N.43 September 2012
Leonardo Becchetti
Stefano Castriota
Pierluigi Conzo
Bank strategies in
catastrophe settings:
empirical evidence and
policy suggestions
Working papers
Bank strategies in catastrophe settings: empirical
evidence and policy suggestions*
Leonardo Becchettia
Stefano Castriotab
Pierluigi Conzoc
July 2012
Abstract
The poor in developing countries are the most exposed to natural catastrophes and
microfinance organizations may potentially ease their economic recovery. Yet, no
evidence on MFIs strategies after natural disasters exists. We aim to fill this gap with
a database which merges bank records of loans, issued before and after the 2004
Tsunami by a Sri Lankan MFI recapitalized by Western donors, with detailed survey
data on the corresponding borrowers. Evidence of effective post-calamity intervention
is supported since the defaults in the post-Tsunami years (2004-2006) do not imply
smaller loans in the period following the recovery (2007-2011) while Tsunami damages
increase their size. Furthermore, a cross-subsidization mechanism is in place: clients
with a long successful credit history (and also those not damaged by the calamity) pay
higher interest rates. All these features helped damaged people to recover and repay
both new and previous loans. However, we also document an abnormal and significant
increase in default rates of non victims suggesting the existence of contagion and/or
strategic default problems. For this reason we suggest reconversion of donor aid into
financial support to compulsory microinsurance schemes for borrowers.
We gratefully acknowledge Cristina Angelico, Carolina Pagano, Eugenia Agostino and Niroshan
Kurera for field data collection and Etimos Foundation, Etimos Lanka and AMF for their logistic
support.
a Department of Economics, University of Rome “Tor Vergata”, becchetti@economia.uniroma2.it.
b Department of Economics, University of Rome “Tor Vergata”, stefano.castriota@uniroma2.it.
c University of Naples "Federico II" & CSEF, pierluigi.conzo@unina.it.
Keywords: Tsunami, disaster recovery, microfinance, strategic default, contagion, microinsurance.
JEL codes: G21, G32, G33.
*
1"
"
1. Introduction
Natural catastrophes cause economic destruction and have severe consequences on
household income, assets, welfare and nutrition. Over the last decades the variability
as well as the frequency and strength of climate-related extremes have increased
alarmingly. There are several reasons for this upward trend, the most relevant ones
being human-driven climate changes and land misuse which have increased the
number and severity of some type of disasters like hurricanes and floods. As it is well
known, low and middle income countries suffer the most from these events due to
unfavorable weather conditions, high population density, poor quality of buildings and
infrastructures, lower insurance protection and, more in general, lower financial
resources required to cope with them (Cummins and Mahul, 2009). These catastrophic
events bring the economic system to an halt in a similar way to a heart attack. In
order to restore financial and economic flows what is needed is a shock therapy (a
defibrillator) which soon restores liquidity of the system. This is why in this dramatic
scenario several authors have tested whether (survival and/or recapitalization of)
microfinance institutions may help to compensate the losses and recover from natural
catastrophes and investigated how the same local credit intermediaries - which are
crucial to restore liquidity - may survive to the shock.
In this respect, many studies document that support from MFIs can be scarce if their
loan portfolios end up being severely damaged by the catastrophe, in which case the
survival of the whole bank serving the poor can be at risk. Collier et al. (2011), using
portfolio-level monthly data of a Peruvian MFI from January 1994 to October 2008,
show that the 1997-1998 El Niño significantly increased loan problems. This is
2"
"
because after a natural disaster a contemporaneous increase in the demand and a fall
in the supply of credit - the latter due to an increase in bad loans - can generate a
significant and long-lasting disequilibrium. Evidence of mismatches between demand
and supply of credit after a natural catastrophe has been provided by Berg and
Schrader (2009) who analyze the effect of volcanic eruptions in Ecuador on the
demand for loans and access to credit. The authors show that, while the former
increased due to volcanic activity, the latter was restricted for new clients.
On the positive side Khandker (2007) documents with household-level panel data that
the 1998 flood in Bangladesh increased vulnerability to poverty reducing both
consumption and assets while microfinance helped to compensate the losses from the
flood. In a similar vein Becchetti and Castriota (2010 and 2011) find that the 2004
Tsunami caused significant economic and psychological losses and document that MFI
recapitalization helped to recover pre-Tsunami welfare levels and achieve convergence
with non-damaged individuals.
Note however that during catastrophes credit mechanisms can worsen also due to
strategic defaults and contagion and MFIs may be particularly vulnerable to these
phenomena in presence of joint liability clauses. This is because under these
contractual arrangements the default of one (or more) borrowers hit by the shock
increases the burden of solvent groupmates not directly affected by the calamity. .
Under these circumstances a “domino effect” can therefore lead to the default of the
entire group and, eventually, of the whole MFI. Bratton (1986) shows that group
lending is better than individual lending in good times, the reverse being true in times
of crisis. Evidence of domino effects is provided by Paxton (1996) in Burkina Faso.
3"
"
As a consequence, if borrowers believe that many clients will default, and that this
would eventually lead the MFI to bankruptcy (or to require higher lending rates in the
future to survive), they may strategically decide to default since microfinance
institutions rely on the promise of future loans to induce repayment. Bond and Rai
(2008) refer to such phenomenon as a borrowers’ run. Evidence in this sense is found
in Goering and Marx (1998) in the case of Childreach in Ecuador where the number of
defaults multiplied as the word spread that few people were paying back. Similar
results are obtained with a different approach by Cassar and Wydick (2010) who carry
out group lending experiments in five countries and demonstrate that players have an
incentive to verify if they believe that a critical number of other group members will
do the same.
Our research aims at studying whether these phenomena are at work after a natural
disaster by investigating the determinants of loan amounts and credit defaults in a Sri
Lankan microfinance organization severely damaged by the 2004 Tsunami and
recapitalized by Western donors after it. In our empirical investigation we rely on a
broad range of controls which provide insights into the credit mechanisms of the
institution and the clients’ repayment incentives. The focus is on the effects of the
Tsunami on the MFI’s operating principles and on the borrowers’ insolvency.
Two main results emerge from the empirical analysis. First, standard lending rules,
which imply that clients do not obtain new loans until they repay old ones, are
suspended in order to help Tsunami victims to recover from the catastrophe. Second,
having been damaged by the 2004 Tsunami has no effect on credit defaults, after
controlling for other confounding elements like socio-demographic and economic
variables and external support and donations. This finding is paralleled after the
4"
"
calamity (years 2004-2006) by a significant and unexpected increase in default rates of
borrowers not affected by the Tsunami and a significant difference in (higher) lending
rates paid by non victims vis à vis victims in the post-Tsunami period.
This evidence suggests that strategic defaults and/or contagion may be in place although our data do not allow us to disentangle the two phenomena. 1 All these
results imply that external support to MFIs with a relevant share of bad loans helps
damaged people to recover from the calamity, but also generates moral hazard
problems for non damaged under the assumption of asymmetric information between
AMF and the latter. Our policy advice is that the problem could be avoided with the
reconversion of donor aid into financial support to compulsory (calamity specific)
microinsurance schemes attached to the loans. This would prevent borrowers
unaffected by future calamities from having negative expectations on their own
financial burden and on the MFI future survival, thereby preventing contagion and
strategic default.
The rest of the paper is organized as follows. Section 2 describes how the database
has been created. Section 3 provides summary statistics and descriptive evidence of
the MFI sample portfolio deterioration after the hazard. Section 4 reports econometric
results over the determinants of loan amounts and credit defaults. Section 5 discusses
the need of compulsory microinsurance schemes attached to bank loans as a possible
solution to moral hazard problems. Section 6 concludes.
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
1
What may be inferred is that, would AMF be able to bridge after the tsunami the
asymmetric information with borrowers, the strategic default rationale would be ruled
out. We do not have however information which allows us to test this hypothesis.
5"
"
2. The database
Our database is created by merging bank records and survey data. It consists of
information on 767 loans issued from 1995 to 2011 to 200 randomly sampled clients
living in the villages of Galle, Matara and Hambantota by Agro Micro Finance, a Sri
Lankan MFI headquartered in the capital Colombo with regional branches in the
South-West of the country.
The Tsunami was an unexpected event, therefore it was impossible to organize
repeated interviews over time, before and after the catastrophe. For this reason we
adopted the Retrospective Analysis of Fundamental Events Contiguous to Treatment
(RETRAFECT) methodology used by McIntosh et al. (2011) which borrows from event
studies used in the finance literature. This methodology relies on cross-sectional
surveys to create a retrospective panel dataset based on fundamental events in the
history of households.
We interviewed MFI borrowers twice: the first time in April 2007 and the second in
December 2011. Interviews were conducted at the monthly society meetings or at the
clients’ homes and made use of professional translators who received intensive
training by the team of researchers and Agro Micro Finance staff members. In April
2007 respondents were asked to declare current and remember past levels of different
wellbeing indicators by making reference to four different periods. We selected periods
easy to remember due to the occurrence of memorable events.
6"
"
The four considered time windows are: (P1) the six month interval before the first
microfinance loan ever obtained; (P2) the period going from the first microfinance loan
to the tsunami date (26th of December 2004); (P3) the period between the tsunami date
and the first microfinance loan after tsunami and (P4) the period from the first
microfinance loan after tsunami to the survey date (April 2007). In December 2011 we
updated the project, which allowed us to collect additional information for a fifth
window (P5) consisting of the six months preceding the interview. Figure 1 shows the
time schedule of the two surveys and the five reconstructed windows. A first step of
the research consisted in merging bank and survey data: in this way when studying
the determinants of credit defaults we are able to provide, for each of the 767 loans
released by the MFI, a number of additional controls.2
More specifically, our records provide official bank information on loan characteristics
such as initial and end dates, duration, amount released, interest rate charged,
whether the loan has been repaid, and the number of previous loans and of previous
defaults. As a complement, the two surveys allow us to collect information on sociodemographic and economic variables, the damages suffered from the Tsunami, and the
support received after the calamity from family members, friends, the Government
and other organizations. This information is important since it can affect the demand
for loans and the default rates.
Another fundamental variable which could influence the two variables of interest is
the initial income of the borrower. In fact, institutions achieving financial
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
In the estimates which follow the retrospective approach is used only to calculate income while all
other data come from official bank files. Our results are robust to the omission of the income variable
and therefore hold also when not using the retrospective approach. Evidence is omitted for reasons of
space and available upon request.
2
7"
"
sustainability could lend higher amounts to wealthier people whose implicit risk is
lower, while organizations achieving outreach might privilege poorer clients.
Similarly, default rates could be influenced by initial income in that, during difficult
times, wealthier people can repay the loan without sacrificing basic needs such as
nutrition and children education. Although at a first glance it is normal to believe that
income is less memorable than other variables, Becchetti and Castriota (2011) find
that it is strongly correlated with memories about average weekly hours of work,
problems in providing daily meals to the family and self-declared satisfaction about
overall economic situation. Answers about these variables are consistent for all the
considered windows. For this reason, when running regressions we include in the
specification the income of the previous window.
Given these database characteristics, from a methodological point of view our work
has a number of strengths with respect to other articles studying the consequences of
the Tsunami on economic and psychological variables (see, for example, Callen, 2009
and Cassar et al., 2011). First, the impact of the hazard is measured at the individual
and not at the village level as in many existing works, thereby preventing location bias
problems. Second, we do not constrain ourselves to considering only whether the
person experienced or not the calamity. In fact, we identify six different types of
possible damages and build a proxy for the intensity of the shock. The six types of
economic and psychological damages are: i) family members dead or injured; damages
to ii) house; iii) office buildings; iv) working tools; v) raw materials; vi) economic
activity in general. 3
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
3
Most borrowers were interviewed at home in the 2007 post-tsunami survey. Damages
of those interviewed at AMF were checked. Hence we could personally verify that the
damage variable were not affected by measurement error.
8"
"
These original features help to solve the identification problem arising from the
impossibility of randomizing ex ante the calamity experience, that is, the causality
link from the Tsunami shock to loan preferences. In fact, it could be argued that
wealthier and less risky borrowers selected areas (in which they have family, house and
economic activities) which were more likely to be inundated by the Tsunami. This
could be the case if rich people were willing to pay an extra price for houses with view
on the ocean or if the closer distance from the coast implied higher revenues (e.g.
coming from profitable businesses like tourism) or lower transportation costs due to
better infrastructures and higher population density.
Such interpretation is hardly plausible since: i) damaged and non damaged
individuals living in the same villages are very similar with respect to observables
(and, arguably, unobservables) (see section 3); ii) people in our sample did not change
residence before and after the calamity; iii) the degree of heterogeneity among
individuals is minimized by the fact that they are all clients of an MFI and received
loans to finance business activities. As a consequence we expect that: i) attendance of
entrepreneurship trainings and monthly borrowers meetings shaped a similar
economic mentality; ii) interviewed borrowers are similar with respect to some
unobservable factors (main suspect of self-selection) like sense of entrepreneurship
and trustworthiness which helped them to pass the screening selection of the bank.
Finally, it could be argued that the most severely hit by the natural calamity left their
village and migrated somewhere else. Although we do not have official data on
migration of clients before and after the Tsunami, the AMF management reports
""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
9"
"
anecdotic evidence in favour of the inexistence of a self-selection bias of the least
damaged individuals, the possible incentive to stay being the possibility to receive new
loans after the calamity. This option would have been difficult to explore if a person
applied for a loan in a new region after having lost all her belongings and without
having previous successful track records.
3. Descriptive statistics and balancing properties of damaged vs. non
damaged before the Tsunami
Table 1 provides a description of the variables used while Table 2 reports summary
statistics. The average loan amount in December 2011 terms is above 66,000 Sri
Lankan Rupees (Rps.), which is a considerable amount based on the local living
standards. AMF’s declared policy, common to many similar institutions, is to start it
with smaller loans in order to test the client’s ability to repay, while increasing over
time the amount lent. From this point of view MFIs privilege financial sustainability
to outreach since it is reasonable to assume that, at the beginning, when they are
starting a new business, clients are more in need of funds but are also riskier. From
Table 3 it is possible to observe that, net of the general upward trend over time, the
average amounts of loans peaked after the Tsunami because of the combined effect of
the increased demand to recover from the damages and bank recapitalization which
generated an inflow of financial resources. In our sample 18% of loans have not been
repaid: such a high share is due to the unexpected 2004 calamity which caused
massive defaults, as shown in Table 3 where the 2005 90% peak is self-explaining.
10"
"
The annualised nominal interest rate is around 37 percent. This rate is not
particularly high if we consider the relatively high inflation rate which has ranged
between 3 and 23 percent in the period under scrutiny, the small average amount and
the relatively short duration of loans which (compared to ordinary banks) boost the
administrative expenses and force MFIs to charge high interest rates on loans
(especially if we consider infra-annual loans) (Hardly et al., 2003). As shown in Table 3
the average interest rate fluctuates over time according to market conditions, but
decreases after the Tsunami because of donors’ constraints on the use of the released
funds. In fact, damaged people were entitled to receive loans at favorable conditions
(6% interest rate). The duration of the loans ranges from one day (0.03 months) for
small amounts to four years for big amounts, for which the authorization of the
regional or even central manager is required. The most common frequency schemes
are based on monthly, followed by weekly and bi-monthly installments, even though
bank managers are free to choose longer or shorter maturities depending on the
amounts released, the type of businesses financed, the credit history and the distance
from the local branch which affects the monitoring costs. Around 11% of loans have
been issued to start a new business (start-up) or launch a new product (spin-off), 82%
to finance ongoing businesses and 6% to recover from the natural calamity.
The “source of the initiative” is a relevant aspect of the lender-borrower relationship
which is able to influence the average amount of loans issued by a bank and the
default rates. The possible “source of initiative” answers in our survey are: (i) AMF
(35% of loans in our sample); (ii) the client, following the suggestion of a borrower who
introduced him to the bank manager (35%); (iii) the client, spontaneously, without the
support of anybody (29%).
11"
"
On the one hand, people who spontaneously look for a loan are likely to be more
proactive and enterprising, which is a signal the bank could use to identify the client’s
profile. On the other hand, individuals who are introduced to AMF by senior clients
benefit from “reputation spillovers”: new members joining a group “inherit” the good or
bad reputation of the coalition, so that collective reputation turns out to be history
dependent (Tirole, 1996). Furthermore, they become immediately part of a group of
people with more similar characteristics and stronger social ties, which could affect
loan amounts and default rates as shown by Cassar et al. (2007) with field
experiments in South Africa and Armenia. As a consequence, whether proactive
borrowers will obtain more/less money and will have higher/lower default rates is an
empirical issue we are going to analyze with econometric regressions in section 4.
The number of previously released and repaid loans ranges from 0 (new clients) to 27,
while that of previous defaults from 0 to 2. The average distance from the closest AMF
branch is 15 km, which is non-negligible given the poor quality of road infrastructures
and the scarcity of own transportation means.
In line with most MFIs, the vast majority of loans have been released to women. Age,
education and family size are in line with regional values. Most borrowers are
involved in manufacturing and trade, while a relevant share has more than one
economic activity (the sum of the mean values of the dummies for the types of activity
exceeds one). The average real monthly income in 2011 terms of the time window
preceding the loan was around 34,000 Rps., ranging from 0 in the aftermath of the
Tsunami for those severely hit by the wave to a maximum of 132,000 Rps. for
successful entrepreneurs.
12"
"
Around half (47%) of loans have been released to people damaged by the Tsunami, the
number of damage types ranging from 0 to 6. Among the six types considered the first
five refer to the direct shock caused by the calamity while the last one (damage to the
economic activity) is indirect and refers to the decrease in market demand. From
Table 2c it emerges that indirect effects are the most common (39%), followed by
damages to raw materials (24%), working tools (18%), office buildings (17%) and house
(11%), while those on family members are rare (1%). Note that the dummy variables
for the damages and Sum of Damages are obviously zero for all loans released before
the Tsunami event.
With respect to external support, only 2% of loans have been provided to people
receiving remittances from abroad, donations and subsidies being more frequent
(respectively 5% and 11%). Finally, while loans provided by other MFIs and other
people are extremely rare (1%), those provided by banks and family members or
friends are more frequent (respectively 14% and 11%).
Table 4 shows parametric tests for difference in means in terms of loans/borrower
characteristics between damaged and non damaged. This is meant to test whether
characteristics of the loans or those of the borrowers were significantly different and
could drive (and bias) the econometric results of section 4. Note that all these variables
are either time invariant or verified as being invariant before and after tsunami and
therefore their values may be considered as pre-tsunami levels. Our tests document
that the null of no difference in observable characteristics between the two groups is
never rejected at 5% level (t-stats are always below 1.96).
13"
"
The respect of balancing properties is likely to be due to the characteristics of our
data. As discussed in the introduction our database is composed of people from the
same villages living at a close distance from each other and all being members of the
same MFI. Participation to one of the two (damaged/undamaged) groups is therefore
likely to be due to casual factors such as natural barriers or small differences in
distance from the coast.
4. Econometric results
4.1 Determinants of loan amounts
We start our empirical analysis by studying the determinants of loan size in our
sample. The estimated specification is:
LS i = α 0 + α1Damageit + ∑ βi X it + ∑ γ t DYear t +∑ δ j Dvillage j + ε i
i
t
j
(1)
The dependent variable (LS) is the loan size expressed in December 2011 terms and
extracted from the AMF electronic database, while Damage is a unit dummy for
borrowers hit by the Tsunami (always equal to zero before the catastrophe) which is
introduced in the second specification (Table 5, column 2). Alternatively in column 3
the dummy is replaced by six dummies related to the type of damage suffered and, in
column 4, by the sum of damages. The X socio-demographic variables control for
gender discrimination, role of seniority and education, household size, business of
activity, initial income (of the time window preceding the loan), damages suffered from
14"
"
the Tsunami and external support received. Regressions include village and time
dummy variables (DYear and Dvillage) (results are omitted for reasons of space but
are available upon request). Standard errors are clustered at the borrower level and
reported in parentheses.
A first main finding is that people hit by the Tsunami (Damage) receive more funds,
the relevant type of damage being the indirect one to the economic activity (Damage:
economic activity), while the index we built to measure the intensity of the damage
does not seem an effective proxy to capture the consequences of the calamity (Table 5).
The economic support AMF received from donors and international organizations was
partly conditioned to the Tsunami victims being financed first, therefore the larger
amounts lent to victims are not unexpected.
However, it appears that AMF did not lend more to those suffering the most since
direct damages (Table 5, column 4) and intensity of the damages (Table 5, column 5)
are not significant.
Turning to financial variables, while AMF clearly states its policy of lending smaller
amounts to new clients and larger amounts to solvent borrowers, econometric results
show that the number of previous loans is irrelevant for the amount released. This
behavior does not closely correspond to patterns observed in microfinance markets
where new clients are offered small loans to test their repayment behavior
(Vogelgesang, 2003).
An apparently counter-intuitive result is the positive - instead of negative – effect of
previous defaults on the amount released by the bank. Ordinary banks and MFIs most
often explicitly forbid to lend money to borrowers until they repay back the amount
15"
"
due. Even if the money is finally repaid, MFIs generally use this piece of information
to update the risk profile of the borrower. As a consequence, the coefficient attached to
past defaults should be at least non-positive. The reason for this unexpected result is
the Tsunami catastrophe which caused serious damages to the businesses and the
properties of historically reliable clients (in our sample there are no defaults until the
calamity occurred). Without further loans clients would have likely been unable to
recover and, in turn, repay the previous loan4.
The purpose for which the loan has been asked matters. Even though when starting a
new business entrepreneurs need more financial resources, lending for a new business
is perceived as riskier by the bank which provides smaller loans. In this case AMF
seems to behave like traditional bank in that it privileges financial sustainability to
outreach.
With respect to the “source of initiative”, individuals who are introduced to AMF by
another client receive the most, meaning that social ties and reputational spillovers
are in place, followed by those who autonomously contact the MFI. Those who get a
credit offer on the initiative of the bank receive the least since are less proactive and
do not belong to well established and homogenous groups. A growing body of literature
has proved the relevance of social networks in household decision-making (Conley and
Udry, 2010) and of personal relationships in credit access, particularly in developing
countries (Okten and Osili, 2004). In line with these intuitions Wydick et al. (2011),
using survey of 465 households living in Western Guatemala, show that access to
credit is closely related to membership of a church network. Our results add to Wydick
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
Given the dramatic event, AMF’s strategy is in contrast with policies adopted by other MFIs in more
normal contexts like Caja Los Andes in Bolivia which does not provide new grants if a client has not
repaid previous loans, as documented by Vogelgesang (2003).
4
16"
"
et al. (2011) in that they study the determinants of access to credit, while our sample
is entirely composed by clients and the dependent variable is the loan amount. Social
ties not only increase access to credit through imitation phenomena as shown by
Wydick et al. (2011), but also increase the average amount of loans through
reputational spillovers.
The distance from the MFI branch does not have any significant effect. This might be
due to two counteracting forces: on the one hand, closer distance may allow better
selection and monitoring of clients while, on the other, due to higher transaction costs
(see Ashraf et al., 2006), lending to clients living farther away could be convenient
only for larger amounts. Either the two effects cancel out or are not at work.
When looking at the significance of other regressors, the negative coefficient attached
to the female gender is surprising since microcredit was born to serve the poor,
especially women. It is difficult to say whether such finding depends on discrimination
or on unobservable gendered differences in financed project characteristics (i.e. women
asking more consumption or small scale loans). Education has a positive effect on loan
amounts, meaning that the bank may interpret it as a signal of lower risk profile. It is
also likely that more educated people set more advanced, sophisticated and expensive
businesses for which a higher amount of money is necessary. The remaining sociodemographic variables are not significant at conventional levels. Initial income does
not play any role: the MFI does not lend more neither to poorer nor to richer clients,
therefore displaying a policy which tries to balance financial sustainability and
outreach. External support in the form of subsidies, donations, remittances and other
loans could have reduced in principle the need of credit, but in our regressions do not
have any impact on the variable under scrutiny.
17"
"
4.2 The determinants of credit defaults
In Table 6 we report findings on the determinants of credit defaults based on the
following specification
Defaulti = α 0 + α1Damageit + ∑ β i X it + ∑ γ t DYear t +∑ δ j Dvillage j + ε i
i
t
j
(2)
where the dependent variable (Default) is a dummy equal to one if the loan has not
been reimbursed, zero otherwise and the other variables are defined as in (1). Given
the discrete nature of the variable the natural candidate for this type of investigation
is a Logit model. Again, standard errors are clustered at the borrower level and
reported in parentheses.
The most interesting result is that the probability of default is neither affected by the
Tsunami victim status nor by the intensity of the damages. This finding must imply
on the descriptive side a significant increase in the default rate also of non victims in
the Tsunami vis à vis the pre-Tsunami period in order to make the victim/non victim
effect not significant. This is indeed what we find. Before the Tsunami the default rate
of victims and non victims is respectively around 23 and 21 percent and not
significantly different between the two groups (consistently with balancing properties
shown in section 3.2). In the Tsunami period (2004-06) the default rate of victims and
non victims raises to 58 and 50 percent, the difference being not statistically different
here as well. This result is unexpected, since the Tsunami should not affect positively
the probability of default of an individual declaring no damages to building, relatives
or economic activity.
18"
"
We identified two possible explanations for this anomaly: strategic defaults and
contagion. It might be the case that, as for Childreach in Ecuador (Goering and Marx,
1998), the word spread that few people were paying back the money and that AMF
was going bankrupt. Another - not mutually exclusive - explanation is contagion, since
during a hard time of local economic downturn the default of one or two members
could have led to the insolvency of the entire group under group lending with joint
liability. Contagion problems could have been particularly serious in the light of the
restricted size of the groups formed by AMF (three members), which, on the one side,
facilitates the creation and the management of groups, but, on the other, increases the
burden for the remaining members in bad times.
When inspecting other financial regressors we find that the interest rate is negatively
correlated with default. Abbink et al. (2006) with laboratory experiments find that, on
the one side, a higher repayment burden intensifies the incentives to free-ride since
shirking allows to save money, but, on the other side, it implies a disciplining effect
given that high-interest loans are less tolerant towards defaulters. Cull et al. (2007),
using data from 124 institutions in 49 countries, compare group-based versus
individual based microfinance institutions and show that, above a certain threshold,
interest rates worsen the quality of portfolio in case of individual loans, but this
relation does not exist for group-based microfinance institutions. Our results differ
from those mentioned above since AMF carries out a cross-subsidization strategy
which consists of increasing the interest rate to solvent clients in order to reduce it to
bankrupt ones5.
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
Table A1 in the Appendix shows the determinants of the interest rate applied by AMF to each loan:
the main drivers of the cost of capital are the client’s distance from the local branch, the amount
released, the duration and the number of successfully repaid loans. Previous defaults do not lead to an
increase in the interest rate.
5
19"
"
Longer maturities reduce default rates, but the frequency of repayments does not
matter. Armendariz and Morduch (2005) with anecdotal evidence from Bangladeshi
microfinance providers and Mcintosh (2008) with more formal analysis of microfinance
contracts offered by FINCA in Uganda find that higher frequency of repayments is
associated with lower default rates. However, this could be due to self-selection since
clients chose their repayment schedule. Field and Pande (2008) use data from a field
experiment with randomized client assignment to a weekly or monthly repayment
schedule and find no significant effect of type of repayment schedule on client
delinquency or default. Our results are consistent with theirs.
In line with expectations, larger loans imply higher default rates. Credit history does
not matter: neither the number of repaid loans nor that of defaults are predictors of
current insolvency. This finding is important since it shows that natural calamities
can lead people to bankruptcy, but do not generate repeated defaults. In other words,
if borrowers receive new support the discontinuity is only temporary and not
permanent. Next, while the nature of credit initiative affects the amount released by
the bank, this is not the case for the default rates.
The distance from the closest AMF branch has no effect on default rates: either closer
distance does not imply better clients’ selection and stricter monitoring6 or, on the
opposite, the selection was so effective that closer and farther clients ended up being
homogeneous with respect to the risk profile: this point is left to future research.
Finally, the default rate of loans issued to finance start-ups, established business or
recovery are the same. Start-ups may show similar mortality rates to other businesses
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
Distance among members of the same group has been shown to affect peer monitoring and, in turn,
repayment rates (Wydick, 1999). The same principle could have, but does not, work for the borrowerlender relationship.
6
20"
"
because of contextual factors - small businesses in a growing developing country which provide consumers’ demand and reduce the minimum efficient size of the firm.
With respect to the significance of the remaining regressors we document that sociodemographic controls do not matter. Gender, age, education and number of house
members have no effect on repayment rates. Age - a proxy for work experience and
wealth - has a negative but not significant effect. Household size could have had a
negative effect due to the large family “fixed costs” during calamities and economic
downturns, but also a positive one due to the available and free workforce. Either none
or both effects are at work here, the final result being null. Past income does not help
reducing default risk: this is probably so because, on the one hand, higher income
allows more savings, but, on the other, it is a proxy for larger activities which are less
flexible on the costs side when the business climate worsens.
4.3 Further evidence on the contagion/strategic default hypothesis
To elaborate more around our contagion/strategic default hypothesis we look at
determinants of defaults for the control group of non damaged only (Table 7) and find
a significant and positive effect of the dummy picking up the post-Tsunami period.
Hence, even though non damaged do not declare any consequence of the tsunami
(including the indirect effect of a demand reduction), they suffer an unexpected
increase in default rates in such period. Hence the jump in default rates documented
with descriptive findings in section 4.2 is confirmed after controlling for confounding
factors in econometric estimates.
21"
"
A second interesting piece of evidence is the comparison of interest rates between
damaged and non damaged in the post-tsunami period. What we find is that the non
damaged pay 8 percent more and the difference is significant (p-value 0.002). The
consequence of this finding is that non damaged which are groupmates of damaged
members have a clear cost in not declaring default, that is, they pay a higher interest
rate and, due to the joint liability clause, they may be asked to contribute to pay the
loan of their unsolved groupmates hit by the Tsunami. The cost of not declaring
strategic default may be a rationale to explain the unexpected increase in default for
non damaged in the post tsunami period.
In our database we do not have information on dropouts and therefore the suspicion
that our findings may be affected by survivorship bias may arise. Survivorship is
generally not balanced between “good” and “bad” borrowers and it may therefore
generate a bias via exclusion of a higher share of bankrupt than succcesful borrowers
from the sample. In such case, with reference to our main two dependent variables, it
would bias downward overall sample default rates while the effect on lending rates
would be uncertain (or it may be assumed to generate an upward bias since we found
that cross-subsidisation from good borrowers is at stake). Note however that it is
reasonable to assume that, if the bias exists, it affects in the same way damaged and
non damaged in the pre-tsunami period (damaged and non damaged have not
significantly different characteristics ex ante) thereby not altering our main results on
the insignificant impact of damaged status on post tsunami defaults. Moreover, in our
specific case we verified that AMF lends also to clients who have a record of past
default and this minimizes the number of dropouts due to misperformance.
22"
"
Last, with regard to the post-tsunami period, we know that the support from foreign
donors is explicitly targeted to loan concession to borrowers defaulting due to the
tsunami. Hence the potential unbalance between damaged and non damaged dropouts
after the tsunami is eliminated by such intervention. All this being considered the
problem may be considered negligible and not affecting our main findings.
5. Ex-ante coping strategies and the need for microinsurance shemes
Loans provided by MFIs after natural calamities have been proved to be a helpful
recovery tool for the victims (Khandker, 2007; Becchetti and Castriota, 2011). Ex-post
recapitalization of a struggling MFI with funds provided by donors, NGOs or
international organizations is a solution which has been adopted, among others, by the
MFI under scrutiny, since neither microinsurance nor contingent repayment schemes
were in place at the time of the Tsunami.
However, relying on non automatic but voluntary external fund schemes to
recapitalize a deteriorated loan portfolio after calamities is risky for a number of
reasons. First, it is not sure whether the institution will find available donors or
partners since, when natural catastrophes occur, the number of potential beneficiaries
gets large and the competition among them keen. Second, recapitalizations necessarily
occur with a delay, which can worsen the already fragile financial situation of current
and potential borrowers looking for new loans.7 Third, because of the delay and of
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
7
As noted by Cummins and Mahul (2009, p.1), “Post-disaster assistance from the
international donor community may be slow and unreliable. In the face of the rising
23"
"
rational/irrational expectations, ex-post solutions do not prevent contagion and moral
hazard problems connected to strategic defaults.
Two similar solutions seem appropriate to prevent these two latter phenomena:
microinsurance schemes attached to loans and contingent repayment systems which
allow rescheduling of savings and installments after natural disasters for affected
members. Since 2002 most MFIs in Bangladesh have been introducing this type of
scheme (Dowla and Barua, 2006), which in a rural Bangladesh context has been
shown to decrease the probability that people skip meals during negative shocks
(Shoji, 2009). However, while the second solution seems adequate in case of natural
catastrophes which occur on a more regular basis like floods in Bangladesh, the first
seems more effective in case of unpredictable and devastating disasters like the 2004
Asian Tsunami since it does not just postpone, but rather cancel, the outstanding debt.
This difference can be of paramount importance when a borrower needs money to
recover from the catastrophe while the repayment of previous loans prevents the issue
of new ones. Furthermore, rescheduling can help to cope with strategic defaults and
contagion but does not prevent credit restrictions - especially to new clients.
Even though with this dataset we are unable to disentangle the relevance of strategic
defaults from that of contagion, a compulsory microinsurance attached to the loans
would have prevented both problems and the AMF portfolio deterioration. In fact, it
should be kept in mind that in the first quarter of 2005, before receiving foreign
support, AMF was technically bankrupt. However, nothing ensures that, if another
""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
frequency and intensity of losses in low- and middle-income countries, the old model of
post-disaster financing and reliance on the donor community is increasingly
inefficient”.
24"
"
catastrophe occurred in the future, further external funds from NGOs and other
donors would be obtained. This problem is even more severe since AMF clients have
experienced international solidarity and refinancing from the bank, therefore they are
likely to expect further assistance and support in case of future natural hazards.
6. Conclusions
Very few evidence on the impact of microfinance as post-calamity recovery mechanism
exists. We use a unique database made of official bank loans and survery submitted in
2007 and 2011 to evaluate the impact of donors recapitalization of a Sri Lankan MFI
after the Tsunami. Our data show that the donors’ intervention was effective in
supporting victims who received large loans at subsidized rates after their postTsunami default. The high default rates among non victim borrowers after the
Tsunami suggest, however, the occurrence of contagion and/or strategic default as it
typically occurs after natural disasters when group lending and joint liability clauses
are at work. The hypothesis of contagion or strategic default is reinforced by evidence
showing that non declaring default for non victims has a cost since their post-Tsunami
lending rate is significantly higher than that of victims due to a cross-subsidisation
mechanism in place.
We suggest that the reconversion of the donors’ fund into a compulsory post-calamity
insurance for all borrowers may maintain the positive post-intervention effects while
solving problems of contagion and strategic default.
25"
"
26"
"
References
[1] Abbink, K., Irlenbusch, B. and Renner, E. (2006), “Interest Rates in Group
Lending: a Behavioural Investigation”, Pacific Economic Review, Vol. 11, No. 2,
pp. 185-199.
[2] Armendariz, B. and Morduch, J. (2005), The Economics of Microfinance,
Cambridge, MA, MIT Press.
[3] Ashraf, N., Karlan, D. and Yin, W. (2006), “Deposit Collectors”, Advances in
Economic Analysis & Policy, Vol. 6, No. 2, pp. 1-22.
[4] Becchetti, L. and Castriota, S. (2010), “The Effects of a Calamity on Income and
Wellbeing of Poor Microfinance Borrowers: the Case of the 2004 Tsunami
Shock”, Journal of Development Studies, Vol. 46, No. 2, pp. 211-233.
[5] Becchetti, L. and Castriota, S. (2011), “Does Microfinance Work as a Recovery
Tool After Disasters? Evidence from the 2004 Tsunami”, World Development,
Vol. 39, No. 6, pp. 898-912.
[6] Berg, G. and Schrader, J. (2009), “Microcredit, Natural Disasters and Lending”,
Working Paper.
[7] Blinder, A. S. (1973), “Wage Discrimination: Reduced Form and Structural
Estimates”, Journal of Human Resources, Vol. 8, pp. 436–455.
[8] Bond, P. and Rai, A.S. (2009), “Borrower Runs”, Journal of Development
Economics, Vol. 88, pp. 185-191.
[9] Callen, M. (2009), “Catastrophe and Time Preference: Evidence from the Indian
Ocean Earthquake”, Working Paper.
[10]
Cassar, A., Crowley, L. and Wydick, B. (2007), “The Effects of Social
Capital on Group Loan Repayment: Evidence from Field Experiments”,
Economic Journal, Vol. 117, pp. 85-106.
[11]
Cassar, A., Healy, A. and von Kessler, K. (2011), “Trust, Risk, and Time
Preferences After a Natural Disaster: Experimental Evidence from Thailand”,
Working Paper.
[12]
Cassar, A. and Wydick, B. (2010), “Does Social Capital Matter? Evidence
from a Five-Country Group Lending Experiment”, Oxford Economic Papers, Vol.
62, No. 4, pp. 715-73.
[13]
Collier, B., Katchova, A.L. and Skees, J.R. (2011), “Loan Portfolio
Performance and El Nino, an Intervention Analysis”, Agricultural Finance
Review, Vol. 71, No. 1, pp. 98-119.
[14]
Conley, T. and Udry, C. (2010), “Learning About a New Technology:
Pineapple in Ghana”, American Economic Review, Vol. 100, No. 1, pp. 35-69.
[15]
Cull, R., Demirguk-Kunt, A. and Morduch, J. (2007), “Financial
Performance and Outreach: a Global Analysis of Leading Microbanks”,
Economic Journal, Vol. 117, p. 107-133.
27"
"
[16]
Cummins, J.D. and Mahul, O. (2009), Catastrophe Risk Financing in
Developing Countries, available on line at the address:
http://siteresources.worldbank.org/FINANCIALSECTOR/Resources/CATRISKb
ook.pdf .
[17]
Dowla, A. and Barua, D. (2006), The Poor Always Pay Back, Bloomfield,
CT, Kumarian Press.
[18]
Field, E. and Pande, R. (2008), “Repayment Frequency and Default in
Micro-Finance: Evidence from India”, Journal of the European Economic
Association, Vol. 6, No. 2-3, pp. 501-509.
[19]
Goering, L. and Marx, G. (1998), “Chaos when Credit is Due”, Chicago
Tribune, March 15.
[20]
Hardly, D.C., Holden, P. and Prokopenko, V. (2003), “Microfinance
Institutions and Public Policy”, IMF WP/02/159.
[21]
Jann, B. (2008), “The Blinder-Oaxaca Decomposition for Linear
Regression Models”, The Stata Journal, Vol. 8, No. 4, pp. 453-479.
[22]
Khandker, S.R. (2007), “Coping with Flood: Role of Institutions in
Bangladesh”, Agricultural Economics, Vol. 36, No. 2, pp. 169-180.
[23]
McIntosh, C. (2008), “Estimating Treatment Effects from Spatial Policy
Experiments: an Application to Ugandan Microfinance”, Review of Economics
and Statistics, Vol. 90, No. 1, pp. 15-28.
[24]
McIntosh, C. Villaran, G. and Wydick, B. (2011), “Microfinance and Home
Improvement: Using Retrospective Panel Data to Measure Program Effects on
Fundamental Events”, World Development, Vol. 39, No. 6, pp. 922-937.
[25]
Oaxaca, R. (1973), “Male-Female Wage Differentials in Urban Labor
Markets”, International Economic Review, Vol. 14, No. 3, pp. 693-710.
[26]
Okten, C. and Osilis, U.O. (2004), “Social Networks and Credit Access in
Indonesia”, World Development, Vol. 32, No. 7, pp. 1225-1246.
[27]
Shoji, M. (2009), “Dies Contingent Repayment in Microfinance Help the
Poor during Natural Disasters?”, Journal of Development Studies, Vol. 46, No.
2, pp. 191-210.
[28]
Tirole, J. (1996), “A Theory of Collective Reputation”, Review of Economic
Studies, Vol. 63, No. 1, pp. 1-22.
[29]
Vogelgesang, U. (2003), “Microfinance in Times of Crisis: The Effects of
Competition, Rising Indebtedness, and Economic Crisis on Repayment
Behavior”, World Development, Vol. 31, No. 12, pp. 2085-2114.
[30]
Wydick, B. (1999), “Can social cohesion be harnessed to repair market
failures? Evidence from group lending in Guatemala”, Economic Journal, Vol.
109, pp. 463-475.
28"
"
Figure 1: Time schedule of the two surveys and the five reconstructed time windows
1st MFI loan
Tsunami date
1st MFI loan
1st survey
2nd survey
ever obtained
December 2004
after Tsunami
April 2007
December 2011
Borrower 1
P1
P2
P3
P4
P5
P4
P5
Borrower 2
P1
P2
P3
Borrower 3
P1
P2
P3
P4
P1= six month interval before first AMF financing.
P2= period ranging from first AMF financing to the tsunami date (December 2004).
P3= period ranging from the tsunami date to the first AMF refinancing.
P4= period ranging from the first AMF refinancing to first the survey date (April 2007).
P5= six month interval before the second survey (December 2011).
Note: Dotted lines indicate non overlapping window borders, continuous lines coincident window borders.
29#
#
P5
Table 1: Description of the variables used
Table 1a: Financial variables
Loan size
Default
Interest rate
Duration
Frequency
Reason: new business
Reason: improve
Reason: recover
Initiative: AMF
Initiative: suggested
Initiative: spontaneously
Previous defaults
Previous repaid loans
Distance AMF
Amount of the AMF loan in December 2011 terms
DV=1 if the loan has not been repaid, 0 otherwise
Annual nominal interest rate on the loan
Duration of the loan in months
Number of installments per month
DV=1 if the loan has been asked to open a new business, 0 otherwise
DV=1 if the loan has been asked to improve an existing business, 0 otherwise
DV=1 if the loan has been asked to recover from the damages, 0 otherwise
DV=1 if , 0 otherwise
DV=1 if , 0 otherwise
DV=1 if , 0 otherwise
Number of previous loans which have not been repaid
Number of previous loans which have been successfully repaid
Distance from the closest AMF branch in km
Table 1b: Socio-demographic variables
Female
Age
Education
House members
Fishery
Manufactory
Trade
Other job
Real income
Matara
Hambantota
Galle
DV=1 if the respondent is female, 0 otherwise
Age of the respondent in years
Education of the respondent in years
Number of people living in the house
DV=1 if the respondent is involved in fishery, 0 otherwise
DV=1 if the respondent is involved in manufactory, 0 otherwise
DV=1 if the respondent is involved in trade, 0 otherwise
DV=1 if the respondent has another
Real total household income in December 2011 terms
DV=1 if the respondent lives in Matara, 0 otherwise
DV=1 if the respondent lives in Hambantota, 0 otherwise
DV=1 if the respondent lives in Galle, 0 otherwise
Table 1c: Damages from the Tsunami and support received
Damaged
Damage: family
Damage: house
Damage: office building
Damage: working tools
Damage: raw materials
Damage: economic activity
Sum of damages
Remittances
Subsidies
Donations and grants
Loans: bank
Loans: MFI
Loans: family/friend
Loans: other
DV=1 if the respondent has been damaged by the Tsunami, 0 otherwise
DV=1 if the respondent reported damages to the family, 0 otherwise
DV=1 if the respondent reported damages to the house, 0 otherwise
DV=1 if the respondent reported damages to the office building, 0 otherwise
DV=1 if the respondent reported damages to the working tools, 0 otherwise
DV=1 if the respondent reported damages to the raw materials, 0 otherwise
DV=1 if the respondent reported damages to the economic activity, 0 otherwise
Number of types of damage from 0 to 6
DV=1 if the respondent receives remittances, 0 otherwise
DV=1 if the respondent receives subsidies, 0 otherwise
DV=1 if the respondent receives donations and grants, 0 otherwise
DV=1 if the respondent has obtained other loans from a bank, 0 otherwise
DV=1 if the respondent has obtained other loans from another MFI, 0 otherwise
DV=1 if the respondent has obtained other loans from family/friends, 0 otherwise
DV=1 if the respondent has obtained other loans from other people, 0 otherwise
30#
#
Table 2: Summary statistics
Variable
Obs.
Mean
Std. Dev.
Min
Max
767
734
767
765
755
767
767
767
733
733
733
767
767
761
66,131
0.19
36.87
10.18
4.12
0.11
0.82
0.06
0.35
0.35
0.29
0.25
3.23
15.16
60,629
0.39
23.63
7.77
6.13
0.32
0.39
0.24
0.48
0.48
0.46
0.46
4.97
9.26
4,459
0
6
0.03
0.05
0
0
0
0
0
0
0
0
0.1
324,720
1
101
47.97
30
1
1
1
1
1
1
2
27
65
767
766
757
767
761
761
761
761
767
767
767
767
0.88
47.40
11.15
4.45
0.02
0.38
0.40
0.14
34,213
0.44
0.28
0.28
0.33
9.58
2.45
1.51
0.14
0.49
0.49
0.35
23,306
0.50
0.45
0.45
0
20
0
1
0
0
0
0
0
0
0
0
1
67
16
10
1
1
1
1
132,978
1
1
1
0.50
0.11
0.31
0.38
0.38
0.43
0.49
1.55
0.12
0.31
0.22
0.35
0.12
0.32
0.12
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
1
6
1
1
1
1
1
1
1
Table 2a: Financial variables
Loan size
Default
Interest rate
Duration
Frequency
Reason: new business
Reason: improve
Reason: recover
Initiative: AMF
Initiative: suggested
Initiative: spontaneously
Previous defaults
Previous repaid loans
Distance AMF
Table 2b: Socio-demographic variables
Female
Age
Education
House members
Fishery
Manufactory
Trade
Other job
Real income
Matara
Hambantota
Galle
Table 2c: Damages from the Tsunami and support received
Damaged
Damage: family
Damage: house
Damage: office building
Damage: working tools
Damage: raw materials
Damage: economic activity
Sum of damages
Remittances
Subsidies
Donations and grants
Loans: bank
Loans: MFI
Loans: family/friend
Loans: other
767
767
767
767
767
767
767
767
767
764
767
766
766
766
760
0.47
0.01
0.11
0.17
0.18
0.24
0.39
1.11
0.02
0.11
0.05
0.14
0.01
0.11
0.01
31#
#
Table 3: Descriptive statistics of selected variables, by year
Year
Obs.
Loan size
Interest rate
Default (%)
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
10
22
39
83
126
75
54
123
133
60
22
18
32,402
27,157
34,042
36,398
37,067
59,216
109,780
82,369
84,160
75,143
127,013
97,053
27
51
40
37
41
31
23
26
51
42
19
20
0
0
0
0.05
0.39
0.90
0.57
0.05
0
0
0
0
32#
#
Table 4: Difference in mean between damaged/non damaged
Variable
Damaged
Non damaged
Difference
T-stat
34,695
0.04
40.51
8.62
4.54
0.15
0.85
0.00
0.35
0.36
0.28
0.03
0.92
14.48
0.45
0.28
0.26
32,215
0.00
37.00
8.44
4.04
0.09
0.90
0.00
0.38
0.25
0.36
0.00
0.88
13.54
0.60
0.16
0.21
-2,479
0.04
-3.51
-0.18
0.50
-0.06
0.05
0.00
0.03
-0.11
0.07
-0.03
-0.04
-0.94
0.15
-0.12
-0.05
-0.56
-1.39
0.88
-0.21
-0.34
-0.99
0.87
N/A
0.37
-1.24
0.88
-0.35
-0.17
-0.58
1.81
-1.45
-0.62
0.86
45.57
11.57
4.30
0.00
0.30
0.45
0.04
36,458
0.03
0.55
0.63
-0.45
0.01
-0.07
0.14
-0.11
2,151
0.48
0.34
1.40
-1.94
-0.60
-0.87
1.69
-1.94
0.51
Table 4a: Financial variables
Loan size
Default
Interest rate
Duration
Frequency
Reason: new business
Reason: improve
Reason: recover
Initiative: AMF
Initiative: suggested
Initiative: spontaneously
Previous defaults
Previous repaid loans
Distance AMF
Matara
Hambantota
Galle
Table 4b: Socio-demographic variables
Female
Age
Education
House members
Fishery
Manufactory
Trade
Other job
Real income
0.82
45.01
10.93
4.76
0.01
0.38
0.30
0.16
34,307
Legend: Data refer to the first two time windows (P1 and P2), before the Tsunami.
33#
#
Table 5: Determinants of loan size
Variables
(1)
(2)
Damaged
(3)
(4)
25,139***
(7,600)
Damage: family
-5,845
(23,313)
8,094
(14,500)
3,103
(10,436)
-16,532
(11,269)
11,044
(11,154)
17,804**
(7,343)
Damage: house
Damage: office building
Damage: working tools
Damage: raw materials
Damage: ec. Activity
Sum of damages
Previous defaults
Distance AMF
Reason: new business
Reason: improve
Previous repaid loans
Previous defaults
Initiative: suggested
Initiative: spontaneously
Female
Age
Education
House members
Fishery
Manufactory
Trade
Other job
(5,758)
-23,358**
(9,612)
79.47
(315.8)
2,054*
(1,086)
1,446
(1,826)
12,430
(15,363)
-2,549
(6,297)
1,147
(5,758)
1,546
(8,963)
6,280
(7,284)
198.8
(291.1)
-12,401
(7,910)
12,408
(7,941)
-2,433***
(684.3)
6,280
(7,284)
21,077***
(6,469)
14,524**
(7,219)
(5,619)
-25,966***
(8,950)
-27.28
(364.0)
2,710**
(1,137)
2,285
(1,708)
11,117
(14,945)
-4,147
(6,230)
-2,957
(5,619)
1,134
(9,112)
34#
#
(5)
6,497
(6,785)
218.6
(266.0)
-12,096*
(7,069)
12,052*
(7,110)
-2,911***
(688.4)
6,497
(6,785)
19,853***
(6,065)
11,846*
(6,926)
(5,245)
-29,013***
(8,482)
25.85
(349.6)
3,238***
(1,054)
1,647
(1,659)
3,542
(14,938)
-6,087
(5,923)
-6,081
(5,245)
2,472
(8,590)
6,875
(7,532)
204.8
(272.1)
-11,485
(7,244)
13,358*
(7,448)
-2,850***
(698.4)
6,875
(7,532)
19,068***
(6,019)
11,821*
(6,701)
(5,384)
-28,200***
(8,811)
-48.08
(357.2)
2,960***
(1,060)
2,174
(1,712)
6,483
(14,253)
-6,585
(6,225)
-5,563
(5,384)
3,033
(8,720)
3,607
(2,581)
6,484
(7,216)
203.7
(289.1)
-10,352
(7,662)
14,891*
(7,853)
-2,535***
(725.7)
6,484
(7,216)
20,233***
(6,688)
13,192*
(7,328)
(5,593)
-27,187***
(8,849)
13.95
(369.4)
2,785**
(1,135)
2,316
(1,717)
9,221
(15,470)
-6,047
(6,437)
-4,678
(5,593)
562.5
(9,232)
(Cont.)
Remittances
-17,836
(21,043)
-15,971*
(8,784)
17,393
(10,977)
10,727
(10,529)
2,244
(15,006)
-37.18
(8,718)
8,385
(17,144)
0.0410
(0.147)
Subsidies
Donations and grants
Loans: bank
Loans: MFI
Loans: family/friend
Loans: other
Real income
Observations
R-squared
749
0.237
-17,733
(23,081)
-12,345
(7,948)
10,977
(10,996)
3,563
(9,485)
4,650
(15,715)
-5,132
(7,545)
-6,715
(14,859)
0.138
(0.153)
702
0.294
-13,177
(19,239)
-15,153*
(7,859)
5,609
(10,870)
1,611
(9,478)
2,584
(14,606)
-4,108
(7,211)
-4,698
(13,564)
0.164
(0.153)
702
0.316
-13,440
(22,469)
-14,563*
(8,289)
6,456
(12,161)
2,387
(9,067)
2,557
(14,715)
-5,430
(7,274)
-5,983
(13,899)
0.150
(0.148)
702
0.309
-16,232
(21,533)
-12,523
(8,058)
5,537
(12,432)
525.5
(9,807)
3,968
(15,689)
-6,531
(7,327)
-7,146
(14,785)
0.143
(0.156)
702
0.299
Legend: The dependent variable is Loan Amount (the amount of the loan in December 2011 terms).
Results come from OLS regressions with standard errors clustered at the borrower level. Regressions
make use of time and village dummy variables (omitted for reasons of space but available upon
request). Robust standard errors are reported in parentheses: *** p<0.01, ** p<0.05, * p<0.1.
#
#
35#
#
Table 6: Determinants of credit defaults
Variables
(1)
(2)
Damaged
Damage: office building
Damage: working tools
Damage: raw materials
Damage: econ. activity
Reason: new business
Reason: improve
Loan amount
Interest rate
Previous repaid loans
Previous defaults
Initiative: suggested
Initiative: spontaneously
0.0338
(0.165)
0.00308
0.00421
0.00750
0.00361
(0.0207)
(0.0204)
(0.0211)
(0.0203)
0.0427
0.0495
-0.131
0.0485
(0.616)
(0.624)
(0.685)
(0.624)
-0.127
-0.114
-0.318
-0.108
(0.601)
(0.605)
(0.679)
(0.627)
1.83e-05*** 1.79e-05*** 1.80e-05*** 1.81e-05***
(4.02e-06)
(3.96e-06)
(4.01e-06)
(4.07e-06)
-0.0304**
-0.0306**
-0.0308**
-0.0303**
(0.0146)
(0.0144)
(0.0151)
(0.0147)
-0.129
-0.125
-0.161
-0.127
(0.105)
(0.103)
(0.111)
(0.105)
-0.556
-0.560
-0.562
-0.559
(0.751)
(0.754)
(0.764)
(0.752)
-0.930*
-0.934*
-0.984*
-0.920*
(0.545)
(0.543)
(0.552)
(0.544)
-0.478
-0.491
-0.388
-0.477
(0.446)
(0.451)
(0.452)
(0.446)
(Cont.)
36#
#
(5)
1.262
(1.359)
-0.819
(0.711)
-0.0170
(0.727)
0.0754
(0.833)
0.0180
(0.616)
0.725
(0.570)
Damage: house
Distance AMF
(4)
0.232
(0.437)
Damage: family
Sum of damages
(3)
(Cont.)
Female
Age
Education
House members
Fishery
Manufactory
Trade
Other job
Remittances
Subsidies
Donations and grants
Loans: bank
Loans: MFI
Loans: family/friend
Loans: other
Real income
-0.134
(0.430)
-0.0194
(0.0155)
-0.0436
(0.0595)
0.00206
(0.120)
1.942**
(0.938)
-0.0367
(0.364)
0.327
(0.368)
0.375
(0.448)
-2.741
(2.938)
0.561
(0.527)
0.479
(0.579)
-0.298
(0.582)
1.146
(0.957)
-0.814**
(0.405)
-0.478
(0.932)
4.78e-06
(6.29e-06)
-0.0175
(0.693)
-0.0113
(0.0210)
-0.121
(0.0774)
0.0545
(0.158)
0.482
(1.577)
-0.381
(0.408)
0.158
(0.433)
-0.736
(0.602)
-1.880*
(1.072)
0.637
(0.711)
0.558
(0.731)
-1.205
(0.894)
0.512
(0.710)
-0.390
(0.501)
0.474
(0.793)
3.27e-06
(7.63e-06)
-0.273***
(0.0592)
-0.00591
(0.0380)
-0.0318
(0.687)
-0.0111
(0.0208)
-0.115
(0.0779)
0.0481
(0.157)
0.414
(1.612)
-0.400
(0.412)
0.141
(0.437)
-0.761
(0.621)
-1.835*
(1.063)
0.630
(0.712)
0.480
(0.740)
-1.221
(0.900)
0.525
(0.700)
-0.388
(0.506)
0.510
(0.795)
3.56e-06
(7.65e-06)
-0.269***
(0.0602)
-0.00692
(0.0380)
0.0168
(0.690)
-0.0113
(0.0207)
-0.125
(0.0834)
0.0402
(0.154)
0.351
(1.568)
-0.415
(0.423)
0.0874
(0.468)
-0.810
(0.634)
-1.818
(1.144)
0.717
(0.787)
0.662
(0.747)
-1.368
(0.954)
0.551
(0.707)
-0.357
(0.526)
0.406
(0.782)
3.53e-06
(8.33e-06)
-0.268***
(0.0600)
-0.0139
(0.0377)
-0.0228
(0.697)
-0.0115
(0.0211)
-0.119
(0.0776)
0.0522
(0.156)
0.471
(1.584)
-0.391
(0.410)
0.151
(0.431)
-0.736
(0.609)
-1.876*
(1.072)
0.630
(0.716)
0.511
(0.762)
-1.213
(0.901)
0.517
(0.704)
-0.396
(0.510)
0.501
(0.784)
3.37e-06
(7.72e-06)
-0.271***
(0.0602)
-0.00620
(0.0381)
717
0.530
660
0.642
660
0.642
660
0.646
660
0.642
Duration
Frequency
Observations
R-squared
Legend: The dependent variable is Default (dummy variable equal to 1 if the loan has not been repaid, 0
otherwise). Results come from Logit regressions with standard errors clustered at the borrower level.
Regressions make use of time and village dummy variables (omitted for reasons of space but available
upon request). Robust standard errors are reported in parentheses: *** p<0.01, ** p<0.05, * p<0.1.
37#
#
Table 7: Determinants of credit defaults, only damaged
Variables
Female
Age
Education
House members
Fishery
Manufactory
Trade
Other job
Remittances
Subsidies
Donations and grants
Loans: bank
Loans: MFI
Loans: family/friend
Loans: other
Real income
Distance AMF
Reason: new business
Reason: improve
Loan amount
Interest rate
Previous repaid loans
Previous defaults
Initiative: suggested
Initiative: spontaneously
Duration
Frequency
DV years 2004-2006
DV years 2007-2011
Coeff.
Robust Std. Err.
0.205
-0.0074
-0.0634
0.243
3.257*
-0.339
-0.387
-0.997
-0.678
0.79
0.88
1.309
-0.563
-7.55E-06
0.0283
-0.72
-0.872
9.02e-06*
-0.0148
-0.255
0.368
-0.506
-0.362
-0.0328
0.0563
3.882***
-1.735
-0.606
-0.0212
-0.0778
-0.174
-1.808
-0.466
-0.49
-0.685
-0.845
-1.155
-0.668
-1.444
-0.524
-8.89E-06
-0.0229
-1.661
-1.618
-4.65E-06
-0.0162
-0.156
-0.522
-0.536
-0.487
-0.0811
-0.0379
-0.622
-1.687
Observations
R-squared
335
0.474
Legend: The dependent variable is Default (dummy variable equal to 1 if the loan has not been repaid, 0
otherwise). Results come from Logit regressions with standard errors clustered at the borrower level.
The regression makes use of village dummy variables (omitted for reasons of space but available upon
request). Robust standard errors are reported in parentheses: *** p<0.01, ** p<0.05, * p<0.1.
38#
#
Table A1: Determinants of loan interest rate
Variables
Female
Age
Education
House members
Fishery
Manufactory
Trade
Other job
Remittances
Subsidies
Donations and grants
Loans: bank
Loans: MFI
Loans: family/friend
Loans: other
Real income
(1)
(2)
(3)
(4)
(5)
2.030
(4.083)
-0.0879
(0.133)
0.461
(0.472)
0.160
(0.878)
-11.63**
(4.985)
1.428
(3.455)
-4.823
(2.942)
-3.220
(4.824)
-7.587**
(3.146)
-0.631
(4.422)
-3.357
(4.245)
1.476
(4.406)
-13.61**
(5.443)
0.180
(4.057)
-5.808
(6.913)
0.000186***
(6.85e-05)
-0.552
(2.221)
-0.0470
(0.0775)
0.283
(0.245)
0.463
(0.449)
-6.054
(5.019)
-0.102
(1.491)
-0.439
(1.417)
-2.255
(2.332)
-3.535
(4.882)
-2.848
(2.316)
2.575
(2.858)
2.327
(1.964)
-9.143
(6.316)
-0.0251
(2.198)
2.503
(6.929)
3.21e-05
(2.85e-05)
-0.886
(2.192)
-0.0439
(0.0768)
0.352
(0.249)
0.419
(0.455)
-6.691
(4.999)
-0.314
(1.453)
-0.759
(1.370)
-2.126
(2.321)
-3.122
(5.088)
-3.191
(2.291)
1.991
(2.904)
2.186
(1.984)
-9.279
(6.240)
0.0216
(2.172)
2.576
(6.944)
3.41e-05
(2.88e-05)
-0.800
(2.242)
-0.0593
(0.0765)
0.300
(0.256)
0.521
(0.458)
-5.786
(4.979)
-0.132
(1.452)
-0.694
(1.378)
-2.153
(2.326)
-2.039
(4.289)
-4.151*
(2.263)
1.804
(2.660)
2.515
(2.050)
-10.30*
(6.209)
-0.287
(2.138)
2.388
(7.004)
3.17e-05
(2.87e-05)
-0.675
(2.240)
-0.0432
(0.0763)
0.294
(0.247)
0.471
(0.455)
-6.198
(4.992)
-0.289
(1.450)
-0.603
(1.368)
-2.315
(2.337)
-3.408
(5.050)
-2.879
(2.327)
2.023
(2.792)
2.049
(2.031)
-9.183
(6.318)
-0.177
(2.209)
2.431
(6.925)
3.22e-05
(2.85e-05)
(Cont.)
39#
#
(Cont.)
Distance AMF
0.248***
0.253***
0.246***
0.249***
(0.0621)
(0.0623)
(0.0616)
(0.0624)
-2.339
-2.334
-2.114
-2.145
(2.679)
(2.694)
(2.801)
(2.752)
-2.149
-2.142
-2.059
-1.904
(2.130)
(2.146)
(2.357)
(2.284)
-4.44e-05*** -4.80e-05*** -4.72e-05*** -4.51e-05***
(1.57e-05)
(1.58e-05)
(1.54e-05)
(1.57e-05)
0.998***
0.958***
0.946***
0.992***
(0.304)
(0.297)
(0.304)
(0.299)
0.0792
0.112
-0.309
0.0938
(2.061)
(2.031)
(1.986)
(2.050)
-1.392
-1.418
-1.470
-1.453
(1.440)
(1.430)
(1.437)
(1.470)
2.611
2.433
2.174
2.505
(1.706)
(1.718)
(1.773)
(1.759)
-1.871***
-1.866***
-1.877***
-1.870***
(0.220)
(0.221)
(0.219)
(0.221)
-0.187
-0.224
-0.183
-0.199
(0.249)
(0.252)
(0.254)
(0.248)
-1.366
(4.166)
2.636
(2.916)
4.371*
(2.344)
-4.752
(2.898)
-0.378
(2.198)
1.743
(2.076)
2.535
(1.953)
0.351
(0.563)
Reason: new business
Reason: improve
Loan amount
Previous repaid loans
Previous defaults
Initiative: suggested
Initiative: spontaneously
Duration
Frequency
Damaged
Damage: family
Damage: house
Damage: office building
Damage: working tools
Damage: raw materials
Damage: econ. activity
Sum of damages
Observations
R-squared
749
0.215
690
0.647
690
0.648
690
0.652
690
0.647
Legend: The dependent variable is the interest loan charged to the borrowers. Results come from OLS
regressions with standard errors clustered at the borrower level. Regressions make use of time and
village dummy variables (omitted for reasons of space but available upon request). Robust standard
errors are reported in parentheses: *** p<0.01, ** p<0.05, * p<0.1.
40#
#