditworthiness of the borrowers.Finally, our econometric test on the relative productivity ofthe loan shows that (iii) the loan significantly affected workedhours and real income for damaged borrowers after recoveryand (iv) the loan is relatively more important when hit bycalamities since the effect on damaged borrowers is higherthan that on non-damaged ones.

Our results provide food for reflection on new ways forusing donors resources after calamities. In parallel to the di-rect provision of food, investment goods, or infrastructure,fundamental to address emergency needs and rebuilding,

recapitalising MFIs under stress after calamities may providean effective liquidity injection by acting as a sort of expansivemonetary policy measure for the poor. Such measure can re-start and stimulate economic activity with significant effectsin terms of both worked hours and income creation.

NOTES

1. Developing countries are particularly vulnerable to the effects ofclimate variability, climate change (Smith, Klein, & Huq, 2003) and othernatural catastrophes like earthquakes and tsunamis. In absolute terms,natural catastrophes disproportionately affect low and middle incomecountries. Available data (see World Bank, 2001) show that, between 1990and 1998, 94% of natural disasters and 97% of deaths from natural

catastrophes occurred in less developed countries (LDCs).

2. http://www.usaid.gov/locations/asia_near_east/tsunami/.

3. The number of families involved and the amount of loans provided arevery similar to those of the MFI we considered in our study in Sri Lanka.Our empirical results on Sri Lankan borrowers confirm the qualitativeevidence on the usefulness of microcredit as a recovery tool already fund

in Thailand by USAID in a very similar setting.

http://-/?-http://-/?-http://www.usaid.gov/locations/asia_near_east/tsunami/http://www.usaid.gov/locations/asia_near_east/tsunami/http://-/?-http://-/?-


14/15

4. Sri Lanka was severely affected by the tsunami of the December 26,2004, even though it was not the most damaged country. The disaster wastriggered by an earthquake measuring 9.0 on the Richter scale whoseepicenter was located in the Indian Ocean near Banda Aceh in Sumatra.The giant wave struck a coastal area stretching over 1,000 kilometers, ortwo-thirds of the countrys coastline (from Jaffna in the North to the westcoast, North of Colombo). Human losses were huge (over 35,000 dead and443,000 displaced people, see Athukorala & Resosudarmo, 2005).Economic losses were also extremely severe since the wave damaged24,000 boats (about 70% of the fishing fleet), 11,000 businesses and 88,500houses, of which more than 50,000 were completely destroyed.

5. The donors still have property rights on the original fund which is nowpart of the equity of the MFI.

6. The length of the third interval is six month for the first quarter of thesample, 10 months for half of it and 15 months up to 75% of the sample.We do not observe significant differences between the two groups in termsof duration of the third time window.

7. The estimation of the significance of a common event in a sample ofnon-synchronous events is the typical focus of event studies in finance (for

a standard treatment see Campbell, Lo, & McKinlay, 1997). In thosestudies non-synchronicity concerns the event date and abnormal returnsare calculated on the basis of a normal return model estimated in theperiod preceding the event window.

8. Some people had more than one economic activity, thus the total sumexceeds one.

9. Under the current OECD rule, earnings are divided by a scale factorA, where A = 1 + 0.5(Nadults 1) + 0.3Nchildren. However, in our samplea large part of consumption is food consumption. It is, therefore,advisable to reduce the extent of economies of scale by increasing weightsin the equivalence scale. We, therefore, decide to follow the standardsuggestion in development studies of giving unit weights to each member(for a discussion of the methodological problems in creating equivalencescales see Deaton & Paxson, 1998).

10. We perform a robustness check with non parametric analysis of thedifference in the change of the considered economic indicators betweentreatment and control samples which confirms significance of parametricfindings. Results are omitted for reasons of space and available uponrequest.

11. Consider as well that, in this question, the interviewer made a cleardistinction between paid and unpaid work and reduced respondentsmnemonic effort by asking worked hours per working days and inweekends.

12. Estimates are omitted for reasons of space and available uponrequest. The above described pattern of relationships will be, however,visible in the SUR estimates which follow.

13. We cannot provide evidence whether the intention to support morethe most damaged borrowers is affected by constraints posed by donorswho provide equity capital to the MFI but we strongly suspect it.

14. Estimates are omitted for reasons of space and available uponrequest.

15. The seemingly unrelated regression (SUR), model (Zellner, 1962), is awell known technique for analyzing a simultaneous pattern of relation-ships among variables in a system of multiple equations with correlatederror terms. We adopt this choice following Davidson and Mackinnon(1993) who argue that the univariate approach should be extended inorder to take into account a more complex pattern of dependence amongvariables and consider SUR a possible solution when all dependentvariables are modeled as correlated and simultaneous. The SUR is used toprovide clearer details of the pattern of relationships among the differentvariables and to show how the loan to income ratio and window length arein turn affected by other factors. After doing this we address in a differentway the two endogeneity problems with estimates in Table 8 where wecontrol for heterogeneity in loan to income ratio and window lengths.

16. The VIF (variance inflation factor) formula is 1/1 R(x) where R(x)is the R2 when the independent variable is regressed on all otherindependent variables (Marquardt, 1970). IfR(x) is low (tends to zero) theVIF test is low (equal to one). A VIF value below 10 (or, morerestrictively, five) is considered acceptable by rules of thumb usuallyadopted in the literature.

17. VIF factor results and the correlation matrix are omitted for reasonsof space and available from the authors upon request.

18. We control whether such introduction generates multicollinearity butwe find that the maximum VIF is below 3.5 and the correlation betweenthe interacted and non interacted loan to income ratio is around 50%.

19. The treatment-effects model evaluates the impact of an endogenouslychosen binary treatment on another endogenous continuous variable,conditional on two sets of independent variables using either a two-stepconsistent estimator or full maximum likelihood (we opt for the secondsolution). In the two equations system (v) and (e) are the error terms of theselection and main equations and are bivariate normal random variables

with zero mean and covariance matrixr q

q 1

. The likelihood function

for the joint estimation is provided by Maddala (1983) and Greene (2003).

REFERENCES

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

Athukorala, P., & Resosudarmo, B. P. (2005). The Indian Ocean tsunami:Economic impact, disaster management and lessons. Asian EconomicPapers, 4(1), 139.

Campbell, J. Y., Lo, A., & McKinlay, C. (1997). The econometrics offinancial markets. Princeton, USA: Princeton University Press.

Davidson, R., & MacKinnon, J. G. (1993). Estimation and inference ineconometrics. New York, NY: Oxford University Press.

Deaton, A., & Paxson, C. (1998). Economies of scale, household sizeand the demand for food. Journal of Political Economy, 106, 897930.

Fafchamps, M., & Gubert, F. (2007). The formation of risk sharingnetworks. The Journal of Development Economics, 83(2), 326350.

Goldberg, N. (2005). Measuring the impact of microfinance: Taking stockof what we know. GrameenFoundation, www.grameenfoundation.org.

Greene, W. (2003). Econometric analysis: 5th edition. Prentice Hall.Hermes, N., & Lensink, R. (2007). The empirics of microfinance: What do

we know?. The Economic Journal, 117, 110.Hoque, S. (2008). Does micro-credit program in Bangladesh increase

households ability to deal with economic hardships? MPRA PaperNo. 6678.

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

Karlan, D. S. (2001). Microfinance impact assessments: The perils of usingnew members as a control group. Journal of Microfinance, 3(2), 7685.

Khandker, S. R. (2007). Coping with flood: Role of institutions inBangladesh. Agricultural Economics, 36, 169180.

http://-/?-http://-/?-


15/15

Maddala, G. S. (1983). Limited-dependent and qualitative variables ineconometrics. Econometric society monographs in quantitative econom-ics. Cambridge, UK: Cambridge University Press.

Marquardt, D. W. (1970). Generalised inverses, ridge regression, biasedlinear estimation, and nonlinear estimation. Technometrics, 12,591612.

McIntosh, C., Villaran, G., & Wydick, B. (in press). Microfinance andhome improvement: Using retrospective panel data to measure

program effects on fundamental events. World Development, 39 (6),922937.

Skees, J., Varangis, P., Larson, D., & Siegel, P. (2002). Can financialmarkets be tapped o help poor people cope with weather risks? WorldBank Policy Research Working Paper No. 2812.

Smith, J. B., Klein, R. J., & Huq, S. (2003). Climate change, adaptivecapacity and development. London, UK: Imperial College Press.

World Bank (2001). World development report 2000/2001: Attackingpoverty. Oxford, UK: Oxford University Press.

Zellner, A. (1962). An efficient method of estimating seemingly unrelated

regression equations and tests for aggregation bias. Journal of theAmerican Statistical Association, 57, 48368.

Available online at www.sciencedirect.com


Becchetti and Castriota_2011_MF After Tsunami Srilanka

Documents

Transcript of Becchetti and Castriota_2011_MF After Tsunami Srilanka