Vulnerability of Microfinance to Natural Disasters:

Evidence from the 2005 Pakistan Earthquake∗

By
Hidayat Ullah Khan# and Takashi Kurosaki$

March 2007

Abstract:
This paper attempts to quantify the ill-effects of covariate shocks such as natural disasters on the
sustainability of microfinance. The vulnerability of microfinance to covariate shocks is well-
known theoretically and in anecdotes, but very few quantitative evidence exists. For our attempt,
we use a unique dataset of about 3,000 micro-borrower households covering over a period of
1994-2006 including the event of a disastrous earthquake in Pakistan. Based on the difference-
in-difference approach, contrasting regions that were hit by the earthquake and regions that were
not, we found that the delay in repayment in the affected areas was 52% higher than that in the
unaffected areas. The observed difference in the repayment delay was decomposed into changes
in borrowers' composition and borrowers' behavior. The decomposition result shows that the
changes in borrowers' behavior accounted for a large portion of the difference, suggesting a
serious difficulty faced by borrowers and microfinance institutions in the earthquake-hit regions.

1. Introduction

This paper empirically evaluates the impact of natural disasters on microfinance

programs in North Pakistan, which was struck by a deadly earthquake in 2005. Microfinance is
regarded as one of the successful poverty reduction policies. Poor households, once obtained
credit from a microfinance institution, will be able not only to enhance their income and
consumption levels, but also to improve their tolerance against idiosyncratic income shocks
through improved access to credit (de Aghion and Morduch, 2005). However, microfinance
schemes are vulnerable to covariate shocks such as natural disasters because the majority of the
borrowers are affected adversely by these shocks, resulting in difficulty in repayment. The

∗
An incomplete version, which is better understood as an extended abstract. The full paper will be
completed by the middle of April.
#
Graduate School of Humanities and Social Sciences, University of Tsukuba. E-mail:
masmaleo@yahoo.com.
$
Institute of Economic Research, Hitotsubashi University. E-mail: kurosaki@ier.hit-u.ac.jp.

vulnerability of microfinance to covariate shocks is well-known theoretically and in anecdotes,

but very few quantitative evidence exists (de Aghion and Morduch, 2005). To mitigate the
repayment burden of disaster-hit borrowers, some authors describe the importance of
rescheduling during disasters (e.g., Meyer, 2002; Norell, 2001). A rare study by Shoji (2006)
estimated the impact of rescheduling on the borrowers' dependence on moneylenders in
Bangladesh where floods often affect borrowers' repayment ability. Nevertheless, we know very
little on the size of repayment delay caused by a very covariate and unexpected natural shock,
like the 2005 Pakistan earthquake.
This paper thus attempts to quantify the ill-effects of a large-scale earthquake on the
sustainability of microfinance. For this attempt, we use a unique dataset of about 3,000 micro
borrowers in North Pakistan, covering over a period of 1994-2006 including the event of the
2005 Earthquake. Pakistan is one of the low income countries in South Asia where income
poverty is prevalent and the majority of the poor have difficulty in their access to efficient
sources of credit (World Bank, 2002). The effect of the microfinance services in Pakistan is
estimated to be positive, although their outreach is not to the expected level, leaving many of
the poor untouched (Montgomery, 2005). The occurrence of the earthquake is thus regarded as a
critical challenge for microfinance schemes in Pakistan.
Based on the difference-in-difference approach, contrasting regions that were hit by
the earthquake and regions that were not, we found that the delay in repayment in the affected
areas was 52% higher than that in the unaffected areas. The observed difference in the
repayment delay was decomposed into changes in borrowers' composition and borrowers'
behavior. The decomposition result shows that the changes in borrowers' composition accounted
for a small portion of the difference.

2. Background

2.1 Microfinance in Pakistan

Microfinance is still relatively new to Pakistan, both from concept and practice point
of view. The country is among the largest potential microfinance markets in the world with
potential borrowers of 10 million adults. This conservative estimate is likely to expand
dramatically with Pakistan's high population growth rates.
The country is in the business of micro-credit on limited basis for many decades but
modern style microfinance began in the late 1990s. The main providers of microfinance at that
time were NGOs and government-supported rural support networks.
Micro finance was declared a priority in the official Poverty Alleviation Strategy in
1999 and a regulatory framework for the promotion of microfinance was established in 2001.

The result was a massive investment of at least US $400 million from 1999 to 2005 largely
funded from multilateral resources such as the World Bank and the Asian Development Bank.
Among various microfinance providers today, the Khushhali Bank, the flagship institution
established by the government in 2000, serves over 175,000 active clients in 2005, more than
the total number of clients reached by all the NGOs and rural support programs before 2001
(Montgomery, 2005). Microfinance institutions in Pakistan now comprise a voluntary network
called Pakistan Microfinance Network (PMN). With the passage of time, the country not only
moved forward its own way but also benefited from the experience of those who were way
ahead and much before in the business.
The economy Pakistan had shown moderate growth during the 1990s, but due to
economic reforms and structural adjustments supervised by the IMF and the World Bank, the
poverty and the income inequality continued to rise (World Bank, 2002). The macro-economic
scene in the country over the last six years (2000-05) has changed drastically by an average
annual rate growth of 6% since 2000. Besides, the debt-to-GDP ratio fell from almost 100% to
60% and exports have also been doubled. Despite this economic turnaround accompanied with
macroeconomic improvements, the absolute level of real income per capita is still very low,
standing at US$ 632 in 2004 evaluated at the official exchange rate and at US$ 2,225 after the
PPP adjustment (UNDP, 2006). The challenge also remains to transfer the benefit of growth
equitably across the board and especially to the poor. This is the economic situation when the
earthquake hit the country.

2.2 The 2005 Pakistan Earthquake

The earthquake on October 8, 2005, left widespread destruction, killing at least 73,000
people, severely injuring another 70,000, and leaving 2.8 million people without shelter. The
affected areas suffered an extensive damage to economic assets and infrastructure, with social
service delivery, commerce, and communications either severely weakened or destroyed.
Beyond the human toll, the overall cost associated with the earthquake is significant. A Joint
Damage and Needs Assessment by a group of experts from the World Bank and the Government
of Pakistan put the value of the direct damage due to the earthquake at US$2.3 billion while
resulting indirect losses estimated at US$576 million.
The estimated costs for relief, livelihood support for victims, and reconstruction cost
are estimated at approximately US$5.2 billion. The earthquake has adversely affected the
economy, especially the fiscal deficit. Fiscal deficits during the three years following the
earthquake are expected to increase by 0.5 to 0.9% of GDP per year. These pressures could pose
difficulties for Pakistan's macroeconomic balances and have the potential to compromise the
achievement of its long-term development goals. Given the magnitude of resources needed for

rehabilitation of earthquake-affected areas, it is unlikely that the government will be able to

fully absorb the fiscal impact of the earthquake without significantly affecting public sector
development activities. Moreover, it is important that the poor in areas not affected by the
disaster (the vast majority) may have adversely affected as well due to increased allocation of
public resources to the earthquake-affected areas at the expense of the rest of the country.

3. Methodology

3.1 Data
We employ a micro dataset maintained for the purpose of financial monitoring of
micro credit intervention. It is comprised of about 3,000 borrower households from one of the
districts severely affected by the earthquake in North Pakistan. The dataset is a pooled cross-
section, which covers the period from 1994 to 2006. As this study uses a unique dataset of
microfinance before and after an unprecedented natural disaster, its analysis can contribute to
the existing literature in an important way.
The information in the dataset is obtained from a member of Pakistan Microfinance
Network (PMN), a voluntary network of microfinance institutions in Pakistan. The names and
identities of the participating households are replaced by computer-generated numbers for the
sake of privacy.
The initial data provided by the microfinance institution were at the installment level
of repayment, having details about every installment, when it was due, when it was actually
paid, etc. We aggregate this information at the loan agreement level for the sake of consistency,
by calculating the borrower-average delay for all installments included in the loan agreement.
Let Yit denote this variable, which is to be used as the dependent variable for the empirical
analysis, where i refers to the loan agreement and t refers to its time period.
We match the loan-agreement-level repayment data thus compiled with the household-
level information, such as income sources of the household, consumption expenditures patterns,
dwelling conditions including the availability of basic necessities, the distance between the
locality of the borrower household and the microfinance institution office, the membership in a
community-based organization (CBO), the amount of household saving deposited at the CBO,
etc. We use these household-level variables (Xit in a vector notation) to control for the changes
in borrowers' composition.
The date of the earthquake is used as a time line to separate data by generating a time
dummy. The data has three types of households as borrowers; first "those who have both
borrowed and repaid before the quake", second "those who borrowed before the earthquake and
their repayment was sandwiched between before and after the earthquake, and third "those who

have borrowed and repaid after the quake". The second category of households is not the focus
of this study as the number of these observations is not large enough due to missing information
of key variables. Therefore, the first and third category of the households, i.e., "borrowed and
paid before" and "borrowed and paid after" the shock are analyzed in this paper. A dummy
variable Dt is created, which takes the value of 1 if the credit transaction occurred after the
earthquake and the value of 0 otherwise.
The additional information about the distance between the locality of each borrower
and the epicenter of the earthquake is obtained from the website of the Earthquake
Reconstruction and Rehabilitation Agency (ERRA). To capture the impacts of the earthquake, a
dummy variable Di is created, which takes the value of 1 if the household was located within a
distance of 75 km from the epicenter of the earthquake, and the value of 0 otherwise 0. The
threshold value of 75km is chosen after consultation with specialists on earthquakes as a useful
line to divide regions into those affected more severely by the earthquake and those affected less
severely. In the empirical analysis, we check the robustness of empirical results with respect to
the choice of Di.

3.2 Empirical Models

We identify the earthquake impact through a difference-in-difference (DID) approach,
treating those borrowers residing in areas far from the epicenter of the earthquake as the
"control group" and those residing in areas near to the epicenter as the "treatment group." An
unconditional DID estimator can be obtained from our dataset through estimating a regression
model:

Yit = a0 + a1Di + a2Dt + a3DiDt + uit, (1)

where a's are parameters to be estimated. If Di and Dt are orthogonal to the error term uit, the
impact of the earthquake on the average delay (Yit) can be inferred by the estimated coefficient
of a3 on the interaction term. This is because:

a3 = {E(Yit | Di=1, Dt=1) - E(Yit | Di=1, Dt=0)} - {E(Yit | Di=0, Dt=1) - E(Yit | Di=0, Dt=0)}. (2)

However, the composition of borrowers may have changed after the earthquake in a
different way between the control and treatment regions. Even without a change in the behavior
of individual borrowers with respect to repayment, such compositional changes are likely to
result in the difference in the aggregate repayment ratio. In other words, it is likely that the
assumption of orthogonality among Di, Dt, and uit is not satisfied and uit is correlated with Di and

Dt. To control for this composition effect, we estimate a regression model:

Yit = b0 + b1Di + b2Dt + b3DiDt + Xitb + eit, (3)

where Xit is a vector of borrowers' characteristics and b is a vector of coefficients (b4, b5, b6, ...)'
that captures the effect of each of Xit on the repayment. If Di and Dt are orthogonal to the error
term eit, the impact of the earthquake on the average delay (Yit) conditional on the household-
level correlates can be inferred by the estimated coefficient of b3 on the cross term, as:

b3 = {E(Yit | Di=1, Dt=1, Xit) - E(Yit | Di=1, Dt=0, Xit)}

- {E(Yit | Di=0, Dt=1, Xit) - E(Yit | Di=0, Dt=0, Xit)}. (4)

For the orthogonality condition to be satisfied, the treatment and control group should
meet the basic criteria for comparison, e.g., both must have similar characters in unobservables.
Our field observations support this: the groups are similar in terms of socio-economic
background and the majority of the member households are poor with more or less the same
asset endowments. In discussing the empirical results, we show other evidence to support the
orthogonality condition.
By subtracting equation (4) from (2), we can obtain the following decomposition of
the unconditional DID:

a3 = b3 - [{E(Xit | Di=1, Dt=1) - E(Xit | Di=1, Dt=0)} - {E(Xit | Di=0, Dt=1) - E(Xit | Di=0, Dt=0)}]b.
(5)
The second term of equation (5) shows the effect of the difference-in-difference in
borrowers' characteristics. Equation (5) thus enables us to decompose the observed difference in
the repayment delay into those attributable to changes in borrowers' behavior (term b3) and
those attributable to changes in borrowers' composition. The composition effect can be
decomposed further into each variable included in vector Xit.
In estimating (3), eight variables are included, whose definitions are given in Table 1.
The variable "single_y" controls for the possibility that a household with a single source of
income is more vulnerable to shocks as compared to the one having alternative sources of
income. In other words, income diversification is a good ex-ante risk coping strategy. The
information about alternative source of income ("alts_y"), i.e., whether the household have
another source of income except the primary source, is also available. By this variable, we can
analyze how the households are able to diversify their income risk in terms of their repayment
behavior. To have comprehensive analysis of income while dealing with covariate shocks, we

use another dummy variable, "outs_y," which will enable us to analyze the diversification on the
basis of spatial distribution of risk.
Community based organizations (CBOs) require their member households to
accumulate the savings that are kept in a bank account operated by the CBO. It is likely that the
households which accumulate higher savings are comparatively more protected against negative
shocks. To control for the effectiveness of savings as a strategy for risk mitigation or coping
with shocks, a variable "savings" is included in the econometric model. The dataset also
provides information about poverty rankings of the poor, i.e., upper and lower quartile of poor,
which are assigned after asking a series of questions from the household about their
expenditures pattern, observing their dwelling conditions, and availability of some basic
necessities. A binary variable ("status") thus composed is included in the econometric model to
capture this effect.
As proxy variables for the access to mutual help and credit transactions, three
variables "distance," "posi," and "members" are included. The variable distance shows us
whether the physical distance between the borrower and lenders matters or not. The variable
posi controls for the effect that holding a manager's position in a CBO affects the repayment
behavior. The variable members, i.e., the number of total CBO members belonging to the CBO
to which the borrower belongs, investigates whether the size of a community network matters in
its ability to tackle with the natural disasters. This also shows the effect of cohesion within a
group, i.e., peer pressure which is used to regulate repayment. The effectiveness peer pressure
can raise a serious question about the very fabric of the community organization in the post-
shock scenario. This variable will be another control for the social aspects of the microfinance,
which is largely ignored in the existing studies.

4. Estimation Results

The OLS regression results of equation (3) are shown in Table 2 with Huber-White
heteroscedasticity-robust standard errors. First, the coefficient on the distance from the epicenter
of the earthquake (Di) is insignificant, which supports our identifying assumption that control
and treatment group are similar. At the same time, the time dummy used for the earthquake
shock (Dt) has a significant (1% level) and positive coefficient. This indicates that repayments
were delayed on average after the earthquake regardless of the place where the borrower resided.
One is tempted to interpret this as an adverse effect of the earthquake on the repayment.
However, since Dt controls for all macroeconomic shocks, the positive coefficient may capture
these macro shocks other than the earthquake.
We therefore concentrate on the coefficient on the cross term (DiDt) to infer the effect

7
of the earthquake on the repayment. The coefficient on the interaction term is highly significant
which has the 1% level of significance and coefficient bears a positive sign, as expected. This
can be interpreted as the earthquake resulted in an increase of 52% delays in repayments on
average. It shows the severity of the earthquake's impact both on borrowers and lenders. This
kind of events threatens the very existence of the intervention meant for the improvement of the
poor's access to credit. Without additional support, microfinance institutions cannot survive
such situations.
We briefly discuss the effects of household-attributes on the repayment. Households
with a single source of income (single_y) have registered about 20% higher delays. This is in
conformity with the notion of high levels of vulnerability among the households depending on a
single source of income. Households with alternative sources of income (alts_y) have reported
about 2% less average delays than the ones without diversification. Income diversification
within the same area thus provides very low level of protection. This needs to be complemented
either by micro-insurance or by diversification based on the geographical distribution of the risk.
Households with income from the source outside the effected area (outs_y) have reported over
34% fewer delays as compared to those not having, confirming the effectiveness of the
diversification over space.
The control used for grass-root-level managers (posi) yields statistically insignificant
results. The physical distance between the borrower and the lender (distance) also has an
insignificant coefficient. The cash savings can be used as an effective ex ante risk management
strategy. Unexpectedly, however, the coefficient on the variable "savings" is significant and
negative. This may be due to the low saving capacity of the poor, because of the concentration
of activities in conventional enterprise.
For every additional member of the community organization (members), an average
2% of the delay is observed. In other words, the group cohesion reduces as the group expands
and results in less pressure from peers. If a substantial repayment delay poses a financial threat,
then the deteriorating group cohesion can jeopardize the building block of the sustainability of
microfinance, i.e., the peer pressure.
As discussed in the subsection on the empirical models, changes in borrowers'
composition also affect the repayment after the earthquake, in addition to the effect captured by
the coefficient on the cross term (DiDt). We implement the decomposition following equation
(5) to examine whether the changes in borrowers' composition indeed affected the overall
repayment ratio and if yes, which variable was the most influential. Our preliminary results
indicate that the composition effects explained only a small portion of the difference, suggesting
that the main source of the increased delay was changes in repayment behavior caused by the
earthquake (the decomposition results will be provided in the next version of this paper).

Overall, the regression results demonstrate a serious difficulty faced by borrowers and
microfinance institutions in the aftermath of the devastating earthquake.

5. Conclusion

Microfinance has helped the poor with ready access to credit that can be used to
increase their income as well as to smooth consumption. However, this tool is prone to covariate
shocks such as natural disasters, although the rigorous evidence for the vulnerability is lacking
in the literature. This paper thus quantified the ill-effects of an unprecedented covariate shock
(the 2005 Pakistan earthquake) on the sustainability of microfinance in North Pakistan. Based
on a unique micro dataset covering over a period of 1994-2006 including the earthquake event,
we found that the delay in repayment in the affected areas was 52% higher than that in the
unaffected areas. The observed difference in the repayment delay was decomposed into changes
in borrowers' composition and borrowers' behavior. The decomposition result showed that the
changes in borrowers' behavior accounted for a large portion of the difference, suggesting a
serious difficulty faced by borrowers and microfinance institutions in the earthquake-hit regions.
From the analysis in this paper, the following policy implications, although tentative,
are derived. First, in order to enhance the sustainability of microfinance institutions, it is
important to protect borrowers against negative and covariate shocks, such as natural disasters.
Microinsurance of the enterprise funded through micro-credit as well as the other income
generating assets of the poor may reduce the vulnerability to a manageable level, if co-insured
at the global level. Second, investment in human capital with diversification across wider
regions is an effective tool to prevent the local shock to affect repayment adversely. Assessment
of poverty thus should take into account the risk factor associated with spatial distribution of
assets as well. A more general lesson is that microfinance institutions should prepare themselves
to tackle with not only idiosyncratic but also covariate shocks.

References:

de Aghion, Beatriz Armendariz and Jonathan Morduch (2005) The Economics of Microfinance.
Cambridge, Mass.: MIT Press.
Montgomery, Heather (2005) "Meeting the Double Bottomline: The Impact of Khushhali Bank's
Microfinance Program in Pakistan." ADBI Policy Paper No.8, November 2005
(http://adbi.adb.org/files/2005.09.28.cpp.khushhali.microfinance.study.pdf).
Meyer, R.L. (2002) "The Demand for Flexible Microfinance Products: Lessons from
Bangladesh." Journal of International Development 14(3): 351-368.
Norell, D. (2001) "How to Reduce Arrears in Microfinance Institutions." Journal of
Microfinance 3(1): 115-130.
Shoji, Masahiro (2006) "Evaluation of Flexible Repayment System in Microfinance: A Case
Study from a Natural Disaster in Bangladesh." Mimeo. Univ of Tokyo.
UNDP (2006) Human Development Report 2006 - Beyond Scarcity: Power, Poverty and Global
Water Crisis.
World Bank (2002) Pakistan Poverty Assessment - Poverty in Pakistan: Vulnerabilities, Social
Gaps, and Rural Dynamics. Report No. 24296-PAK, October 2002.

10

Table 1: Definition and Summary Statistics of Empirical Variables

Variable Definition
NOB. Mean Std.Dev. Min. Max.

Dependent variable:

Yit The average delays in repaying

2644 6.710 1.554 0 10.115
each installment (days).

DID related variables:

Di Distance from the epicenter of 2950 0.317

the earthquake (dummy equal to 1
If within an area of 75km from
the epicenter).

Dt Time dummy equal to 1 if the 2950 0.286

credit transaction occurred after
the earthquake.

Explanatory variables in Xit:

single_y Dummy for a household with 2950 0.293

a single source of income.

alts_y Dummy for a household with

2911 0.392
alternative sources of income.

outs_y Dummy for a household with

2950 0.207
income sources located outside

the affected area.

savings Household savings (Pak. Rs.).

2940 12201 16262 0 95000

posi Dummy for a CBO manager.

2950 0.373

status Dummy for a household belonging

2950 0.560
to the upper quantile of the poor.

distance Distance between the borrower and 2950 12.42 5.70 4 28

the lender (km).

members Number of members in a CBO to 2940 28.36 16.87 10 110

which the household belongs.

11

Table 2: Difference-in-Difference Estimation Results

Variable Coefficient Robust t-statistics

standard error
Di 0.324 0.274 1.18
Dt 0.00766*** 0.00260 2.95
Di Dt 0.526*** 0.107 4.92
single_y 0.200*** 0.0548 3.64
alts_y -0.0179*** 0.00517 -3.47
outs_y -0.342*** 0.0247 -13.86
savings 7.15e-06*** 1.98e-06 3.61
posi 0.0310 0.0674 0.046
status -2.391*** 0.227 -10.52
distance 0.0903 0.249 0.36
members 0.0185*** 0.00181 10.23
constant 9.238*** 0.351 26.33

Number of observations = 2596.

Prob > F = 0.0000
R-squared = 0.3422
Adj R-squared = 0.3394
Root MSE = 1.2643

Estimated by OLS with Yit (the average repayment delays in days) as the dependent variable, using
Huber-White heteroscedasticity-robust standard errors.

12