Accepted Manuscript

Title: Financial fragility and natural disasters: An empirical

analysis

Author: Jeroen Klomp

PII: S1572-3089(14)00053-9
DOI: http://dx.doi.org/doi:10.1016/j.jfs.2014.06.001
Reference: JFS 294

To appear in: Journal of Financial Stability

Received date: 14-8-2013

Revised date: 31-1-2014
Accepted date: 2-6-2014

Please cite this article as: Klomp, J.,Financial fragility and natural
disasters: An empirical analysis, Journal of Financial Stability (2014),
http://dx.doi.org/10.1016/j.jfs.2014.06.001

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
 Financial fragility and natural disasters:
An empirical analysis

Jeroen Klomp
Wageningen University, The Netherlands1

t
ip
This version, May 2014.

cr
us
Abstract
Using data for more than 160 countries in the period 1997 to 2010, we explore the impact of
large-scale natural disasters on the distance-to-default of commercial banks. The financial

an
consequences of natural catastrophes may stress and threaten the existence of a bank by
adversely affecting their solvency. After extensive testing for the sensitivity of the results, our
M
main findings suggest that natural disasters increase the likelihood of a banks’ default. More
precisely, we conclude that geophysical and meteorological disasters reduce the distance-to-
default the most due to their widespread damage caused. In addition, the impact of a natural
d

disaster depends on the size and scope of the catastrophe, the rigorousness of financial
te

regulation and supervision, and the level of financial and economic development of a
particular country.
p
ce
Ac

Keywords: Banking sector; Financial fragility; Natural disasters.

JEL Codes: G21; Q54

1
Corresponding author: Jeroen Klomp, Wageningen University, Development Economics Group, P.O. Box
8130, 6700 EW, Wageningen, The Netherlands. E-mail: jeroen.klomp@wur.nl. Telephone: +31317482950.

1
Page 1 of 35
1. Introduction
In 2001 an alarming report was published by the UNEP Finance Initiatives on banking in the
wake of large-scale natural catastrophes2. The main message of this report was that the
growing trend in the frequency and intensity of these severe natural events have the potential
to stress and threaten banks to the point of impaired viability or even insolvency, primarily

t
ip
caused by increasing the share of non-performing loans, raising the leverage or through the
occurrence of a bank run immediately after the disaster. It is, for instance, a well-known fact

cr
that there is an outflow of foreign private capital shortly after a disaster has struck as the
uncertainty about future repayment increases (Yang, 2008; David 2011). During the last three

us
decades, there have been about 10,000 natural disasters worldwide, affecting more than 7
billion people, and causing over the $2 trillion in estimated damages, and these numbers are
still steadily increasing (EM-DAT, 2013). According to the Basel Committee, natural

an
disasters are considered as an operational risk as it adversely affects the smooth functioning of
the various components of the financial system, i.e., financial markets, and payments,
M
settlement and clearing systems (BSBC, 2010).
One common characteristic of any large-scale natural disaster is that it adversely
affects large parts of the domestic financial sector at the same moment. To manage these
d

correlated shocks, banking regulators and supervisors require that banks maintain adequate
te

capital reserves. According to the UNEP report, supervisors should include the exposure of a
bank to natural disasters in their assessment of these reserve requirements, for example, by
p

connecting internal lending rates with local climatic conditions. Only, banks may be uncertain
ce

about their actual exposure to natural disasters, as assessing this complete risk can be quite
difficult. For instance, banks are not only affected by the direct impact of a natural
catastrophe, but also by the spillovers from the interbank market since banks are highly
Ac

connected due to their lending activities (Goldberg, 2009; Cetorelli and Goldberg, 2011;
Charnobai and Rachev, 2006)3. As a result, holding too little capital reserves threatens the
solvency of the lender when a catastrophe occurs. However, as banks are typically highly

2
The UNEP finance initiative is a partnership between United Nations Environment Program and about three
hundred banks, insurance companies and investment firms.
3
Moreover, up so far most research on the impact of disasters on the fragility of the financial system is done
using data on microfinance institutions (cf. Collier, 2011; Collier and Skees, 2012; Berg and Schrader, 2012;
Klomp, 2012). However, two important differences exist between MFIs and the more traditional commercial
banks. First, banks are much more interconnected due to interbank market lending. Second, MFIs are only active
in developing countries where the stock of physical capital at risk during a disaster is much smaller. These
differences could influence the impact of natural events on financial fragility.

2
Page 2 of 35
leveraged with low capital-to-asset ratios, holding excessive reserves represents significant
opportunity costs for lenders (Van Greuning and Bratanovic, 2009).
The existing empirical literature on the impact of natural disasters on the fragility and
performance of commercial banks is rather limited and inconclusive. A first attempt was
made by Steindl and Weinrobe (1983). These authors explore if the amount of deposits
received by a sample of US savings and loan associations and commercial banks reacted to

t
ip
major floods. Their main results do not provide any significant support for the popular view
that shortly after a natural disaster a bank run occurs. In contrast, in some econometric

cr
specifications, the authors find even evidence of an increase in the amount of deposits
received by these financial institutions in the aftermath. This outcome is rather

us
counterintuitive as one might expect at least some withdrawals of deposits to finance
emergency expenditures and reinvestments. Skidmore (2001) gives one potential explanation
for this outcome in the long run. He explores the effect of an increase in the probability of a

an
future economic loss on savings. His empirical evidence point out that damages caused by
natural disasters are positively correlated with household savings rates. This result suggests
M
that households have attempted to self-insure against some catastrophic events as insurance
markets have not provided a sufficient level of protection against possible losses arising from
natural catastrophes.
d

In addition, using a panel model including more than 100 countries, Noy (2009)
te

reports a contraction in the amount of credit supplied by banks to the private sector in the
aftermath of a disaster as banks become more concerned with the uncertainty of repayment in
p

the future. However, countries with more developed credit markets appear to be more robust
ce

and better able to endure natural disasters. Likewise, the results from David (2011) point out
that bank lending activities reduce rapidly after a climatic disaster in developing countries.
Berg and Schrader (2012) explore the impact of volcano eruptions in Ecuador on the loan
Ac

demand and access to credit using data from microfinance institutions. Their results
demonstrate that while credit demand increases due to volcanic activity, access to credit is
restricted. Hosono et al. (2012) go one step beyond and find that that a adverse natural
disaster shock to bank lending capacity reduces client firms’ activity even in an economy with
well-developed financial markets and institutions.
Using a simulation approach for Peruvian microfinance institutions, Collier et al.
(2013) indicate that natural catastrophes can be considered as a type of systemic risk. Their
results indicate that there is a drop in the capital ratio, equity and loan origination immediately
following a disaster. The conclusion is strengthened by Collier and Skees (2013) who find, by

3
Page 3 of 35
using a sample of more than 900 microfinance institutions, that there is a drop in the capital
ratio when the number of people affected by a natural catastrophe increases.
These studies so far provide only some circumstantial evidence on the relationship
between bank survival and natural disasters as they only study specific risk factors (cf capital
ratios, loan portfolio quality) or activities (cf. credit supply, deposits). In turn, our
contribution to the literature is instead that we examine more directly to what extent large-

t
ip
scale natural disasters are accountable for changes in the default risk faced by commercial
banks. For this purpose, we use a dynamic panel model including about 180 large-scale

cr
natural disasters over 160 countries in the period 1997 to 2010. Our measure of default risk is
based on the distance-to-default taken from the Database on Financial Development and

us
Structure reported by the World Bank4. The distance-to-default reflects the number of
standard deviations that a bank’s return on assets has to drop below its expected value before
equity is depleted and the bank is insolvent. In addition, we construct several measures on the

an
frequency and intensity of natural disasters based on the information provided by EM-DAT.
We address the potential endogeneity problems of the economic consequences of natural
M
disasters by presenting a system-GMM model.
After extensive testing for the sensitivity of the results, our main findings suggest that
natural disasters increase the likelihood of a banks’ default. More precisely, we conclude that
d

geophysical and meteorological disasters reduce the distance-to-default the most due to their
te

widespread damage caused. In addition, the impact of a natural disaster depends on the size
and scope of a natural disaster, the rigorousness of financial regulation and supervision, and
p

the level of financial and economic development of a particular country.

ce

The remainder of the paper is organized as follows. In the next section, we discuss our
theoretical foundation underlying our hypothesis. In section three, we describe our data and
methodology used, while in section four, we present our estimation results on the relationship
Ac

between natural disasters and the default risk in the banking sector. Finally, we end in section
five with our conclusion and discussion.

2. Theoretical foundation
According to the existing literature, the occurrence of natural disasters affects the default risk
of banks through various channels (cf. Collier, 2011; Collier and Skees, 2012; Berg and
Schrader, 2012). To illustrate the impact of large-scale natural disasters on the default risk, we

4
One advantage of measuring financial fragility using the z-score is that it is rather objective compared to more
subjective credit ratings (Demirgüç-Kunt and Detragiache et al., 2008)

4
Page 4 of 35
use the different risk dimensions identified by the International Monetary Fund (IMF)’s core
set of Financial Soundness Indicators or CAMEL indicators—i.e., capital adequacy, asset
quality, management, earnings and profitability, and liquidity (IMF, 2000). The first
dimension of risk refers to capital adequacy. According to the IMF (2000), capital adequacy
ultimately determines the robustness of banks to withstand adverse shocks to their balance
sheets. The reserves of a bank may deplete due to the large write-off of loan losses or the

t
ip
occurrence of a bank run after immediately after a natural disaster. As a result, disasters may
make banks over-leveraged or even insolvent.

cr
The second risk dimension is related to asset quality. When a major disaster strikes,
the degree of asset quality decreases significantly by the death or disability of borrowers.

us
Physical unscathed borrowers may be inhibited to attain earnings and pay interest or
principles. An increasing non-performing loans ratio in the aftermath of a disaster therefore
signals a deterioration of the quality of the credit portfolio, which increases the fragility of the

an
bank. Besides, asset risk may be also enhanced by the destruction of the collateral used by
borrowers to obtain a loan.
M
The third risk dimension that is distinguished by the IMF is associated with
managerial qualities. A high ratio of expenses to total revenues may indicate that financial
institutions are not operating efficiently due to management deficiencies. The increase in the
d

operational risk after a disaster arises from the disruption of the operational activities through
te

physical damages, e.g. the destruction of offices, equipment or information systems, or

through indirect implications like the inability of clients to reach offices. The fourth risk
p

dimension is related to the profitability of a bank. After a natural disaster, the profitability of a
ce

bank may under pressure due to a combination of the large write-offs, the decrease in
efficiency and the increased interbank interest rate as the uncertainty of repayment increases.
The final dimension the IMF distinguishes is liquidity risk. Insufficient liquidity may
Ac

threaten the survival of a bank after the occurrence of a large-scale natural disaster, notably so
in case of severe maturity mismatches. Liquidity risk is tightened in the aftermath of a disaster
by missing savings and immediate withdrawals of existing deposits to replace lost physical
capital or afford medical care. Those who are not in possession of deposits will probably
demand emergency loans, whether they are already existing clients or non-clients.
However, the impact of natural disasters on financial soundness of banks goes beyond
these direct impacts. For instance, reputational risk may arise from an institution’s inability to
fulfil its contractual obligations in the aftermath of a disaster, e.g. to disburse deposits to
savers and may cause a bank run. Additionally, banks are often part of a larger financial

5
Page 5 of 35
conglomerate including an insurance branch, which also have to incur substantial losses after
a disaster.
The theoretical relationship between natural disasters and the default probability of a
bank can be illustrated using the framework proposed by Boyd and Runkle (1993). Assume
that the probability of default of a particular bank can be represented by

t
ip
Pt  P0 e t   0,   0 (1)

cr
Where Pt is the default probability of a bank at time t. This probability is accumulating over
time due to the recurrence of natural disasters which depletes the banks’ reserves and

us
increases the leverage, primarily through the channels described above. This natural disaster
risk is represented by αη, where α measures the inverse of the adaptive capacities of a country

an
to the consequences of natural disasters. This capacity consists out of multiple dimensions
relevant for the default risk of a bank. First, a more developed financial system may be better
able to mitigate the consequences of natural disasters through more diversification
M
opportunities or the existence of a well-functioning re-insurance or catastrophic bonds market
(cf. McDermott et al., 2011)5. Second, there are various structural government policies for
reducing the adverse impact of natural disasters including mitigation policies—i.e., the
d

reduction of the risk of adverse natural events themselves through improved policies and
te

environmental standards —and adaptation policies— i.e., the restructuring of the economy
away from disaster-prone activities and upgrading the physical infrastructure to withstand the
p

impact of disasters (cf. Freeman et al., 2003). Finally, appropriate banking regulation and
ce

supervision may induce more rigorous capital reserves to withstand adverse (disaster) shocks
(Collier and Skees, 2013; Klomp and De Haan, 2011).
Furthermore, the parameter η represents the frequency or intensity of natural disaster,
Ac

for instance, measured by the annual growth rate of the physical damage caused by a natural
disaster. Today the physical damage increases by about 8 percent annually (EM-DAT, 2013).
Moreover, the initial probability P0 depends on the distance-to-default. We define a bank
failure when the profit loss -π exceeds the bank equity E. This initial probability of a failure is
then given by

5
A smaller value of α represents more adaptive capacities.

6
Page 6 of 35
 k
P0 (   E )  P0  r  k      r  dr (2)


Where r is the return-on-assets and k the negative of the equity-asset ratio. This equation
illustrates that the initial default risk depends on the distribution ϕ. If the return on assets is

t
normally distributed, equation (2) can be rewritten as

ip
z
P0  r  k    N (0,1)dz

cr
(3)


where

us
k
z (4)


an
Where ρ and σ are respectively the true mean and standard deviation of the distribution of the
return-on-assets. The variable z represents the number of standard deviations below the mean
M
by which the return on assets would have to fall in order to eliminate equity. This so-called z-
score is a direct measure of the distance-to-default. Combining equation (1)-(4) demonstrates
that an increase in frequency or intensity of natural disasters accumulates default risk within
d

the banking sector and reduces the distance-to-default.

te

3. Data and methodology

p
ce

3.1 Data
As our aim is to estimate the impact of natural disasters on the banks’ distance-to-default, we
have to quantify them both. The data on natural disasters and their impact are documented in
Ac

the “Emergency Events Database” (EM-DAT, 2013) with data collected by the Centre for
Research on the Epidemiology of Disasters (CRED). The EM-DAT database recorded about
8.000 natural disasters wordwide between 1995 to 2010. A natural disaster is recognized
when a natural situation or event which overwhelms local capacity, necessitating a request for
external assistance. For a disaster to be entered into the EM-DAT database at least one of the
following criteria must be fulfilled: (1) 10 or more people reported killed; (2) 100 people
reported affected; (3) declaration of a state of emergency; or (4) call for international
assistance.

7
Page 7 of 35
 However, as already pointed out by Gassebner et al. (2010), many of the disasters
recorded in the EM-DAT dataset seemed to have caused few casualties or damages. For
example, only 10 percent of the disasters involve deaths of more than a hundred people.
Given this distribution of the disaster data, it is conceivable that many of the disasters
included in EM-DAT will not have any impact on the default risk in the banking sector. For a
disaster to have an empirically impact, it should be of a magnitude that can directly cause

t
ip
widespread damage to the economy and financial system within a country6.
For this reason, we decided to confine our empirical analysis to disasters which meet

cr
any of the following criteria: (i) the estimated amount of damages is no less than one-percent
of GDP and ii) the estimated amount of damages is no less than a half-percent of the total

us
assets owned by commercial banks7. The adoption of this decision rule reduces the number of
natural disasters for our analysis to about 180 disasters8.
We distinguish between four broad groups of natural disasters: 1) hydrological

an
disasters including floods and wet mass movements; 2) meteorological disasters concerning
storms and hurricanes; 3) geophysical disasters entails earthquakes, tsunamis and volcanic
M
eruptions; and 4) climatic disasters including extreme temperatures, droughts and wildfires.
Figure 1 shows the distribution of disasters between 1995 and 2010. The graph illustrates that
hydrological disasters are the most common natural disaster (38 percent), while less than 10
d

percent of the major natural disasters is categorized as a geophysical disaster. On average,

te

there are about 12 major natural disasters annually.

p

[Figure 1 about here]

ce

To determine the impact of a natural disaster on the default risk of banks, we construct for
each country-year the following count variable, suggested by Noy (2009), that takes the
Ac

timing of a disaster in the course of a year into account. This allows disasters occurring early
in the year to have a different contemporaneous impact than those that happen near the end of
the year.

6
Obviously, larger countries have a higher probability of experiencing a natural event (Skidmore and Toya,
2002; Gassebner et al., 2010). When, for instance, the United States is hit by a hurricane the consequences for
the economy as a whole is likely to be smaller than when an island like Cuba, is hit by the same hurricane.
7
We use last year’s banking assets and last year’s GDP since the current year’s banking assets and GDP have
been affected by the disaster itself.
8
The concern that two episodes may have completely different impact because of their intensity and location is
partially controlled by imposing that disasters affect a minimum number of people and cause a minimum damage
to capital and wealth. Thus, disasters that occur in the middle of the desert are not considered as disasters under
our measure.

8
Page 8 of 35
   (12  M it ) /12     M it 1 /12  ( post )disaster year
disasterit  
0 otherwise

It is calculated as the sum of (12 - Mit)/12 in a disaster year and Mit/12 in a post-disaster year,

t
where M is the month of the disaster. In all other years its value is set to zero9. Using a count

ip
variable, a country that is hit more than once by a major disaster in the same year will suffer a
sharper increase in the default risk faced by banks than a country which suffers only a single

cr
incident10.
Another motivation for using a count measure of disaster events is to reduce the

us
potential endogeneity of the disaster measure with respect to economic development. The
consequences of disaster events, in terms of physical damage rely to some extent on the

an
economic conditions present (cf. Kellenberg and Mobarak, 2008). For instance, the total
damage may be positive related to the level of income and the depth of the financial sector
(McDermott et al., 2011). For this reason, it may be questionable to assume the exogeneity of
M
the consequences of disasters with respect to economic development. The use of a count
measure reduces the potential influence of endogeneity on our results.
d

Our measure for the default risk of banks is based on the z-score of the banking sector
within a particular country-year. The z-score is the ratio of return-on-assets plus capital-asset-
te

ratio to the standard deviation of return on assets. Specifically, the z-score indicates the
number of standard deviations that a bank’s return on assets has to drop below its expected
p

value before equity is depleted and the bank is insolvent (see Roy, 1952, Hannan and
ce

Henwick, 1988, Boyd, et al., 1993 and De Nicolo, 2000). If profits are assumed to follow a
normal distribution, it can be shown that the z-score measures the distance-to-default. Thus, a
higher z-score indicates that a bank is less fragile. The data on the z-score is taken between
Ac

1997 and 2010 from the Database on Financial Development and Structure reported by the

9
An alternative to the count variable that we employ is to use a dummy variable which takes the value one if at
least one disaster satisfying our decision rule occurred during the calendar year. However, the advantage of a
count variable is that it allows us to obtain a more precise estimate of the impact of disasters on the default risk
in a particular country-year. Besides, the “frequency effect” captured by a count variable would be lost if one
just uses a dummy variable. In effect, the count variable allows us to retain more information about disasters
than the use of a dummy variable. Nevertheless, as part of our robustness tests, we have also estimated the model
using a dummy variable to represent the occurrence of disasters satisfying our decision rule. Yet, the results
remain in line with our main findings throughout this paper.
10
The count measure gives equal weight to the disaster events. This has the advantage of reducing the potential
influence of outlier events at the upper end of the disaster distribution.

9
Page 9 of 35
World Bank. Figure 2 shows the distance-to-default between 1997 and 201011. Although
globally the z-score is rather stable in our period of analysis, the long-run trend differs
between industrialized and developing countries. While in OECD countries the average z-
score of the banking sector decreases in our period of analysis making them more fragile,
whereas in developing and emerging market economies there is an upward trend12.
As preliminary test for the exogeneity of the natural disasters in a country, we

t
ip
compare the distance-to-default before and after the occurrence of a natural disaster. The z-
score is 18.05 before a natural disaster has occurred and 15.24 afterwards. According to a

cr
Chi-squared test this difference is significant at the 5 percent level. However, this
nonparametric test is only suggestive, as unobserved country heterogeneity, as well as other

us
confounding variables are not taken into account.

[Figure 2 about here]

3.2 Empirical model

an
M
In this section, we build our model in which we are able to estimate the relationship between
large-scale natural disasters and default risk of banks. We use a dynamic model based on a
panel including more than 160 countries between 1997 and 2010 (Table A1 in the appendix
d

lists the countries included in our analysis). Since some of the data are not available for all
te

country-years, the panel data are unbalanced. The model is given as follows
p

(5)
ln distanceit  i  t   ln distanceit 1   j xitj 1   disasterit   it
ce

where distanceit is the distance-to-default based on the z-score in country i at time t

Ac

(taken in logarithms)13. We include the lagged dependent variable to control for auto-
regressive tendencies. The vector xj is a vector of (lagged) control variables containing j
elements, while disaster is our constructed disaster variable outlined above to capture the
exposure to large-scale natural catastrophes. The parameter αi is a country-specific intercept,
while ηt is a time fixed effects. Using country specific-intercepts places the emphasis for
identification of effects on the within country variation over time. This approach reduces the

11
Weighted by total bank assets-to-GDP
12
Table A3 in the appendix reports some time series properties of our distance-to-default measure.
13
To be precise, in the regressions, we use as the dependent variable ln(1+z-score) to smooth out higher values
of the z-score and to avoid truncating the dependent variable at zero.

10
Page 10 of 35
influence of any potential selection bias that might arise, for example, if poorer countries were
over-represented in the disaster data. The final term εit is the error term.
An alternative way of writing the previous equation is

(6)
ln distanceit  ln distanceit 1  i t  ( 1)ln distanceit 1   j xitj 1   disasterit  it

t
ip
In this specification, the dependent variable is the log-change of the z-score. Using this

cr
specification we are able to estimate the parameter γ by a difference-in-difference
methodology, where banks in countries affected by a large-scale natural disaster are the

us
“treated”, and those that are not are the “controls”. Our hypothesis is that distance-to-default
decreases due to the economic and financial consequences of large-scale natural disasters (γ <
0).

an
To reduce the omitted variable bias, we include control variables suggested by
previous studies on financial fragility. Similar to the specifications used by Beck et al. (2006),
Demirgüç-Kunt and Detragiache (1998) and Klomp and De Haan (2009), we consider
M
variables related to the macroeconomic environment, monetary policy and financial sector
development to capture the role of structural policies and institutions. First, we control for
d

macroeconomic factors by including inflation, economic growth, current account balance,

trade openness and exchange rate changes (see also Beck et al., 2006). Adverse shocks that
te

influence the economy will increase the fragility of the financial system, for example, by
p

affecting the solvency of borrowers, by increasing uncertainty, or by unexpected and

excessive exposure to foreign exchange risk. As already shown in Figure 1, there are
ce

important differences in the fragility of the banking sector between OECD and developing
countries. To control for this notion, we include real GDP per capita. In addition, the state of
Ac

the economy of many developing and emerging market countries relies primarily on
commodity prices. To capture the idea that this might affect the stability of the financial
system, we include the terms of trade.
According to Demirgüç-Kunt and Detragiache (1998), high short-term real interest
rates affect bank balance sheets adversely if banks cannot increase their lending rates quickly
enough and hence increase banking risk. We add the annual real interest rate taken from the
IMF. Furthermore, Calvo et al. (1993) conclude that capital flows are sensitive to changes in
the level of the world interest rate. Large capital inflows and capital flight may affect the
stability of the financial sector. Frankel (1999) argues that since the 1990s international

11
Page 11 of 35
private capital inflows have rapidly increased, raising financial vulnerability and the
transmission of financial crises. It is a well-known fact that there is an outflow of foreign
private capital shortly after a disaster as the uncertainty of future repayment raises (Yang,
2008). To test whether the default risk is related to sudden capital outflows or changes in the
foreign exchange reserves, we include the net international financial flows and the ratio of
M2-to-foreign exchange reserves.

t
ip
In addition, Demirguc-Kunt and Detragiache (1998) report that the probability of
financial fragility is positively related with weaker institutions. Countries lacking a sound

cr
legal system and good governance might have more financial system problems due to
corruption or inefficient enforcement of law and government ineffectiveness (La Porta et al.,

us
1998; Levine and Zervos, 1998; Barth et al., 2004; Fernandez and González, 2005). To
capture this, we include the Polity IV score which represents the democratic quality of the
political institutions present in a country (Marshall and Jaggers, 2012). In addition, we control

an
for the degree of capital regulation and supervision by including the first principle component
using use data of the bank regulation and supervision survey held by Barth et al. (2004; 2008).
M
The survey consists of about 175 questions, classified into twelve categories. We have
considered only the questions regarding capital regulation and supervisory control, as these
issues tend to be the most important in restraining banking risk (Klomp and De Haan, 2012).
d

Next, we include a measure to capture financial liberalization. Improperly

te

implemented financial liberalization is likely to raise default risk as financial institutions are
allowed more opportunities for risk-taking in a liberalized financial market (Kaminsky and
p

Reinhart, 1999)14. In contrast, proper financial liberalization may reduce default risk of banks
ce

due to more opportunities to diversify their risk. We proxy financial liberalization by

including the first principal component of the indicators on international capital market
controls, ownership of banks and interest rate controls taken from the Economic Freedom
Ac

Dataset (Gwartney et al., 2012) and financial freedom reported by the Heritage foundation.
Likewise, we include the growth rate of the domestic credit supplied to the private sector. A
credit boom together with improper liberalized financial markets leads often to a systemic
banking crisis. One can argue that the growth rate of credit measures the degree of de factor
liberalization, while the indices on financial freedom impinge on the de jure level of
liberalization. In addition, we add the credit-to-GDP to measure the financial depth of the

14
As an alternative, we measured financial liberalization by the first principal component of the indicators of
credit controls, interest rate controls, capital account restrictions, and security market policy taken from Abiad et
al. (2008). However, including this variable reduces our dataset dramatically, by more than 40 percent. However,
the results on the disaster variables are statistical similar.

12
Page 12 of 35
banking system in a country. McDermott et al. (2011) find that for countries with low levels
of financial sector development, natural disaster events exert a larger negative impact on
economic recovery.
Furthermore, we control for banking sector concentration. The direction of this
variable is theoretically not direct clear. First, De Nicolo et al. (2004) find that highly
concentrated banking systems exhibit higher levels of systemic risk due to a competition

t
ip
effect. In contrast, Beck et al. (2006) report that banking crises are less likely in more
concentrated banking systems mainly caused by enhanced efficiency. Moreover, we use the

cr
natural logarithm of real total banking assets in a particular country-year to control for the size
of the financial sector. Finally, we take two variables into account which represent the

us
operational efficiency of the banking sector: liabilities-to-assets and cost-income ratio. Table
A2 in the appendix provides an overview of all variables, their definition as well as their
source.

4. Empirical results
an
M
4.1 Baseline results
Let us now turn to the results. In the first part of Table 1, we present different econometric
d

specifications using the OLS estimator. To obtain robust standard errors we use the jacknife
te

procedure with 1.000 replicators. In view of the unequal distribution of the natural disasters
across countries, we cluster the standard errors on country level. For instance, more than 50
p

percent of the disasters occur in only 10 percent of the countries.

ce

In section 3.2 we mentioned a long list of variables which theoretically may explain
the changes in the distance-to-default of banks. However, due to reasons of data availability,
using all suggested control variables in one specification reduces our dataset dramatically. We
Ac

are left with less than 100 countries, which increases the potential of a sample selection bias
due to the exclusion of many developing countries, while the majority of the damage caused
by natural disasters is produced in these particular countries (Freeman et al., 2003). To
overcome this problem we selected our set of control variables by applying the general-to-
specific method. This method does not rely on economic theory, but is a widely used method
in applied econometrics to decide on model specification (see Hendry, 1993). We first
estimate a model including all control variables. Next, we drop the least significant variable

13
Page 13 of 35
and estimate the model again. We repeat this procedure until only variables that are
significant at a 10 percent level remain15.
Using this approach, we are left with about 170 countries. The estimation results in
column (1) indicate that a large credit supply relative to the size of the economy makes the
financial sector more fragile and decreases the distance-to-default. Conversely, real GDP per
capita affect the distance-to-default in the opposite direction. When the income level of a

t
ip
country is increasing, this will lead to a safer financial system. Finally, it turns out that a more
open economy, measured by the contribution of trade to GDP, makes the financial sector

cr
more resilient.
In columns (2) and (3) we add our broad natural disaster count measure, including the

us
four different types of natural events outlined above, to our baseline specification. The results
demonstrate that large-scale natural disasters have a significant negative impact on the z-score
and increase the fragility of the banking system. One additional disaster reduces the distance-

an
to-default by about 5 percentage-points. Our results give some empirical support for the
policy proposed by the UNEP that bank supervisors should include the exposure of financial
M
institutions to large-scale natural disasters in their assessment. It may be therefore important
that bank regulators and supervisors require that banks maintain adequate capital reserves to
absorb negative disaster shocks, for example, by connecting portfolio risk of borrowers with
d

the climatic and biophysical conditions in particular regions.

te

In addition, the various types of natural disasters included in the broad count measure
are rather diverse and differ in a number of respects. First, there are 20 times more floods than
p

volcano eruptions. Second, there is a large difference between the duration of a natural
ce

disaster. While an earthquake takes only a few minutes, the duration of a flood can take up to
three months. Third, the types of disasters are not equally divided around the globe. For
instance, Alexander (1993) shows that most hurricanes occur within the tropics between
Ac

latitude 30o North and South, but not within +/-5o of the equator, where the atmospheric
disturbances tend to be insufficient to cause them (cf. Skidmore and Toya, 2002). Finally, the
disasters considered differ in the degree of predictability. Climatic and hydrological disasters
are more frequent and can better be forecasted compared to geophysical disasters. So, it is
easier to take preparation measures for, for instance, floods by building dykes, than taking
physical precautionary measures for volcano eruptions or earthquakes. Therefore, in columns

15
The complete results of the general-to-specific approach are available upon request. As a robustness test, we
also estimated the models throughout this study using all the control variables mentioned in section 3.2. Yet, the
results do not significantly differ from the outcomes reported.

14
Page 14 of 35
(4) and (5) we split our complete set of disasters into four more homogenous groups:
hydrological, meteorological, geophysical and climatic disasters16. As some natural disasters
such as storms and floods often occur in tandem, simultaneous inclusion allows isolation of
the effects of each disaster (Loayza et al., 2012). The results indicate that a large part of the
significant impact reported on our broad disaster measure is attributed by the consequences of
geophysical and meteorological disasters. The main explanation for this main finding is that

t
ip
these kind of disasters cause the majority of the total physical damage per affected worldwide
(EM-DAT, 2013). As a consequence, these disasters may have the strongest influence on the

cr
repayment performance of borrowers due to the widespread destruction of productive capital
compared to, for instance, climatic disasters. Meanwhile, to make re-investments, clients

us
often demand additional loans and may reduce the amount of money on savings accounts. So,
in the aftermath of a disaster the loans-to-deposit ratio increases. Finally, meteorological
disasters occur more frequently giving affected clients only little time to cope and recover
from these shocks.

an
In the regressions so far, we assumed that the number of large-scale natural disasters is
M
exogenous. Though, the criteria on which we define a large-scale natural disaster are more
endogenous. For instance, as already mentioned above, the total damage may be positive
related to the level of income and the depth of the financial sector. When we fail to explicitly
d

control for these factors, our results might be spurious. Addressing the potential endogeneity
te

problem formally, we use the generalized method of moments (GMM) estimator developed
by Arellano and Bond (1991). In this approach, the endogeniety problem is solved by
p

estimating an instrumental regression for the first-difference equation using the second and
ce

higher-order lags of the endogenous and dependent variable and the first-difference of the
exogenous variables as instruments.
This approach, however, has drawbacks. First, differencing the equation removes the
Ac

long run cross-country information present in the levels of the variables. Second, if the
independent variables display persistence over time, their lagged levels will be poor
instruments for their differences. Under additional assumptions, it is possible to construct an
alternative GMM estimator that overcomes these problems. Specifically, more moment
conditions are available if we assume that the explanatory variables are uncorrelated with the
individual effects (see Arellano and Bover, 1995). In this case, lagged differences of these
variables and of the dependent variable may also be valid instruments for the levels equation.

16
The correlation between the different large-scale natural disaster measures ranges between -0.04 and 0.07.

15
Page 15 of 35
The estimation then combines the set of moment conditions available for the first-differenced
equations with the additional moment conditions implied for the levels equation (Blundell and
Bond, 1998). Finally, as long as the model is over-identified, validity of the assumptions
underlying both the difference and the system estimators can be tested through Sargan tests of
orthogonality between the instruments and the residuals and through tests of second- or higher
order residual autocorrelation17.

t
ip
The results in columns (6) and (7) of Table 1 point in the same direction as the
previous OLS results. We still find that the significant negative impact of natural catastrophes

cr
on the distance-to-default which is mainly caused by the consequences of geophysical and
meteorological disasters. Turning to the bottom part of Table 1, the Sargan test provides no

us
evidence of misspecification, while the serial correlation tests point to first- but no second-
order autocorrelation of the residuals, which is in accordance with the assumptions underlying
the selection of instruments18.

an
So, to sum up our results so far, we find that large-scale meteorological and
geophysical disasters have a significant negative impact on the distance to default of the
M
banking sector19. An issue often raised in the debate over financial fragility is that the
consequences of adverse shocks may be more serious when political institutions and policies
needed to support the efficient and smoothing functioning of the banking sector are not well
d

developed (cf. Klomp and De Haan, 2014). To control for this we make the impact natural
te

disasters conditional on the level of democratic quality and on the degree of capital regulation
and supervision. This is done by including an interaction term in the main specification and
p

estimate the following model

ce

ln distanceit  ln distanceit 1  i t  ( 1)ln distanceit 1   j xitj 1   1 disasterit 

 2institutionit   3 (disaster  institutionit )   it (7)
Ac

17
Specifically, we use the GMM estimator implemented by Roodman (2006) in Stata, including Windmeijer’s
(2005) finite sample correction.
18
Alternatively, we used the Hansen test to explorer the validity of our instruments used. The p-values of the
Hansen over-identifying restriction test indicate also that the GMM instruments are valid.
19
In the results presented so far, we assumed an one-year impact of a disaster. To explore the long run impact,
we have estimated our main econometric specification of column (5) including our disaster variable up to five
lags. The results indicate that natural disasters have only a temporarily impact on the default risk of a bank. After
one year most of the impact of the disaster has disappeared. These results point out that natural disasters have
probably a more important impact on the liquidity position of a bank than affecting on the solvency.
Alternatively, banking sectors that become highly unstable after a large shock may be restructured leading
potentially to less fragility.

16
Page 16 of 35
Where institution refers respectively to our democratic quality or capital regulation indicator.
The conditional effect of large-scale natural disasters on financial fragility can be calculated
by the derivation of equation (7) with respect to disaster.

 ln distance
  1   3institution
disaster

t
ip
We report in Table 2 the OLS-FE estimation results on the interaction term using the various

cr
large-scale disaster measures20. The findings show that more rigorous capital regulation and
supervision may restrain the impact of a natural disaster, in particular in case of geophysical

us
and meteorological disasters. One potential explanation is that a single geological or
meteorological disaster causes the most widespread damage. Therefore, larger capital reserves
make it more likely that a bank may withstand a disaster shock. Moreover, we find no

an
evidence that more democratic regimes are better able to restrain the impact of disasters on
the distance-to-default of banks.
M
[Insert Tables 1 and 2 about here]
d

4.2 Sensitivity analysis

te

By including a large number of countries which vary in the size and fragility of their financial
sector, it is possible that the results presented so far are determined by the heterogeneity of the
p

sample. That is, natural disasters have a significant impact on the fragility of the banking
sector in one country (group) and are insignificant in another. To control for this we perform a
ce

number of sample splits using the OLS fixed effects model21. First, economic development
influences the impact of disasters in two opposite ways. On the one hand, developing
Ac

countries are more vulnerable for disasters due to a combination of a high frequency of
disasters, large shares of the population living in disaster prone areas and the strong
dependence on the agricultural sector. Besides re-insurance and risk diversification
opportunities are limited in low-income countries. On the other hand, the commercial banking
sector is much larger and more interconnected in industrialized countries than elsewhere and
have larger stocks of physical capital at risk during a disaster. For instance, Melecky and

20
Given space constraints, we do not report the results of the models for our system-GMM model. However, the
results are similar to the reported OLS-FE results and are available upon request.
21
Using system-GMM gives similar results which are available upon request.

17
Page 17 of 35
Raddatz (2011) argue that in high-income countries, the major financial consequences of
large-scale natural disasters are usually passed through to the insurance and banking sector
because of a fiscal constraint faced by the government, while developing countries rely more
on government assistance. To test if the impact of a large-scale natural disaster on financial
fragility depends on the degree of economic development, we split in columns (1)-(3) of
Table 3 our sample, in accordance to the classification from the IMF, into developing

t
ip
countries, emerging market countries and industrialized countries. The results show that our
broad measure of natural disasters has only a significant impact in emerging market countries.

cr
One rational explanation for this result is that these countries are in transition and have
already invested substantially in their capital stock. At the same time, these countries are still

us
relying on relative disaster prone sectors such as agriculture.
In more detail, as already mentioned above, the different types of disasters are not
equally divided around the globe. For instance, droughts are most common in sub-Saharan

an
Africa, while floods occur more frequently in Asia and Latin America. To disaggregate the
latter results some further, we explore if the impact of each kind of disaster depends on the
M
degree of economic development. The results indicate that large-scale hydrological disasters
have the most severe impact in low-income countries, while geophysical disasters have
mainly an impact in high-income countries. This strengthens our view that the impact of
d

specific natural disasters on the fragility of the financial sector may differ among countries
te

based on the degree of economic development.

In addition, economic development is closely related to financial development. In
p

more financially developed countries, banks have more opportunities to diversify their risk or
ce

are able to re-insure their risk. To have a closer look on the conditional impact of large-scale
natural catastrophes on the level of financial development, we split our sample into financially
developed and financially constrained countries, based on the median score of the ratio
Ac

between credit to the private sector to GDP ratio. A low level of private credit may induce
that the population is facing credit constraints because savings and loans are not efficiently
matched. The results in columns (4) and (5) of Table 3 demonstrate that natural disasters have
a larger impact in financially constrained countries as they have less opportunities to deal with
natural disaster shocks. According the Basel Committee this is in particularly the case in
developing countries which tend to have less diversified loan portfolios, funding sources and
geographic scope (BCBS, 2010). Therefore, bank supervisors and regulators should facilitate
policies that try to deepen financial markets in non-disaster periods.

18
Page 18 of 35
[Insert Table 3 about here]

Meanwhile, Rasmussen (2004) emphasizes the importance of considering the number of

natural disasters in relation to the size of the economy. Disasters affect the economy in ways
that are quite different. For example, expressed in terms of damage per person affected,
geophysical and meteorological disasters are by far the most devastating of all disaster types

t
ip
considered, while droughts and floods tend to leave more people affected, with less economic
damage recorded per affected person. So far we did not take into account the scope or

cr
intensity of a disaster. As an additional robustness test, we normalize our disaster measure by
the magnitude of the disaster relative to the size of the economy or banking sector. We use

us
two different measures for the intensity of a disaster: (1) the amount of direct damage divided
by the previous year’s GDP and (2) the amount of direct damage divided by the total banking
assets in the previous year22. So, for each natural disaster we compute the following variable

disasterit    an
ln    G  (12  M it ) /12    G   M it 1 /12   ( post )disaster month
M
 otherwise
d

Where G measures the magnitude of a disaster. Consequently the log is taken to avoid
that the empirical results are driven by extreme values23. As already mentioned above, the
te

consequences of disaster events depend to some extent on economic development. When we

fail to explicitly control for these factors, our results might be spurious. Addressing the
p

potential endogeneity problem formally, we use the system-GMM approach outlined above.
ce

The estimation results are in the first part of Table 4. As the disaster measures are
taken in logarithms, we can interpret the coefficients as elasticities. An increase of one
Ac

percent in the share of the share of physical damage to GDP decreases the change in the z-
score by 0.6 percentage-points, while an increase of one percent in the share of physical
damage to banking sector assets reduces the z-score by 0.5 percentage-points. Although, the
economic impact seems rather small, as already mentioned above, the last thirty years the
physical damage increases by about 8 percent each year.

22
We divide the two measures by last year’s banking assets and last year’s GDP since the current year’s banking
assets and GDP have been affected by the disaster itself
23
Taking logs, however, leaves the problem of how to treat the country-years for which no event occurred (as
the log of zero is undefined). In order to retain these valuable observations while not distorting the distributional
properties of the natural disaster measure, following Loayza et al., 2012, these observations are assigned a low
number ε that is just below the lowest disaster measure for which an event was reported and is common across
all types of natural disasters.

19
Page 19 of 35
 In addition, one can argue that our results so far are driven by an endogenous selection
bias as we have only included a selected number of large-scale disasters based on their
damage created. As a robustness analysis, we re-estimate the model including all disasters
reported in the EM-DAT dataset. The results in the bottom part of Table 4 show that our
previous results may partially be driven by the used decision rule, as we are not able to find
any significant effect for any of the natural disasters measures including also small-scale

t
ip
disasters. One explanation is that more than 90 percent of the disasters affect less than 100
persons. Therefore, our results point out that there is actually a threshold effect in the impact

cr
of natural disasters on the fragility of the financial sector. As a sensitivity test, we have tried
to find a kink in the relationship between natural disasters and banking fragility to explore the

us
minimum threshold of a disaster. The results clearly indicate that the relationship is rather
linear. The larger the damage relative to GDP, the greater is the decrease in the z-score
(results are available upon request).

an
Moreover, as many activities of a bank do not stop at the border, the default risk of
banks may be affected by the exposure to the global impact of disasters. To test this notion,
M
we include in Table 5 the total number of large-scale natural disasters globally and total
global damage (as a share of global GDP) in a particular year. The results provide any
evidence of a global impact of natural disasters24. Therefore, natural disasters may have most
d

likely only an impact on the domestic activities of a bank25.

te

It is widely documented that financial fragility may drive the occurrence of banking
crises (cf. Demirgüç-Kunt and Detragiache, 1998). To explore if large-scale natural disasters
p

may cause systemic banking crises, we estimate the following logit model26
ce

bankcrisit    t   j xitj 1   disasterit   it

(8)
Ac

where bankcris is one in country i at year t when an onset of a banking crisis recognized in
that year (taken from Laeven and Valencia, 2013)27. The results in the second part of Table 5

24
However, one disadvantage is that we in this specification have to drop the time fixed effects because of the
multicolinearity with the total number of disasters globally which is the same for all countries.
25
The latter results are based on the use of the OLS-FE estimator. However, using a system-GMM model points
to the same conclusions (results are available upon request).
26
We employ Beck et al.’s (1998) remedy to test and correct for temporal dependence by adding a cubic splines,
along with a count variable for the number of years since the last banking crisis.
27
According to Laeven and Valencia (2013), a banking crisis is recognized as systemic if two conditions are
met: (i) Significant signs of financial distress in the banking system (as indicated by significant bank runs, losses
in the banking system, and/or bank liquidations) and (ii) Significant banking policy intervention measures in
response to significant losses in the banking system.

20
Page 20 of 35
indicate that, although large-scale natural disasters increase financial fragility, natural
catastrophes can not explain the occurrence of a domestic banking crisis.

[Insert Tables 4 to 5 about here]

t
5. Conclusions

ip
This study is one of the first attempts to directly relate the impact of large-scale natural
disasters on the default risk faced by commercial banks. We use a dynamic panel model

cr
including about 180 natural disasters over 160 countries in the period 1995 to 2010. Our
measure on default risk is based on the distance-to-default that indicates the number of

us
standard deviations that a bank’s return on assets has to drop below its expected value before
equity is depleted and the bank is insolvent, the so-called z-score.
After extensive testing for the sensitivity of the results, our main findings suggest that

an
natural disasters increase the likelihood of a banks’ default. More precisely, we conclude that
geophysical and meteorological disasters reduce the distance-to-default the most due to their
M
widespread damage caused. The results of this study clearly indicate that natural disasters
may be a substantial threat to the liquidity, however, not directly to the solvency of the
commercial banking sector, as we do not find any impact of natural events on the onset of
d

systemic banking crises. To remain stable, policy makers should address strict reserve
te

requirements and connect them to the local climate and biophysical conditions present in
countries which are frequently hit by a natural disaster.
p

Besides, it turns out that the impact of a natural disaster depends on the degree of
ce

financial development of a particular country. This provides some evidence that bank
supervisors and regulators should facilitate policies that make it possible to deepen the
financial markets and make them function smoothly also after a natural disaster. For instance,
Ac

banks should be able to insure their risk by using catastrophic bonds or through a well-
functioning re-insurance market.

21
Page 21 of 35
References
Alexander, D., 1993. Natural disasters. New York. Chapman and Hall.
Abiad, A., Detragiache, E., Tressel, T., 2008. A new database of financial reforms. IMF
Working Paper 08/266.
Arellano, M., Bond, S., 1991. Some tests of specification for panel data: Monte Carlo
evidence and an application to employment equations. Review of Economic Studies

t
ip
58, 277–297.
Arellano, M., Bover, O., 1995. Another look at the instrumental variable estimation of

cr
error-components models. Journal of Econometrics 68, 29–51.
Arestis, P., Demetriades, P., 1997. Financial development and economic growth: As-

us
sessing the evidence. The Economic Journal 107, 783-799.
Barth, J., Caprio, G., Levine, R., 2004. Bank regulation and supervision: what works
best? Journal of Financial intermediation 13, 205-248.

an
Barth, J., Caprio, G., Levine, R., 2008. Bank regulations are changing: for better or worse?
World Bank Policy Research Working Paper 4646.
M
Beck, T., Demirgüc¸ -Kunt, A., and Levine, R., 2006. Bank concentration, competition,
and crises: First results. Journal of Banking and Finance, 30(5), 1581–1603.
Berg, G., Schrader, J., 2012. Access to credit, natural disasters, and relationship
d

lending. Journal of Financial Intermediation 21, 549-568.

te

Borensztein, E., Cavallo, E., and Valenzuela, P., 2009. Debt Sustainability under Cata-
strophic Risk: The Case for Government Budget Insurance. Risk Management and
p
ce

Insurance Review 12, 273‐294.

Ac

Boyd, J., Graham, S., Hewitt, R., 1993. Bank holding company mergers with nonbank
financial firms: Effects on the risk of failure. Journal of Banking and Finance 17, 43–
63.
Boyd, J., Runkle, D., 1993. Size and performance of banking firms: Testing the predict-
tions of theory. Journal of Monetary Economics 31, 47-67.
Calvo, G., Leiderman, L., Reinhart, C., 1993. Capital inflows and real exchange rate
appreciation in Latin America: The role of external factors. IMF Staff Papers 40, 108-
151.

22
Page 22 of 35
Charnobai, A., Rachev, S., 2006. Applying robust methods to operational risk modeling,
Journal of Operational Risk 1, 27-41.
Cetorelli, N., Goldberg, L., 2011. Global Banks and International Shock Transmission:
Evidence from the Crisis. IMF Economic Review 59, 41–76.
Čihák, M., 2007. Central Bank Independence and financial instability, IMF.
Collier, B., and Skees, J., 2012. Increasing the resilience of financial intermediaries

t
ip
through portfolio-level insurance against natural disasters. Natural Hazards 64, 1-18.
Collier, B., Katchova, A., and Skees, J., 2011. Loan portfolio performance and El Niño,

cr
an intervention analysis. Agricultural Finance Review 71, 98-119
Collier, B., Skees, J., 2013. Exclusive finance: How unmanaged systemic risk continues to

us
limit financial services for the poor in a booming sector. Mimeo.
Collier, B., Miranda, J., and Skees, J., 2013. Natural disasters and credit supply shocks in de
veloping and emerging economies. Wharton Working Paper 2013-03. University of
Pennsylvania.

an
David A., 2011. Natural Disasters? Global Economy Journal 11, 1-38.
M
De Gregorio, J., Guidotti, P., 1995. Financial development and economic growth. World
development 23, 433-448.
De Nicolo, G., 2000. Size, Charter Value, and Risk in Banking: An International Per-
d

spective. International Finance Discussion Papers 689. Washington, DC: Board of

te

Governors of the Federal Reserve System.

De Nicolo, G., Bartholomew, P., Zaman, J., Zephirin, M., 2004. Bank consolidation,
p

internationalization, and conglomeration: Trends and implications for financial risk.

ce

Financial Markets, Institutions, and Instruments 13, 173–217

Demirguc-Kunt, Asli, and Enrica Detragiache. 1998. The Determinants of Banking Cri-
ses: Evidence from Developing and Developed Countries. IMF Staff Papers 45, 81–
Ac

109.
Demirgüç-Kunt, A., Detragiache, E., 2010. Basel core principles and bank soundness: Does
compliance matter? Journal of Financial Stability, forthcoming.
Demirgüç-Kunt, A. Detragiache, E., Tressel, T., 2008. Banking on the principles: Compliance
with Basel Core Principles and Bank Soundness. Journal of Financial Intermediation 17,
511-542.
EM-DAT, 2013. The OFDA/CRED International Disaster Database Université Catholique de
Louvain, Brussels.Fernandez,
Fernandez, A., González, F., 2005. How accounting and auditing systems can counter-

23
Page 23 of 35
 act risk-shifting of safety nets in banking: Some international evidence. Journal of
Financial Stability 1, 466–500.
Frankel, J., 1999. Preventing banking crises. In Caprio, G., Hunter, C., Kaufman, G.,
Leipsiger, D. (eds.), Preventing bank crises: Lessons from recent global bank failures,
pp. 3-12, World Bank Publications, Washington DC.
Freeman, P.K., Keen, M. and Mani, M., 2003, Dealing with Increased Risk of Natural

t
ip
Disasters: Challenges and Options, IMF Working Paper WP/03/197 (Washington:
International Monetary Fund).

cr
Gassebner, M., Keck, A., Teh, R., 2010. Shaken, Not Stirred: The Impact of Disasters
on International Trade. International Economic Review 18, 351-368.

us
Goldberg, L., 2009. Understanding banking sector globalization. IMF Staff Papers 56,
171–197.
Hannan, T., Hanweck, G., 1988. Bank Insolvency Risk and the Market for Large Certif-

an
icates of Deposit. Journal of Money, Credit, and Banking 20, 203–11.
Hanson, S., Pesaran, M., Scheurman, T., 2008. Firm heterogeneity and credit risk diver-
M
sification. Journal of Empirical Finance 15, 583-612.
Hendry, D., 1993. Econometrics, Alchemy or Science? Essays in Econometric Method-
ology, Blackwell Publishers, Oxford, UK.
d

Jaggers, K., Gurr, T., and Marshall, M. 2010. Polity IV, Political regime characteristics
te

and Transitions, 1800-2010. Ann Arbor, MI, Inter-university Consortium for Political
and Social Research.
p

Hosono, K., Miyakawa, D., Uchino, T., Hazama, M., Ono, A., Uchida, H., and Uesugi, I.,
ce

2012. Natural disasters, damage to banks, and firm investment RIETI Discussion
PaHoper 12-062.
Kaminsky, G., Reinhart, C., 1999. The twin crises: The causes of banking and balance-
Ac

of-payment problems. American Economic Review 89, 473-500.

Keefer, P., 1999. When do special interests run rampant? Disentangling the role of elec-
tions, incomplete information and checks and balances in banking crises, World Bank
Policy Research Working Paper No. 2543, Washington DC.
Kellenberg, D., Mobarak, A., 2008. Does rising income increase or decrease damage
risk from natural disasters?. Journal of Urban Economics 63, 788-802.
Klomp, J., 2012. Microfinance risk and natural disasters: A panel data-analysis. Mimeo.
Klomp, J., De Haan, J., 2014, Bank regulation, the quality of institutions and banking risk in
emerging and developing countries: An empirical analysis. Mimeo.

24
Page 24 of 35
Klomp, J., De Haan, J., 2012. Banking risk and regulation: Does one size fit all. Forth-
coming Journal of Banking and Finance 36, 3197-3212.
Klomp, J., De Haan, J. 2009. Central bank independence and financial stability. Journal
of Financial Stability 5, 321-338.
Klomp, J., 2010. Causes of banking crises revisited. The North American Journal of
Economics and Finance 21, 72-87.

t
ip
La Porta, R., Lopez-de-Silanes, F., Shleifer, A.,Vishny, R., 1998. Law and finance.
Journal of Political Economy 106, 1113–1155.

cr
Laeven, L., Valencia, F., 2013. Systemic banking crises database. IMF Economic Review 61,
225-270.

us
Levine, R., 1997. Financial development and economic growth: views and agenda.
Journal of economic literature 35, 688-726.
Levine, R., 2005. Finance and growth: Theory and evidence. Handbook of economic
growth 1, 865-934.

an
Levine, R., Zervos, S., 1998. Stock Markets, Banks, and Economic Growth. American
M
Economic Review 88, 537–58.
Loayza, N., Olaberria, E., Rigolini, J., and Christiaensen, L., 2012. Natural disasters and
growth: Going beyond the averages.World Development 40: 1317-1338.
d

Marshall, M. Jaggers, K., 2012. Polity IV project: Political regime characteristics and transi-
te

tions, 1800-2012. http://www.systemicpeace.org/polity/polity4.htm

McDermott, T. 2011. Disasters and Development: Natural Disasters, Credit Constraints and
p

Economic Growth. Mimeo.

ce

Melecky, M., Raddatz, C. 2011. How do governments respond after catastrophes.

Natural Disaster Shocks and the Fiscal Stance. World Bank Policy Research Working
Paper, 5564.
Ac

Noy, I., 2009. The macroeconomic consequences of disasters, Journal of Development

Economics 88, 221-231.
Noy, I., Nualsri, A., 2011. Fiscal storms: Public spending and revenues in the after
math of natural disasters. Environment and Development Economics 16, 113-128.
Oosterloo, S., De Haan, J., 2004. Central banks and financial stability: a survey. Journal
of Financial Stability 1, 257-273.
Raddatz, C., 2009. The Wrath of God: Macroeconomic Costs of Natural Disasters.
World Bank Policy Research Working Paper 5039.
Roy, A., 1952. Safety First and the Holding of Assets. Econometrica 20, 431–49.

25
Page 25 of 35
Skidmore, M., Toya, H., 2002. Do natural disasters promote long run growth? Eco-
nomic Inquiry 40, 664-687.
Steindl, F., Weinrobe, M., 1983. Natural hazards and deposit behavior at financial insti-
tutions. Journal of Banking and Finance 7, 111-118.
Van Greuning, H., Bratanovic, S., 2009. Analyzing banking risk: a framework for as-
sessing corporate governance and risk management. World Bank Publications.

t
ip
Windmeijer, F., 2005. A finite sample correction for the variance of linear efficient two-
step GMM estimators. Journal of Econometrics 126, 25-51.

cr
Yang, D., 2008. Coping with Disaster: The Impact of Hurricanes on International Financial
Flows, 1970-2002. Advances in Economic Analysis & Policy, Berkeley Electronic

us
Press 8, 1903-1903.

an
M
d
p te
ce
Ac

26
Page 26 of 35
 Table 1: Distance to default - Natural disasters
Δ ln zscore
OLS System GMM
(1) (2) (3) (4) (5) (6) (7)

t
lagged z-score -0.393 ** -0.394 ** -0.398 ** -0.394 ** -0.400 ** -0.281 ** -0.305 **

ip
(0.033) (0.033) (0.032) (0.032) (0.032) (0.118) (0.117)
Real GDP per capita 0.146 * 0.148 * -0.022 0.131 * -0.022 -0.037 -0.046

cr
(0.078) (0.078) (0.103) (0.076) (0.103) (0.094) (0.072)
Credit-to-GDP -0.099 ** -0.099 ** -0.118 ** -0.097 ** -0.119 ** -0.009 -0.001
(0.033) (0.033) (0.034) (0.032) (0.035) (0.073) (0.067)

us
Trade openness 0.114 ** 0.117 ** 0.087 0.115 ** 0.085 0.260 ** 0.280 **
(0.053) (0.053) (0.057) (0.053) (0.057) (0.132) (0.103)

an
All large-scale natural disasters -0.051 * -0.054 * -0.069 **
(0.030) (0.029) (0.034)
Hydrological disasters -0.058 -0.068 -0.071
(0.063) (0.064) (0.066)
M
Meteorological disasters -0.122 ** -0.125 ** -0.142 **
(0.055) (0.056) (0.066)
Geophysical disasters -0.142 * -0.168 ** -0.189 **
(0.078) (0.079) (0.088)
d

Climatic disasters -0.018 -0.017 -0.019

(0.175) (0.175) (0.195)
p te
ce

Country fixed effects YES YES YES YES YES YES YES
Year fixed effects NO NO YES NO YES YES YES
Observations 1962 1962 1962 1962 1962 1959 1959
Ac

Countries 169 169 169 169 169 169 169

R-squared 0.051 0.054 0.058 0.057 0.057
F-test (p-value) 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Arelanno-Bond AR(1) p-value - - - - - 0.000 0.000
Arelanno-Bond AR(2) p-value - - - - - 0.599 0.999
Sargan test (p-value) - - - - - 0.311 0.188
**/* Indicating significance levels of respectively 5 and 10 percent. Bootstrapped standard errors are shown between brackets.

27
Page 27 of 35
 Table 2: Conditional impact of large-scale natural disasters on institutional policies
Δ ln zscore
All disasters Hydrological Meteorological Geophysical Climatic
(1) (2) (3) (4) (5)
Democratic quality

t
Natural disasters -0.082 ** -0.067 -0.127 * -0.133 * -0.025

ip
(0.051) (0.062) (0.065) (0.071) (0.028)
Natural disastesr x institutions 0.011 0.003 0.007 0.015 0.002
(0.011) (0.013) (0.013) (0.014) (0.003)

cr
Capital regulation and supervision
Natural disasters -0.089 ** -0.059 -0.132 ** -0.144 ** -0.017

us
(0.042) (0.062) (0.061) (0.072) (0.021)
Natural disasters x institutions 0.004 * 0.001 0.002 * 0.007 * 0.002
(0.002) (0.001) (0.001) (0.004) (0.004)

an
**/* Indicating significance levels of respectively 5 and 10 percent. Bootstrapped standard errors shown between
brackets. Including control variables found significant in column (1) of Table 1
M
Table 3: Sample split analysis
d

Δ ln zscore
Developing Emerging market Developed Financial Financial
countries countries countries developed constraint
te

(1) (2) (3) (4) (5)

All large-scale disasters -0.041 -0.088 ** 0.114 -0.027 -0.091 *
p

(0.045) (0.042) (0.087) (0.026) (0.051)

ce

Observations 602 886 474 1079 883

R-squared 0.117 0.038 0.053 0.021 0.098
F-test (p-value) 0.000 0.000 0.000 0.000 0.000
Ac

Hydrological disasters -0.161 * -0.126 ** 0.053 -0.043 -0.074 *

(0.087) (0.059) (0.051) (0.094) (0.044)
Meteorological disasters 0.014 -0.054 ** 0.021 0.035 -0.102 *
(0.122) (0.025) (0.038) (0.057) (0.057)
Geophysical disasters 0.079 -0.126 -0.076 * -0.114 * -0.142
(0.050) (0.131) (0.041) (0.061) (0.156)
Climatic disasters -0.300 -0.012 - -0.031 -0.037
(0.282) (0.050) (0.058) (0.151)

Observations 602 886 474 1079 883

R-squared 0.126 0.038 0.053 0.021 0.097
F-test (p-value) 0.000 0.000 0.000 0.000 0.000

28
Page 28 of 35
**/* Indicating significance levels of respectively 5 and 10 percent. Bootstrapped standard errors are shown between
brackets. Including control variables found significant in column (1) of in Table 1

Table 4: Distance to default - Intensity of the disaster

Δ ln zscore

t
All Hydrological Meteorological Geophysical Climatic

ip
(1) (2) (3) (4) (5)
physical damage / GDP - System GMM -0.006 * -0.002 ** -0.003 ** -0.015 0.000
Large-scale disasters (0.004) (0.001) (0.002) (0.016) (0.015)

cr
us
Observations 1962 1962
Countries 169 169
Arelanno-Bond AR(1) p-value 0.000 0.000
Arelanno-Bond AR(2) p-value 0.412 0.461

an
Sargan test (p-value) 0.474 0.491
physical damage / banks assets - System GMM
Large-scale disasters -0.005 *
(0.003)
M
Observations 1962 1962
Countries 169 169
d

Arelanno-Bond AR(1) p-value 0.000 0.000

Arelanno-Bond AR(2) p-value 0.472 0.488
te

Sargan test (p-value) 0.377 0.319

All size disasters - OLS-FE
All disasters 0.000 -0.001 -0.001 -0.004 0.000
p

(0.002) (0.004) (0.003) (0.003) (0.001)

ce

Observations 1962 1962

R-square 0.053 0.054
F-test (p-value) 0.000 0.000
Ac

Countries 169 169

**/* Indicating significance levels of respectively 5 and 10 percent. t Bootstrapped standard errors are shown between
brackets. Including control variables found significant in column (1) of in Table 1

29
Page 29 of 35
 Table 5: Global impact and banking crises
Δ ln zscore Banking crisis dummy

Global number of Global damage (as Number of Damage (as share

disasters share of GDP) disasters of GDP)
(1) (3) (4) (5)

t
All large-scale disasters -0.0005 -0.0002 0.025 0.019

ip
(0.003) (0.001) (0.720) (0.087)

Observations 1962 1962 1611 1611

cr
R-squared 0.052 0.051 - -
F-test (p-value) 0.000 0.000 0.000 0.000

us
Hydrological disasters -0.0005 -0.0003 0.073 0.073
(0.006) (0.006) (0.919) (0.092)
Meteorological disasters -0.0011 -0.0008 -0.898 0.039

an
(0.008) (0.006) (2.145) (0.168)
Geophysical disasters -0.0012 -0.0011 0.487 0.051
(0.008) (0.007) (1.483) (0.136)
M
Climatic disasters -0.0002 0.0000 -0.074 -0.286
(0.015) (0.001) (0.761) (8.215)

Observations 1962 1962 1611 1611

d

R-squared 0.052 0.051 - -

F-test (p-value) 0.000 0.000 0.000 0.000
te

**/* Indicating significance levels of respectively 5 and 10 percent. t-values are shown between
brackets. Including control variables found significant in column (1) of in Table 1
p
ce
Ac

30
Page 30 of 35
 Figure 1: Number and impact of large-scale natural disasters

t
ip
cr
us
an
M
Figure 2: Distance-to-default
d
te

19

18
p

17
ce

16
z-score

15
Ac

14

13

12
1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Year
Industrial countries Developing and emerging market countries Global

The graph shows the distance-to-default measured by the asset weighted z-score. A higher score indicates a
larger default probability.

31
Page 31 of 35
Appendix

Table A1: Countries included

Albania Djibouti Lesotho Saudi Arabia

t
Algeria Dominica Liberia Senegal

ip
Angola Dominican Republic Libya Serbia
Antigua and Barbuda Ecuador Lithuania Seychelles
Argentina Egypt, Arab Rep. Luxembourg Sierra Leone

cr
Armenia El Salvador Macao SAR, China Singapore
Australia Equatorial Guinea Macedonia, FYR Slovak Republic
Austria Eritrea Madagascar Slovenia

us
Azerbaijan Estonia Malawi South Africa
Bahamas, The Ethiopia Malaysia Spain
Bahrain Finland Maldives Sri Lanka
Bangladesh France Mali St. Kitts and Nevis
Barbados Gabon Malta St. Lucia

an
Belarus Gambia, The Mauritania St. Vincent and the Grenadines
Belgium Georgia Mauritius Sudan
Belize Germany Mexico Suriname
Benin Ghana Moldova Swaziland
M
Bhutan Greece Mongolia Sweden
Bolivia Grenada Montenegro Switzerland
Bosnia and Herzegovina Guatemala Morocco Syrian Arab Republic
Botswana Guinea Mozambique Tajikistan
Brazil Guyana Namibia Tanzania
d

Brunei Darussalam Haiti Nepal Thailand

Bulgaria Honduras Netherlands Togo
te

Burkina Faso Hong Kong SAR, China New Zealand Trinidad and Tobago
Burundi Hungary Nicaragua Tunisia
Cambodia Iceland Niger Turkey
p

Cameroon India Nigeria Turkmenistan

Canada Indonesia Norway Uganda
Cape Verde Ireland Oman Ukraine
ce

Central African Republic Israel Pakistan United Arab Emirates

Chad Italy Panama United Kingdom
Chile Jamaica Papua New Guinea United States
China Japan Paraguay Uruguay
Ac

Colombia Jordan Peru Vanuatu

Congo, Dem. Rep. Kazakhstan Philippines Venezuela, RB
Congo, Rep. Kenya Poland Vietnam
Costa Rica Korea, Rep. Portugal Yemen, Rep.
Cote d'Ivoire Kuwait Qatar Zambia
Croatia Kyrgyz Republic Romania Zimbabwe
Cyprus Lao PDR Russian Federation
Czech Republic Latvia Rwanda
Denmark Lebanon Samoa

32
Page 32 of 35
 Table A2: Data sources used and descriptive statistics
Variable Description Source Mean s.d.
Real GDP per capita Log of the real GDP per capita in constant US$ of WDI (2012) 7.70 1.46
2000
Growth rate of GDP per capita Annual growth rate of the real GDP per capita WDI (2012) 3.11 4.35
Credit-to-GDP The amount of credit supplied to the private sector WDI (2012) 3.46 0.95
as a share of GDP in logarithms

t
Growth rate of credit Annual growth rate of credit-to-GDP WDI (2012) 0.05 0.18

ip
M2-to-reserves The money supply as a share of the reserves from WDI (2012) 7.20 41.85
the central bank
Exchange rate change Annual change in the official exchange rate WDI (2012) 0.04 0.18

cr
Inflation ln[1+inflation] where inflation is the annual change WDI (2012) 0.08 0.13
in the GDP deflator

us
International capital flows The sum of capital outflows and inflows WDI (2012) 4.54 6.64

Real interest rate Real interest rate IMF (2012) 0.07 0.12
Banking concentration The herfindahl-hirsch index of the banking assets Beck et al. (2012) 4.11 0.35
in logarithms

an
Size of the financial sector Sum of banking sector assets as percentage of Beck et al. (2012) 64.31 56.31
GDP
Polity IV Polity IV score Polity IV 4.73 5.84
Cost-income ratio Total cost divided by total revenue Beck et al. (2012) 60.07 19.67
M
Liabilities to assets Log of bank liabilities divided by total assets Beck et al. (2012) -0.02 0.45
Trade openness Log of import plus export as a share of GDP\ WDI (2012) 4.16 0.65
Current account Export minus import as a share of GDP WDI (2012) -3.61 11.01
Terms of trade Log of net barter terms of trade WDI (2012) 4.66 0.29
Financial liberalization First principle component of various sub-indices on Own computation 0.07 1.04
d

financial liberalization taken from the Fraser

Institute and Heritage Foundation (see main text
for the indicators).
te

Capital regulation and supervision First principle component of various questions on Own computation 0.12 1.09
capital regulation and supervisory control taken
from Barth et al. (2004; 2008)
p
ce
Ac

33
Page 33 of 35
 Table A3: Time series properties distance-to-default
z-score ln(z_score) ∆ z-score
Observations 2388 2385 2183
Mean 18.679 2.769 0.014
Standard deviation 21.441 0.639 0.319
Median 16.102 2.841 0.011

t
Maximum 467.01 6.148 3.139

ip
Minimum -17.178 -0.633 -3.013
Skewness 14.482 -0.256 0.578

cr
Kurtosis 290.212 4.873 23.126
ADF constant and trend 616.381 577.259 1136.192

us
an
M
d
p te
ce
Ac

34
Page 34 of 35
Research highlights “Financial fragility and natural disasters: An empirical analysis”

 We explore the impact of natural disasters on the distance-to-default of banks.

 Natural catastrophes may affect the liquidity and solvency position of banks.
 We find that disasters increase the likelihood of a banks’ default.

t
 This impact depends on the scope of a disaster and the degree of economic

ip
development.

cr
us
an
M
d
p te
ce
Ac

35
Page 35 of 35